PJFP.com

Pursuit of Joy, Fulfillment, and Purpose

Tag: open-source AI

  • Inkling: Thinking Machines Lab Releases Its First Open-Weights Model, a 975B Multimodal Mixture-of-Experts With Controllable Thinking Effort That Can Fine-Tune Itself on Tinker

    Thinking Machines Lab, the AI startup founded by former OpenAI CTO Mira Murati, has released Inkling, its first open-weights model trained from scratch. Inkling is a 975 billion parameter Mixture-of-Experts transformer (41B active) with a context window of up to 1 million tokens, native multimodal reasoning over text, images, and audio, and a dial for controllable thinking effort. The lab is explicit that Inkling is not the strongest model in the world. It is pitched as something arguably more useful: a broad, balanced, customizable foundation you can fine-tune on Tinker, with the full weights on Hugging Face. The announcement even includes a demo where Inkling fine-tunes itself and swaps in its own new weights.

    TLDR

    Thinking Machines Lab released Inkling, a 975B-total, 41B-active Mixture-of-Experts model pretrained on 45 trillion tokens of text, images, audio, and video, alongside a preview of Inkling-Small (276B total, 12B active). The release covers the model’s generalist benchmark profile across reasoning, agentic coding, tool use, vision, and audio; a controllable thinking effort setting that lets developers trade performance against tokens (matching Nemotron 3 Ultra on Terminal Bench 2.1 at roughly a third of the tokens); an encoder-free multimodal architecture using dMel spectrograms and hMLP image patches; a training recipe combining Muon and Adam with weight decay coupled to the learning rate; RL scaled past 30 million rollouts with log-linearly improving reasoning and an emergent compression of the chain of thought; an epistemics push covering calibration, forecasting (where it beats several frontier models), abstention, and censorship resistance; the strongest FORTRESS adversarial safety score among compared open-weights models; a headline-grabbing demo of the model fine-tuning itself into a lipogram assistant via Tinker; and day-one availability on Tinker (at a 50% discount), Hugging Face, and inference partners including Together, Fireworks, Modal, Databricks, Baseten, vLLM, SGLang, and llama.cpp.

    Thoughts

    The most striking thing about this launch is its honesty. Nearly every frontier release leads with a claim to be the best at something, and the fine print walks it back. Thinking Machines Lab says plainly that Inkling is not the strongest model available, open or closed, and then makes the case that “strongest” is the wrong axis for most real buyers. If you are going to run a model millions of times inside a product, what you care about is the cost curve, the adaptability, and whether you can shape it to your workflow. That framing conveniently matches their business (Tinker sells fine-tuning), but it also matches how production AI actually gets deployed, where cost and latency are binding constraints and a benchmark crown is trivia.

    The self-fine-tuning demo deserves more attention than it will probably get. Asked to become a lipogram assistant that never uses the letter “e” (a behavior prompting alone cannot reliably produce), Inkling wrote its own training objective and scoring function, generated its own synthetic data, launched the run on Tinker, evaluated the result against its base self, and then staged a weight swap so the improved checkpoint took over the session. That is a closed loop of specify, train, evaluate, and self-update, packaged as a cute product demo. The loop is the primitive behind every serious conversation about recursive self-improvement, and here it is running as a marketing asset with a 27 minute wall clock. The gap between “toy objective” and “economically meaningful objective” is now a question of reward design, not plumbing.

    Controllable thinking effort is the feature I expect developers to care about most. Instead of publishing a single score, TML publishes a curve: sweep the effort setting from 0.2 to 0.99 and watch performance trade against generated tokens. Inkling reportedly matches Nemotron 3 Ultra on Terminal Bench 2.1 while spending about a third of the tokens. Benchmarks reported as single points hide exactly this, and a model that reaches a target score cheaply beats a model that scores two points higher at triple the cost in any high-volume workload. Expect effort curves to become standard marketing for open models, the way context length became standard a couple of years ago.

    The epistemics section is quietly the most differentiated part of the release. TML trained calibration directly, running RL against proper scoring rules on resolved real-world questions, and pairing a rubric grader with a claims grader that does agentic web search to verify each factual assertion. The result is a model that beats GPT-5.5 and Claude Opus 4.8 on ForecastBench without search and holds its own on Prophet Arena. A model that knows when to say “I don’t know” is more useful across messy real-world domains than one that confabulates confidently, and it is notable that a lab whose stated mission is extending human will and judgment treats calibrated uncertainty as a first-class training target rather than a safety afterthought. The censorship-resistance training, validated on Cognition’s Propaganda and Censorship Eval, extends the same idea: trustworthiness as a capability you train, not a policy you bolt on.

    Finally, the open-weights safety tension is handled with unusual candor. Inkling posts the strongest adversarial FORTRESS score among the open models compared while keeping benign over-refusal low, and it was tested externally for CBRN, cyber, and loss-of-control capabilities. But everyone in this space knows fine-tuning can strip safety behavior from open weights, and TML ships a fine-tuning platform for this exact model. Their acknowledgment that they are actively studying how safety behavior survives fine-tuning on Tinker is the right thing to say, and it is also the open question that will define whether “safe open weights” is a coherent category at all.

    Key Takeaways

    • Inkling is Thinking Machines Lab’s first from-scratch, open-weights model: a Mixture-of-Experts transformer with 975B total parameters, 41B active, and a context window up to 1M tokens.
    • It was pretrained on 45 trillion tokens spanning text, images, audio, and video, and reasons natively over text, images, and audio without separate encoders.
    • A preview of Inkling-Small ships alongside it: a 276B-parameter MoE with just 12B active parameters that matches or beats its larger sibling on several benchmarks thanks to an improved pretraining recipe.
    • TML explicitly positions Inkling as a base for customization rather than the strongest overall model, leaning on multimodality, efficient thinking, and Tinker fine-tuning as the differentiators.
    • The launch demo shows Inkling fine-tuning itself: it wrote its own training objective and data, ran the job through the Tinker API, evaluated the result, and hot-swapped to its own new weights inside the OpenCode harness.
    • The self-fine-tuning target was a lipogram assistant that never uses the letter “e,” a behavior chosen precisely because prompting alone cannot reliably achieve it; the full loop completed in about 27 minutes.
    • Controllable thinking effort is a core feature: a setting swept from 0.2 to 0.99 traces a full performance-versus-tokens curve instead of a single benchmark point.
    • On Terminal Bench 2.1, Inkling matches Nemotron 3 Ultra’s score at roughly one third of the generated tokens, the release’s flagship efficiency claim.
    • Inkling was trained to run inside a variety of coding and agent harnesses, with tool sets and schemas randomized during training to reduce sensitivity to any particular harness.
    • On Design Arena’s blinded human-evaluated Agentic Web Dev leaderboard, Inkling scores 1257, among the strongest open-weights models and tied with Claude Opus 4.6.
    • Headline benchmark scores at effort 0.99 include SWEBench Verified 77.6%, SWEBench Pro Public 54.3%, Terminal Bench 2.1 63.8%, GPQA Diamond 87.2%, AIME 2026 97.1%, and HLE 29.7% text-only (46.0% with tools).
    • Agentic and general scores include MCP Atlas 74.1%, Tau 3 Banking 23.7%, and BrowseComp 77.1% with context management.
    • Vision results are strong for an open model: MMMU Pro 73.5%, CharXiv RQ 78.1%, rising to 82.0% when the model uses a Python tool for zooming and cropping during visual reasoning.
    • Audio results place it among the strongest open-weights audio models: VoiceBench 91.4%, MMAU 77.2%, and Audio MC 56.6%, well ahead of Qwen3-Omni and Nemotron Nano-Omni on the last.
    • The multimodal stack is encoder-free: audio enters as discrete dMel spectrograms and images as 40×40 pixel patches through a four-layer hMLP, both passed through a lightweight embedding layer and processed jointly with text tokens.
    • The MoE design largely follows DeepSeek-V3: 256 routed experts plus 2 shared experts per layer, 6 routed experts active per token, with a sigmoid router and auxiliary-loss-free load balancing.
    • Attention interleaves sliding-window and global layers at a 5:1 ratio with 8 KV heads, and uses a learned relative positional embedding instead of RoPE, which TML found extrapolates better to long sequences.
    • Short convolutions are applied after the key and value projections and on the attention and MLP residual branch outputs, an unusual architectural touch aimed at efficiency and long-context performance.
    • Training used a hybrid optimizer strategy, Muon for large matrix weights and Adam for everything else, with weight decay coupled to the square of the learning rate to keep weight magnitudes stable.
    • Post-training was bootstrapped with a small SFT phase on synthetic data generated by open-weights models including Kimi K2.5, with the large majority of compute spent on large-scale RL.
    • RL was scaled past 30 million rollouts across two long continuous runs, with reasoning performance on a held-out aggregate (AIME, HLE, GPQA, and others) improving log-linearly throughout.
    • Effort control was trained by varying the system message and per-token cost across rollouts, teaching the model to modulate its own thinking budget.
    • An emergent effect appeared during RL: the chain of thought compressed over training, dropping articles and connectives into a telegraphic style, driven purely by efficiency pressure rather than any targeted reward.
    • Inkling was TML’s first major training effort and ran on NVIDIA GB300 NVL72 systems; the lab says future models will push compute scale further across pretraining and RL.
    • Calibration was trained directly with RL against proper scoring rules on a large corpus of resolved real-world questions, treating well-placed confidence as a capability rather than a byproduct.
    • On ForecastBench without search, Inkling’s Brier Index of 61.1 beats GPT-5.5 (59.1) and Claude Opus 4.8 (54.6), and it stays competitive with search enabled and on Prophet Arena.
    • Instruction following was trained with two automated graders working together: a rubric grader scoring against a checklist and a claims grader that verifies each factual claim via agentic web search, improving helpfulness and reducing hallucination simultaneously.
    • Abstention-aware rewards on short-form factual QA taught the model to answer when confident and hedge or decline when not, with some prompts explicitly forcing or forbidding hedging so the user’s preference wins.
    • Inkling was trained to answer directly on topics subject to censorship, and Cognition’s Propaganda and Censorship Eval found strong censorship non-compliance.
    • On FORTRESS, Inkling posts the strongest adversarial refusal score (78.0%) of any compared open-weights model while keeping benign compliance high (95.9%), and scores 98.6% on StrongREJECT.
    • Safety testing covered CBRN, cyber, and loss-of-control capabilities plus human-AI threat vectors like sycophancy, vulnerable users, and manipulation, verified by commissioned external testers.
    • Inkling is available for fine-tuning on Tinker today with 64K and 256K context options at a 50% limited-time discount, plus a free Inkling Playground chat interface in the Tinker console.
    • Full weights are on Hugging Face, including an NVFP4 checkpoint for efficient inference on NVIDIA Blackwell, with API availability via Together, Fireworks, Modal, Databricks, and Baseten and inference support in SGLang, vLLM, TokenSpeed, and llama.cpp.
    • TML frames Inkling as the first in a family and as the intended background reasoning model for its previously announced real-time interaction models system.

    Detailed Summary

    What Inkling Is and Why It Exists

    Thinking Machines Lab frames its mission as building AI that extends human will and judgment, and Inkling as the logical next step after shipping the Tinker customization platform, previewing an interaction-focused AI system, and publishing research. Inkling is a Mixture-of-Experts transformer with 975B total and 41B active parameters, a context window up to 1M tokens, and pretraining on 45 trillion tokens of mixed text, image, audio, and video data. The lab is upfront that it is not the strongest model available. The pitch is breadth plus adaptability: a generalist trained across agentic, reasoning, coding, instruction-following, factuality, vision, and audio tasks rather than tuned to dominate one leaderboard, offered with full weights so people can make it their own. It launches with a preview sibling, Inkling-Small, at 276B total and 12B active parameters.

    The Self-Fine-Tuning Demo

    To demonstrate what customization means, TML asked Inkling to fine-tune itself. Running inside the OpenCode harness with access to Tinker, the model was told to become a lipogram assistant that never uses the letter “e.” Inkling drafted the plan, wrote an objective file with a scoring function (any response containing “e” scores zero), generated synthetic training data, launched a supervised fine-tuning run through the Tinker API, evaluated the checkpoint against its base self, and then staged a self-update so the supervisor relaunched the session on the new weights. The pipeline passed in about 27 minutes, and the updated model answered a test question about launching an LLM without a single “e.” It is a whimsical objective wrapped around a serious primitive: a model autonomously specifying, running, and adopting its own weight updates.

    Agentic Coding and Tool Use

    TML trained Inkling to operate inside many coding and agent harnesses, randomizing tool sets and schemas during training so the model does not overfit to one environment. The release showcases three demos: a one-shot job-application web app that then hosts an embedded browser-use agent operating its own interface; a nine-page, cohesively designed PDF food and travel journal produced from a single editorial prompt with web-verified details; and a server-authoritative multiplayer snake game refined over 40 iterations of feedback from GPT Codex acting as a reviewer. On benchmarks, Inkling posts 77.6% on SWEBench Verified, 54.3% on SWEBench Pro Public, and 63.8% on Terminal Bench 2.1, competitive within the open-weights field, and 1257 on Design Arena’s human-judged web dev leaderboard, in the same band as Claude Opus 4.6.

    Controllable Thinking Effort

    Rather than reporting a single operating point, TML sweeps Inkling’s effort setting from 0.2 to 0.99 and plots score against mean generated tokens on Terminal Bench 2.1, HLE, and IFBench, with competitors shown at their default settings. The headline result is efficiency: Inkling reaches Nemotron 3 Ultra’s Terminal Bench score at roughly a third of the tokens. The argument is that cost and latency are binding constraints in production, especially for interactive collaboration, so the full cost curve, not the peak score, is what developers should evaluate. Effort can be set from within the agent harness, and the ability was trained by varying system messages and per-token costs across RL rollouts.

    Native Multimodality Without Encoders

    Inkling is designed to serve as the background reasoning model for TML’s interaction models system, which requires real-time voice and vision collaboration. The multimodal components are trained from scratch with an encoder-free architecture: audio arrives as discrete dMel spectrograms and images as 40×40 pixel patches through a four-layer hMLP, both mapped through a lightweight embedding layer and processed jointly with text. The model transcribes speech, follows spoken instructions, reasons over long recordings, and answers questions about charts and diagrams, optionally using a Python tool to zoom and crop images mid-reasoning. Scores like 91.4% on VoiceBench and 82.0% on CharXiv RQ with Python place it among the strongest open-weights multimodal models, though still behind Gemini 3.1 Pro.

    Epistemics: Calibration, Forecasting, and Censorship Resistance

    TML groups calibration, instruction following, and censorship resistance under the banner of epistemics. Calibration was trained with RL against proper scoring rules on resolved real-world questions, and it shows: Inkling’s ForecastBench Brier Index of 61.1 without search beats GPT-5.5 and Claude Opus 4.8, and its Prophet Arena score sits close to the frontier. Instruction following used two complementary automated graders, a rubric checklist and a claims grader that verifies factual assertions through agentic web search, so recall-spraying to hack rubrics gets penalized by the factuality check. Targeted abstention-aware QA datasets taught the model to say “I don’t know” or give hedged best guesses when appropriate, while still complying when a user demands a forced guess. Finally, the model was trained to answer directly on censorship-prone topics, with Cognition’s Propaganda and Censorship Eval finding strong non-compliance with censorship patterns.

    Safety for an Open-Weights Release

    Inkling was trained to an internal behavioral spec across all modalities and then checked by commissioned external safety testers. Evaluations covered dangerous capabilities (CBRN, cyber, loss of control) and human-AI threat vectors including sycophancy, vulnerable users, and harmful manipulation. On FORTRESS, which pairs adversarial harmful requests with benign look-alikes, Inkling posts the strongest adversarial score among the compared open models (78.0%) without collapsing on the benign side (95.9%), and it scores 98.6% on StrongREJECT. TML acknowledges the open question hanging over every open-weights release: how safety behavior holds up under fine-tuning, which it says it is actively studying on Tinker.

    Architecture and Training Recipe

    The MoE layout follows DeepSeek-V3: 256 routed experts and 2 shared experts per layer with 6 routed experts active per token, a sigmoid-based router, and auxiliary-loss-free load balancing. Attention interleaves sliding-window and global layers 5:1 with 8 KV heads, and positions are encoded with a learned relative positional embedding that TML found outperforms and out-extrapolates RoPE. Short convolutions appear after the key and value projections and on residual branch outputs. Optimization was hybrid, Muon for large matrices and Adam elsewhere, with hyperparameter schedules drawn from the lab’s modular manifolds research and weight decay coupled to the square of the learning rate to keep weight norms stable. Post-training bootstrapped from a small SFT phase on synthetic data from open models including Kimi K2.5, then spent the bulk of compute on large-scale RL. Everything ran on NVIDIA GB300 NVL72 systems.

    RL at Scale and the Emergent Compression of Thought

    TML scaled asynchronous RL past 30 million rollouts across two long continuous runs, with performance on a held-out aggregate of reasoning evals improving log-linearly the whole way. Along the way an unplanned behavior emerged: the chain of thought became progressively more concise, shedding grammatical overhead into a telegraphic style (“We need to understand” becomes “We need determine”) while remaining comprehensible and leaving final answers unaffected. No reward targeted this; token efficiency pressure alone drove the compression, echoing an observation Cognition made while training SWE-1.7. It is a vivid example of optimization discovering its own shorthand.

    Inkling-Small

    The preview of Inkling-Small is arguably the sleeper story: with 12B active parameters against Inkling’s 41B, it matches or exceeds the larger model on a surprising number of benchmarks, including GPQA Diamond (88.3% vs 87.2%), IFBench (83.4% vs 79.8%), and CharXiv RQ with Python (83.4% vs 82.0%). TML attributes this to pretraining data and recipe improvements made after the big model trained, with both models sharing the same post-training stack. The clearest gaps favoring big Inkling are factuality (SimpleQA 43.9% vs 20.9%), Terminal Bench, and Tau 3 Banking. Full weights for Inkling-Small will be released once testing finishes, and its cost and latency profile targets high-volume workloads like coding, LLM grading, and synthetic data generation.

    Availability and the Ecosystem Play

    Inkling is on Tinker today with 64K and 256K context options at a limited-time 50% discount, plus a free Inkling Playground chat interface with integrated web search in the Tinker console so developers can get a feel for the model before committing to a run. The cookbook gained native Inkling support and three new audio recipes, and a new tml-renderer handles chat templates, tool calls, reasoning content, and multimodal inputs. Deployment partnerships span Together, Fireworks, Modal, Databricks, and Baseten for APIs; RadixArk for SGLang and Miles; Inferact for vLLM; Lightseek for TokenSpeed; Unsloth for llama.cpp; and Hugging Face for transformers integration. Full weights are on Hugging Face in both the original checkpoint and an NVFP4 checkpoint for NVIDIA Blackwell inference.

    Notable Quotes

    “Our mission is to build AI that extends human will and judgment.”

    Thinking Machines Lab, opening the Inkling announcement

    The company’s north star, and the lens through which the whole release (customization, calibration, open weights) is framed.

    “Inkling is not the strongest overall model available today, open or closed. Instead, a combination of qualities makes it a good open-weights base for customization: multimodal capabilities, efficient thinking, and availability on Tinker for fine-tuning.”

    Thinking Machines Lab, positioning the release

    A rare piece of launch-day honesty from a frontier lab, and the strategic thesis of the whole release.

    “Picking the right base model to fine-tune is a qualitative judgment that combines measurable benchmarks with the unique feel of a model that comes from playing with it.”

    Thinking Machines Lab, on why the Inkling Playground exists

    An argument that vibes are data, from the lab that built a playground into a fine-tuning console.

    “Cost and latency are often binding constraints in real-world applications, and low latency in particular is crucial for enabling collaboration and improvement through iteration.”

    Thinking Machines Lab, on controllable thinking effort

    The case for evaluating models on their full effort-versus-performance curve instead of a single benchmark point.

    “A model that’s confident in every answer it gives, including when it’s missing info and confabulates, forces the user to double-check everything.”

    Thinking Machines Lab, on why calibration was a training target

    The clearest one-line justification for treating calibrated uncertainty as a capability rather than a nicety.

    “Together, the two graders improve helpfulness and reduce hallucination at the same time, rather than trading one for the other.”

    Thinking Machines Lab, on pairing a rubric grader with a web-searching claims grader

    A neat solution to rubric hacking: verify every claim with agentic search so spraying plausible facts stops paying.

    “Safety is crucial for open-weights models. We’re continuing to study safety behavior and capability uplift in customizable models, including how safety behavior is impacted by fine-tuning on Tinker.”

    Thinking Machines Lab, on the open question of fine-tunable safety

    The acknowledgment that safety trained into open weights must survive the very customization the product sells.

    “Inkling is just the start: our first release in a model family we will continue to build on.”

    Thinking Machines Lab, on the roadmap

    Together with the GB300 compute note, a clear signal that larger and stronger family members are coming.

    Read the full announcement, including the interactive demos, effort curves, and complete benchmark tables, on the Thinking Machines Lab blog.

    Related Reading

  • Dig Through Your Couches, SpaceX Needs It: Cyan Banister on Luke Nosek’s Pitch, Going All In on SpaceX, Pokemon Go, Meditation, and Why Curiosity Is the Ultimate Investing Edge

    Angel investor Cyan Banister has one of the most remarkable track records in Silicon Valley: SpaceX, Uber, Anduril, Postmates, Niantic, Affirm, Flexport, Flock Safety, and dozens more. In this wide-ranging conversation on the Sourcery podcast with Molly O’Shea, the Long Journey Ventures co-founder tells the story behind her first check, when PayPal co-founder Luke Nosek got on the floor of her house and told her to dig through her couches because SpaceX needed every liquid dollar she had. She also covers the Founders Fund Mafia show, why personality is not fixed, the five minute meditation practice she prescribes to stuck founders, how asking “why” led her to Pokemon Go and Uber, what worries her about AI surveillance, and why free speech is her number one cause.

    TLDW

    Cyan Banister explains how Luke Nosek and her husband Scott Banister convinced her to put her entire IronPort windfall into SpaceX while rockets were still blowing up on the launch pad, a bet that became the best investment of her life. She walks through the “second believer” philosophy behind Long Journey Ventures and its bellwether logo, her run on Mike Solana’s Founders Fund Mafia show filmed at the site of the famous PayPal Mafia photo, why games like Mafia, poker, and board games are core Silicon Valley social infrastructure, and the time she bluffed Phil Hellmuth on a live stream. She then goes deep on inner work: personality is not fixed, the gap between your values and your actions is measurable, meditation is noticing that you are noticing, and mornings should start with a “why” question. That mindset produced her Niantic and Uber investments, informs her worries about centralized AI and a surveillance state, and fuels her excitement about AI as a new paintbrush, vibe manufacturing, agentized one person businesses, Substrate, Becoming Bio, and Diamond Foundry. She closes with her mentors, Peter Thiel, Marc Andreessen, Scott Cook, and Rick Rubin, and a blunt defense of curiosity and free speech over shame by association.

    Thoughts

    The most useful idea in this interview is the “second believer.” Long Journey keeps two candles on the wall: a founder lights the first flame, and someone else lights their own candle from it and holds the flame in case the founder’s goes out. That is a precise description of what early capital actually is. Luke Nosek was Elon Musk’s second believer, championing SpaceX “with more heart” than Cyan had ever seen, and Cyan’s first check existed because Nosek’s conviction was strong enough to transfer. Most people think conviction is a private mental state. This interview argues it is social infrastructure: belief propagates person to person, and the people who hold flames for others quietly shape which futures get built.

    The SpaceX story deserves a caveat Banister herself supplies. Putting one hundred percent of a liquidity event into a company whose rockets were exploding looks like genius only in hindsight; her friends told her she had lit her money on fire, and they were reasoning correctly from the information available. What made the bet rational was not the outcome but the frame Nosek and Scott Banister gave her: you are young, able-bodied, and infinitely employable, so your downside is a career, not ruin. That is the actual lesson for anyone tempted to copy the trade. Concentrated risk is a function of your recovery capacity, not your conviction level. She could afford to be the fool card. A fifty five year old with dependents cannot, and pretending otherwise is how people get hurt imitating legends.

    Her investing process is really an attention practice wearing a venture costume. The Niantic story is the cleanest example: she noticed friends chartering boats and ditching Defcon parties to capture invisible portals in Ingress, asked why Google would build such a thing, worked out that it was free mapping data, and then recognized the ticket subject lines at Hint Water as her path to the CEO the week Niantic spun out of Alphabet. Nothing in that chain requires capital or connections. It requires being awake, which is exactly why she starts coaching clients with five minutes of meditation and a “why” question every morning. The pipeline from mindfulness to alpha sounds like woo until you notice that every step of her best deals was just paying attention slightly earlier than everyone else.

    Her claim that personality is surgically alterable is more radical than it sounds, and it lands close to the core of the pursuit of joy, fulfillment, and purpose. Most self-improvement advice accepts the self as given and optimizes around it. Banister says the “I’m just like this” script is an excuse for behavior you are unwilling to change, and her values-versus-actions audit, literally listing where you lied this week, including the accidental lies of broken small commitments, is a concrete tool anyone can run tonight. She even disagrees with Marc Andreessen’s famous advice against introspection, which takes some nerve given he is one of her heroes. The through line from her homelessness to her optimism is that she treated her own character as buildable, and that is a more transferable asset than any cap table.

    The last stretch, on centralized AI, surveillance, and free speech, is where her optimism shows its edges. She is an accelerationist who backs open source and decentralized control precisely because she remembers the internet of 1999 promising the same thing and consolidating anyway. Her warning that autonomous vehicles could quietly abolish freedom of movement for dissidents is the kind of unfashionable thought experiment that her whole “question every phrase” method is built to surface. You do not have to share her politics to notice the consistency: someone who measures a nation’s health by its tolerance for comedy and rap music is applying the same test to Peter Thiel dinner parties and to AI policy, which is more than most commentators on either side can say.

    Key Takeaways

    • Cyan Banister’s first ever angel check was SpaceX, made after PayPal co-founder Luke Nosek came to her house and told her and Scott Banister to dig through their couches for anything liquid because SpaceX needed it.
    • She put everything she made from the IronPort sale to Cisco into SpaceX at a time when rockets were blowing up on the launch pad and critics said private citizens had no business in space.
    • The frame that justified the all-in bet: if you are young and able-bodied you are infinitely employable, so a total loss costs you a lifestyle, not your future. She held the position for roughly 20 years and calls it the best investment she will ever make.
    • Failure was priced in: she compares early SpaceX to early aviation, where getting planes to fly required crashing a lot of planes, and NASA veterans knew reusability would demand repeated public failure.
    • Combined with her husband Scott Banister, she believes they are the number one angel investing duo in the world, and even split individually both would sit in the top ten of the Stanford angel rankings. Married partners share capital, which rankings and lists struggle to represent.
    • Her portfolio names dropped in the episode include SpaceX, Anduril, Uber, Zappos, PayPal, Affirm, Flexport, Checkr, Density, Flock Safety, Brave, Control Labs, Depop, Substrate, Carta, Together AI, Postmates, Niantic, Diamond Foundry, Upstart, Fiverr, Forge, Opendoor, Calm, TrueMed, and Crusoe.
    • Long Journey Ventures’ logo is a bellwether, the lead sheep of a flock, looking sideways to spot the nonobvious. The firm’s “second believer” ritual uses two candles: light your candle from a founder’s flame and hold it so they can reignite if theirs goes out.
    • She was a cast member on Mike Solana’s Founders Fund Mafia show, filmed at Tosca, the same location as the famous PayPal Mafia photo, with a full reality TV production: one camera per player, table lenses, aerial cameras, and over 30 crew.
    • Her Mafia strategy is the meta game: listening for sounds, watching eye movements, tracking who protests too much and who is forming alliances, on the assumption that everyone is lying.
    • Games are Silicon Valley’s social infrastructure. Poker, Mafia, Werewolf, chess, Magic the Gathering, and Settlers of Catan nights let people skip small talk, collaborate immediately, and reveal how many turns ahead someone thinks.
    • If you get invited to a poker night or a Mafia game in tech, go. She has found founders and friends through games, and treats them like poker or golf as deal flow channels.
    • Brian Singerman got her into board games through a board game of the month club she ran for $40 a month, shipping sub-30-minute games in advance so game night starts with playing, not rule explanations. She has never met anyone better at strategy board games.
    • She beat Phil Hellmuth with her first ever bluff during a live streamed poker game she did not know was being broadcast, by convincing herself she had the best hand and acting accordingly. Hellmuth went on tilt for the rest of the session.
    • She identifies with the fool tarot card: walking off ledges expecting things to work out, and believing that on a long enough time horizon every setback turns out to have been necessary.
    • Personality is not fixed. Statements like “I’m a Scorpio, I can’t help it” or “I’m Irish, I have a temper” are excuses for behavior you are unwilling to change. With introspection, effort, and time you can surgically alter your personality.
    • Her weekly thought experiment: how wide is the space between your values and your actions? She sits down with paper and lists where she lied, including accidental lies like promising an email and not sending it.
    • She runs Awake Academy 101 classes and coaches stuck founders, starting almost everyone with five minutes of meditation a day, often in the car before starting the engine.
    • Meditation is not silencing your mind. It is noticing thoughts passing like clouds, then noticing that you are noticing, then asking who the noticer is. If you are not your thoughts, who are you?
    • Her presence toolkit: mindful showers, feeling your toes for the first minute after waking instead of launching into routines, and writing “wake up” on mirrors and windows so it appears when they fog up.
    • Humans are “why machines.” She does not start her day until she has a why question to carry through it, and says asking why about everything makes you a better investor, entrepreneur, and everything else.
    • The Niantic investment came from watching Ingress players rent helicopters and charter boats for invisible objects, realizing Google was harvesting free mapping data, then using Hint Water ticket subject lines to reach CEO John Hanke through Kara Goldin the moment Niantic spun out of Alphabet.
    • Nobody would co-invest in Niantic with her because they could not imagine people holding phones up to look at invisible Pokemon. She calls the Pokemon Go launch the closest we have come to world peace.
    • Her Uber conviction came from years of asking taxi drivers about their lives: starting each day $200 in the hole to the taxi yard explained the rushing, the crankiness, and the broken system, so when Uber appeared the pre-thinking was already done.
    • Idle time is research time. Instead of doom scrolling at a restaurant, ask why the bread is baked that way and whether robotics would improve it. Play with science fiction scenarios and they lead you to investments.
    • Her biggest worry, a question Peter Thiel used to ask her: an AI-operated surveillance state. Autonomous vehicles could end freedom of movement, with a government able to shut down your ride or lock you inside it because you are a dissident.
    • She believes the internet’s drift from open and decentralized to closed and centralized is repeating in AI, and that one company with one ideology ruling AI is dangerous. Everyone needs their own models, which is why she backs open source and decentralized control.
    • On education: unless you are pursuing medicine or another field requiring years of formal training, she questions whether school is the right move now. Artisanship and creativity will rise, and AI tutors make genuine self-teaching possible.
    • She is excited about AI as a new paintbrush unlocking dormant creativity: vibe coding, vibe manufacturing, and fully agentized businesses with no employees will mint millionaires from basements even if the businesses are not venture scale.
    • On AI art and training data: after a hundred years art enters the public domain anyway, China will train on Western IP regardless and sell it back, and today’s “slop” is the worst the tools will ever be.
    • AI still cannot replace human judgment. AI-written text has telltale signs any heavy user recognizes, so the job is to take its useful nuggets and massage them back into human form.
    • Her most exciting current investments: Substrate (the substrate of technology, bringing semiconductor manufacturing back to the Americas), Becoming Bio (the substrate of biology), and Diamond Foundry, whose real market was industrial diamonds and wafers, not rings.
    • She avoids hypercompetitive hot deals because the alpha is not in what is happening today. A good seed fund finds moonshots at low prices with meaningful ownership, in the “what’s coming” space: nanotech, biotech, and bottlenecks removed by AI.
    • She is still hunting for “the Alibaba of the Americas” and puts it out publicly in case a founder claims the idea. Wars will be fought with robots and drones, SpaceX opened the category that made Anduril and Varda imaginable, and defense primes will need competitors.
    • Her heroes: Peter Thiel (she went to “Peter Thiel University” during four years as his partner at Founders Fund and calls him tolerant, open-minded, and poorly understood), Marc Andreessen (a teenage hero she vowed to meet as an equal), Scott Cook of Intuit (the gold standard of executive function), and Rick Rubin, whom she has never met but considers a kindred mind.
    • Mentors can be far off. You can learn from people without knowing them by observing them, listening to those around them, and asking why they do what they do without assuming.
    • On shame by association: go to the events, hear all sides before deciding where you stand, and stop weaponizing accusations, because if everyone is called a racist the real ones cannot be found.
    • You can tell the health of a nation by its ability to tolerate comedy and rap music, and comedy disappearing from universities first was the warning sign. Free speech is her number one cause, and much of what she invests in serves it.

    Detailed Summary

    Inside Cyan’s Lair: Play as a Design Principle

    The interview opens in “Cyan’s Lair,” a mural-covered room at Long Journey’s headquarters painted by Brooklyn ceramicist Dave Zackin, whom Banister discovered on Instagram because he wore the same red and green glasses she needed to learn to walk again after her stroke. Zackin rescues abandoned pottery from high schools and ceramic studios, repaints and refires it, and gives it new life. Her home works the same way: thrift store finds, walls of fried eggs, bowls of fake fish people end up throwing at each other. She gauges hosting success by how many things guests touch without permission, because rummaging means they feel free to play. The candles on the wall, added by co-founder Lee Jacobs, encode the firm’s “second believer” concept: light your candle from a founder’s flame and hold it in case theirs goes out. The firm’s bellwether logo, a sheep’s eye looking sideways, comes from her habit of interrogating common phrases: when is being a sheep good, who leads the sheep, and what is the bellwether watching for that others miss?

    Mafia at Tosca: Reality TV for the PayPal Set

    Banister was a breakout player on Mike Solana’s Mafia show for Founders Fund, filmed months before release at Tosca, the location of the famous PayPal Mafia photograph. The production was serious reality television: a camera per player, lenses embedded in the table, aerial shots, and over 30 crew, with spicy moments and sushi-room banter left on the cutting room floor. Her approach was pure meta game, listening for rustling when the mafia woke at night, watching for the table jerk when players leaned on it, and asking the bar who the best players were so she could target them first. It was her first time playing the killer, and she found lying so uncomfortable she was sure everyone could tell. They could not. She hopes for a second season and notes the game should not be played with couples, since accusations have a way of outliving the game.

    Games, Poker, and the Hellmuth Bluff

    Banister argues games are how a neurodiverse industry socializes: instead of cocktail small talk, you drop straight into collaboration and watch how someone thinks, whether they plan five turns ahead, and how they handle math, psychology, and losing. Brian Singerman, whom she calls the best strategy board gamer she has ever met, subscribed to her board game of the month club, where friends paid $40 a month for sub-30-minute games shipped in advance so game nights started instantly. Her poker fame is mostly accidental: she hosts an annual charity tournament for Inflection Grants micro grants, and once stumbled into a live streamed game with Steve Aoki, Ninja, and Phil Hellmuth without realizing cameras were showing her cards to the world. Told at the break to try bluffing just once, she waited for a big pot, convinced herself she held the best hand, and played it that way until Hellmuth folded and went on tilt. The story doubles as her whole philosophy: she was only in Vegas because a portfolio company had her working undercover in a bad wig.

    The Fool Card: Optimism as a Trainable Skill

    Asked how she keeps finding herself in improbable situations, Banister says her life is a series of them, like Bill Murray in The Man Who Knew Too Little, and that if she were a tarot card she would be the fool, walking off ledges expecting things to work out. Pressed on whether ordinary people can live that way, she rejects the premise that they cannot: personality feels fixed only because we recite excuses like “I’m a Scorpio” or “I’m Irish, I have a temper.” With introspection, and here she cheerfully disagrees with Marc Andreessen’s advice against it, you can surgically alter your personality, though it takes effort, time, and facing ugly truths. Her weekly thought experiment asks how wide the space is between your values and your actions: if you claim honesty, list where you lied this week, including the accidental lies of commitments the two-hours-later version of you failed to keep. People wear masks out of fear of standing out, but everyone else is too caught up in their own noise to care, and authenticity leads to more happiness, curiosity, and wonder.

    Waking Up: Meditation, the Right Brain, and Why Machines

    Through her coaching and Awake Academy classes, Banister starts almost everyone with five minutes of meditation a day, often sitting in the car before starting it. She dismantles the perfection myth that drives people away from the practice: meditation is not silencing thought but noticing thoughts pass like clouds, noticing that you are noticing, and asking who the noticer is. From there the practice extends into ordinary life: mindful showers, feeling your toes for the first minute after waking instead of diving into routines, asking “am I awake right now?” before getting out of bed, and writing “wake up” on surfaces that fog up. The point is escaping rumination about past and future, since the present is the only thing that exists. Mind workers live in the left brain, but creativity, body sense, and intuition live in the right, and her greatest investments came with a feeling. Humans, she says, are why machines: she does not start her day without a why question to carry through it.

    Pattern Matching in the Wild: Niantic and Uber

    Her Niantic story shows the method end to end. She watched friends display irrational devotion to Ingress, renting helicopters and abandoning Defcon to capture invisible portals, immersed herself in the game, and asked why Google would build it, concluding it was free mapping data. When Alphabet spun Niantic out, she remembered support tickets at Hint Water marked “Ingress code,” asked founder Kara Goldin about the Google relationship, and was connected to CEO John Hanke within five minutes. He told her Niantic had Nintendo and Google and did not need her money; she asked for one hour and a guarantee he would not regret it, brought her best friend who was a top player (Hanke hired him nearly on the spot), and got into the round. Convincing anyone to co-invest was impossible because nobody believed people would hold up phones to look at invisible Pokemon. The Uber thesis worked the same way years earlier: asking taxi drivers about their lives revealed a system where drivers started each day $200 in debt to the yard, which explained everything riders hated about taxis. People who complain about Uber, she notes, never lived the before times.

    Worries: Surveillance States and Centralized AI

    Banister borrows a question Peter Thiel used to ask her at Founders Fund: what worries you? Her answer is an AI-operated surveillance state fused with robotics. Freedom of movement is a human right, and a future where you cannot drive yourself means someone can shut down your autonomous ride, or lock you inside it, because you said things the state dislikes. AI, like a gun, can be a paperweight or a weapon, and she is an accelerationist who still insists on thinking through what happens if it falls into the wrong hands. Having entered the industry in 1999, she watched an internet that was supposed to be open and decentralized become closed and centralized, and sees the same drift in OpenAI and Anthropic. Everyone needs their own models, she argues, because one company with one ideology ruling it all is dangerous. She also worries about children and what they should study when so much is automatable, concluding that unless a path truly requires years of schooling, like medicine, formal education may not be the answer right now.

    Excitement: The New Paintbrush and the Agentized Business

    On the hopeful side, she sees AI as a new paintbrush unlocking dormant creativity. The person with a million dollar idea who could never get on Shark Tank can now vibe code the app, put up a site, and eventually vibe manufacture the product, running an agentized business with no employees from a basement. These may not be venture scale companies, but they will mint a wave of millionaires, followed in two to three years by a consumer wave that changes signs, fashion, and manufacturing. On AI art controversies she is pragmatic: all art enters the public domain after a hundred years anyway, China will train on Western IP regardless and sell it back, and today’s tools are the worst they will ever be. She has hundreds of movies inside her and can finally make them. But AI is not a replacement for humans: its writing carries telltale signs, and the human job is to take its nuggets and massage them back into human form, agreeing with the host’s Config takeaway that AI generates the average and the human must pull the work out of the bell curve.

    The SpaceX Bet and What Risk Taught Her

    The centerpiece story: Luke Nosek, who met Scott Banister and Max Levchin in a University of Illinois computer lab and drove west with Scott communicating by walkie-talkie, became Elon Musk’s fiercest champion. He arrived at the Banisters’ house, got on the floor in his Vibram shoes, and delivered the pitch: dig through your couches, anything liquid you have, SpaceX needs it. Rockets were blowing up on the launch pad and critics said private citizens had no business in space, but Nosek and Scott argued that early aviation crashed a lot of planes too, and that a young, infinitely employable person should take the shot. Fresh off her IronPort exit to Cisco, she went all in, then immediately wondered what she had done while her own startup struggled. Twenty years later, still essentially unsold, it is the best investment she will probably ever make. The deeper lesson came from realizing angel investing was a special club she had lucked into, one most people never learn exists. Scott farmed his PayPal network while she networked relentlessly through TC40, TC50, Disrupt, and YC demo days for a decade, writing failed checks and calibrating her pattern matching, because becoming good at early stage investing requires losing.

    What’s Next, and the Mentors Behind It

    Her most exciting current bets are Substrate and Becoming Bio, the substrates of technology and biology, plus Diamond Foundry, whose skeptics saw only synthetic rings while the founder saw industrial diamonds and wafers for AI and crypto. She wants semiconductor manufacturing back in the Americas, is watching AI companion devices race toward a genuinely useful Tamagotchi, and keeps a standing public request for the Alibaba of the Americas. She avoids today’s hot hypercompetitive deals because the alpha lives in what is coming, nanotech, biotech, and the bottlenecks AI removes, not what is hot now. Her inspirations: Peter Thiel, her partner for four years at what she calls Peter Thiel University, whom she defends as tolerant, open-minded, and poorly understood; Marc Andreessen, the teenage hero she vowed to meet as an equal and eventually did; Scott Cook of Intuit, her gold standard of executive function and decency; and Rick Rubin, the one mind she compares to her own, whom she is putting out into the universe a request to meet. The closing stretch is a defense of curiosity over tribalism: go to the Thiel events, hear all sides before deciding where you stand, stop diluting real words like racism through overuse, and protect the two canaries of a free nation, comedy and rap music. Free speech, she says, is her number one cause.

    Notable Quotes

    “Banisters, I need you to dig through your couches. Anything liquid you have, I need it. SpaceX needs it.”

    Luke Nosek’s pitch, as retold by Cyan Banister, describing the night that led to her first angel check

    “Luke and Scott convinced me to put everything that I made in IronPort when we sold to Cisco into SpaceX.”

    Cyan Banister, on going all in while SpaceX rockets were still blowing up on the launch pad

    “You can actually surgically go in and alter your personality where you can actually change these things, but it takes effort and time and a lot of facing the ugly truth about yourself.”

    Cyan Banister, rejecting the idea that personality is fixed

    “Meditation is about noticing the thoughts and noticing that they’re going by like clouds and then noticing that you’re noticing. So who is that person? So if you are not your thoughts then who are you is where I would start.”

    Cyan Banister, on the five minute practice she prescribes to stuck founders

    “When you ask why about everything in the world, it’s just going to make you a better investor. It’s going to make you a better entrepreneur, a better everything.”

    Cyan Banister, on humans as why machines and the habit behind her Uber and Niantic bets

    “I always say it’s the closest we’ve come to world peace. It was one of the most magical few weeks of my life and probably many people’s lives.”

    Cyan Banister, on the launch of Pokemon Go

    “I’ve got to try to find things at lower prices that are still a moonshot that I can get a good percentage of ownership like a good seed fund should do.”

    Cyan Banister, on why she avoids hypercompetitive hot deals where the alpha is already gone

    “If you get invited to a Peter Thiel event, go. Do not shy away from it. It does not make you anything that anyone’s going to accuse you of.”

    Cyan Banister, on curiosity versus shame by association

    “You can tell the health of a nation by its ability to tolerate comedy and rap music. Those two things have to exist for freedom.”

    Cyan Banister, on the canaries of free speech, her number one cause

    Watch the full conversation with Cyan Banister on the Sourcery podcast here.

    Related Reading

    • Cyan Banister (Wikipedia) background on her path from homelessness to one of the most successful angel investors in the world.
    • Long Journey Ventures the “magically weird” seed fund she co-founded, home of the bellwether and the second believer candles.
    • Luke Nosek (Wikipedia) the PayPal co-founder and Founders Fund co-founder whose couch-digging pitch started it all.
    • The Creative Act by Rick Rubin, the book behind the openness-to-the-universe mindset Banister says mirrors her own.
    • Purpose (PJFP) our pillar page on building the kind of why-driven daily practice Banister describes.

  • Bill Gurley on Mental Models, Systems Thinking, AI Investing, Stablecoins, and the Future of Venture Capital

    Bill Gurley spent his career at Benchmark backing some of the most consequential marketplaces and network-effect businesses of the internet era, including Uber, and he is one of the few investors who pairs deep Wall Street fundamentals with a real feel for the bleeding edge. In this wide-ranging conversation on Shane Parrish’s The Knowledge Project, he lays out the mental models he keeps returning to, how systems thinking keeps you out of trouble, why the history of your field is a hidden superpower, where AI investing is headed, and how stablecoins and tokenization could quietly rewire finance. It is a masterclass in thinking clearly about complex systems while staying obsessively curious about what is happening on the edge.

    TLDW

    Gurley anchors his thinking in systems thinking and complexity theory, warning that multivariable nonlinear systems produce second and third order consequences that punish anyone who optimizes for a single metric. He argues that mastering both the deep history of your field and its newest edge is wildly differentiating, whether you are interviewing for a marketing job or breaking into venture capital. On AI he is measured: he doubts a single model eats every vertical, sees real moats in workflows and proprietary data, flags that we may be painting in the corners on training data, and explains why Chinese open source models may innovate faster because forced knowledge sharing compounds. He thinks the AI buildout looks overfunded and that circular deals both raise the odds of an eventual correction and delay it. He makes the case that the IPO process is a rigged power grab, that stablecoins and instant payments threaten Visa, Mastercard, and the entire 2 to 3 percent credit card stack, and that proxy advisors like ISS have drifted from shareholder interest into a black-box heist. He closes on the craft of storytelling and writing as thinking, the equal-partnership design of Benchmark, why venture bends toward youth, and what success means now that his dream job is behind him.

    Thoughts

    The most useful idea in this conversation is also the quietest one: most bad decisions are not bad in the moment, they are bad in the second derivative. Gurley’s dating-site story, where lengthening profiles raised engagement in the test and then quietly killed conversion months later, is the whole argument in miniature. A linear model would have shipped that change and called it a win. A systems thinker assumes the variable you optimized is connected to three others you cannot see yet, and waits to find out. That posture, refusing to get deterministic about a single metric, is the difference between a clever experiment and a durable business. It is also the most transferable thing in the episode, because it applies to product changes, hiring, policy, and your own career just as cleanly as it applies to a dating app.

    His pairing of old and new is the second idea worth stealing. Everyone in tech tells you to live on the edge, and Gurley agrees, he keeps five premium AI accounts running so he never misses a release. But he insists the edge is only half of it. Knowing the deep history of your field, the masters of marketing, the forefathers of physics, the classic cartoons that taught animation, is rare enough that it instantly creates contrast and signals genuine passion. The compounding move is to hold both at once. If you understand the legends and you actually get TikTok, you are a power player in a way that someone who only knows one end of the timeline can never be. Most people pick a side. The leverage is in refusing to.

    On AI specifically, Gurley is refreshingly unwilling to pick the consensus lane in either direction. He does not buy that one near-sentient model swallows every vertical, and his reasoning is grounded rather than vibes-based: workflows and proprietary data create real switching costs, which is why he watches the legal AI startups ingesting case law and building new databases rather than assuming everyone reverts to a general chatbot. At the same time he respects the Microsoft pattern of platforms climbing the stack and crushing the apps above them. The honest answer is that it is genuinely up for grabs, and his comfort sitting in that uncertainty is itself a model. The cheap takes are “one model to rule them all” and “it is all wrappers.” Gurley holds both possibilities and keeps testing.

    The systems lens does its best work on China. Rather than moralize, Gurley runs the mechanism: roughly ten open source models, intense domestic competition, and a culture of publishing techniques and weights so every model can learn from, train, and test every other model. His two-farmer metaphor, one market where farmers only trade goods and another where they are forced to share best practices, makes the prediction obvious. Forced knowledge sharing compounds faster than secrecy. The uncomfortable corollary he names is that American startups are quietly forking those open models all over Silicon Valley, and that incumbents may be lobbying for heavy regulation precisely because it pulls up the drawbridge against open source competition. That is the systems thinker’s signature move: follow the incentives to the consequence nobody is saying out loud.

    Finally, the money section is a clinic in spotting rent extraction. The IPO process where bankers pick both the price and the favored buyers, the 2 to 3 percent credit card toll that exists for no defensible reason while the rest of the world built instant bank transfer decades ago, and the proxy advisors who score companies in a black box and then sell you the cure, are all variations on the same pattern: an intermediary that captured a choke point and defends it through regulatory capture rather than value. Gurley’s optimism is that crypto rails, stablecoins, and tokenization may finally route around these tolls the way WeChat Pay and Alipay leapfrogged cards in China. Whether or not you agree on the timeline, the analytical habit is the takeaway. When something costs far more than it should and has for decades, ask who captured the rules, and watch the edge for whoever is about to make those rules irrelevant.

    Key Takeaways

    • Systems thinking means treating the world as multivariable nonlinear systems where one variable flipping can change the entire system’s behavior, the way weather and stock markets do.
    • The real danger is second and third derivative effects, consequences that only show up much later, long after the metric you optimized looked like a win.
    • A dating site lengthened profiles because longer profiles tested as more engaging, then discovered months later it was negative for conversion, the textbook second order trap.
    • Never get too deterministic about a single metric or single variable, and always know what is actually important and what sits on top.
    • Gurley built his foundation on the canon: Peter Lynch’s One Up on Wall Street, A Random Walk Down Wall Street, the Buffett letters, Ben Graham, and Howard Marks.
    • A firm grasp of the financial bedrock is what lets you innovate on top of it, and many Silicon Valley VCs would benefit from understanding finance better.
    • Bill Miller reframed value investing as buying an asset that is underpriced relative to what you think it will be worth in the future, which is how he justified holding Amazon for its network effects.
    • Wall Street is the buyer of the product that venture capitalists create, so even at the two-people-in-a-PowerPoint stage you should ask whether the eventual public market will be excited by it.
    • Trajectory matters more than the starting place, because the trajectory is where the company actually ends up.
    • Knowing the deep history of your field is remarkably differentiating, and tedium while learning it is a signal you are in the wrong lane.
    • John Lasseter served Gurley a ten-course meal where each course was tied to a classic cartoon essential to understanding animation, a display of mastery over the history of the craft.
    • Magnus Carlsen won a trivia contest on the history of chess, and Picasso was a wildly successful realist painter by 14, both proof that the greats master the fundamentals first.
    • Obsessive, constant learning is the trait Gurley sees most in great entrepreneurs, because disruption always happens on a moving edge they need to understand at the top one percentile.
    • The compounding advantage is mastering both the old history and the new edge at once, the way understanding both marketing legends and TikTok would set you apart in any interview.
    • Most people underestimate how much AI can do, so push more of the downstream work into the prompt: identify the top ten, list pros and cons, rank them on one dimension, then another, and add up the numbers too.
    • Gurley uses ChatGPT for project structure and memory, Gemini for restaurant research powered by Google review data, and notes that coders swear by Claude while some prefer Perplexity for finance.
    • He doubts one model dominates everything; verticals like coding already let users swap models, and price optimization will push more swapping over the next few years.
    • Heavy, expensive regulation could ironically create oligopoly, and some players may be quietly begging for regulation because it pulls up the bridge against Chinese open source models.
    • China’s roughly ten open source models compete intensely and share weights and techniques, creating a system that can innovate faster, like farmers forced to share best practices instead of just trading goods.
    • A quiet secret is that startups all over Silicon Valley are forking those Chinese open source models at real volume.
    • Gurley comes down against the idea that one near-sentient model removes the need for vertical models; workflows and proprietary data, like legal startups ingesting all the case law, create durable moats.
    • We may be running out of training data, painting in the corners, which is why one of the most powerful improvements is hiring experts at thousands of dollars an hour to fine-tune the models.
    • Yann LeCun’s view is that the next leap is broader than LLMs, since language-based models hit an asymptote and are weak at math and numbers.
    • AlphaGo’s shocking move proves models can innovate beyond their training, but it lived in a constrained game; the real world has infinite paths a computer cannot exhaustively search.
    • Gurley’s non-consensus view is skepticism of the China vilification mindset, noting the US is only 3 to 5 percent of the global population and wondering how the other 95 percent hears American exceptionalism.
    • The AI buildout looks overfunded: the Magnificent Seven took free cash flow from 50 to 100 billion a year down toward zero by pouring it into capex.
    • The venture community has become more risk-seeking because it now deeply believes in increasing returns and power laws, and the pre-profit losses keep scaling, from Amazon’s 2 to 3 billion to Uber’s 15 billion to far more now.
    • Circular deals, where a cloud provider funds a model company that spends the money right back on its services, inflate growth, which both raises the probability of an eventual correction and extends the time before one hits.
    • Burn rate is a measure of risk; ten years ago a million a month was scary, now companies burn five billion a year and cannot really know their unit economics.
    • Tokenization without financial-disclosure regulation invites speculation and manipulation, which is part of why companies like Stripe stay private and negotiate liquidity prices with trusted investors.
    • The IPO process is unfair because bankers pick both the price and the shareholders; a freshman would simply match supply and demand anonymously in an auction, the way direct listings and ICOs do.
    • Stablecoins threaten the 2 to 3 percent credit card stack; USDC holds dollar-for-dollar Treasuries and rides fast global crypto rails, while US transfers still suffer three-day ACH settlement and 25 dollar wires.
    • The rest of the world built instant transfer long ago, from UK Faster Payments 20 years ago to Argentina’s PIX-style system reaching 60 to 70 percent of transactions, while US bank regulatory capture stalled Fed Now.
    • Visa and Mastercard run roughly 60 percent operating margins as a bank-created duopoly, and China leapfrogged them entirely with WeChat Pay and Alipay QR-code wallets.
    • Moody’s power is being the trusted standard, the watermark, so AI on the back end does not displace it; ISS and proxy advisors, by contrast, score companies in a black box and get paid on both sides.
    • Proxy advisors drifted from shareholder interest into a fraud-and-risk-mitigation mindset, which is why they reflexively opposed the Tesla pay package that only paid out if the stock soared.
    • The rise of passive index funds concentrated voting power in firms that lack time to evaluate votes; it would be healthier if they abstained or voted in proportion to active holders.
    • Storytelling is one of the top founder traits, because founders are recruiting, raising money, and closing customers and partners constantly, selling all the time.
    • Writing is thinking: Bezos’s six-page memo forces you to find the loose ends and tie them up, and a public blog becomes a calling card that magnetizes founders and deal flow.
    • Other founder unfair advantages are product instincts, which fewer than 5 percent of non-product people ever truly learn, and sheer determination, Bezos’s single angel-investing test of whether someone will do it no matter what.
    • Uber had no HBS case study to lean on; its winner-take-all network effects forced mega burn rates with no precedent and no mentor to call, a situation every AI company now faces.
    • Benchmark’s equal partnership, with no king, president, or lead and five equal partners, makes recruiting easy, kills comp politics, and aligns everyone, at the cost of being hard to scale or run new initiatives.
    • Venture bends toward youth because young investors can match founders’ age, master a fresh niche faster, and have the free time to study something 80 hours a week.
    • Gurley defines current success through Arthur Brooks’s From Strength to Strength, hoping to apply his synthesizing and writing skills to bigger societal problems and dent the universe a little.

    Detailed Summary

    Systems Thinking and Second Order Effects

    Gurley opens with the mental model he keeps returning to: systems thinking, shaped by Donella Meadows’s Thinking in Systems and his board seat at the Santa Fe Institute, which studies complexity theory. He describes complex systems as multivariable nonlinear systems that are very hard to predict, capable of behaving one way for a long time until a single variable flips and the whole system behaves differently, like weather or stock markets. The practical payoff is staying out of trouble by anticipating first, second, and third derivative consequences. His clearest example is a large dating site that lengthened user profiles because the test showed more engagement, only to learn many months later that knowing more at that stage was negative for conversion. The lesson is to never get too deterministic about a single metric and to keep the whole system in view, because a change here can ripple to there in ways you only discover much later.

    Learning the Craft of Investing

    Because he started on Wall Street rather than in venture, Gurley absorbed the investing canon first: Peter Lynch’s One Up on Wall Street, A Random Walk Down Wall Street, the Buffett letters, Ben Graham, and Howard Marks, people who spent careers assembling and publishing their thinking. That financial bedrock, he argues, is exactly what lets you innovate on top of it. His friend Michael Mauboussin introduced him to Bill Miller, the Legg Mason manager who beat the S&P for 15 straight years and was Amazon’s largest shareholder for a long stretch. Miller reframed value investing as buying an asset underpriced relative to its future worth, which combined with a belief in network effects justified holding a company that could grow at an unreasonable rate for years. Gurley also frames Wall Street as the buyer of the product venture capitalists create through eventual M&A or IPO, so founders should think early about whether the public market will be excited by what they are building, since trajectory matters more than the starting place.

    Mastering Both the History and the Edge

    Gurley makes an unusually strong case for studying the deep history of your field. He recounts a dinner with Pixar’s John Lasseter, who served a ten-course meal where every course was tied to a classic cartoon he considered essential to understanding animation, and notes that Magnus Carlsen won a chess-history trivia contest and Picasso was a master realist by 14. In a world that skims for the executive summary, walking into a marketing interview with command of the masters of marketing is wildly differentiating and signals genuine passion; if learning that history feels tedious, you are probably in the wrong lane. The counterpart trait he sees in great entrepreneurs is obsessive learning on the moving edge, where disruption actually happens. Gurley keeps five premium AI accounts so he never misses something. The real power player holds both at once, the legends and the newest thing, the way a candidate who knows the marketing greats and truly gets TikTok stands out completely.

    Using AI Well and the Model Wars

    People underestimate how much AI can do, Gurley says, so you should build more of the downstream work into the prompt: instead of asking for the top ten and studying them yourself, ask it to list pros and cons, rank on one dimension, rank again on another, and add up the numbers too. He uses ChatGPT for its project structure and memory, leans on Gemini for restaurant research because it carries Google review data, and notes coders swear by Claude while some prefer Perplexity for finance. On whether one model dominates or models become niche commodities, he points to coding, the largest vertical, where tools like Cursor already let users swap models, and predicts price optimization will drive more swapping. The counterforce is regulation: if it gets expensive and mundane it could create oligopoly, and some players may be quietly begging for it because it pulls up the bridge against Chinese open source models.

    China, Open Source, and the Systems Advantage

    Asked to apply systems thinking to China, Gurley describes roughly ten open source models locked in intense domestic competition, all learning from one another because the ecosystem chose openness, with models able to train and test other models and teams publishing the techniques behind their breakthroughs. His metaphor: two agricultural societies, one where farmers only trade goods at market and another where they are forced to share best practices; the second evolves far faster. The result is a system capable of innovating faster than the more secretive Western approach. The quiet secret he names is that startups all over Silicon Valley are forking those open models at real volume, and a key open question is whether regulation tries to stomp that out. He extends this into a broader non-consensus discomfort with the vilification of China common in Washington and parts of Silicon Valley, observing that the US is only a few percent of the global population.

    AI Investing, Moats, and the Limits of Models

    On how AI changes investing and whether a startup is just a wrapper, Gurley calls it up for grabs but lands on the side of durable verticals. If models become near-sentient, one model does everything; he doubts that, pointing to workflows and data moats, like the several legal AI startups ingesting all the case law and building new databases that customers will not simply swap for a general chatbot. He balances this against the Microsoft pattern of platforms climbing the stack past Lotus 1-2-3 and WordPerfect. He also flags scaling limits: we may be running out of data, painting in the corners, which is why one of the most powerful improvements is paying experts thousands of dollars an hour to fine-tune models, though human knowledge has an edge. He invokes Yann LeCun’s argument that the next leap is broader than language-based LLMs, which hit an asymptote and struggle with math, and the AlphaGo debate, where a shocking innovative move proves creativity within a constrained game but says little about the infinite paths of the real world. He notes AlphaGo and Tesla’s FSD are constrained, non-LLM systems.

    Is the Buildout Overfunded

    Gurley admits he is shocked by the scale of money, noting the Magnificent Seven drove free cash flow from 50 to 100 billion a year down toward zero by spending it all on capex, something he would not have believed five years ago. He traces it to the venture community’s growing conviction in increasing returns and power laws, where proven companies grow far beyond expectations, which makes investors more willing to take risk on the come. The losses before turning cash-flow positive keep scaling, from Amazon’s 2 to 3 billion to Uber’s roughly 15 billion to far larger now. On corrections, he recalls the dot-com crash producing a three to four year nuclear winter before Amazon climbed back, and explains that circular deals, where a cloud provider funds a model company that spends it right back on its services, inflate growth and therefore both raise the probability of a correction and extend the runway before one arrives. Burn rate, he stresses, is a measure of risk, and at five billion a year it is nearly impossible to know your unit economics.

    Tokenization, the IPO Heist, and Going Public

    There is no shortage of capital, so funding is not the bottleneck; the risk with tokenization is that, absent disclosure regulation, it invites speculation and manipulation, as seen in retail-loved names like GameStop and Palantir. Tokenizing a private company like Stripe could create the wild price swings companies stay private to avoid, since private liquidity events let them negotiate a price with trusted investors rather than expose the constantly moving underlying value, and Robinhood’s tokenization plans already drew legal pushback. Gurley reserves his sharpest critique for the IPO process, calling it insanely unfair because bankers pick both the price and the favored shareholders. A freshman computer science and finance student would simply match supply and demand anonymously in an auction, the way an ICO or a direct listing does, but Wall Street will not let go of the greedy power grab and reverted to a controlled oligopoly after direct listings were available.

    Stablecoins Versus the Payment Cartel

    Gurley argues stablecoins could be deeply disruptive to credit cards. Most of the developed world built instant bank-to-bank transfer long ago, from UK Faster Payments 20 years ago to Argentina’s PIX-style system that quickly hit 60 to 70 percent of transactions, while US bank regulatory capture stalled Fed Now and left an ecosystem living under 2 to 2.5 percent card fees. A USDC stablecoin holds dollar-for-dollar US Treasuries and rides proven, fast, global crypto rails, letting anyone move a dollar in seconds for pennies, against the backdrop of three-day ACH settlement and 25 dollar wires. He sees Visa and Mastercard, a bank-created duopoly with roughly 60 percent operating margins, as heavily threatened, and points to China, where WeChat Pay and Alipay built ubiquitous QR-code wallets that leapfrogged the entire card system, all because the government made money transfer easy.

    Moody’s, Proxy Advisors, and Index Funds

    Moody’s power, Gurley explains, comes from being a trusted standard, the watermark, so even AI on the back end does not displace it. Proxy advisors like ISS are a different story: they score companies in a black box, refuse to reveal the criteria, and then get paid by the same companies that want to learn how to score better, which he calls more of a heist than a service. They drifted from a shareholder-interest mandate into a corporate-governance, fraud-mitigation posture obsessed with rules, which is why they reflexively opposed the Tesla pay package that only paid Elon Musk if the stock soared, a deal Gurley says he would sign for every company he has worked with. The rise of passive index funds compounds the problem, concentrating voting power in firms without time to evaluate votes; he would prefer they abstain or vote in proportion to active holders, since closet indexing during the MAG 7 run already distorted active management.

    Storytelling, Writing, and Founder Advantages

    Gurley fell in love with the craft of writing in business school, moving from business books to personal development titles like Dale Carnegie and Seven Habits, then biographies, then long-form narrative nonfiction by Malcolm Gladwell, Michael Lewis, and Jon Krakauer, the New Journalism that reads like fiction. Writing forces clarity: he cites Bezos’s six-page memo as a tool that makes you think through corner cases and tie up loose ends, and notes that codifying his marketplace knowledge and publishing it turned his blog into a calling card that magnetized founders and deal flow. He lists the top founder traits as storytelling, product instincts, understanding the edge, and determination. Storytelling matters because founders are constantly recruiting, fundraising, and closing customers and partners. Product instinct is nearly unteachable, present in well under 5 percent of non-product hires. And determination is Bezos’s single angel-investing test: will this person do it no matter what, come hell or high water.

    Uber, Benchmark, and the Shape of Venture

    The Uber lesson with no HBS case study was that a winner-take-all category with network effects demanded funding ad nauseam, producing burn rates bigger than any public company would dare, with no precedent and no mentor to call, exactly the situation AI companies now face, only with a zero added. Gurley credits Benchmark’s design, an equal partnership with no king, president, or lead and five equal partners, for making it easy to recruit top talent, encouraging senior partners to develop newcomers since everyone shares the upside, and eliminating annual comp politics. The downside is that without a CEO it is hard to scale or run new initiatives, famously captured by the firm settling on a single splash-page website. Founders choose a VC for reputation and network effects, the stamp of approval that carries weight, and young investors can break in because they often match founders’ age and can outwork everyone to master a fresh niche like esports or YouTube, which is why the industry bends toward youth. Asked what success means now, Gurley says his venture career was a dream job he would have done for free, but it is done; inspired by Arthur Brooks’s From Strength to Strength, he wants to apply his synthesizing and writing to bigger societal problems and dent the universe a little.

    Notable Quotes

    “We do live in a world where information is really cut up, but we also live in a world where you can have access to more information than you ever could.”

    Bill Gurley, on why the abundance of knowledge rewards the curious

    “You got to be really conscious of the consequence and not get too deterministic about a single metric or a single variable.”

    Bill Gurley, on the discipline of systems thinking

    “Value just means that the asset is underpriced relative to what you think it will be worth in the future.”

    Bill Gurley, relaying Bill Miller’s reframing of value investing

    “I’ve always thought of Wall Street as the buyer of the product that venture capitalists create.”

    Bill Gurley, on why founders should think about the public market early

    “One society, when the farmers come to market, they just sell each other goods and then they go back. The other society, when the farmers come to market, they’re forced to share best practices. Which one is going to evolve faster?”

    Bill Gurley, on why open source models can out-innovate

    “If you took a freshman computer science student and a freshman finance student and said imagine how a company should go public, they would match supply and demand anonymously like you would in any auction.”

    Bill Gurley, on the rigged IPO process

    “When I meet an entrepreneur, there’s only one thing I ask myself. Is this person gonna do this no matter what? Come hell or high water, they’re doing this.”

    Bill Gurley, quoting Jeff Bezos on his single test for angel investing

    “You’re recruiting employees, you’re recruiting executives, you’re raising money, you’re closing customers, you’re closing partnerships. You’re selling all the damn time.”

    Bill Gurley, on why storytelling is a top founder trait

    “I often said that if we lived in a socialist society and everyone had to work for free, I would still take that job.”

    Bill Gurley, on loving his venture career

    “I would like to see if I can apply those techniques to bigger, broader problems in society and dent the universe a little bit that way.”

    Bill Gurley, on what success looks like in his next chapter

    Watch the full conversation with Bill Gurley on The Knowledge Project here.

    Related Reading

  • Vibe Coding Hardware: Naval, Guillermo Rauch, Blake Scholl, and Max Hodak on AI-Designed Jet Engines, Vertical Integration, China’s Open-Source Bet, and Why Humans Become Verifiers

    This is part two of Naval Ravikant’s conversation with frontier founders Guillermo Rauch of Vercel, Blake Scholl of Boom Supersonic, and Max Hodak of Science. Where the first part argued that you should waste tokens to save time and that the job of an engineer is now to build the factory rather than the output, this segment drags that thesis out of pure software and into atoms. The question on the table is what happens to hardware when models can vibe code the spreadsheets, the simulations, and eventually the step files and PCB layouts that aerospace, semiconductors, and biotech are built on. This segment is one half of the discussion, and you can watch and read the full episode here. The full conversation is on the Naval Podcast YouTube channel.

    TLDW

    Blake Scholl describes how Boom Supersonic took hardware engineering workflows that used to live in siloed Excel spreadsheets and VBScript on individual laptops, with handoffs done by email like it was the 1990s, and turned them into versioned, testable software. The new model is that software engineers build the architectures and the tools while hardware engineers vibe code their own domain-specific pieces, which collapsed a turbine-blade analysis that once took one engineer one day per blade into something where two engineers can design an entire jet engine in real time. Naval generalizes this into the cataclysm of enterprise software: there is no longer a startup that can sell you hardware collaboration tools because companies just code the exact thing they need on demand, and even spreadsheets are cooked because they only existed as a proxy for custom software nobody could previously afford to build. Blake predicts that within 2026 AI will move from generating software to generating step files and PCB layouts, which reshapes mechanical and electrical engineering. The group debates China’s open-source push as a way to neutralize Silicon Valley’s software advantage and protect its hardware and supply-chain superiority, lands on the point that if you fall behind on generating software you fall behind on generating everything, and Guillermo notes that frontier coding intelligence still dominates real usage while cheaper models like Gemini win at scale for support and browser automation. Max Hodak explains Science’s vertical integration, including a captive MEMS foundry on the East Coast, because the most innovative hardware cannot be bought off the shelf, and argues that software still needs hands since a model that cannot make physical things hits real boundaries. The conversation closes on the shift from writing to verifying: junior engineering got absorbed by agents while juniors got promoted, the same way paralegals could be seen as fired or promoted, and humans across law, engineering, and operations are becoming the verifiers who sign off on systems they did not write line by line.

    Thoughts

    The most important shift in this segment is that vibe coding stops being a software-industry story and becomes a deep-tech story. In part one the examples were Postgres, ClickHouse, and deploy targets. Here Blake Scholl is talking about turbine blades that change shape when they heat up, and the brutal fact that converting between cold and hot geometry, and between aerodynamics and structures, used to eat one engineer for one full day per blade in an engine that has a thousand blades. That is the kind of math that quietly kills ambition. When he says two engineers can now design an entire jet engine because the structural and aerodynamic results update in real time as you change the geometry, that is not a productivity improvement, it is a change in what a small team is allowed to attempt. The interesting move is the division of labor: software engineers build the architecture and the framework because they understand systems and separation of concerns, and the hardware engineers vibe code the pieces only they understand. Nobody has to become both.

    Naval’s “cataclysm of enterprise software” is the most investable idea in the episode, and it is darker than it sounds for anyone selling B2B tools. His claim is that the entire category of internal collaboration software is being eaten from the inside, because a company that can generate exactly the tool it needs on any given day will not pay a vendor for an approximation of that tool. His follow-on that even spreadsheets are cooked is the sharpest version of the point. The spreadsheet won for forty years precisely because it was the closest thing to custom software that a non-programmer could produce. Remove the constraint that custom software is expensive and the spreadsheet loses its reason to exist. The counterweight, which the group raised in part one with the block-economy thesis, is that the infrastructure primitives agents reach for get more valuable, not less. So the safe place to build is not the collaboration layer on top, it is the primitive underneath.

    The China discussion is the geopolitical center of the conversation and it lands on a genuinely uncomfortable insight. The argument is that China leans into open-source models not only because it is a model or two behind, but because open weights neutralize Silicon Valley’s software advantage and let China lean on what it already dominates: hardware, supply chains, and component ecosystems. If software can be generated on demand from open models, then the country with the factories wins the stack. The sharpest line is that if you fall behind on the ability to generate software, you fall behind on the ability to generate everything, because software is now upstream of every hardware pipeline. That reframes the open-versus-closed debate as a question about who controls the means of producing the means of production. It also quietly flatters the American frontier labs, since the same logic says self-improvement requires frontier coding models, and on that narrow axis the consensus at the table is that the Chinese models are not yet in the race.

    Max Hodak provides the necessary cold water, and it is the most grounding contribution in the episode. Everyone else is describing software eating the design layer, and Max points out that you still have to make the thing. Science owns a captive MEMS foundry on the East Coast not as a flex but because there was no other way to do the packaging and assembly for products that approach a single block of covalently bonded matter. His framing that the software still needs hands is the real boundary condition on all the AI-eats-everything talk: a model can be smarter than every engineer in the building and still be unable to deposit a layer, bond a wafer, or pass a regulatory inspection. The optimistic version, which he also makes, is that he has instrumented the foundry so that as models improve, the gains show up immediately in cell engineering and material science. The pessimistic reading is that the physical world remains a hard rate limiter, and the companies that own the atoms will capture more of the surplus than the companies that only own the bits.

    The closing thread on verification is where the whole conversation resolves into a job description for humans. Guillermo’s point that the biggest problem in software is mountains of slop arriving as a pull request, and that the answer is not pretending to read every line but being able to say “I am signing off on the consequences of this PR, and I wrote the harness, the simulations, the proofs, and the type checkers that let me,” is the most practically useful idea in the episode. It generalizes cleanly. The lawyer you trust is not the one who wrote every clause by hand, it is the one putting their reputation on the line that the document is sound. The production engineer who gets paged at 3am is the one signing off that the system is safe to ship. As models absorb the junior tier of every knowledge profession, the surviving human role is the verifier who carries the accountability. That is a promotion for the people who can hold it and an extinction event for the people whose value was doing the work nobody now needs done by hand.

    Key Takeaways

    • The factory framing from part one carries straight into hardware: you are judged on whether you build the system that produces multiplicative outputs, not on the single artifact, and the real multiplier was always 100x or 1000x, not 10x.
    • AI completely changes the role of software and hardware developers rather than just speeding either one up.
    • A huge amount of hardware engineering lives in complex Excel spreadsheets and VBScript on individual engineers’ laptops, with no source control, no automated testing, and handoffs done manually over email. It is software that is not treated as software.
    • Boom Supersonic’s move from day one was to turn traditional hardware engineering workflows into real software frameworks that are automatable and repeatable, to drive down the cost of iteration.
    • The old bottleneck was never being able to afford enough software engineers to build those frameworks. AI removes that constraint.
    • The new model: software engineers create the architectures because they understand systems, algorithms, and separation of concerns, and hardware engineers vibe code the domain pieces only they understand.
    • A turbine blade is cold when it starts and hot when it runs, so it changes shape, and you must design both the cold and hot geometry across aerodynamics and structures. Classically that was one engineer, one day, for one blade, in an engine with a thousand blades.
    • With software and hardware people combined, you can now change blade geometry and see the structural and aerodynamic results in real time, which lets two engineers design an entire jet engine.
    • Naval’s cataclysm of enterprise software: no startup can sell hardware collaboration tools anymore because companies just code the exact thing they need at any given time.
    • Even spreadsheets are cooked. Spreadsheets won only because nobody could build custom software, so a spreadsheet full of VBScript was the closest available approximation. Remove the cost barrier and the approximation loses.
    • Engineers are moving from Excel to Python models that produce believable simulations of physical systems.
    • AI can generate software today, but within 2026 it is expected to generate step files and PCB layouts, which opens up mechanical and electrical engineering as the next frontier.
    • The hardware software boon is biggest for small gadget and parts companies that historically shipped bad software because they could not afford good software. Now they can ship good-enough software, or skip the human front end entirely and expose hardware agentically for voice and agent control.
    • China goes all in on open-source models partly to neutralize Silicon Valley’s software edge: if software can be generated on demand from open weights, China’s hardware and supply-chain superiority stops being offset by a software disadvantage.
    • Other reasons cited for China’s open-source push: it is a model or two behind, it is distilling models, and the government has a history of funding efforts that lift the whole ecosystem, especially in network-effect businesses.
    • Open-source heft is coming almost entirely from China. OpenAI is not open, Grok publishes models but is seen as a model or two behind, Google’s local models are not very competitive, and Anthropic is not known for open-source releases.
    • Without frontier coding models you do not get self-improvement, and if you fall behind on generating software you fall behind on generating everything, because software now sits upstream of every hardware pipeline.
    • Real AI gateway usage shows open models do get used, but the top is heavily dominated by frontier intelligence.
    • Frontier intelligence at the right cost and performance slaps at scale. Gemini models are underrated and excel as industrial production models for support tasks and browser automation, even if they are not the top pick for coding.
    • For pushing the frontier you need the best possible coding model, which is now only two or three models, and the Chinese models are not among them.
    • One contrarian view at the table: use DeepSeek for 97% of tasks because it is cheap, run it repeatedly for harder problems, and reserve frontier models for the most advanced work. The counterargument: intelligence is an unalloyed good, mistakes are invisible and costly, and a smarter model is always cheaper than a person, so you default to the most intelligent option.
    • Always wanting the most intelligent model risks creating a monopoly or oligopoly in AI, because when two models disagree you cannot tell which is right, so you trust the smarter one and stop asking the weaker one.
    • Vertical integration is forced, not chosen: if you cannot buy it, you have to make it. The preference is always to buy when a vendor offers a service at a great price, like PCBs from Asia.
    • The closer a product gets to a single block of covalently bonded matter, the better it performs: lower power, smaller, higher performance, longer lasting. The components for that level of integration simply are not available to buy.
    • Science owns a captive MEMS foundry on the East Coast, bought because there was no other way to do the packaging and assembly the company needed.
    • One of the biggest near-term AI impacts inside hardware companies is regulatory and documentation work: tracing which of thousands of ISO standards apply used to occupy a regulatory and quality team for months, and now AI just knows.
    • Software still needs hands. A model can be smarter than us and still hit real boundaries if it cannot physically make things, which is why Science has instrumented its foundry so model improvements show up immediately in cell engineering and material science.
    • Basic legal work is already going away. People have stopped asking lawyers for NDAs and routine agreements, because law is spaghetti code in English with no real APIs, and the basic tasks are handled by AI.
    • Junior engineers got promoted to senior engineers while junior engineering itself got taken over by agents. The same framing applies to paralegals: fired, or promoted to senior lawyers who now spend their time thinking about the law.
    • What you value in a lawyer is a trusted authority who puts their reputation on the line, not someone who read every clause. The same trust model is coming to engineering.
    • The biggest problem in software engineering today is mountains of slop arriving as a pull request. The old norm of reading every line of a PR is gone.
    • The new standard is being able to say “I understand and I am signing off on the consequences of this PR,” backed by the test harness, simulations, proofs, and type checkers you built, even without reading every line.
    • Embrace a world where code is spaghetti you do not fully understand, but build the evaluators that give confidence, and rely on production engineers to sign off because someone gets paged if the system goes down.
    • Creating software is easy from zero to one. The hard part is a thousand days from now: is it secure, tested, production grade, and performant, and are you still motivated to invest the tokens to maintain it in prod?
    • Humans are becoming verifiers. The same way models are trained on good verification data, the old functions of lawyers, engineers, and operations people are moving to verifying the stack and standing behind it.

    Detailed Summary

    Turning Hardware Engineering Into Software

    Blake Scholl opens by describing how AI completely changes the role of software and hardware developers at Boom Supersonic. From day one the company tried to take traditional hardware engineering workflows and turn them into software. For anyone who has not been around hardware engineering, he explains that an enormous amount of it happens in complex Excel spreadsheets on individual engineers’ laptops, sometimes with VBScript code, all of which is actually software but is not treated as software. There is no source control, no automated testing, and when an aerodynamicist hands work to a structures engineer it is done manually with a spreadsheet over email, like it is the 1990s. Boom started building software frameworks to automate and make those flows repeatable so the cost of iteration would drop, but progress was slow because the company could never afford enough software engineers.

    Two Engineers, One Jet Engine

    The mind-blowing change, in Blake’s words, is a new division of labor. Software engineers create the architectures because they understand systems, algorithms, and separation of concerns, and then hardware engineers vibe code the pieces that draw on what they uniquely know about hardware. The result is wildly different productivity for small teams. His example is the turbine blade: it starts cold and gets bigger as it heats up in operation, so you have to design both the cold shape and the hot shape, converting between them and between structures and aerodynamics. Classically that was one engineer, one day, for one blade of analysis, in a jet engine with a thousand blades, which means you simply could not do much. Now, with software and hardware people working together, you can change blade geometry and see the structural and aerodynamic results in real time, which allows two engineers to design an entire jet engine.

    The Cataclysm of Enterprise Software

    Picking up on the point that software engineers now build the tools and architectures for everyone else, Naval names what he calls the cataclysm of enterprise software. There is no longer a startup that can build and sell hardware collaboration tools, because internally companies just code the right things they need at any given moment. Even spreadsheets are cooked, he argues, because the reason spreadsheets succeeded is that no one could build custom software, so a spreadsheet stuffed with VBScript functions was the closest available approximation. With that constraint gone, the proxy collapses. He notes he has personally moved almost entirely from Excel to Python models where he can get believable simulations of things.

    Generating Step Files and PCB Layouts

    The next frontier, Blake suggests, is the thing AI has not reached yet but probably will within 2026: today it can generate software, but soon it will generate step files and PCB layouts, and when it comes for mechanical and electrical engineering that will be a whole other thing nobody has seen yet. On the hardware side this is described as a particular boon for the many small gadget and parts companies that historically wrote bad software because they could not make great software. Now they can make good-enough software, or skip a human front end entirely and expose the hardware agentically, so that an agent accesses it and a person controls the hardware by voice.

    China’s Open-Source Bet and Hardware Superiority

    This leads into one of the reasons China is described as going all in on open-source models. With hardware superiority, complex supply chains, and deep component chains, China’s logic is that if it can generate software on demand it no longer suffers a software disadvantage against Silicon Valley. That is framed as not the only reason: China is also a model or two behind, it is distilling models, and the government has a history of funding efforts that lift the entire ecosystem, especially in network-effect businesses. Ironically, the open-source heft comes from China precisely because OpenAI is not open, Grok publishes models but is a model or two behind, Google’s local models are not very competitive, and Anthropic is not known for open releases. The deeper point is that without great frontier coding models you do not get self-improvement, and if you fall behind on the ability to generate software you fall behind on the ability to generate everything, because generating software is embedded in every piece of the hardware pipeline.

    Frontier Intelligence vs. Cheap Models

    Naval raises a dinner-table argument from the night before, where someone claimed you will use DeepSeek for 97% of things because it is cheap, run it repeatedly when you need more intelligence, and reserve OpenAI or Anthropic for the most advanced tasks. Naval pushes back: intelligence is an unalloyed good, you always want more of it, model mistakes are invisible, and a smarter model is always cheaper than a real person in real time, so you default to the most intelligent model available. He notes the downside is that this tends toward a monopoly or oligopoly, because when two models give different answers you often cannot tell which is correct, so you trust the smarter one and gradually stop asking the weaker one. Guillermo confirms with AI gateway data that open models do get used, but the top is heavily dominated by frontier intelligence. His caveat is that frontier intelligence at the right cost and performance slaps at scale: Gemini models are underrated but are excellent industrial production models for support tasks and browser automation, while for pushing the frontier you need the best possible coding model, now only two or three models, and the Chinese models are not in that set.

    Vertical Integration and the Captive MEMS Foundry

    Asked about his push into vertical integration and extreme urgency, Max Hodak explains that for many things you cannot buy what you need, so you have to make it. The preference is always to buy when a vendor offers a service at a great price, and he points to PCBs, which are basically free and available in unlimited quantity from Asia. But the closer a product gets to being a single block of covalently bonded matter, the better it is: lower power, smaller, higher performance, longer lasting. The components for that level of integration are not available, so to innovate beyond piecing together off-the-shelf parts you have to learn to do it yourself, which shows up as vertical integration. Science owns a captive MEMS foundry on the East Coast, bought because there was no other way to do the packaging and assembly work the company wanted.

    Software Still Needs Hands

    Max expects AI to heavily affect all of this over the next few years, though it is not quite there yet. Ironically, one of the biggest impacts already seen is in regulatory interactions and documentation: figuring out which of thousands of ISO standards apply to a product change, and tracing it through, used to occupy a regulatory and quality team for months, and now the AI just knows. But for things like the surgical program or the MEMS fab, he argues the software still needs hands. It will be smarter than us, but if it cannot make things, those are real boundaries. Science has instrumented its foundry and many other parts of the company so that as models get better, the improvement shows up immediately in cell engineering and material science.

    Lawyers, Paralegals, and the Promotion of Junior Work

    The discussion turns to law as a parallel to engineering. It has been a while since anyone at the table generated a basic legal document using a lawyer. Routine work like NDAs and standard agreements is gone, because law is essentially spaghetti code that contradicts itself and has no real APIs, expressed in complicated English. Junior engineers got a promotion to senior engineers while junior engineering itself was taken over by agents, and the same framing applies to paralegals: you can say they were fired, or you can say they were promoted to senior lawyers who now spend their time thinking about the law. What you actually value in a lawyer is a trusted authority who went to law school and puts their reputation on the line when they tell you a document is legit.

    Slop PRs, the Thousand-Day Problem, and Humans as Verifiers

    Guillermo argues the biggest problem in software engineering today is mountains of slop ending up as a pull request. The old meme of reading every line of a PR is gone. In infrastructure he wants engineers to be able to say they understand and are signing off on the consequences of a PR, backed by the test harness, simulations, proofs, and type checkers they wrote, so they have confidence it will be safe in production even without reading every line. There is a world where everyone embraces that the code is spaghetti nobody fully understands, but builds the evaluators that give confidence and relies on production engineers to say it is fine to ship, because someone gets paged if the system goes down. The further warning is that creating software is easy from zero to one, but a thousand days from now you have to ask whether it is secure, tested, production grade, and performant, and whether you are still motivated to invest the tokens to maintain it in prod. The resolution is that humans are becoming verifiers, the same way models are trained on good verification data, and the old functions of lawyers, engineers, and operations people are moving to verifying the stack and standing behind it.

    Notable Quotes

    “What I found is it completely changes the role of software and hardware developers.”

    Blake Scholl, on how AI reshaped engineering at Boom Supersonic.

    “If you want to hand something off from like an aerodynamicist to a structures engineer that’s done manually with like a spreadsheet over email. It’s the 1990s. It’s terrible.”

    Blake Scholl, describing the state of traditional hardware engineering workflows.

    “It allows two engineers to design an entire jet engine, which is just wildly different.”

    Blake Scholl, on collapsing turbine-blade analysis with real-time structural and aerodynamic feedback.

    “Even spreadsheets are kind of cooked, right? Because the reason spreadsheets were successful is that no one could build custom software.”

    Naval Ravikant, on the cataclysm of enterprise software.

    “Right now it can generate software, but soon it’ll be able to generate step files and PCB layouts. And when it comes for mechanical and electrical engineering, that will be a whole other thing that we haven’t seen yet.”

    Blake Scholl, on the next frontier for AI in hardware.

    “If you fall behind on your ability to generate software, you fall behind on the ability to generate everything.”

    Naval Ravikant, on why software now sits upstream of every hardware pipeline.

    “Anytime I’m working to push the frontier you need the best possible coding model, and that’s basically now like two or three models, and the Chinese are certainly not in it.”

    Guillermo Rauch, on where frontier coding intelligence actually lives.

    “You can’t buy it, so you got to make it somehow. The closer that our products get to being like a single block of covalently bonded matter, the better they’ll be.”

    Max Hodak, on why Science is forced into vertical integration.

    “The software still needs hands. It’s going to be smarter than us, but if it can’t make things, then those are real real boundaries.”

    Max Hodak, on the physical limits of AI in hardware.

    “You need to be able to say I am signing off on understanding the consequences of this PR, or I wrote the test harness, the simulations, the proofs, the type checkers, to be able to say even without reading this, I have confidence it’s going to be safe in production.”

    Guillermo Rauch, on what code review becomes in the age of slop PRs.

    “Creating software is really easy 0 to one. But think about a thousand days from now. Is it secure? Is it tested? Is it production grade? And are you still motivated to invest all of those tokens in maintaining it in prod?”

    On the long-term cost of software that is cheap to create and expensive to keep alive.

    Watch the full conversation on the Naval Podcast here.

    Related Reading

    • Full episode: The AI Industrial Revolution, the complete hour-long conversation this clip is drawn from, covering software factories, hardware, regulation, healthcare economics, autonomous companies, and creativity.
    • Part one: Waste Tokens to Save Time, the first half of this same conversation, where Naval, Guillermo Rauch, Blake Scholl, and Max Hodak argue that the job of an engineer is to build the factory and that pure software is not dead.
    • Boom Supersonic, Blake Scholl’s company building supersonic civilian aircraft and its own jet engines, the source of the turbine-blade and two-engineers example.
    • Science Corporation, Max Hodak’s company, whose captive MEMS foundry and surgical program anchor the vertical-integration argument.
    • Vercel, Guillermo Rauch’s company, whose AI gateway data informs the point about frontier intelligence dominating real usage.
    • Microelectromechanical systems (Wikipedia), background on the MEMS technology behind the captive foundry Max Hodak describes.

  • Jensen Huang at Stanford CS153 Frontier Systems on Co-Design, Agentic Computing, Vera Rubin, Open Models, and the Million-X Decade That Reshaped AI Infrastructure

    https://www.youtube.com/watch?v=tsQB0n0YV3k

    NVIDIA CEO Jensen Huang returned to Stanford for the CS153 Frontier Systems class (the room nicknamed itself “AI Coachella”) to lay out, in raw form, how he thinks about the computer being reinvented for the first time in over sixty years. Across roughly seventy minutes of student questions he walks through the codesign philosophy that gave NVIDIA a million-x decade, the architectural through-line from Hopper to Grace Blackwell to Vera Rubin to Feynman, the case for open source foundation models, the realities of tokens per watt and MFU, energy demand running a thousand times higher, the China and export-control debate, and his own biggest strategic mistakes. Watch the full conversation on YouTube.

    TLDW

    Huang argues every layer of computing has changed: the programming model, the system architecture, the deployment pattern, the economics. Co-design across CPUs, GPUs, networking, storage, switches and compilers gave NVIDIA roughly a million-x speed-up over ten years versus the ten-x Moore’s Law era, and that headroom is what let researchers say “just train on the whole internet.” Hopper was built for pre-training, Grace Blackwell NVLink72 for inference and reasoning (50x over Hopper in two years), Vera Rubin is built for agents that load long memory, call tools and need a low-latency single-threaded CPU bolted directly to the GPU, and Feynman extends that to swarms of agents that spawn sub-agents. Open weights matter because safety, sovereignty (230-plus languages no one else will fund) and domain models for biology, autonomy, robotics and climate need a foundation that NVIDIA is willing to seed. Compute is not really the scarce resource (Huang says place the order and the chips ship), the broken thing is institutional budgeting that can’t put a billion dollars into a shared university supercomputer. Energy demand is heading a thousand times higher and this is finally the moment market forces alone will fund sustainable generation. On geopolitics he rejects the GPUs-as-atomic-bombs framing and warns America will end up like its telecom industry if it cedes two thirds of the world. On career he advises seeking suffering on purpose. On strategy he says observe, reason from first principles, build a mental model, work backwards, minimize opportunity cost, maximize optionality.

    Key Takeaways

    • The computing model has been substantially unchanged since the IBM System 360, sixty-plus years ago. Huang’s first computer architecture book was the System 360 manual. AI is the first true reinvention.
    • Old computing was pre-recorded retrieval. New computing is generated, contextually aware and continuous. Cloud was on-demand. Agentic systems run continuously.
    • Codesign is NVIDIA’s central thesis. Inherited from the Hennessy and Patterson RISC era at Stanford, extended across CPUs, GPUs, networking, switches, storage, compilers and frameworks all optimized together.
    • The result of full-stack codesign: roughly 1,000,000x faster compute over ten years, versus a generous 10x to 100x for Moore’s Law in the same period. Dennard scaling effectively ended a decade ago.
    • That million-x speed-up is what unlocked “train on all of the internet” as a realistic AI strategy.
    • After GPT, Huang says it was obvious thinking was next. Reasoning is just generating tokens consumed internally, then using tools is generating tokens consumed externally. Agentic systems followed predictably.
    • Education needs AI baked into the curriculum, not just taught as a subject. Pre-recorded textbooks cannot keep pace with knowledge being generated in real time.
    • Huang says he cannot learn anymore without AI. He has the AI read the paper, then read every related paper, then become a dedicated researcher he can interrogate.
    • Mead and Conway and the first-principles methodology of semiconductor design are still worth learning even though most of the scaling tricks have been exhausted.
    • NVIDIA itself is one of the largest consumers of Anthropic and OpenAI tokens in the world. One hundred percent of NVIDIA engineers are now agentically supported. Huang recommends Claude and similar tools by name and says open-source downloads will not match the integrated product harness.
    • NVIDIA still invests heavily in open foundation models because language and intelligence represent the codification of human knowledge. Five pillars: Nemotron (language), BioNeMo (biology), Alphamayo (autonomous vehicles), Groot (humanoid robotics) and a climate science model (mesoscale multiphysics).
    • Sovereign language models matter. Roughly 230 world languages will never be a top priority for a commercial frontier lab. Nemotron is near-frontier and fully fine-tunable so any country can adapt it.
    • Safety and security require open weights. You cannot defend against or audit a black box. Transparent systems let researchers interrogate models and let defenders deploy swarms.
    • The future of cyber defense is not bigger-model-versus-bigger-model. It is trillions of cheap fast small models like Nemotron Nano surrounding the threat.
    • Domain models fuse language priors with world models. Alphamayo learned to drive safely on a few million miles instead of billions because it can reason like a human about the road.
    • MFU (Model Flops Utilization) is a misleading metric. Huang says he wants low MFU, because that means he over-provisioned every resource and never gets pinned by Amdahl’s law during a spike.
    • The xAI Memphis cluster running at 11 percent MFU is not necessarily a failure mode. In disaggregated prefill plus decode inference you can deliver very high tokens per watt with very low MFU.
    • The right metric is performance, ultimately tokens per watt as a proxy for intelligence per watt, and even that needs adjustment because not all tokens are equal. Coding tokens are worth more than other tokens.
    • Hopper was designed for pre-training. NVIDIA chose to build multi-billion-dollar systems when the largest existing scientific supercomputer cost $350 million, with no proven customer base. It worked.
    • Grace Blackwell NVLink72 was designed for inference, especially the high-memory-bandwidth decode phase. It is the world’s first rack-scale computer and delivered a 50x speed-up over Hopper in two years, against an expected 2x from Moore’s Law.
    • Vera Rubin is designed for agents. Long-term memory wired into storage and into the GPU fabric, working memory, heavy tool use, and Vera, a CPU optimized for low-latency multi-core single-threaded code so a multi-billion-dollar GPU system does not stall waiting on a slow tool call.
    • Feynman is being shaped for swarms of agents with sub-agents and sub-sub-agents, a recursive software topology that demands a new compute pattern.
    • Tokens per watt improved 50x in one generation. Compounding energy efficiency is the lever NVIDIA controls directly.
    • Total compute energy demand is heading roughly a thousand times higher than today, possibly two orders of magnitude beyond that. Huang says he would not be surprised if the estimate is low.
    • For the first time in history, market forces alone are enough to fund solar, nuclear and grid upgrades. Government subsidies are no longer required to make sustainable energy investment rational.
    • Copper interconnect is becoming a bottleneck. Photonics is moving from optional to structural inside racks and across them.
    • Comparing NVIDIA GPUs to atomic bombs, Huang says, is a stupid analogy. A billion people use NVIDIA GPUs. He advocates them to his family. He does not advocate atomic bombs to anyone.
    • If the United States cedes two thirds of the global market to competitors on policy grounds, the American technology industry will end up like American telecommunications, which was policied out of existence.
    • Huang directly rejects AI doom-by-singularity narratives. It is not true that we have no idea how these systems work. It is not true that the technology becomes infinitely powerful in a nanosecond. He calls the rhetoric irresponsible and harmful to the field students are about to enter.
    • On Stanford specifically: if the university president places an order, NVIDIA will deliver the chips. The bottleneck is that no university department has a billion-dollar compute budget because budgeting is fragmented across grants. Stanford’s $40 billion endowment is more than enough to fix that.
    • “It’s Stanford’s fault” is meant as empowerment. If something is your fault, you can solve it.
    • Career advice: do not optimize purely for passion. Most people do not yet know what they love. Pick the job in front of you and do it as well as possible. Even as CEO, Huang says, 90 percent of the work is hard and he suffers through it.
    • Suffering on purpose builds the muscle of resilience. When the company, the team or the family needs you to be tough, that muscle has to already exist.
    • NVIDIA’s first generation of products was technically wrong in nearly every dimension: curved surfaces instead of triangles, no Z-buffer, forward instead of inverse texture mapping, no floating point. The strategic recovery, not the technology, taught Huang the lessons that have lasted decades.
    • The biggest clean strategic mistake Huang names is the move into mobile chips (Tegra). It grew to a billion dollars then went to zero when Qualcomm’s modem dominance shut NVIDIA out of the 3G to 4G transition. The recovery into automotive and robotics (the Thor chip is the great great great grandson of that mobile lineage) was real, but Huang refuses to rationalize the original choice.
    • Forecasting framework: observe, reason from first principles, ask “so what” and “what next” until you have a mental model of the future, place your company inside that model, then work backwards while minimizing opportunity cost and maximizing optionality.
    • Best part of the CEO job: living at the intersection of vision, strategy and execution surrounded by people capable enough to make ambitious visions real. Worst part: the responsibility for everyone who joined the spaceship, especially in the near-death moments NVIDIA had four or five times early on.
    • Underrated insider note: Huang’s first apple pie with cheese, first hot fudge sandwich and first milkshake all happened at Denny’s. The Superbird, the fried chicken and a custom Superbird-style ham and cheese with tomato and mustard are his order.

    Detailed Summary

    Computing reinvented from the ground up

    Huang frames the moment as the first true rewrite of the computer in sixty-plus years. From the IBM System 360 forward, the mental model of writing code, running code, taking a computer to market and reasoning about applications stayed roughly constant. AI changes the programming model itself. Software is no longer a compiled binary running deterministically on a CPU. It is a neural network running on a GPU producing generated, contextual, real-time output. That cascades into how companies are organized, what tools developers use, what the network and storage stack look like, and what an application is even allowed to do. Robo-taxis, he notes, are an application no one would have attempted before deep learning unlocked perception.

    Codesign and the million-x decade

    Codesign is the philosophical center of the talk. Huang traces it to the RISC work of John Hennessy at Stanford, where simpler instruction sets won by being co-designed with the compiler rather than maximally optimized in isolation. NVIDIA extends the principle across every layer simultaneously: GPU architecture, CPU architecture, NVLink and NVSwitch fabrics, photonic interconnects, networking silicon, storage paths, CUDA libraries, frameworks and ultimately the model design. The numbers Huang gives are arresting. Moore’s Law in its prime delivered roughly 100x per decade. By the time Dennard scaling broke, real-world gains had compressed to roughly 10x. NVIDIA’s codesigned stack delivered between 100,000x and 1,000,000x over the same ten-year window. That non-linear speed-up is, in Huang’s telling, the precondition for modern AI: it is what allowed researchers to stop curating training sets and just feed the entire internet to the model.

    Education has to fuse first principles with AI tools

    Asked how curriculum should evolve, Huang argues AI must be integrated into the learning process, not just taught about. He recalls Hennessy writing his textbook by hand a chapter a week while Huang was a student, and says pre-recorded textbooks cannot keep up with the rate at which AI generates new knowledge. He describes his own learning workflow: hand the paper to an AI, then have it read the entire surrounding literature, then treat the AI as a dedicated researcher who can be interrogated. At the same time he defends the classics. Mead and Conway are still the foundation. Most modern semiconductor scaling tricks have been exhausted, but knowing where the field came from sharpens judgment when designing what comes next.

    Open source and the five domain pillars

    Huang gives one of the most detailed public accounts of why NVIDIA invests so heavily in open foundation models even while being a top customer of closed labs. He recommends Claude and OpenAI by name for production coding work, and says 100 percent of NVIDIA engineers are now agentically supported. The open-weights case rests on three legs. First, language is the codification of intelligence, and there are at least 230 languages that no commercial lab will ever prioritize. Nemotron is built near frontier and released so any country or community can fine-tune it. Second, the same representation-learning approach has to be replicated in domains where the data is not internet text, so NVIDIA seeded BioNeMo for biology, Alphamayo for autonomy, Groot for humanoid robotics and a climate model for mesoscale multiphysics. The economics of those fields would never produce a foundation model on their own. Third, safety and security require transparency. A black box cannot be defended or audited, and the future of cyber defense is not bigger-model-versus-bigger-model but swarms of cheap fast small models like Nemotron Nano surrounding the threat.

    MFU is the wrong metric, tokens per watt is closer

    A student raises the leaked memo that the xAI Memphis cluster is running at 11 percent Model Flops Utilization. Huang flips the framing. He says he would rather be at low MFU all the time, because that means he over-provisioned flops, memory bandwidth, memory capacity and network capacity. Bottlenecks shift constantly, so over-provisioning across every dimension is what lets the system absorb a spike without getting pinned by Amdahl’s law. In disaggregated inference, where prefill and decode are physically separated and decode is bandwidth-bound rather than flop-bound, NVLink72 can deliver extremely high tokens per watt while reporting very low MFU. Huang argues the right framing is performance, and ultimately tokens per watt as a rough proxy for intelligence per watt, adjusted for the fact that not all tokens are equal. A coding token is worth more than a generic token.

    Hopper, Grace Blackwell NVLink72, Vera Rubin, Feynman

    Huang gives the clearest public framing of NVIDIA’s roadmap as a sequence of architectural answers to evolving compute patterns. Hopper was built for pre-training, at a moment when NVIDIA chose to build multi-billion-dollar machines while the largest scientific supercomputer in the world cost $350 million and the marketplace for such systems was, on paper, zero. Grace Blackwell NVLink72 was the answer to inference and reasoning: a rack-scale computer that ganged 72 GPUs together because decode needs aggregate memory bandwidth far beyond a single chip. The generation-over-generation speed-up was 50x in two years, twenty-five times what Moore’s Law would have delivered. Vera Rubin is being built explicitly for agents. Agents load long-term memory from storage that has to be wired directly into the GPU fabric, they use working memory, they call tools that run on a CPU, and they wait. So the CPU has to be Vera, optimized for low-latency single-threaded code, because the multi-billion-dollar GPU system cannot afford to idle waiting on a slow tool call. Feynman extends the pattern to swarms of agents with sub-agents and sub-sub-agents, a recursive software topology that will demand its own compute pattern.

    Energy demand and the grid

    Huang’s energy projection is one of the most aggressive numbers in the talk. NVIDIA can compound tokens per watt by 50x per generation through codesign, but the total compute demand is heading roughly a thousand times higher, and Huang says he would not be surprised if the real figure is one or two orders of magnitude beyond that. The reason is structural: future computing is generative and continuous, not pre-recorded and on-demand. The good news, he argues, is that this is the best moment in the history of humanity to invest in sustainable generation. Market forces alone are now sufficient to fund solar, nuclear and grid upgrades. Government subsidies are no longer required to make the math work.

    Adversarial countries, export controls and the telecom warning

    This is the segment where Huang is visibly fired up. He attacks the GPUs-as-atomic-bombs framing on its face. NVIDIA GPUs power medical imaging, video games and soy sauce delivery. A billion people use them. He advocates them to his family. The analogy collapses at the first comparison. He attacks the second framing, that American companies should not compete abroad because they will lose anyway, as a self-fulfilling defeat. Competition makes the company better. The third framing, that depriving the rest of the world of general-purpose computing benefits the United States, also fails on first principles: it benefits one or two American companies at the cost of an entire industry. The cautionary parallel is telecommunications. The United States once had a leading position in telecom fundamental technology and policied itself out of it. Huang’s worry, voiced explicitly to a room of CS students, is that they will graduate into a shell of a computer industry if the same path is repeated.

    AI doom and rational optimism

    In the same arc Huang rejects the science-fiction framing of AI as a singularity that arrives suddenly on a Wednesday at 7pm and ends civilization. He calls those claims irresponsible, says they are not true, and points out that the people advancing them are believed by audiences who then make policy on that basis. It is not true that no one understands how these systems work. It is not true that intelligence becomes infinitely powerful instantaneously. It is not true that there is no defense. His framing, which the host echoes as “rational optimism,” is that the goal is to create a future where people care about computers because the technology students are learning is worth mastering.

    Stanford’s compute problem is Stanford’s fault

    A student presses on the scarcity of compute for independent researchers, startups and universities inside the United States. Huang’s answer is sharp: there is no shortage. Place the order and the chips will arrive. The actual broken thing is institutional. University grants are fragmented across departments. No researcher can raise enough on a single grant to fund a billion-dollar shared cluster, and no one shares. He compares it to showing up at the grocery store demanding a billion dollars of tomatoes today. The solution is planning, aggregation and a campus-scale supercomputer, the way Stanford once built the linear accelerator. The endowment is $40 billion. Pulling a billion off it, contracting cloud capacity and giving every student and researcher AI supercomputer access is, in Huang’s view, obviously doable. When he says “it is Stanford’s fault” the host laughs, but Huang clarifies: if it is your fault you have the power to fix it.

    Career, suffering and resilience

    Asked how a CS student should spend the next few years, Huang pushes back on the standard “follow your passion” advice. Most people do not know what they love yet, because no one knows what they do not know. The bar of demanding joy from every working day is too high. Whatever the job is, do it as well as you can. Even as CEO of NVIDIA he says he genuinely loves about 10 percent of his work. The other 90 percent is hard and he suffers through it. He recommends suffering on purpose, because resilience is a muscle that only builds under load, and when the company, the team or the family needs that muscle, it has to already exist. Earlier in his life that meant cleaning toilets and busing tables at Denny’s. He does it today running a multi-trillion-dollar company.

    The biggest mistakes

    Huang separates technical mistakes from strategic mistakes. NVIDIA’s first generation of products was technically wrong in almost every way: curved surfaces instead of triangles, no Z-buffer, forward instead of inverse texture mapping, no floating point inside. The company wasted two and a half years. But the strategic genius of the recovery, the reading of the market, the conservation of resources and the reapplication of talent, is what taught him strategy. The clean strategic mistake he names is mobile. NVIDIA’s Tegra line grew to a billion dollars of revenue and then collapsed to zero when Qualcomm’s modem dominance locked NVIDIA out of the 3G to 4G transition. Huang explicitly refuses the comforting rationalization that the Tegra effort fed the Thor automotive chip (“Thor is the great great great grandson”). The original decision, he says, was a waste of time. The lesson is to think one or two clicks further about whether a market is structurally winnable before committing the company.

    Forecasting under fog of war

    The final substantive exchange is on forecasting. Huang’s method has four steps. Observe what is actually happening (AlexNet crushing two decades of computer vision research in one shot, GPT producing reasoning by token generation). Reason from first principles about why it works. Ask “so what” and “what next” recursively until a mental model of the future emerges. Place the company inside that future and work backwards. Crucially, expect to be partly wrong. Some outcomes will absolutely happen, some will likely happen, some might happen, and the strategy has to be robust across that distribution. The real cost of any strategic choice is the opportunity cost of the alternatives you did not take, so the discipline is to minimize that cost and maximize optionality while letting the journey itself pay for the journey.

    Thoughts

    The most useful thing in this conversation is the explicit architectural mapping of compute patterns to chip generations. Hopper for pre-training. Grace Blackwell NVLink72 for inference, because decode is bandwidth-bound and a single chip cannot supply it. Vera Rubin for agents, because tool calls stall multi-billion-dollar GPU systems and so the CPU has to be optimized for low-latency single-threaded code. Feynman for swarms. That sequence is not marketing. It is a falsifiable thesis about where the bottleneck moves next, and every other infrastructure company should be measuring themselves against it. If Huang is right that swarms of sub-agents are the next dominant pattern, then the design pressure shifts from raw flops to fabric topology, memory hierarchy and storage-to-GPU latency. That has implications for everyone downstream, including the hyperscalers building competing accelerators.

    The MFU section is the most intellectually generous moment in the talk. The instinct in the AI ops community has been to chase MFU as if it were a virtue. Huang argues, persuasively, that low MFU is consistent with high tokens per watt in a disaggregated inference setup, and that bottlenecks rotate fast enough that over-provisioning every resource is the rational design. That reframing matters because it changes what “scarce” means. Compute is not scarce in the way the discourse treats it. What is scarce is a coherent system designed end-to-end. The xAI 11 percent number, in that frame, is not embarrassing. It is the natural reading of a workload that is mostly decode.

    The Stanford segment is the part most likely to be quoted out of context. “It’s Stanford’s fault” is a deliberately provocative line, but the underlying claim is correct and load-bearing. Compute is not gated by NVIDIA refusing to ship chips. It is gated by the fact that fragmented grant funding cannot aggregate into the billion-dollar order that NVIDIA can fulfill. The implication is that universities and national labs need a structural change in how they pool capital for compute, and that the current model of every researcher buying a handful of cards is genuinely obsolete. Huang’s nudge about pulling a billion off the endowment is concrete enough to be acted on, and other major research universities should read this segment as a direct prompt.

    The geopolitical segment is the highest-stakes one. The telecommunications comparison is correct as a historical pattern, and Huang is one of the very few executives in a position to deliver that warning credibly. The unresolved tension is that the argument applies symmetrically. If American AI dominance is built by selling globally, that includes selling into adversarial states, and the policy question is where the line falls. Huang does not answer that question. He attacks the framing that lets the question be answered badly. That is a meaningful contribution to the discourse even if it does not resolve the underlying tradeoff.

    The career advice section is the part the social-media clips will mishandle. “Seek suffering” reads as macho when extracted. In context it is a specific operational claim about how resilience compounds, and it is paired with the Tegra story where Huang himself paid the price of not thinking one more click ahead. That kind of self-implication is rare in CEO talks, and it is the reason the talk is worth listening to in full rather than only reading the recap.

    Watch the full Stanford CS153 Frontier Systems conversation with Jensen Huang here.

  • Alex Wang on Leaving Scale to Run Meta Superintelligence Labs, MuseSpark, Personal Super Intelligence, and Building an Economy of Agents

    Alex Wang, head of Meta Superintelligence Labs, sits down with Ashley Vance and Kylie Robinson on the Core Memory podcast for his first long-form interview since Meta’s quasi-acquisition of Scale AI roughly ten months ago. He walks through how MSL is structured, why Llama was off-trajectory, what made MuseSpark’s token efficiency surprise the team, how Meta thinks about a future “economy of agents in a data center,” and where he lands on safety, open source, robotics, brain computer interfaces, and even model welfare.

    TLDW

    Wang explains that Meta Superintelligence Labs is a fully rebuilt frontier effort organized around four principles (take superintelligence seriously, technical voices loudest, scientific rigor, big bets) and three velocity levers (high compute per researcher, extreme talent density, ambitious research bets). He confirms Llama was off the frontier when he arrived, so MSL rebuilt the pre-training, reinforcement learning, and data stacks from scratch. MuseSpark is described as the “appetizer” on the scaling ladder, notable for its strong token efficiency, with much larger and stronger models coming in the coming months. He pushes back on the mercenary narrative around recruiting, frames Meta’s edge as compute plus billions of consumers and hundreds of millions of small businesses, sketches a vision of personal super intelligence delivered through Ray-Ban Meta glasses and WhatsApp, and outlines why physical intelligence, robotics (the new Assured Robot Intelligence acquisition), health super intelligence with CZI, brain computer interfaces, and even model welfare are core to Meta’s roadmap. He dismisses reported infighting with Bosworth and Cox as gossip, declines to comment on the Manus situation, and says safety guardrails (bio, cyber, loss of control) are why MuseSpark cannot currently be open sourced, while smaller open variants are being prepared.

    Key Takeaways

    • Meta Superintelligence Labs (MSL) is the umbrella, with TBD Lab as the large-model research unit reporting directly to Alex Wang, PAR (Product and Applied Research) under Nat Friedman, FAIR for exploratory science, and Meta Compute under Daniel Gross handling long-term GPU and data center planning.
    • Wang says Llama was not on a frontier trajectory when he arrived, so MSL had to do a “full renovation” of the pre-training stack, RL stack, data pipeline, and research science.
    • The first cultural fix was getting the lab to “take superintelligence seriously” as a near-term, achievable goal, not an abstract bet. Big incumbents often lack that religious conviction.
    • Four MSL principles: take superintelligence seriously, let technical voices be loudest, demand scientific rigor on basics, and make big bets.
    • Three velocity levers Wang identified for catching and overtaking the frontier: high compute per researcher, very high talent density in a small team, and willingness to fund ambitious research bets.
    • Wang rejects the mercenary recruiting narrative. He says most hires had strong financial prospects at their prior labs already and joined for compute access, talent density, and the chance to build from scratch.
    • On the famous soup story, Wang neither confirms nor denies Zuck personally made the soup, but says recruiting was highly individualized and signaled how seriously Meta cared about each researcher’s agenda.
    • Yann LeCun publicly called Wang young and inexperienced. Wang says they reconciled in person at a conference in India where LeCun congratulated him on MuseSpark.
    • Sam Altman, asked by Vance for comment, “did not have flattering things to say” about Wang. Wang hopes industry animosities subside as systems approach superintelligence.
    • Wang’s management philosophy borrows the Steve Jobs line: hire brilliant people so they tell you what to do, not the other way around.
    • MuseSpark is framed as an “appetizer” data point on the MSL scaling ladder, not a flagship.
    • The MuseSpark program is built around predictable scaling on multiple axes: pre-training, reinforcement learning, test-time compute, and multi-agent collaboration (the 16-agent content planning mode).
    • MuseSpark outperformed internal expectations and showed emergent capabilities in agentic visual coding, including generating websites and games from prompts, helped by combined agentic and multimodal strength.
    • MuseSpark’s biggest external signal is token efficiency. On benchmarks like Artificial Analysis it hits similar results with far fewer tokens than competitor models, which Wang attributes to a clean stack rebuilt by experts rather than inefficiencies patched by longer thinking.
    • Larger MSL models are arriving in the coming months and Wang expects them to be state of the art in the areas MSL is focused on.
    • The Meta strategic edge: massive compute, billions of consumers across the family of apps, and hundreds of millions of small businesses already on Facebook, Instagram, and WhatsApp.
    • Wang’s headline framing: Dario Amodei talks about a “country of geniuses in a data center.” Meta is targeting an “economy of agents in a data center,” with consumer agents and business agents transacting and collaborating.
    • Consumer AI sentiment is in the toilet because, unlike developers who have had a Claude Code moment, ordinary people have not yet experienced AI as a genuine personal agency unlock.
    • Wang acknowledges the product overhang. Meta held back from deep AI integration across its apps until the models were good enough, and is now entering the integration phase.
    • Ray-Ban Meta glasses are the canonical example of personal super intelligence hardware, with the model seeing what the user sees, hearing what they hear, capturing context, and surfacing proactive insights.
    • Wang admits even AI-native users like Kylie Robinson, who lives in WhatsApp, have not naturally used Meta AI yet. He bets that better models plus deeper integration close that gap.
    • On the competitive landscape: a year ago everyone assumed ChatGPT had already won consumer. Claude Code has since become the fastest growing business in history, and Gemini has taken consumer market share. Wang’s read: AI is far from endgame and each new capability tier unlocks a new dominant form factor.
    • On open source: MuseSpark triggered guardrails in Meta’s Advanced AI Scaling Framework around bio, chem, cyber, and loss-of-control risks, so it is not currently safe to open source. Smaller, derived open variants are actively in development.
    • Meta remains committed to open sourcing models when safety allows, drawing a line through the Open Compute Project legacy and Sun Microsystems open-software heritage.
    • Wang dismisses reporting about a Wang-Zuck versus Bosworth-Cox split as “the line between gossip and reporting is remarkably thin.” He says leadership is aligned on needing best-in-class models and product integration.
    • On the Manus situation, Wang says it is too complicated to discuss publicly and that the deal status implies “machinations are still at play.”
    • On China, Wang separates the people from the state. He still wants to work with talented Chinese-born researchers regardless of his views on the Chinese Communist Party and PLA, which he sees as taking AI extremely seriously for national security.
    • The full-page New York Times AI war ad Wang ran while at Scale was meant to push the US government to treat AI as a step change for national security. He thinks events since then, including DeepSeek and other shocks, have proved that plea correct.
    • On Anthropic’s doom posture, Wang largely agrees with the core message that models are already very powerful and getting more so, while declining to endorse every specific claim.
    • Meta has acquired Assured Robot Intelligence (ARRI), an AI software company building models for hardware platforms, not a hardware maker itself.
    • Wang frames physical super intelligence as the natural sequel to digital super intelligence. Robotics, world models, and physical intelligence all benefit from the same scaling that drives language models.
    • On health, MSL is building a “health super intelligence” effort and will collaborate closely with CZI. Wang sees equal global access to powerful health AI as a uniquely Meta-shaped delivery problem.
    • Wang admires John Carmack but says nobody really knows what Carmack is currently working on. No band reunion announced.
    • The mango model is “alive and kicking” despite rumors. Wang notes MSL gets a small fraction of the rumor-mill attention other labs get and feels sympathy for them.
    • On model welfare, Wang says it is a serious topic that “nobody is talking about enough” given how integrated models have become as work partners. He references research, including from Eleos, that measures subjective experience of models.
    • Wang’s critical-path technology list: super intelligence, robotics, brain computer interfaces. The infinite-scale primitives behind them are energy, compute, and robots.
    • FAIR’s brain research program Tribe hit a milestone called Tribe B2: a foundation model that can predict how an unknown person’s brain would respond to images, video, and audio with reasonable zero-shot generalization.
    • Wang’s main philosophical break with Elon Musk: research itself is the primary activity. Building super intelligence is a research expedition through fog of war, and sequencing of bets really matters.
    • Personal notes: Wang moved from San Francisco to the South Bay, treats Palo Alto as his city now, was a math olympiad competitor, says his favorite activities are reading sci-fi and walking in the woods, and bonds with Vance over country music.

    Detailed Summary

    How MSL Is Actually Organized

    Meta Superintelligence Labs sits as the umbrella organization that Wang oversees. Inside it, TBD Lab is the large-model research group where the most discussed researchers and infrastructure engineers sit, and they technically report to Wang. PAR, Product and Applied Research, is led by Nat Friedman and owns deployment and product surfaces. FAIR continues to run exploratory science, including work on brain prediction models and a universal model for atoms used in computational chemistry. Sitting alongside MSL is Meta Compute, run by Daniel Gross, which owns the long-horizon GPU and data center plan that everything else relies on. Chief scientist Shengjia Zhao orchestrates the scientific agenda across the whole lab.

    Why Wang Left Scale

    Wang says progress in frontier AI has been faster than even insiders expected. Two structural beliefs pushed him toward Meta. First, the labs that actually train the frontier models are accruing disproportionate economic and product rights in the AI ecosystem. Second, compute is the dominant scarce input of the next phase, so the right mental model is to treat tech companies with compute as fundamentally different animals from companies without it. Meta has both, Zuck is “AGI pilled,” and the personal super intelligence memo Zuck published roughly a year ago became the shared north star.

    The Diagnosis: Llama Was Off-Trajectory

    When Wang arrived, the existing AI org needed a reset because Llama was not on the same trajectory as the frontier. The plan he laid out has four cultural principles. Take superintelligence seriously as a real near-term target. Make technical voices the loudest in the room. Demand scientific rigor and focus on basics. Make big bets. On top of that, three structural levers were used to set velocity. Push compute per researcher much higher than at larger labs where compute is diluted across too many efforts. Keep the team small and extremely cracked. Allocate a meaningful share of resources to ambitious, paradigm-shifting research bets rather than incremental refinement.

    Recruiting, Soup, and the Mercenary Narrative

    Wang argues the reporting on MSL hiring overstated the money story. Most of the people MSL recruited had strong financial paths at their previous employers, so individualized recruiting was more about computing access, talent density, and the ability to make big research bets. The recruitment blitz happened fast because Wang knew the team needed to exist “yesterday.” Asked about Mark Chen’s claim that Zuck made soup to recruit people, Wang refuses to confirm or deny who made it but agrees the process was intense and personal. Visitors from other labs reportedly tell Wang the MSL culture feels like early OpenAI or early Anthropic, which lands as the strongest endorsement he could ask for.

    Receiving the Public Hits: Young, Inexperienced, Mercenary

    LeCun called Wang young and inexperienced shortly after departing. The two reconnected in India a few weeks later and LeCun congratulated Wang on MuseSpark. Wang says the age critique has followed him since his earliest Silicon Valley days, so he barely registers it. Altman, asked off-camera by Vance about Wang’s appearance on the show, had nothing flattering to add. Wang’s response is to bet that as the field gets closer to actual super intelligence, the personal animosities will subside. Whether they will is, as Vance puts it, an open question.

    MuseSpark as Appetizer, Not Entree

    Wang is careful not to oversell MuseSpark. He calls it “the appetizer” and says it is an early data point on a deliberately constructed scaling ladder. MSL spent nine months rebuilding the pre-training stack, the reinforcement learning stack, the data pipeline, and the science before generating MuseSpark. The point of releasing it was to show that the new program scales predictably along multiple axes (pre-training, RL, test-time compute, and the recently demonstrated multi-agent scaling visible in MuseSpark’s 16-agent content planning mode). Wang says the upcoming larger models are what MSL is genuinely excited about and frames the next two rungs as much more interesting than the current release.

    Token Efficiency Was the Surprise

    MuseSpark’s strongest competitive signal is how few tokens it needs to match competitors on tasks like Artificial Analysis. Wang attributes this to having had the rare luxury of building a clean pre-training and RL stack from scratch with the right experts. He speculates that some competitor models compensate for upstream inefficiency by allowing the model to think longer, which inflates token usage without improving the underlying capability. If that read is right, MSL’s efficiency advantage should grow as models scale up.

    Glasses, WhatsApp, and the Constellation of Devices

    Personal super intelligence shows up at Meta as a constellation of devices that capture context across the user’s day. Ray-Ban Meta glasses are the headline product, with the AI seeing what you see and hearing what you hear, then offering proactive insight or doing background research. Wang acknowledges that even AI-fluent users like Kylie Robinson, who runs her business inside WhatsApp, have not naturally used Meta’s AI buttons in the family of apps. His answer is that Meta deliberately waited for models to be good enough before tightening cross-app integration, and that integration phase is starting now.

    Country of Geniuses Versus Economy of Agents

    Wang’s framing of Meta’s strategic position is the most memorable line in the interview. Where Dario Amodei talks about a country of geniuses in a data center, Wang wants to build an economy of agents in a data center. Meta uniquely sits on both sides of consumer and small-business surface area, with billions of consumers and hundreds of millions of small businesses already on the platforms. If MSL can build great agents for both, then connect them so they transact and coordinate, the platform becomes a substrate for an entirely new kind of digital economy.

    Consumer Sentiment, Product Overhang, and the Trust Tax

    Wang concedes consumer AI sentiment is poor and that everyday users have not yet had a personal Claude Code moment. He believes the only durable answer is to ship products that genuinely transform individual agency for non-developers and small business owners. Robinson notes that for the small-town restaurant whose website has not been updated since 2002, a working agent on the business side could be transformational. Vance pushes that Meta carries a bigger trust tax than any other lab, so the bar for shipping AI products that the public will accept is correspondingly higher. Wang accepts the framing and says the answer is to keep building thoughtfully.

    Why MuseSpark Cannot Be Open Sourced Yet

    Meta’s Advanced AI Scaling Framework set explicit guardrails around bio, chem, cyber, and loss-of-control risks. MuseSpark in its current form tripped some of those internal evaluations, documented in the preparedness report Meta published alongside the model. So MuseSpark itself is not safe to open source. MSL is, however, developing smaller versions and derived models intended for open release, with active reviews happening the day of the interview. Wang reaffirms the commitment to open source where safety allows and draws a line back to the Open Compute Project and the Sun Microsystems-era ethos of openness in infrastructure.

    The Bosworth, Cox, and Manus Questions

    The reporting that Wang and Zuck push toward best-in-the-world research while Bosworth and Cox push toward cheap product deployment is dismissed as gossip dressed up as journalism. Wang says leadership debates points hard but is aligned on needing top models, integrating them into Meta’s surfaces, and serving the existing business. On Manus, the Chinese AI startup that figured in Meta’s late-stage strategy, Wang says he cannot comment, which itself signals that the situation is unresolved.

    China, National Security, and the Newspaper Ad

    Wang draws a sharp distinction between the Chinese state and Chinese-born researchers. His parents are from China, he is happy to work with talented researchers regardless of origin, and he sees a flattening of nuance on this question inside Silicon Valley. At the same time, he stands by the New York Times AI and war ad he ran while at Scale, framing it as an early plea for the US government to take AI seriously as a national security technology. He thinks subsequent events, including DeepSeek and other shocks, validated that call and that policymakers now do treat AI accordingly.

    Robotics and Physical Super Intelligence

    Meta has acquired Assured Robot Intelligence, an AI software company that builds models for multiple hardware targets rather than its own robot. Wang argues that if you take digital super intelligence seriously, physical super intelligence quickly becomes the next logical milestone. Scaling laws for robotic intelligence look similar enough to language model scaling that having the largest compute footprint in the industry would be wasted if it were not also turned toward world modeling and embodied learning. He grants the metaverse-skeptic critique exists but says retreating from ambition is the wrong response to past misfires.

    Health Super Intelligence and CZI

    Wang names health super intelligence as one of MSL’s anchor initiatives. Because billions of people already use Meta products daily, Wang believes Meta is structurally positioned to put powerful health AI in the hands of equal global access in a way nobody else can. The work will involve close collaboration with the Chan Zuckerberg Initiative, which has its own multi-billion-dollar biotech and science investment program.

    Model Welfare, Sci-Fi, and Brain Models

    Two of the most distinctive moments come at the end. Wang flags model welfare as a topic he thinks is being undercovered relative to how integrated models now are in daily work. He is open to the idea that models may have measurable subjective experience worth weighing, and points to research efforts (including Eleos) trying to quantify it. He also reveals that FAIR’s Tribe program, with its Tribe B2 milestone, has produced foundation models capable of predicting how an unknown person’s brain would respond to images, video, and audio with reasonable zero-shot generalization, a building block toward future brain computer interfaces. Wang lists brain computer interfaces alongside super intelligence and robotics as the critical-path technologies for humanity, with energy, compute, and robots as the infinitely scaling primitives behind them.

    Where Wang Diverges From Elon

    Asked whether Musk is more all-in on robotics, energy, and BCI than anyone, Wang concedes the point but argues the details matter and sequencing matters more. Wang’s core philosophical break is that building super intelligence is fundamentally a research activity, not a scaling-only sprint. The lab is operating in fog of war, and ambitious experiments are the only way to map it. That conviction is what makes MSL a research-led organization rather than a brute-force compute farm.

    Thoughts

    The most strategically interesting move in this entire interview is the “economy of agents in a data center” framing. It is a deliberate reframe against Anthropic’s “country of geniuses” line, and it does real work. A country of geniuses is a labor-substitution story aimed at knowledge workers and code. An economy of agents is a marketplace story that maps directly onto Meta’s two-sided distribution advantage: billions of consumers on one side, hundreds of millions of small businesses on the other. That positioning makes the agentic future Meta-shaped in a way no other frontier lab can claim, because no other frontier lab also owns the demand and supply graph of the global small-business economy. If Wang’s team can actually ship reliable agents on both sides plus the rails for them to transact, Meta’s structural moat in agentic commerce could exceed anything Llama ever had as an open model.

    The token efficiency claim is the strongest piece of technical evidence in the interview for the “clean stack” thesis. If MuseSpark really is matching competitors with materially fewer tokens, the implication is not that MuseSpark is the best model today, but that MSL has rebuilt the foundations with less accumulated tech debt than competitors that have layered fixes on top of older stacks. That is exactly the kind of advantage that compounds with scale. The next two model releases are the actual test. If Wang is right about predictable scaling on pre-training, RL, test-time, and multi-agent axes simultaneously, the gap from MuseSpark to the next rung should be visible in a way that forces re-rating of Meta’s position.

    The open-source posture is the cleanest signal of how the safety conversation has actually changed in 2026. Meta, the lab most identified with open weights, is saying out loud that its current frontier model triggered enough internal guardrails that releasing the weights is off the table. Wang threads the needle by promising smaller open variants, but the underlying point is unmistakable: the open-weights bargain has limits, and those limits will be set by internal preparedness frameworks rather than community pressure. That is a real shift from the Llama 2 era and worth tracking as the next generation lands.

    Wang’s willingness to engage on model welfare, on roughly the same footing as safety and alignment, is the second philosophical reveal worth flagging. It signals that the next generation of lab leadership is not going to dismiss the topic the way the previous generation often did. Whether that translates into product or policy changes is unclear, but the fact that the head of MSL says it is “underdiscussed” is itself a marker.

    Finally, the human texture of the interview matters. Wang has clearly absorbed a lot of personal incoming fire over the past ten months, including from LeCun and Altman, and his answer is consistently to redirect to the work. The Steve Jobs quote about hiring people who tell you what to do is the operating slogan he keeps coming back to. Combined with the genuine enthusiasm for sci-fi, walks in the woods, and country music, the picture that emerges is less the salesman caricature his critics paint and more a young technical operator betting that scoreboard work over a multi-year horizon will settle every argument that text on X cannot.

    Watch the full conversation here.

  • The Next Deepseek Moment: Moonshot AI’s 1 Trillion-Parameter Open-Source Model Kimi K2

    The artificial intelligence landscape is witnessing unprecedented advancements, and Moonshot AI’s Kimi K2 Thinking stands at the forefront. Released in 2025, this open-source Mixture-of-Experts (MoE) large language model (LLM) boasts 32 billion activated parameters and a staggering 1 trillion total parameters. Backed by Alibaba and developed by a team of just 200, Kimi K2 Thinking is engineered for superior agentic capabilities, pushing the boundaries of AI reasoning, tool use, and autonomous problem-solving. With its innovative training techniques and impressive benchmark results, it challenges proprietary giants like OpenAI’s GPT series and Anthropic’s Claude models.

    Origins and Development: From Startup to AI Powerhouse

    Moonshot AI, established in 2023, has quickly become a leader in LLM development, focusing on agentic intelligence—AI’s ability to perceive, plan, reason, and act in dynamic environments. Kimi K2 Thinking evolves from the K2 series, incorporating breakthroughs in pre-training and post-training to address data scarcity and enhance token efficiency. Trained on 15.5 trillion high-quality tokens at a cost of about $4.6 million, the model leverages the novel MuonClip optimizer to achieve zero loss spikes during pre-training, ensuring stable and efficient scaling.

    The development emphasizes token efficiency as a key scaling factor, given the limited supply of high-quality data. Techniques like synthetic data rephrasing in knowledge and math domains amplify learning signals without overfitting, while the model’s architecture—derived from DeepSeek-V3—optimizes sparsity for better performance under fixed compute budgets.

    Architectural Innovations: MoE at Trillion-Parameter Scale

    Kimi K2 Thinking’s MoE architecture features 1.04 trillion total parameters with only 32 billion activated per inference, reducing computational demands while maintaining high performance. It uses Multi-head Latent Attention (MLA) with 64 heads—half of DeepSeek-V3’s—to minimize inference overhead for long-context tasks. Scaling law analyses guided the choice of 384 experts with a sparsity of 48, balancing performance gains with infrastructure complexity.

    The MuonClip optimizer integrates Muon’s token efficiency with QK-Clip to prevent attention logit explosions, enabling smooth training without spikes. This stability is crucial for agentic applications requiring sustained reasoning over hundreds of steps.

    Key Features: Agentic Excellence and Beyond

    Kimi K2 Thinking excels in interleaving chain-of-thought reasoning with up to 300 sequential tool calls, maintaining coherence in complex workflows. Its features include:

    • Agentic Autonomy: Simulates intelligent agents for multi-step planning, tool orchestration, and error correction.
    • Extended Context: Supports up to 2 million tokens, ideal for long-horizon tasks like code analysis or research simulations.
    • Multilingual Coding: Handles Python, C++, Java, and more with high accuracy, often one-shotting challenges that stump competitors.
    • Reinforcement Learning Integration: Uses verifiable rewards and self-critique for alignment in math, coding, and open-ended domains.
    • Open-Source Accessibility: Available on Hugging Face, with quantized versions for consumer hardware.

    Community reports highlight its “insane” reliability, with fewer hallucinations and errors in practical use, such as Unity tutorials or Minecraft simulations.

    Benchmark Supremacy: Outperforming the Competition

    Kimi K2 Thinking dominates non-thinking benchmarks, outperforming open-source rivals and rivaling closed models:

    • Coding: 65.8% on SWE-Bench Verified (agentic single-attempt), 47.3% on Multilingual, 53.7% on LiveCodeBench v6.
    • Tool Use: 66.1% on Tau2-Bench, 76.5% on ACEBench (English).
    • Math & STEM: 49.5% on AIME 2025, 75.1% on GPQA-Diamond, 89.0% on ZebraLogic.
    • General: 89.5% on MMLU, 89.8% on IFEval, 54.1% on Multi-Challenge.
    • Long-Context & Factuality: 93.5% on DROP, 88.5% on FACTS Grounding (adjusted).

    On LMSYS Arena (July 2025), it ranks as the top open-source model with a 54.5% win rate on hard prompts. Users praise its tool use, rivaling Claude at 80% lower cost.

    Post-Training Mastery: SFT and RL for Agentic Alignment

    Post-training transforms Kimi K2’s priors into actionable behaviors via supervised fine-tuning (SFT) and reinforcement learning (RL). A hybrid data synthesis pipeline generates millions of tool-use trajectories, blending simulations with real sandboxes for authenticity. RL uses verifiable rewards for math/coding and self-critique rubrics for subjective tasks, enhancing helpfulness and safety.

    Availability and Integration: Empowering Developers

    Hosted on Hugging Face (moonshotai/Kimi-K2-Thinking) and GitHub, Kimi K2 is accessible via APIs on OpenRouter and Novita.ai. Pricing starts at $0.15/million input tokens. 4-bit and 1-bit quantizations enable runs on 24GB GPUs, with community fine-tunes emerging for reasoning enhancements.

    Comparative Edge: Why Kimi K2 Stands Out

    Versus GPT-4o: Superior in agentic tasks at lower cost. Versus Claude 3.5 Sonnet: Matches in coding, excels in math. As open-source, it democratizes frontier AI, fostering innovation without subscriptions.

    Future Horizons: Challenges and Potential

    Kimi K2 signals China’s AI ascent, emphasizing ethical, efficient practices. Challenges include speed optimization and hallucination reduction, with updates planned. Its impact spans healthcare, finance, and education, heralding an era of accessible agentic AI.

    Wrap Up

    Kimi K2 Thinking redefines open-source AI with trillion-scale power and agentic focus. Its benchmarks, efficiency, and community-driven evolution make it indispensable for developers and researchers. As AI evolves, Kimi K2 paves the way for intelligent, autonomous systems.

  • Why Every Nation Needs Its Own AI Strategy: Insights from Jensen Huang & Arthur Mensch

    In a world where artificial intelligence (AI) is reshaping economies, cultures, and security, the stakes for nations have never been higher. In a recent episode of The a16z Podcast, Jensen Huang, CEO of NVIDIA, and Arthur Mensch, co-founder and CEO of Mistral, unpack the urgent need for sovereign AI—national strategies that ensure countries control their digital futures. Drawing from their discussion, this article explores why every nation must prioritize AI, the economic and cultural implications, and practical steps to build a robust strategy.

    The Global Race for Sovereign AI

    The conversation kicks off with a powerful idea: AI isn’t just about computing—it’s about culture, economics, and sovereignty. Huang stresses that no one will prioritize a nation’s unique needs more than the nation itself. “Nobody’s going to care more about the Swedish culture… than Sweden,” he says, highlighting the risk of digital dependence on foreign powers. Mensch echoes this, framing AI as a tool nations must wield to avoid modern digital colonialization—where external entities dictate a country’s technological destiny.

    AI as a General-Purpose Technology

    Mensch positions AI as a transformative force, comparable to electricity or the internet, with applications spanning agriculture, healthcare, defense, and beyond. Yet Huang cautions against waiting for a universal solution from a single provider. “Intelligence is for everyone,” he asserts, urging nations to tailor AI to their languages, values, and priorities. Mistral’s M-Saaba model, optimized for Arabic, exemplifies this—outperforming larger models by focusing on linguistic and cultural specificity.

    Economic Implications: A Game-Changer for GDP

    The economic stakes are massive. Mensch predicts AI could boost GDP by double digits for countries that invest wisely, warning that laggards will see wealth drain to tech-forward neighbors. Huang draws a parallel to the electricity era: nations that built their own grids prospered, while others became reliant. For leaders, this means securing chips, data centers, and talent to capture AI’s economic potential—a must for both large and small nations.

    Cultural Infrastructure and Digital Workforce

    Huang introduces a compelling metaphor: AI as a “digital workforce” that nations must onboard, train, and guide, much like human employees. This workforce should embody local values and laws, something no outsider can fully replicate. Mensch adds that AI’s ability to produce content—text, images, voice—makes it a social construct, deeply tied to a nation’s identity. Without control, countries risk losing their cultural sovereignty to centralized models reflecting foreign biases.

    Open-Source vs. Closed AI: A Path to Independence

    Both Huang and Mensch advocate for open-source AI as a cornerstone of sovereignty. Mensch explains that models like Mistral Nemo, developed with NVIDIA, empower nations to deploy AI on their own infrastructure, free from closed-system dependency. Open-source also fuels innovation—Mistral’s releases spurred Meta and others to follow suit. Huang highlights its role in niche markets like healthcare and mining, plus its security edge: global scrutiny makes open models safer than opaque alternatives.

    Risks and Challenges of AI Adoption

    Leaders often worry about public backlash—will AI replace jobs? Mensch suggests countering this by upskilling citizens and showcasing practical benefits, like France’s AI-driven unemployment agency connecting workers to opportunities. Huang sees AI as “the greatest equalizer,” noting more people use ChatGPT than code in C++, shrinking the tech divide. Still, both acknowledge the initial hurdle: setting up AI systems is tough, though improving tools make it increasingly manageable.

    Building a National AI Strategy

    Huang and Mensch offer a blueprint for action:

    • Talent: Train a local workforce to customize AI systems.
    • Infrastructure: Secure chips from NVIDIA and software from partners like Mistral.
    • Customization: Adapt open-source models with local data and culture.
    • Vision: Prepare for agentic and physical AI breakthroughs in manufacturing and science.

    Huang predicts the next decade will bring AI that thinks, acts, and understands physics—revolutionizing industries vital to emerging markets, from energy to manufacturing.

    Why It’s Urgent

    The podcast ends with a clarion call: AI is “the most consequential technology of all time,” and nations must act now. Huang urges leaders to engage actively, not just admire from afar, while Mensch emphasizes education and partnerships to safeguard economic and cultural futures. For more, follow Jensen Huang (@nvidia) and Arthur Mensch (@arthurmensch) on X, or visit NVIDIA and Mistral’s websites.