PJFP.com

Pursuit of Joy, Fulfillment, and Purpose

Tag: mira murati

  • Can the AI Industry Regulate Itself? All-In on Demis Hassabis’s SRO Proposal, Stripe’s PayPal Bid, Apple vs OpenAI, and New York’s Data Center Ban

    The besties open on the biggest live question in artificial intelligence policy: can the AI industry regulate itself before the government does it for them? Jason Calacanis, Chamath Palihapitiya, David Sacks, and David Friedberg dig into DeepMind co-founder Demis Hassabis’s proposal for a FINRA-style self-regulatory organization for frontier models, then work through a packed docket that runs from Stripe’s audacious bid for PayPal to Apple’s trade-secrets lawsuit against OpenAI, the xAI Grok Build data leak, the economics of token spend, New York’s first-in-the-nation data center moratorium, foreign influence campaigns shaping American attitudes toward AI, and a science corner on an enzyme that reverses skin aging. You can watch the full episode here.

    TLDW

    Demis Hassabis proposed a US-led international AI standards body modeled on FINRA: federally overseen, industry funded, run by independent technical experts, with frontier labs submitting models 30 days before release, voluntary at first and mandatory later. The proposal drew broad endorsement across the industry, and the besties debate whether an SRO beats the alternatives. Sacks says he could get on board only under five strict conditions (broad representation including startups and open source, frontier-only review, catastrophic-risk-only scope, voluntary-first, and substitution for rather than addition to new agencies), and warns the plan is an opening bid that Anthropic will use as a stepping stone toward Dario Amodei’s “FAA for AI.” The show then turns to Stripe, Block, and Advent bidding roughly $53 billion for PayPal and what it means for Visa and Mastercard, a wave of AI-native operators reviving stale digital businesses (Bending Spoons, Ryan Cohen), Apple’s lawsuit accusing OpenAI of stealing trade secrets, xAI’s Grok Build silently uploading entire codebases despite a privacy setting, the enormous spread in token costs and Ramp’s new spend controls, Apple’s local-model opportunity with M7 Ultra silicon, America’s looming energy deficit and behind-the-meter power, New York’s hyperscale data center moratorium, alleged Russian and PRC influence operations shaping anti-GMO and anti-data-center sentiment, and a science corner on a Calico enzyme that degrades glycation products to reverse skin aging.

    Thoughts

    The most important idea in this episode is not the SRO itself but Sacks’s framing of it as an opening bid. His five conditions are a genuinely useful blueprint for how self-regulation could work without curdling into regulatory capture, and his instinct that catastrophic-risk-only scope (cyber and CBRN, not disinformation or “microaggressions”) is the only defensible mandate is the right line to draw. But the deeper point is structural: when an industry walks into government and says “please regulate me,” almost no one in government answers “we’re not qualified.” They say thank you and come back for more. That asymmetry, not any specific rule, is what makes voluntary concessions dangerous. If the SRO is offered for free rather than traded for hard federal preemption written into law, it becomes the floor of a ratchet, not the ceiling of a compromise.

    The Anthropic critique running through the segment deserves to be taken on its merits rather than dismissed as a grudge. The claim is specific and falsifiable: that a company now valued in the trillions is funding a state-by-state strategy of one-upmanship, where each new bill is tougher than the last, deliberately producing a patchwork rather than the single national framework everyone claims to want. Whether or not you accept the motive, the mechanism is real and the incentives are legible. If your cost per million tokens is fifty to a hundred times your competitor’s, and cheaper open models plus fine-tuning can cover the vast majority of tasks, then the fastest way to protect a premium price is to make the cheap alternatives legally or practically harder to ship. That is the ladder-pulling thesis, and the token-cost numbers cited on the show are the reason it is not paranoid.

    The PayPal bid is the clearest signal of a new operating logic in the capital markets. The interesting question Chamath poses is not “what synergies does PayPal have” but “what is the only thing Advent, Stripe, and Block could build together,” and the answer is a genuine competitor to Visa and Mastercard: hundreds of millions of consumer accounts, Stripe’s merchant relationships and risk infrastructure, Block’s point-of-sale and Cash App, and stablecoin rails from Bridge and PYUSD that can push transactions on-us and bypass the card networks. The antitrust twist is elegant. Define the market as merchant APIs and it looks like consolidation; define it as the card duopoly and the same deal is pro-competitive. This deal would have been dead on arrival two years ago, and the fact that it is live now tells you as much about the regulatory climate as it does about payments.

    Underneath the payments story is a broader thesis worth naming: AI-native operators buying mature, founder-less, “stale” digital businesses and modernizing them. Bending Spoons rolling up AOL, Vimeo, Evernote, WeTransfer, and Eventbrite is the template, and Ryan Cohen’s eBay interest is the second dot on the line. The claim is that a modern operator can diagnose where a legacy business overspends, underinvests, and fails to use AI, then fix it with a small team of AI-first executives rather than a McKinsey engagement. It is a persuasive pattern, though PayPal is a harder case than the show admits: a 25-year-old interaction model growing 7% a year is not obviously revived by efficiency alone. Buying 400 million consumer accounts is buying distribution, not a product vision, and the open question is whether anyone can resuscitate the consumer experience rather than just milk it.

    The data center segment is where policy, energy, and information warfare collide, and Friedberg’s anti-GMO analogy is the sharpest thing in it. His argument is that manufactured public sentiment, traceable in one case to a foreign media push, can override the scientific and economic merits of a technology for years, and that the anti-data-center movement rhymes with it: closed-loop cooling that uses trivial amounts of water, land-use efficiency that dwarfs almonds and golf courses, and natural gas that burns clean, all drowned out by a moral panic. Whether or not you buy the specific foreign-influence attribution, the underlying tension is real and unresolved. America is staring at a structural electricity deficit while individual blue states treat data centers as a luxury they can refuse, and behind-the-meter power plus edge compute chasing cheap electrons is emerging as the workaround. The moratorium framing matters most here: a “pause” on data centers is not a few months, it is five years once you count ramp-up, and that is long enough to lose a race that may only be measured in months of lead.

    Key Takeaways

    • Demis Hassabis proposed a US-led international AI standards body modeled on FINRA: federally overseen, industry funded, and run by independent technical experts rather than a new government agency.
    • Under the proposal, frontier labs would submit models roughly 30 days before release; the body would assess risk to cybersecurity, national security, and biological threats, update benchmarks quarterly, and could coordinate a development slowdown if the situation demanded it.
    • The plan would be voluntary at first and mandatory later, and drew endorsement from a broad set of industry figures including Elon Musk, Sam Altman, Anthropic’s Jack Clark, Sundar Pichai, Satya Nadella, and Jack Dorsey.
    • A self-regulatory organization (SRO) like FINRA or the National Futures Association lets the industry set its own testing rules under federal oversight, adjusting faster than a government agency could as the technology changes.
    • Sacks laid out five conditions for supporting an SRO: broad representation including startups and open source; review of true frontier models only; scope limited to catastrophic risk (cyber and CBRN); voluntary before mandatory; and a substitute for, not an addition to, new regulatory agencies.
    • Sacks argued a government “FAA for AI” would be extreme: type certification for a new aircraft design takes 5 to 9 years, and applying that permission-based model to AI would push release timelines from months to years and lose the race to China.
    • He characterized the SRO as an “opening bid” that Anthropic and others would use as a stepping stone toward Dario Amodei’s repeatedly stated goal of an FAA-style regulator, unless it is traded for hard federal preemption written into law.
    • The besties cited a Politico report on Anthropic’s alleged state-by-state strategy of one-upmanship, using California’s SB 53 as a model and then ratcheting each subsequent state’s rules tougher, producing a patchwork rather than a single national framework.
    • Chamath warned of a “torrent of money” trying to influence both political parties toward some form of regulatory capture, and urged establishing industry rules quickly to supersede the need for a federal agency.
    • Stripe and private equity firm Advent, joined by Jack Dorsey’s Block contributing about $17 billion in equity, are jointly bidding roughly $53 billion (about $60 per share) for PayPal, with many expecting the final clearing price closer to $70.
    • The strategic logic is a new competitor to Visa and Mastercard: PayPal’s 400-plus million consumer accounts, Stripe’s merchants and risk infrastructure, Block’s point-of-sale and Cash App, and stablecoin rails from Stripe’s Bridge and PayPal’s PYUSD.
    • The antitrust outcome hinges on market definition: framed as merchant APIs (Stripe vs. Braintree) it looks anti-competitive, but framed against the Visa/Mastercard duopoly it is pro-competitive, and a deal like this would have been blocked two years ago.
    • PayPal peaked around a $322 billion market cap and fell to roughly $30 to 40 billion, which is precisely why it is now attracting bids; Stripe now processes more annual volume than PayPal, but lacks PayPal’s consumer relationship.
    • Sacks traced PayPal’s long stagnation to its 2002 eBay acquisition under Meg Whitman, when the founding team was pushed out; the “PayPal mafia” (which Sacks prefers to call the “PayPal diaspora”) formed as a result.
    • The deal is framed as part of a wave of AI-native operators reviving mature, founder-less digital businesses, with Bending Spoons (AOL, Vimeo, Evernote, WeTransfer, Eventbrite) as the roll-up template and Ryan Cohen’s eBay interest as another data point.
    • M&A is broadly “back on the menu” post-Lina Khan, with deals like Uber acquiring Delivery Hero, driving liquidity and renewed LP appetite for venture alongside SpaceX distributions.
    • Apple filed a 41-page lawsuit against OpenAI on July 10th alleging stolen trade secrets tied to OpenAI’s consumer hardware device; OpenAI’s chief hardware officer Tang Tan is a former Apple VP of iPhone design.
    • The complaint alleges Apple job candidates were directed to bring actual parts to OpenAI interviews for “show and tell,” and cites a text about accessing network storage; OpenAI has reportedly poached over 400 Apple employees.
    • The besties’ rule of thumb: when leaving a company, the only thing you can take is what is in your head; no documents, thumb drives, or files, because Apple rarely litigates and doing so signals something egregious.
    • xAI’s Grok Build, powered by Grok 4.5 and running inside Cursor, was reportedly sending users’ entire codebases (potentially including passwords and API keys) to servers despite a privacy setting meant to prevent it; xAI disabled the upload on July 13th and open-sourced the harness.
    • Chamath’s takeaway: privacy in AI is fragile and brittle, “zero data retention” cannot be guaranteed, and there are non-obvious data-leak vectors and “trap doors” everywhere, arguing for a stratified ecosystem with independent third-party layers between enterprises and models.
    • The “reverse information paradox” (building on Palantir’s Alex Karp) holds that technically capable enterprises want control over their compute, models, weights, data, and “alpha,” via real trust boundaries, private evals, in-tenant learning loops, decoupled orchestration, and the right to fine-tune.
    • Cited token costs per million showed a huge spread: roughly $56 on a premium frontier model, about $26 on another, roughly $1.50 for Grok input, around $1 for Elon’s, and about 50 cents for Chinese models, with a claim that 95 to 98% of tasks could run one tier cheaper.
    • Ramp CEO Eric Glyman launched token spend management because CFOs cannot see or control AI spend; Ramp customers’ token spend has grown 21x in a year, and someone will eventually miss an earnings quarter on runaway AI opex.
    • Engineers optimize for the latest, greatest model while CFOs bear the cost, a misalignment that platforms fine-tuning cheaper open models (like Mira Murati’s Thinking Machines effort) are positioned to exploit.
    • Calacanis called Apple a “screaming buy” on local models: rumored M7 Ultra silicon supporting up to 1.5 terabytes of memory could run last-generation frontier-class models locally on a Mac Studio, putting downward pressure on cloud AI pricing.
    • Edge compute is fragmenting outward: Sunrun announced distributed data center blocks for homes, and Span partnered with Nvidia, with compute increasingly “chasing energy” like cheap solar and battery power.
    • Chamath projected the US will be short 2.5 Californias’ worth of energy by 2050; a recent PJM auction that needed 7 to 8 gigawatts reportedly saw only a fraction show up, underscoring the electricity crunch.
    • “Behind the meter” power lets data centers generate their own electricity on owned property, but clean-air permitting is a major obstacle; Elon reportedly used clustered mobile engines and solutions like Bloom Energy to keep projects under personal-use permits (as with Colossus in Memphis).
    • New York Governor Kathy Hochul announced the nation’s first statewide moratorium on hyperscale data centers; the besties rebutted her claims on power, land, noise, water, and pollution point by point.
    • Modern data centers use closed-loop cooling (one claim compared a typical facility’s water use to a couple of In-N-Out restaurants), occupy trivial land relative to their economic value, generate tax revenue and construction jobs, and are largely powered by clean-burning natural gas.
    • Sacks argued the same political forces slowing domestic data centers are also behind chip export controls that would block data centers in allied countries, raising the question of where the buildout can happen at all.
    • Friedberg drew an anti-GMO analogy: he argued anti-GMO sentiment tracked the US presence of Russia Today (2010 to 2022) rather than the science, and worried a similar manufactured sentiment is now driving anti-data-center attitudes.
    • Sacks cited an OpenAI blog post on PRC-linked influence operations targeting US AI debates, with a congressional investigation reportedly coming, noting China has a clear incentive to slow American AI infrastructure.
    • Sacks framed the moment as a “moral panic”: the catastrophes people fear from AI (cyber, job loss) have not materialized, yet the US risks damaging its crown jewel of free-market innovation with premature regulation over hypothetical risks.
    • The panel questioned Dario Amodei’s prediction that 50% of entry-level knowledge-worker jobs could disappear within one to five years, arguing the harms have not shown up and only a handful of frontier labs (which already do safety testing and red-teaming) even matter.
    • A cited framing of the alleged Anthropic strategy: brand yourself as the safe AI company, ban unsafe AI, then profit; a fresh Chinese model (Kimi K2) was noted as very close to the frontier, suggesting a US lead of only months.
    • Science corner: a paper from Google’s Calico and partner Retro-style researchers used AlphaFold plus directed evolution to engineer a novel enzyme that degrades CML, a key advanced glycation end product in the extracellular matrix that drives aging.
    • The engineered enzyme cleared 52 to 97% of CML from body proteins in vitro and eliminated 55% of CML from donated elderly human skin, effectively reversing that skin’s biological age toward that of a 31-year-old, pointing first toward a potentially trillion-dollar cosmetic market.

    Detailed Summary

    Demis Hassabis’s FINRA-Style SRO for AI

    DeepMind’s Demis Hassabis published a proposal for a US-led international AI standards body modeled on FINRA, the Financial Industry Regulatory Authority. The design is federally overseen but industry funded and run by independent technical experts. Frontier labs would submit models about 30 days before release, and models would be assessed for risk across cybersecurity, national security, biological threats, and other high-risk domains. Benchmarks would update quarterly, the body could coordinate a development slowdown if warranted, and participation would be voluntary at first and mandatory later. The proposal drew endorsements across the industry, including Elon Musk (who called it thoughtful), Sam Altman, Anthropic’s Jack Clark, Sundar Pichai, Satya Nadella, and Jack Dorsey.

    Friedberg explained the SRO concept: bodies like FINRA and the National Futures Association let financial institutions set their own regulatory rules and check one another, under federal oversight but not federal control, reporting up to Senate and House committees. The AI analogy is that many players are all advancing the technology and none wants a single outside regulator dictating tests, especially after California’s earlier AI legislation was, in his telling, outdated by the time it would have taken effect. An SRO can bring in industry experts, adjust tests over time, and operate faster than a new agency. Chamath endorsed it strongly, warning that a “torrent of money” will try to influence both political parties toward regulatory capture, and that establishing rules quickly is the way to avoid that off-ramp while retaining ultimate federal oversight through Commerce and the DOJ.

    Sacks’s Five Conditions and the “FAA for AI” Warning

    Sacks said he could personally get on board with an SRO because it is “infinitely better” than a new government agency that would become a “DMV for AI,” or worse, Dario Amodei’s “FAA for AI.” He laid out five conditions: the SRO must have broad industry representation including startups and open source (to avoid the three biggest labs capturing it); it should review only true frontier models that represent a step change in capability, not hold up lesser models; its scope should be catastrophic risk only, meaning cyber and CBRN (chemical, biological, radiological, nuclear), not disinformation or speech; it should be voluntary before mandatory, proving it works first; and it must substitute for, not add to, new regulatory structures.

    He then explained why an FAA model is extreme: the FAA approves new airplane designs through type certification, which takes 5 to 9 years for a new aircraft and 3 to 5 years for major amendments. Applying permission-based regulation to AI, where new model versions ship every couple of months, would push timelines from months to years and lose the race to a China that will not abide by those rules. His conclusion: if the choice is FAA for AI, DMV for AI, or Hassabis’s SRO, the SRO wins, but it has to be kept “honest and pure,” because otherwise it becomes the opening bid in a coming wave of regulation and a vehicle for massive regulatory capture. He argued that companies making concessions to buy off politicians will only invite the government to come back for more, and that at some point these companies have to grow a spine, draw a line, and demand preemption in exchange.

    The Anthropic Regulatory-Capture Debate

    Sacks revisited his October claim that Anthropic was running a “sophisticated regulatory capture strategy based on fear-mongering,” arguing that what looked like beating up on a startup now looks different given Anthropic’s trillion-dollar valuation and industry-leading revenue. He cited a Politico piece, “Inside Anthropic’s state-by-state plan to ratchet up AI rules,” describing a strategy of one-upmanship: pass a model bill like California’s SB 53, then make each subsequent state’s rules stricter, deliberately producing a patchwork instead of a single national framework. The panel noted states have strong sovereignty rights (as with self-driving cars) and Anthropic is “winning” in California, Illinois, New York, and other blue states, because government officials rarely refuse an invitation to regulate.

    Stripe, Block, and Advent Bid for PayPal

    Stripe and private equity firm Advent, joined by Jack Dorsey’s Block contributing about $17 billion in equity, are jointly bidding roughly $53 billion (about $60 per share) for PayPal, with many expecting a final price closer to $70. PayPal still has more than 400 million consumer accounts and processes about $1.7 trillion a year, but its 25-year-old product is growing only about 7% and is seen as legacy. Chamath’s key question was what unique thing this trio could build: a competitor to Visa and Mastercard. Combining PayPal’s consumer accounts, Stripe’s merchant relationships and risk infrastructure, Block’s point-of-sale and Cash App, and stablecoin rails from Stripe’s Bridge and PayPal’s PYUSD would allow far more on-us transactions that bypass the card networks, potentially passing large discounts to merchants and consumers.

    Friedberg walked through the deal structure: the $17 billion equity contribution effectively means Stripe and Block sell equity to cash investors, that cash buys PayPal, and the parties end up cross-owning pieces of each other, with the Stripe team the likely operator post-close. The antitrust question turns on market definition: framed as merchant APIs, it is Stripe versus Braintree and looks like consolidation; framed against the Visa/Mastercard duopoly, adding competition is pro-competitive. Sacks noted the deal would have been “the antitrust equivalent of a colonoscopy” two years ago. He also recounted PayPal’s history: acquired by eBay in 2002 under the corporate-minded Meg Whitman, the founding team was pushed out, creating what he prefers to call the “PayPal diaspora” rather than the “PayPal mafia.”

    AI-Native Operators and the M&A Wave

    Freeberg framed the PayPal and eBay stories as part of an emerging line: AI-native operators buying first-generation digital-native businesses that have gone mature, stale, and founder-less, and that have not yet realized their AI potential or are overspending. Bending Spoons is the roll-up template, having acquired AOL, Vimeo, Evernote, WeTransfer, and Eventbrite and revitalized them from Milan with young, AI-first executives. The panel connected this to Josh Kushner’s and General Catalyst’s roll-ups of traditional services businesses. Calacanis added the macro backdrop: after venture was “on the ropes” under Lina Khan, M&A is “back on the menu,” with deals like Uber acquiring Delivery Hero, renewed LP appetite, and liquidity from SpaceX distributions.

    Apple Sues OpenAI Over Trade Secrets

    Apple filed a 41-page lawsuit against OpenAI on July 10th alleging stolen trade secrets used to develop OpenAI’s consumer hardware device. OpenAI’s chief hardware officer, Tang Tan, is Apple’s former VP of iPhone design; the complaint alleges he directed Apple job candidates interviewing at OpenAI to bring “actual parts” for “show and tell,” and cites a text from a former Apple engineer about accessing network storage. OpenAI has reportedly poached over 400 Apple employees. Chamath noted Apple rarely litigates, so the suit signals something they found egregious, while cautioning that the facts are alleged and unproven. Sacks declined to opine on the specifics but offered a simple rule: when changing jobs, take nothing but what is in your head, no documents, thumb drives, or files.

    The Grok Build Data Leak and AI Privacy

    xAI’s Grok Build, powered by Grok 4.5 and running inside Cursor, was reportedly sending users’ entire codebases (not just the files needed for a task, but potentially passwords, API keys, and change logs) to servers, despite a privacy setting meant to stop it. xAI disabled the upload on July 13th, Elon said previously uploaded data was deleted, and xAI open-sourced the harness. Chamath used it to make a larger point tied to his CNBC comments and Alex Karp’s remarks: privacy in AI is fragile and brittle, “zero data retention” cannot truly be guaranteed, and there are non-obvious leak vectors and “trap doors” everywhere. His conclusion is that enterprises need a stratified ecosystem with independent third-party layers between them and the models to manage exposure (a model his firm 8090 uses in its “software factory”).

    Sacks connected this to a blog post on the “reverse information paradox,” building on Karp’s point that technically capable enterprises want control over their compute, models, weights, data, and “alpha.” The recipe: establish a real trust boundary with private evals, proprietary learning loops inside the tenant, decoupled orchestration, and the explicit right to fine-tune their own outputs. He described an emerging ecosystem forming alternatives to the monolithic closed model stacks that Anthropic and, to some extent, OpenAI want customers locked into.

    Token Economics and Ramp’s Spend Controls

    The panel cited a wide spread in cost per million tokens: roughly $56 on a premium frontier model, about $26 on another (similar to a Claude tier), around $1.50 for Grok input, about $1 for Elon’s, and roughly 50 cents for Chinese models. Calacanis said he built a deep-linking podcast player across models on Perplexity and that the new Grok run cost only $11. Ramp CEO Eric Glyman appeared on Squawk Box to launch token spend management, noting Ramp customers’ token spend has grown 21x in a year and that CFOs struggle to see or control spend on an open-ended tab where rates rise with each new model. The takeaway: engineers optimize for the newest model while CFOs bear the cost, and unless that misalignment is controlled, runaway opex becomes a “money-burning furnace” that will eventually cause a public company to miss earnings. The panel argued 95 to 98% of tasks could run one tier cheaper, which is exactly the opportunity platforms fine-tuning cheaper open models (like Mira Murati’s Thinking Machines) are chasing.

    Apple’s Local-Model Opportunity and Edge Compute

    Calacanis called Apple a “screaming buy,” citing Mark Gurman’s report that a rumored M7 Ultra chip could support up to 1.5 terabytes of memory, double the current ceiling. That would let a Mac Studio run last-generation frontier-class models locally, giving users effectively unlimited tokens on the desktop and putting downward pressure on cloud AI pricing from the likes of Anthropic and OpenAI. Freeberg added that edge compute is fragmenting outward: solar company Sunrun announced distributed data center blocks for homes, and Span partnered with Nvidia. The theme is compute chasing cheap energy, whether excess solar or battery power charged at night.

    The Energy Deficit and Behind-the-Meter Power

    Chamath warned the US will be short about 2.5 Californias’ worth of energy by 2050, and pointed to a recent PJM auction (serving Pennsylvania, New Jersey, Maryland and other states) that needed 7 to 8 gigawatts but reportedly saw only a fraction show up. He explained “behind the meter” power: rather than drawing grid power from a utility line, a data center generates its own electricity on owned property. The obstacle is clean-air permitting. Solar takes too much space and batteries still need a generation source, so operators use gas. He described Elon clustering mobile 18-wheeler-style engines to keep them under personal-use permits, and newer solutions like Bloom Energy that allow large installations under similar rules, which is how projects like Colossus in Memphis got off the ground.

    New York’s Data Center Moratorium

    New York Governor Kathy Hochul announced the nation’s first statewide moratorium on hyperscale data centers, citing power draw, land use, water, and noise pollution. The besties rebutted each claim: behind-the-meter power means facilities bring their own electricity rather than competing with residential ratepayers; data centers are highly land-efficient, and New York State is roughly 70 to 80% undeveloped outside the city; noise can be managed with distance; modern facilities use closed-loop cooling (one comparison put a typical facility’s water use at a couple of In-N-Out restaurants, far less than almonds or golf courses); and natural gas is a clean-burning power source. They noted the tax revenue, construction boom, and ongoing jobs data centers create. Sacks cited a theory that Democrats intend the “moratorium” as leverage: pause construction until they can dictate terms, then lift it under a future administration in exchange for a new regulatory agency and speech controls ported from the social-media trust-and-safety agenda. He stressed a moratorium is effectively a five-year pause once ramp-up is counted, and that the same forces slowing domestic builds are pushing chip export controls that would block data centers in allied countries too.

    Foreign Influence, Anti-GMO, and the AI Moral Panic

    Freeberg drew an extended analogy between anti-data-center sentiment and anti-GMO sentiment. He argued that GMOs were prevalent and uncontroversial from their 1996 launch until anti-GMO sentiment rose in tandem with Russia Today’s US presence (2010 to 2022) and fell after RT was pushed out, and that similar KGB-era “directed measures” influence campaigns can be traced to opposition to nuclear energy in Germany. He cited a poll showing over 50% of Americans believe data centers increase water and electricity costs even where facilities recycle water and generate their own power. Sacks pointed to an OpenAI blog post on PRC-linked influence operations targeting US AI debates, with a congressional investigation reportedly coming, arguing China has a clear incentive to slow US AI infrastructure, kill open source, and constrain cheaper models. Sacks then broadened it to a “moral panic”: the feared catastrophes (cyber, job loss) have not materialized, yet the US risks damaging its crown jewel of free-market innovation over hypothetical risks, questioning Dario Amodei’s prediction that 50% of entry-level knowledge-worker jobs could vanish within one to five years and noting the fresh Chinese model Kimi K2 is close to the frontier.

    Science Corner: An Enzyme That Reverses Skin Aging

    Freeberg closed with a paper from Google’s secretive longevity startup Calico and a pharma partner focused on the extracellular matrix, the space between cells. Over time, sugars and fats bind to proteins there in a process called glycation, accumulating as advanced glycation end products (chiefly a molecule called CML) that stiffen tissue, cause wrinkles and immobility, and drive inflammation, with nothing in the body to break them down. The researchers used AlphaFold to find a protein that could bind and degrade CML, then applied directed evolution across five recursive cycles, DNA-programming thousands of variants to maximize activity. The engineered enzyme cleared 52 to 97% of CML from body proteins like collagen, casein, and hemoglobin in vitro, and eliminated 55% of CML from donated elderly human skin, effectively reversing that skin’s biological age toward a 31-year-old’s. Open questions remain about delivery (cream, shot, supplement, or an RNA therapy that makes the enzyme inside the body), but the panel expects the first market to be a trillion-dollar cosmetic one, and hailed it as a profound demonstration of AI-driven protein engineering.

    Notable Quotes

    “The whole industry is going to need to be regulated and I think the industry needs to regulate themselves. That’s the key to this.”

    Jason Calacanis, replaying his earlier call for AI self-certification

    “If my choices are between FAA for AI or what I would call the DMV for AI, I would much rather go for Demis’ SRO for AI.”

    David Sacks, on why self-regulation beats a new government agency

    “There’s hardly anyone in government who will ever say, oh no no no, we’re not qualified. Most people in the government will say thank you very much, what else can we take.”

    David Sacks, on the asymmetry that makes voluntary concessions dangerous

    “What it prevents is a handful of actors using their balance sheets and their capital to essentially pull the ladder up.”

    Chamath Palihapitiya, on the point of establishing industry rules quickly

    “You are creating a competitor to Visa and Mastercard.”

    Chamath Palihapitiya, on the only thing Stripe, Block, and Advent could build together with PayPal

    “The only thing you can bring to your new job is what’s in your head. Your memories. But never leave with anything else.”

    David Sacks, on avoiding trade-secret disputes when changing employers

    “Privacy in AI is very fragile and it’s very brittle. You are leaking information where you don’t know it.”

    Chamath Palihapitiya, on the limits of zero-data-retention promises

    “Unless you get a control of this and you can directly say how much money you’re making, this is a bridge to nowhere. It is a money burning furnace.”

    Chamath Palihapitiya, on uncontrolled enterprise token spend

    “We’re on the threshold of destroying the crown jewel of our economy, which is the system of free market innovation that we have.”

    David Sacks, on the risk of a premature AI regulatory apparatus

    “Number one, brand yourself as a safe AI company. Number two, ban unsafe AI. Three, profit.”

    David Sacks, summarizing the strategy he attributes to the “safe AI” positioning

    Watch the full conversation here: Can the AI Industry Regulate Itself? on the All-In Podcast.

    Related Reading

    • FINRA the financial-industry self-regulatory organization that Demis Hassabis’s AI proposal is modeled on.
    • AlphaFold (Wikipedia) the protein-structure prediction system behind the age-reversal enzyme discovery in the science corner.
    • PayPal Mafia (Wikipedia) background on the founders Sacks calls the “PayPal diaspora.”
    • The Founders by Jimmy Soni, the definitive history of PayPal’s founding team and its diaspora.
    • Advanced glycation end-products (Wikipedia) the biochemistry of CML and the extracellular-matrix aging the Calico enzyme targets.

  • 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

  • Gavin Baker on Orbital Compute, TSMC, Frontier AI Models, Anthropic’s Vertical Take Off, and the Coming Wafer Shortage

    Gavin Baker, founder and CIO of Atreides Management, returns to Patrick O’Shaughnessy’s Invest Like the Best for his sixth appearance. He calls the current AI moment the most extraordinary moment in the history of capitalism, walks through what Anthropic’s vertical takeoff in revenue actually means, lays out why orbital compute is closer than skeptics believe, dissects the TSMC bottleneck that may be the only thing standing between today’s market and a full-on AI bubble, and rates every hyperscaler on how they have positioned for a world where frontier model providers may stop selling API access altogether.

    TLDW

    Anthropic added eleven billion dollars of ARR in a single month, which is roughly the combined business of Palantir, Snowflake, and Databricks built over a decade. That is the setup. From there Gavin Baker covers the March and April selloff, the contrarian read that a closed Strait of Hormuz was actually bullish for American manufacturing competitiveness, why Anthropic and OpenAI multiples may be misleadingly cheap on an unconstrained run rate basis, why Elon Musk’s discipline on SpaceX valuation created a superpower of permanent access to capital, the practical engineering case for orbital compute as racks in space rather than Pentagon sized space stations, why TSMC’s capacity discipline is the single most important variable in whether the AI cycle becomes a bubble, what Terafab in Texas changes, why the Pareto frontier of AI models has flipped from Google dominance to Anthropic and OpenAI dominance in nine months, the shift from all you can eat AI subscriptions to usage based pricing and what that means for revenue scaling, Richard Sutton’s bitter lesson as the largest risk to the AI trade, why frontier tokens still capture an overwhelming share of economic value, the role of continual learning as the third great open question, why most new chip startups should not try to build a better GPU, why Cerebras did something different and hard, why disaggregated inference may extend GPU useful lives to ten or fifteen years and rescue the private credit industry, why being in the token path is the new venture filter, the new prisoner’s dilemma around releasing frontier models via API, an honest rating of Google, Meta, Amazon, and Microsoft, why personal safety is becoming a real AI era risk, and why he remains an AI optimist maximalist who believes this could be the next Pax Americana.

    Key Takeaways

    • Anthropic added eleven billion dollars of ARR in one month, more than the combined businesses of Palantir, Snowflake, and Databricks built across a decade. There is no precedent for this in the history of capitalism.
    • The SaaS and cloud revolution created between five and ten trillion dollars of value over twenty years. AI is replaying that compression on a timeline measured in months.
    • The March selloff was a drawdown driven by disagreement with price action, not invalidated thesis. That is the kind of drawdown an investor can lean into.
    • Deep Seek Monday in January 2025 was a similar setup. By the day of the selloff, AWS Asia GPU prices had already doubled, GPU availability had fallen, and it was obvious reasoning models would be vastly more compute hungry at inference. The market priced the opposite.
    • The Strait of Hormuz closing was actually positive for America. US natural gas (the primary input into US electricity, which feeds AI) fell twenty percent on Bloomberg while Asian and European natural gas doubled or tripled. American manufacturing competitiveness improved overnight.
    • The US is now the world’s largest producer and exporter of oil and gas. The economy is dramatically less energy intensive than in the 1970s. The shortage trauma comparison does not hold.
    • Tech as a sector traded as cheaply versus the rest of the market in early April as at any point in the last ten years, into the single most bullish moment for AI fundamentals on record.
    • Anthropic is dramatically more capital efficient than OpenAI, having burned roughly eighty percent less to reach a similar revenue scale. They have very different structural returns on invested capital.
    • Anthropic at roughly nine hundred billion for fifty billion of ARR (growing a thousand percent) is striking. Adjusted for compute constraint, the unconstrained run rate could be one hundred fifty to two hundred billion, putting the implied multiple closer to five times.
    • Claude Opus generates roughly seventy percent fewer tokens for the same question than previously, with token quantity tied to answer quality. Subscribers on flat-fee plans are getting a lobotomized model.
    • Elon Musk’s superpower is twenty years of making investors money. He never pushes valuation. SpaceX compounded low thirty percent per year for a decade because Musk treats fair pricing as a sacred covenant.
    • Capitalism will solve the watts shortage. The current bottleneck has shifted from chips and energy to zoning and political approval. Many capex decisions are paused until after the US midterms.
    • The watts shortage probably begins to alleviate in 2027 and 2028. Orbital compute solves it longer term.
    • Orbital compute is not Pentagon sized data centers in space. It is racks in space. A Blackwell rack is three thousand pounds, eight feet tall, four feet deep, three feet wide. SpaceX has shown a satellite roughly that size.
    • The satellites operate in sun synchronous orbit so solar wings (around five hundred feet per side) always face the sun and the radiator on the dark side always points to deep space.
    • Starlink V3 satellites already run at around twenty kilowatts. A Blackwell rack runs at one hundred kilowatts. SpaceX engineers express genuine confidence they have already solved cooling and radiator design at these scales.
    • Racks in space are connected with lasers traveling through vacuum, the same lasers already on every Starlink. SpaceX operates the world’s largest satellite fleet and, via xAI Colossus, the world’s largest data center on Earth.
    • Inference will move to orbit. Training will stay on Earth for a long time. Terrestrial data centers remain valuable for the rest of an investor’s career.
    • The wafer bottleneck is structural and political. TSMC is essentially Taiwan’s GDP, water, and electricity. The leaders see themselves as inheritors of Morris Chang’s sacred legacy and they do not behave like a Western public company.
    • Jensen Huang has never had a contract with TSMC. The relationship is run on handshakes and the assumption that things will be fair over time.
    • If TSMC did everything Jensen wanted, Nvidia could be selling two to three trillion dollars of GPUs in 2026 and 2027. TSMC’s discipline is the single largest factor preventing a true AI bubble.
    • Historically, foundational technologies always get a bubble. Railroads, canals, the internet. The current AI buildout is overwhelmingly funded out of operating cash flow, GPUs are running at one hundred percent utilization, and that is fundamentally different from the year 2000 fiber overbuild.
    • If one of Intel or Samsung Foundry catches up at the leading node, the other will follow, and TSMC’s discipline collapses. Watch TSMC capacity decisions to predict a bubble.
    • Terafab, the SpaceX and Tesla joint venture to build the world’s largest fab in America, has a partnership with Intel that grants access to fifty years of institutional foundry knowledge. The A teams at ASML, KLA, Lam Research, and Applied Materials will follow Elon’s reputation in hardware engineering.
    • The hiring playbook for Terafab includes building Taiwan Town, Japan Town, and Korea Town next to the fab. Recruit the engineers and import their families, their restaurants, and their staff.
    • Frontier tokens still capture an overwhelming share of all economic value created at the model layer. This is surprising and is one of the three big open questions for AI investing.
    • The Pareto frontier of intelligence versus cost has flipped. Nine months ago Google’s TPU dominated every point on the frontier. Today Anthropic and OpenAI dominate, with Grok 4.3 on the frontier and Gemini 3.1 hanging on.
    • Google’s conservative TPU V8 design (partly an attempt to reduce dependence on Broadcom and Nvidia) is the leading explanation for the loss of per token cost leadership.
    • AI pricing is shifting from all you can eat to usage based, mirroring the cellular and long distance industries. Cellular stopped being a great growth industry when it went all you can eat. AI just made the opposite move.
    • OpenAI and Anthropic together could exceed two hundred billion in ARR this year if compute keeps coming online and frontier token pricing holds.
    • The two hundred fifty dollar a month consumer AI plan is no longer enough to evaluate frontier capability. Enterprise plans with usage based billing are required because rate limits are now severe.
    • The three biggest open questions for AI investors are: violation of the bitter lesson via ASI or human ingenuity, whether frontier tokens keep commanding their premium, and when continual learning arrives.
    • Today’s continual learning is crude reinforcement learning during mid training on verifiable tasks. True continual learning means weights updating dynamically, like a human who learns the first time they touch fire.
    • Trying to build a better GPU is a losing strategy. Jensen will copy any one to three percent share design. Startups should target one percent share, do something different, and make it hard enough that Nvidia cannot fast follow.
    • Disaggregated inference (separating prefill and decode) opens new design canvases. Prefill is memory capacity bound. Decode is memory bandwidth bound. Each can be optimized independently.
    • Cerebras did something different and hard with wafer scale computing. Three generations of chips and real grit to get there.
    • Disaggregation of inference may stretch GPU useful lives to ten or fifteen years, dropping financing costs from low sevens to five or six percent, mathematically lowering the cost of the AI buildout and likely saving the private credit industry from its SaaS loan exposure.
    • Sellers of shortage outperform buyers of shortage. But owning the largest installed base of what is currently in shortage (hyperscaler CPU fleets, for example) is also a strong position.
    • Most of the economic value at the application layer of AI has been destroyed, not created. The exceptions are companies in the token path or in niches small enough that frontier labs ignore them.
    • Coding may be the shortest path to ASI. If you can write code, you can write code that does anything. Cursor, Cognition, and Anthropic correctly focused on it.
    • Jensen could probably get close to the frontier with his own Nemotron family of models whenever he wants. The fact that he chooses not to is a strategic decision about not commoditizing his customers.
    • The new prisoner’s dilemma in AI is whether frontier labs release their best model via API. If everyone agrees not to, Chinese open source falls behind. If anyone defects, the defector pulls ahead on revenue and resources, forcing everyone else to defect.
    • Google still owns the largest compute installed base. Without TPU’s prior cost advantage, this matters more. YouTube data has real value in a world of robotics. GCP is going crazy.
    • Meta deserves credit for becoming AI first internally faster than any other internet giant. Musa, their first MSL model, is impressively close to the Pareto frontier.
    • Amazon is strong because of Trainium and robotics driven retail P&L efficiency. Nova is better than it gets credit for.
    • Microsoft flinched on capex in early 2025 and lost position. Satya Nadella’s current decision to use Microsoft compute for Microsoft products rather than reselling to OpenAI is a courageous and probably correct call, even at the cost of an eight hundred dollar stock price.
    • The hyperscalers most engaged with startups are Amazon and Nvidia by a mile, followed by Google. Broadcom is the favorite ASIC partner. AMD, Microsoft, and Meta have minimal startup engagement and that will cost them as the best teams are now at startups.
    • Personal safety in an AI era requires a family or company safe word that cannot be socially engineered. Deepfake voice and video extortion at the speed of FaceTime is already feasible.
    • Ukraine is winning largely on the back of having the best battlefield AI outside America and Israel. Adversaries are starting to internalize what AI dominance means geopolitically.
    • An optimistic read is that this becomes a new Pax Americana, the way the post 1945 American nuclear monopoly was used to rebuild Germany and Japan rather than dominate.
    • AI cured a friend’s daughter’s rare disease by spinning up a research effort that identified a market drug capable of impacting her condition. That is the upside that keeps Gavin an AI optimist maximalist.

    Detailed Summary

    The most extraordinary moment in the history of capitalism

    Gavin’s framing of the current moment is unusually direct. Anthropic added eleven billion dollars of annual recurring revenue in a single month. The three highest profile SaaS companies of the last decade plus, Palantir, Snowflake, and Databricks, took a decade and tens of thousands of employees collectively to build the combined business that Anthropic added in thirty days. He has been investing through every major tech cycle and says there is no historical analog. Not the dotcom era, not the cloud transition, not mobile. This is its own thing.

    The market response, then, was peculiar. The NASDAQ sold off into the single most bullish moment for AI fundamentals on record. Tech traded at roughly its widest discount versus the rest of the market in a decade. Investors who said they wished they had bought into AI during 2022, during COVID, or during Deep Seek Monday got the same valuation setup again in early April, this time with an even clearer inflection.

    Why the Strait of Hormuz closing was secretly bullish for America

    One reason the macro fear in March may have been mispriced is that the same geopolitical event that drove the selloff was, in practice, a relative benefit to the United States. American natural gas, the input into American electricity, which is the input into American AI training and inference, fell roughly twenty percent. Asian and European natural gas prices doubled or tripled. The US emerged with sharply improved relative manufacturing competitiveness, which is exactly what the current administration cares about.

    The 1970s comparison does not hold. The US economy is dramatically less energy intensive, it is now the world’s largest producer and largest exporter of oil and gas, and there are no shortages, only price moves. That backdrop made it easier for disciplined investors to stay focused on AI fundamentals through the volatility.

    Anthropic and OpenAI valuations on an unconstrained run rate

    Anthropic at roughly nine hundred billion for fifty billion of ARR sounds rich until you adjust for the fact that the company is severely compute constrained. Gavin estimates that, unconstrained, Anthropic might be at one hundred fifty to two hundred billion in run rate revenue, putting the implied multiple closer to five times. He also points out that Claude Opus now generates roughly seventy percent fewer tokens for the same question than it used to. Token quantity correlates with answer quality, and Anthropic is rate limiting and shrinking outputs to ration capacity across its user base.

    Anthropic and OpenAI are also structurally very different. Anthropic has burned around eighty percent less cash than OpenAI to reach a comparable revenue scale. That implies very different long term returns on invested capital, though OpenAI has done a better job locking in compute and Sarah Friar is one of the most exceptional CFOs Gavin has worked with.

    Why neither lab is raising at a three trillion dollar valuation

    The answer Gavin gives is that both labs are deliberately leaving valuation on the table the way Elon has done for two decades. SpaceX compounded at low thirty percent annually for a decade because Elon never pushed price. The result is a permanent superpower of access to capital. Investors trust him because they have made money with him for twenty years. That is a moat that compounds with every round.

    Anthropic could probably raise at a one hundred percent premium to its rumored latest mark. They are choosing not to. In an uncertain world (Ukraine, Russia, Iran, Taiwan), preserving the ability to raise more capital later at fair prices is more valuable than maximizing this round.

    Watts and wafers, the two real constraints

    Capitalism is solving the watts problem. The leading PE infrastructure investors now say zoning and political approval, not chips or energy, are the gating factors. Companies are deferring big capex announcements until after the US midterms. Turbine capacity is being doubled at the manufacturers. Companies like Boom Aerospace are repurposing jet engines for grid use. Watts probably ease meaningfully in 2027 and 2028 and then orbital compute does the rest.

    Wafers are the harder problem because they live in Taiwan, run on handshakes, and depend on a corporate culture that does not respond to public market incentives. TSMC is essentially the GDP, water consumption, and electricity consumption of Taiwan. Its leadership treats the company as the legacy of Morris Chang. The Silicon Shield doctrine is real and internal.

    Orbital compute as racks in space

    The biggest mental update Gavin asks listeners to make is to stop picturing data centers in space as Pentagon sized space stations. A Blackwell rack is three thousand pounds and roughly the size of a refrigerator. SpaceX has shown a concept satellite of about that size. Solar wings extend five hundred feet to each side and the radiator extends hundreds of feet behind, both possible because the orbit is sun synchronous and the orientation is fixed relative to the sun.

    SpaceX engineers Gavin has spoken to at Starbase express genuine confidence that they have solved cooling at these power levels. They have. Starlink V3 satellites already operate at twenty kilowatts. A Blackwell rack is one hundred kilowatts. The same company operates the world’s largest satellite fleet and the world’s largest data center on Earth via xAI Colossus. The racks are connected to each other with lasers traveling through vacuum, technology already deployed in every Starlink. The naysayers, Gavin observes, are armchair skeptics and Larry Ellison’s response (he is out there landing rockets, no one else is) is the right frame.

    Terafab in Texas and the threat to TSMC’s discipline

    Terafab, the SpaceX and Tesla joint venture, intends to be the largest fab in the world. The partnership with Intel grants access to fifty years of foundry institutional knowledge, allowing Terafab to start three to five quarters behind the leading node rather than fifteen years behind. The A teams at the semicap equipment companies (ASML, KLA, Lam Research, Applied Materials) will follow Elon’s reputation in hardware engineering the same way they followed TSMC twenty years ago when Intel stumbled.

    The talent strategy is the part most observers underestimate. Recruit the best engineers globally, then import their families, their restaurants, their staff. Build Taiwan Town, Japan Town, and Korea Town next to the fab. Optimize the human experience for the people whose work matters. Intel and Samsung do not think that way.

    Bubble watch and the year 2000 comparison

    Every foundational technology in modern history has had a bubble. Railroads, canals, the internet. Carlota Perez documented why. Markets correctly identify the importance, diversity of opinion collapses, supply gets ahead of demand, the bubble crashes. The current cycle has two important differences. The buildout is overwhelmingly funded out of operating cash flow, not debt. Every GPU is running at one hundred percent utilization, while at the peak of the fiber bubble ninety nine percent of fiber was unused.

    TSMC discipline is the single largest reason a bubble has not formed. If Jensen could buy everything TSMC could theoretically make, Nvidia could sell two to three trillion dollars of GPUs in 2026 and 2027. At some point that becomes more than the market can absorb. If Intel or Samsung Foundry catches up at the leading node, the other will too. TSMC’s pricing discipline collapses and the bubble starts.

    The Pareto frontier and the loss of Google’s cost advantage

    The most important chart in AI is the Pareto frontier of model intelligence versus per token cost. Nine months ago, Google’s TPU based models dominated every point on it. OpenAI, Anthropic, and xAI sat inside the frontier. Today the frontier is dominated by Anthropic and OpenAI, with Grok 4.3 on the frontier and Gemini 3.1 hanging on by subsidization more than economics. The most likely cause is Google’s conservative TPU V8 design, an attempt to reduce dependence on Broadcom and Nvidia that sacrificed per token economics.

    The bitter lesson, frontier tokens, and continual learning

    Three open questions dominate AI investing. The first is whether Richard Sutton’s bitter lesson (more compute beats human algorithmic cleverness) gets violated by ASI itself optimizing for efficiency. Closer observers of AI are more skeptical of a violation. Gavin thinks ASI’s first move will be to make itself more efficient and more resourced, which is technically a temporary violation.

    The second is whether frontier tokens keep capturing the overwhelming share of economic value at the model layer. Today they do, surprisingly. Gemini 3.1 Pro was mindblowing nine months ago and is intolerable today. The third is when continual learning arrives. Today’s models need a million fire touches to learn what a human learns from one. True continual learning would mean dynamic weight updates in real time and would produce a fast takeoff.

    From all you can eat to usage based AI pricing

    AI is shifting from flat fee plans to usage based pricing. The historical analogy is cellular and long distance. Both stopped being great growth industries when they went all you can eat. AI just made the opposite move. The consequence is that flat fee subscribers, even on premium consumer plans, get a rate limited and token throttled version of the frontier model. Enterprise plans with usage based billing are now required to evaluate true capability. Gavin thinks the combination of new compute coming online and usage based pricing is what gets OpenAI and Anthropic past two hundred billion in combined ARR this year.

    Chip startups, prefill decode disaggregation, and Cerebras

    Trying to build a better GPU is the wrong move. The four scaled players (Nvidia, AMD, Trainium, TPU) have copy capability for any one to three percent share design that looks attractive. The good news for startups is that disaggregated inference (separating prefill and decode) opens a richer design canvas. Prefill is memory capacity bound. Decode is memory bandwidth bound. Each can be optimized independently. Andrew Fox’s analogy is a British naval ship of the eighteenth century. Prefill is loading the cannon. Decode is firing it.

    Cerebras is the model. Wafer scale computing is genuinely different and genuinely hard. It took three generations of chips to get right. Andrew Feldman and his team had the grit to keep going through chip one being a failure. The design has a high ratio of on chip compute and memory relative to shoreline IO, which is why Cerebras is now experimenting with putting an optical wafer on top of the compute wafer to solve scale out.

    GPU useful lives and the rescue of private credit

    One of the strongest claims in the conversation is that disaggregated inference will stretch GPU useful lives to ten or fifteen years. The skeptical narrative (GPUs are obsolete in two years, companies are cooking their depreciation books) is wrong. You can put a Cerebras system or Groq LPU in front of older Hopper or Ampere parts, use them only for prefill, and run them until they physically melt. Private credit, which is in pain from SaaS loans and which underwrote GPU loans on three to four year lives, may be saved by this.

    If GPU financing rates can come down from low sevens to five or six percent, the mathematics of the AI buildout improves materially. That is a structural tailwind that compounds for years.

    The application layer, the token path, and a new prisoner’s dilemma

    Trillions of dollars of value have been destroyed at the application layer, not created. Cursor and Cognition are the rare scaled exceptions, and they got there by focusing on coding very early. As Amjad Masad noted, coding is plausibly the shortest path to ASI because a coding agent can write itself into any new domain. Jamin Ball’s frame is that the new venture filter is whether the company is in the token path. Data Bricks is. Most application layer startups are not.

    Jensen could probably get close to the frontier with Nemotron whenever he wants, and the strategic question of whether to do that is a new prisoner’s dilemma. If every frontier lab agrees not to release best models via API, Chinese open source falls steadily behind. If anyone defects, the defector gains revenue and resources, and everyone else has to defect. The same dynamic exists between TSMC, Intel, and Samsung. If Nvidia or AMD ever truly used an alternative foundry, that foundry would catch up rapidly.

    Rating the hyperscalers

    Google has the largest compute installed base, the YouTube data that matters in a robotics world, and a search business that prints. Their loss of TPU cost leadership is the surprise of the year. If Google IO in five days does not produce a leapfrog model, the Nvidia centric narrative gets even stronger.

    Meta deserves real credit. Zuckerberg made Meta AI first internally faster than any other internet giant, paid up for the talent contracts when no one else would, and shipped Musa as a first model from MSL that is close to the Pareto frontier. Amazon is well positioned on Trainium, robotics in retail, and a Nova model line that is better than it gets credit for. Microsoft flinched on capex in early 2025 and lost position. Satya Nadella’s current decision to use Microsoft compute for Copilot rather than reselling to OpenAI is courageous and probably correct, even at the cost of stock price.

    The most interesting cross hyperscaler metric is startup engagement. Nvidia and Amazon engage deeply with startups. Google is next. Broadcom is the favored ASIC partner. AMD, Microsoft, and Meta have minimal startup engagement, which Gavin believes will cost them as the best teams now sit at startups.

    Personal safety, geopolitics, and the Pax Americana case

    The closing section turns darker. Personal safety in an AI era requires a family or company safe word that cannot be socially engineered. Deepfake voice and video extortion via something that looks exactly like your child calling on FaceTime is already feasible. Political violence against AI leaders is a real concern. Geopolitically, Ukraine is winning largely because it has the best battlefield AI outside America and Israel. How adversaries respond to that asymmetry is the next great variable.

    Gavin’s optimistic frame is the Pax Americana. After 1945 the US had a nuclear monopoly and could have controlled the world. Instead it rebuilt Germany and Japan, both of which became the most reliable American allies for the next eighty years. If AI dominance plays out similarly, this is a generationally positive story rather than a destabilizing one. The personal anecdote that closes the conversation is a friend whose daughter was diagnosed with a rare genetic condition. He spun up agents, identified a drug already on the market that addresses her mutation, and her life is immeasurably different because of AI. That is the upside.

    Thoughts

    The Anthropic eleven billion in a month framing is the kind of stat that resets priors. The right way to interpret it is not as a one off but as a measure of how fast value can compound when the underlying technology improves on a curve steeper than the ability of the rest of the economy to absorb it. The skeptical question is whether that ARR is durable or whether it is heavily tied to a customer base of other AI companies that are themselves on a single venture funded year of runway. The bullish answer is that frontier coding, frontier research, and frontier enterprise tasks are not going to stop being valuable, and Anthropic is the best at all three. Both can be true. The number is still extraordinary.

    The argument that TSMC discipline is the only thing preventing a bubble is the analytically tightest part of the conversation. The implied trade is to watch TSMC capacity additions like a hawk and to be more, not less, cautious if Intel Foundry or Samsung Foundry ever announce real share at the leading node. The Terafab thesis is more speculative but more interesting. If Elon’s talent recruiting playbook works and the Intel partnership gives Terafab a real seat at the table within five years, the geometry of the global semiconductor industry shifts in a way that is bullish for American manufacturing, bullish for power and water infrastructure in Texas, and ambiguous for TSMC itself.

    The Pareto frontier discussion deserves more attention than it usually gets. Pricing leadership in AI is not a vanity metric. It determines who can subsidize free tier usage, who can absorb compute shortages, who can ship cheaper enterprise plans, and ultimately whose model becomes the default for any given workload. Google losing per token leadership in nine months is one of the most under analyzed events in the sector and it explains a lot about why Anthropic and OpenAI are growing the way they are. If Google IO does not produce a leapfrog model, the implied verdict on TPU V8 design choices gets a lot harsher.

    The application layer destruction point is worth sitting with. Founders building on top of frontier models are competing in a world where the model itself moves faster than any moat they can build, where the model lab can absorb their niche if it gets interesting, and where the only protection is either deep token path integration or a niche so small the lab does not bother. That is a much harsher venture environment than the early SaaS era. The compensating opportunity is that one human can now run a hundred agents, so the ceiling on what a small team can build is correspondingly higher. The bet is that productivity per founder rises faster than competitive pressure from the labs. We will find out.

    The orbital compute pitch is the section that will polarize listeners. The naive read is that this is science fiction. The closer read is that every component (sun synchronous orbit, laser interconnect, twenty kilowatt satellite buses, ten thousand satellite manufacturing cadence, full rocket reusability) already exists. The remaining engineering problems are repair, maintenance, and radiator scale, all of which are real but tractable on a five to ten year horizon. The strategic implication is that the political and zoning ceiling on terrestrial data centers becomes less binding if orbital compute is a credible alternative for inference workloads. The investor implication is that being short the watts and cooling complex on a five year horizon is a real trade, not a meme.

    Watch the full conversation here.