PJFP.com

Pursuit of Joy, Fulfillment, and Purpose

Tag: DeepSeek

Global Madness Unleashed: Tariffs, AI, and the Tech Titans Reshaping Our Future

As the calendar turns to March 21, 2025, the world economy stands at a crossroads, buffeted by market volatility, looming trade policies, and rapid technological shifts. In the latest episode of the BG2 Pod, aired March 20, venture capitalists Bill Gurley and Brad Gerstner dissect these currents with precision, offering a window into the forces shaping global markets. From the uncertainty surrounding April 2 tariff announcements to Google’s $32 billion acquisition of Wiz, Nvidia’s bold claims at GTC, and the accelerating AI race, their discussion—spanning nearly two hours—lays bare the high stakes. Gurley, sporting a Florida Gators cap in a nod to March Madness, and Gerstner, fresh from Nvidia’s developer conference, frame a narrative of cautious optimism amid palpable risks.

A Golden Age of Uncertainty

Gerstner opens with a stark assessment: the global economy is traversing a “golden age of uncertainty,” a period marked by political, economic, and technological flux. Since early February, the NASDAQ has shed 10%, with some Mag 7 constituents—Apple, Amazon, and others—down 20-30%. The Federal Reserve’s latest median dot plot, released just before the podcast, underscores the gloom: GDP forecasts for 2025 have been cut from 2.1% to 1.7%, unemployment is projected to rise from 4.3% to 4.4%, and inflation is expected to edge up from 2.5% to 2.7%. Consumer confidence is fraying, evidenced by a sharp drop in TSA passenger growth and softening demand reported by Delta, United, and Frontier Airlines—a leading indicator of discretionary spending cuts.

Yet the picture is not uniformly bleak. Gerstner cites Bank of America’s Brian Moynihan, who notes that consumer spending rose 6% year-over-year, reaching $1.5 trillion quarterly, buoyed by a shift from travel to local consumption. Conversations with hedge fund managers reveal a tactical retreat—exposures are at their lowest quartile—but a belief persists that the second half of 2025 could rebound. The Atlanta Fed’s GDP tracker has turned south, but Gerstner sees this as a release of pent-up uncertainty rather than an inevitable slide into recession. “It can become a self-fulfilling prophecy,” he cautions, pointing to CEOs pausing major decisions until the tariff landscape clarifies.

Tariffs: Reciprocity or Ruin?

The specter of April 2 looms large, when the Trump administration is set to unveil sectoral tariffs targeting the “terrible 15” countries—a list likely encompassing European and Asian nations with perceived trade imbalances. Gerstner aligns with the administration’s vision, articulated by Vice President JD Vance in a recent speech at an American Dynamism event. Vance argued that globalism’s twin conceits—America monopolizing high-value work while outsourcing low-value tasks, and reliance on cheap foreign labor—have hollowed out the middle class and stifled innovation. China’s ascent, from manufacturing to designing superior cars (BYD) and batteries (CATL), and now running AI inference on Huawei’s Ascend 910 chips, exemplifies this shift. Treasury Secretary Scott Bessent frames it as an “American detox,” a deliberate short-term hit for long-term industrial revival.

Gurley demurs, championing comparative advantage. “Water runs downhill,” he asserts, questioning whether Americans will assemble $40 microwaves when China commands 35% of the global auto market with superior products. He doubts tariffs will reclaim jobs—automation might onshore production, but employment gains are illusory. A jump in tariff revenues from $65 billion to $1 trillion, he warns, could tip the economy into recession, a risk the U.S. is ill-prepared to absorb. Europe’s reaction adds complexity: *The Economist*’s Zanny Minton Beddoes reports growing frustration among EU leaders, hinting at a pivot toward China if tensions escalate. Gerstner counters that the goal is fairness, not protectionism—tariffs could rise modestly to $150 billion if reciprocal concessions materialize—though he concedes the administration’s bellicose tone risks misfiring.

The Biden-era “diffusion rule,” restricting chip exports to 50 countries, emerges as a flashpoint. Gurley calls it “unilaterally disarming America in the race to AI,” arguing it hands Huawei a strategic edge—potentially a “Belt and Road” for AI—while hobbling U.S. firms’ access to allies like India and the UAE. Gerstner suggests conditional tariffs, delayed two years, to incentivize onshoring (e.g., TSMC’s $100 billion Arizona R&D fab) without choking the AI race. The stakes are existential: a misstep could cede technological primacy to China.

Google’s $32 Billion Wiz Bet Signals M&A Revival

Amid this turbulence, Google’s $32 billion all-cash acquisition of Wiz, a cloud security firm founded in 2020, signals a thaw in mergers and acquisitions. With projected 2025 revenues of $1 billion, Wiz commands a 30x forward revenue multiple—steep against Google’s 5x—adding just 2% to its $45 billion cloud business. Gerstner hails it as a bellwether: “The M&A market is back.” Gurley concurs, noting Google’s strategic pivot. Barred by EU regulators from bolstering search or AI, and trailing AWS’s developer-friendly platform and Microsoft’s enterprise heft, Google sees security as a differentiator in the fragmented cloud race.

The deal’s scale—$32 billion in five years—underscores Silicon Valley’s capacity for rapid value creation, with Index Ventures and Sequoia Capital notching another win. Gerstner reflects on Altimeter’s misstep with Lacework, a rival that faltered on product-market fit, highlighting the razor-thin margins of venture success. Regulatory hurdles loom: while new FTC chair Matthew Ferguson pledges swift action—“go to court or get out of the way”—differing sharply from Lina Khan’s inertia, Europe’s penchant for thwarting U.S. deals could complicate closure, slated for 2026 with a $3.2 billion breakup fee at risk. Success here could unleash “animal spirits” in M&A and IPOs, with CoreWeave and Cerebras rumored next.

Nvidia’s GTC: A $1 Trillion AI Gambit

At Nvidia’s GTC in San Jose, CEO Jensen Huang—clad in a leather jacket evoking Steve Jobs—addressed 18,000 attendees, doubling down on AI’s explosive growth. He projects a $1 trillion annual market for AI data centers by 2028, up from $500 billion, driven by new workloads and the overhaul of x86 infrastructure with accelerated computing. Blackwell, 40x more capable than Hopper, powers robotics (a $5 billion run rate) to synthetic biology. Yet Nvidia’s stock hovers at $115, 20x next year’s earnings—below Costco’s 50x—reflecting investor skittishness over demand sustainability and competition from DeepSeek and custom ASICs.

Huang dismisses DeepSeek R1’s “cheap intelligence” narrative, insisting compute needs are 100x what was estimated a year ago. Coding agents, set to dominate software development by year-end per Zuckerberg and Musk, fuel this surge. Gurley questions the hype—inference, not pre-training, now drives scaling, and Huang’s “chief revenue destroyer” claim (Blackwell obsoleting Hopper) risks alienating customers on six-year depreciation cycles. Gerstner sees brilliance in Nvidia’s execution—35,000 employees, a top-tier supply chain, and a four-generation roadmap—but both flag government action as the wildcard. Tariffs and export controls could bolster Huawei, though Huang shrugs off near-term impacts.

AI’s Consumer Frontier: OpenAI’s Lead, Margin Mysteries

In consumer AI, OpenAI’s ChatGPT reigns with 400 million weekly users, supply-constrained despite new data centers in Texas. Gerstner calls it a “winner-take-most” market—DeepSeek briefly hit #2 in app downloads but faded, Grok lingers at #65, Gemini at #55. “You need to be 10x better to dent this inertia,” he says, predicting a Q2 product blitz. Gurley agrees the lead looks unassailable, though Meta and Apple’s silence hints at brewing counterattacks.

Gurley’s “negative gross margin AI theory” probes deeper: many AI firms, like Anthropic via AWS, face slim margins due to high acquisition and serving costs, unlike OpenAI’s direct model. With VC billions fueling negative margins—pricing for share, not profit—and compute costs plummeting, unit economics are opaque. Gerstner contrasts this with Google’s near-zero marginal costs, suggesting only direct-to-consumer AI giants can sustain the capex. OpenAI leads, but Meta, Amazon, and Elon Musk’s xAI, with deep pockets, remain wildcards.

The Next 90 Days: Pivot or Peril?

The next 90 days will define 2025. April 2 tariffs could spark a trade war or a fairer field; tax cuts and deregulation promise growth, but AI’s fate hinges on export policies. Gerstner’s optimistic—Nvidia at 20x earnings and M&A’s resurgence signal resilience—but Gurley warns of overreach. A trillion-dollar tariff wall or a Huawei-led AI surge could upend it all. As Gurley puts it, “We’ll turn over a lot of cards soon.” The world watches, and the outcome remains perilously uncertain.

March 21, 2025
Is DeepSeek’s Terms of Use Normal or Sketchy AF? A Plain English Breakdown

Let’s dive into the DeepSeek Terms of Use and Privacy Policy, updated as of January 20 and February 14, 2025, respectively, and figure out if this is just standard tech company stuff or something that raises red flags. DeepSeek, a Chinese AI company run by Hangzhou DeepSeek Artificial Intelligence Co., Ltd. and its affiliates, offers generative AI services—think chatbots and text generation tools. Their legal docs outline how you can use their services and what they do with your data. But is it par for the course, or does it feel sketchy? Let’s break it down in plain English and weigh the vibes.

What’s DeepSeek Offering?

DeepSeek’s services let you interact with AI models that churn out text, code, or tables based on what you type in (your “Inputs”). You get responses (called “Outputs”), and the company uses big neural networks trained on tons of data to make this happen. They’re upfront that the tech’s always evolving, so they might tweak, add, or kill off features as they go. They also promise to keep things secure and stable—at least as much as other companies do—and let you complain or give feedback if something’s off.

Normal or Sketchy?

This part’s pretty standard. Most tech companies, especially in AI, have similar setups: you input stuff, they spit out answers, and they reserve the right to change things. The “we’ll keep it secure” promise is boilerplate—vague but typical. Nothing screams sketchy here; it’s just how these platforms roll.

Signing Up and Your Account

You need an account to use DeepSeek, and they want your email or a third-party login (like Google). They say it’s for adults, and if you’re under 18, you need a guardian’s okay. You’ve got to give real info, keep your password safe, and not hand your account to anyone else. If you lose it or someone hacks it, you can ask for help, but you’re on the hook for anything done under your name. You can delete your account, but they might hang onto some data if the law says so.

Normal or Sketchy?

Totally normal. Every app from Netflix to X has account rules like this—real info, no sharing, your fault if it gets compromised. The “we keep data after you delete” bit is standard too; laws often force companies to hold onto stuff for compliance. No red flags yet.

What You Can and Can’t Do

Here’s where they lay down the law: you get a basic right to use the service, but they can yank it anytime. You can’t use it to make hateful, illegal, or creepy stuff—like threats, porn, or fake celebrity accounts (unless it’s labeled parody). No hacking, no stealing their code, no reselling their service. If you share AI-generated content, you’ve got to check it’s true and tag it as AI-made. They can scan your inputs and outputs to make sure you’re playing nice.

Normal or Sketchy?

This is par for the course. Every platform has a “don’t be a jerk” list—X, YouTube, you name it. The “we can revoke access” part is standard; it’s their service, their rules. Checking your content isn’t weird either—AI companies like OpenAI do it to avoid legal headaches. The “label it as AI” rule is newer but popping up more as fake content worries grow. Nothing sketchy; it’s just them covering their butts.

Your Inputs and Outputs

You own what you type in and what the AI spits out, and you can use it however—personal projects, research, even training other AI (cool, right?). But they might use your inputs and outputs to tweak their system, promising to scramble it so no one knows it’s yours. They warn the outputs might be wrong, so don’t bet your life on them—especially for big stuff like legal or medical advice.

Normal or Sketchy?

Mostly normal, with a twist. Letting you own outputs and use them freely is generous—some AI companies (looking at you, certain competitors) claim rights to what their models make. Using your data to improve their AI is standard; Google and others do it too, with the same “we’ll anonymize it” line. The “outputs might suck” disclaimer is everywhere in AI—nobody wants to get sued over a bad answer. The twist? They’re based in China, and data laws there can be murky. Not sketchy on its face, but the location might make you squint.

Who Owns the Tech?

DeepSeek owns all their code, models, and branding. You can’t use their logos or try to copy their tech without permission. Simple enough.

Normal or Sketchy?

Bog-standard. Every company guards its intellectual property like this. No surprises, no sketchiness.

If Something Goes Wrong

If you think they’re ripping off your ideas or breaking rules, you can complain via email or their site. They’ll look into it. If you break their rules, they can warn you, limit your account, or ban you—no notice required. They’re not liable if the service flops or gives you bunk info, and you’ve got to cover their back if your screw-up costs them money.

Normal or Sketchy?

Normal, if a bit harsh. The “we can ban you anytime” clause is in every terms of service—X has it, so does every game app. The “we’re not responsible” and “you pay if you mess up” bits are classic corporate shields. It’s not cuddly, but it’s not sketchy—just self-protective.

Privacy Stuff

They collect your account details, what you type, your device info, and rough location (via IP). They use it to run the service, improve their AI, and keep things safe. They might share it with their team, service providers (like payment processors), or cops if the law demands it. You’ve got rights to see, fix, or delete your data, but it’s stored in China, and they don’t take kids under 14.

Normal or Sketchy?

Mostly normal, with a catch. Data collection and sharing are what every tech company does—X grabs your IP and tweets, Google slurps everything. Rights to access or delete are standard, especially with privacy laws like GDPR influencing global norms. The China storage is the catch—data there can be subject to government snooping under laws like the National Intelligence Law. Not sketchy by design, but it’s a wild card depending on your trust level.

Legal Fine Print

Chinese law governs everything, and disputes go to a court near their HQ in Hangzhou. They can update the terms anytime, and if you keep using the service, you’re cool with it.

Normal or Sketchy?

Normal-ish. Picking their home turf for law and courts is typical—X uses U.S. law, others pick wherever they’re based. The “we can change terms” bit is everywhere too. The China angle might feel off if you’re outside that system, but it’s not inherently sketchy—just inconvenient.

The Verdict

DeepSeek’s terms and privacy rules are mostly par for the course. They’re doing what every AI and tech company does: setting rules, grabbing data, dodging liability, and keeping their tech theirs. The “you own outputs” part is a nice perk, and the content rules align with industry norms as AI gets more regulated. The sketchy vibes creep in with the China factor—data storage and legal oversight there aren’t as transparent as, say, the U.S. or EU. If you’re chill with that, it’s standard fare. If not, it might feel off. Your call, but it’s not a screaming red flag—just a “hmm, okay” moment.

March 6, 2025
The AI Revolution Unveiled: Jonathan Ross on Groq, NVIDIA, and the Future of Inference
TL;DR

Jonathan Ross, Groq’s CEO, predicts inference will eclipse training in AI’s future, with Groq’s Language Processing Units (LPUs) outpacing NVIDIA’s GPUs in cost and efficiency. He envisions synthetic data breaking scaling limits, a $1.5 billion Saudi revenue deal fueling Groq’s growth, and AI unlocking human potential through prompt engineering, though he warns of an overabundance trap.

Detailed Summary

In a captivating 20VC episode with Harry Stebbings, Jonathan Ross, the mastermind behind Groq and Google’s original Tensor Processing Unit (TPU), outlines a transformative vision for AI. Ross asserts that inference—deploying AI models in real-world scenarios—will soon overshadow training, challenging NVIDIA’s GPU stronghold. Groq’s LPUs, engineered for affordable, high-volume inference, deliver over five times the cost efficiency and three times the energy savings of NVIDIA’s training-focused GPUs by avoiding external memory like HBM. He champions synthetic data from advanced models as a breakthrough, dismantling scaling law barriers and redirecting focus to compute, data, and algorithmic bottlenecks.

Groq’s explosive growth—from 640 chips in early 2024 to over 40,000 by year-end, aiming for 2 million in 2025—is propelled by a $1.5 billion Saudi revenue deal, not a funding round. Partners like Aramco fund the capital expenditure, sharing profits after a set return, liberating Groq from financial limits. Ross targets NVIDIA’s 40% inference revenue as a weak spot, cautions against a data center investment bubble driven by hyperscaler exaggeration, and foresees AI value concentrating among giants via a power law—yet Groq plans to join them by addressing unmet demands. Reflecting on Groq’s near-failure, salvaged by “Grok Bonds,” he dreams of AI enhancing human agency, potentially empowering 1.4 billion Africans through prompt engineering, while urging vigilance against settling for “good enough” in an abundant future.

The Big Questions Raised—and Answered

Ross’s insights provoke profound metaphorical questions about AI’s trajectory and humanity’s role. Here’s what the discussion implicitly asks, paired with his responses:
- What happens when creation becomes so easy it redefines who gets to create?
- Can an underdog outrun a titan in a scale-driven game?
- What’s the human cost when machines replace our effort?
- Is the AI gold rush a promise or a pipe dream?
- How do we keep innovation’s spirit alive amid efficiency’s rise?
- What’s the price of chasing a future that might not materialize?
- Will AI’s pursuit drown us in wasted ambition?
- Can abundance liberate us without trapping us in ease?
Jonathan Ross’s vision is a clarion call: AI’s future isn’t just about faster chips or bigger models—it’s about who wields the tools and how they shape us. Groq’s battle with NVIDIA isn’t merely corporate; it’s a referendum on whether innovation can stay human-centric in an age of machine abundance. As Ross puts it, “Your job is to get positioned for the wave”—and he’s riding it, challenging us to paddle alongside or risk being left ashore.
February 20, 2025
The DeepSeek Revolution: Financial Markets in TurmoilA Sputnik Moment for AI and Finance

On January 27, 2025, the financial markets experienced significant upheaval following the release of DeepSeek’s latest AI model, R1. This event has been likened to a modern “Sputnik moment,” highlighting its profound impact on the global economic and technological landscape.

Market Turmoil: A Seismic Shift

The unveiling of DeepSeek R1 led to a sharp decline in major technology stocks, particularly those heavily invested in AI development. Nvidia, a leading AI chip manufacturer, saw its shares tumble by approximately 11.5%, signaling a potential loss exceeding $340 billion in market value if the trend persists. This downturn reflects a broader market reassessment of the AI sector’s financial foundations, especially concerning the substantial investments in high-cost AI infrastructure.

The ripple effects were felt globally, with tech indices such as the Nasdaq 100 and Europe’s Stoxx 600 technology sub-index facing a combined market capitalization reduction projected at $1.2 trillion. The cryptocurrency market was not immune, as AI-related tokens experienced a 13.3% decline, with notable losses in assets like Near Protocol and Internet Computer (ICP).

DeepSeek R1: A Paradigm Shift in AI

DeepSeek’s R1 model has been lauded for its advanced reasoning capabilities, reportedly surpassing established Western models like OpenAI’s o1. Remarkably, R1 was developed at a fraction of the cost, challenging the prevailing notion that only vast financial resources can produce cutting-edge AI. This achievement has prompted a reevaluation of the economic viability of current AI investments and highlighted the rapid technological advancements emerging from China.

The emergence of R1 has also intensified discussions regarding the effectiveness of U.S. export controls aimed at limiting China’s technological progress. By achieving competitive AI capabilities with less advanced hardware, DeepSeek underscores the potential limitations and unintended consequences of such sanctions, suggesting a need for a strategic reassessment in global tech policy.

Broader Implications: Economic and Geopolitical Considerations

The market’s reaction to DeepSeek’s R1 extends beyond immediate financial losses, indicating deeper shifts in economic power, technological leadership, and geopolitical influence. China’s rapid advancement in AI capabilities signifies a pivotal moment in the global race for technological dominance, potentially leading to a reallocation of capital from Western institutions to Chinese entities and reshaping global investment trends.

Furthermore, this development reaffirms the critical importance of computational resources, such as GPUs, in the AI race. The narrative that more efficient use of computing power can lead to models exhibiting human-like intelligence positions computational capacity not merely as a tool but as a cornerstone of this new technological era.

DeepSeek’s Strategic Approach: Efficiency and Accessibility

DeepSeek’s strategy emphasizes efficiency and accessibility. The R1 model was developed using a pure reinforcement learning approach, a departure from traditional methods that often rely on supervised learning. This method allowed the model to develop reasoning capabilities autonomously, without initial reliance on human-annotated datasets.

In terms of cost, DeepSeek’s R1 model offers a significantly more affordable option compared to its competitors. For instance, where OpenAI’s o1 costs $15 per million input tokens and $60 per million output tokens, DeepSeek’s R1 costs $0.55 per million input tokens and $2.19 per million output tokens. This cost-effectiveness makes advanced AI technology more accessible to a broader audience, including developers, businesses, and educational institutions.

Global Reception and Future Outlook

The global reception to DeepSeek’s R1 has been mixed. While some industry leaders have praised the model’s efficiency and performance, others have expressed skepticism regarding its rapid development and the potential implications for data security and ethical considerations.

Looking ahead, DeepSeek plans to continue refining its models and expanding its offerings. The company aims to democratize AI by making advanced models accessible to a wider audience, challenging the current market leaders, and potentially reshaping the future landscape of artificial intelligence.

Wrap Up

DeepSeek’s R1 model has not merely entered the market; it has redefined it, challenging established players, prompting a reevaluation of investment strategies, and potentially ushering in a new era where AI capabilities are more evenly distributed globally. As we navigate this juncture, the pertinent question is not solely who will lead in AI but how this technology will shape our future across all facets of human endeavor. Welcome to 2025, where the landscape has shifted, and the race is on.

January 27, 2025