Jensen Sees $1 Trillion. YC Just Had Its Best Batch Ever. And a Judge Says the Pentagon Is "Punishing" Anthropic.

Hi, I'm Kenneth. Welcome to this week's edition of Tech Venture Navigator.

This week's edition covers two of the most packed weeks in AI and venture that I can remember. Nvidia GTC happened. YC W26 Demo Day happened. GPT-5.4 shipped. And a federal judge yesterday told the Pentagon that banning Anthropic "looks like an attempt to cripple" the company. Let me get straight into it.

Brought to you by Kondo.

Jensen Huang delivered a nearly three-hour keynote at SAP Center to 30,000+ attendees from 190 countries. The headlines:

$1 trillion in purchase orders. Last year, Nvidia projected $500B in high-confidence demand through 2026. Huang now sees $1T through 2027 for Blackwell and Vera Rubin combined. "Now, does it make any sense? That's what I'm going to spend the rest of the keynote talking about."

Groq 3 LPU unveiled. Nvidia's first chip from Groq, the startup it mostly acquired for $20B in December. The Language Processing Unit is built for inference, not training. Shipping Q3 2026. A full rack of 256 LPUs sits alongside the Vera Rubin system. This is Nvidia's answer to everyone asking "what about inference costs?"

Vera Rubin platform detailed. Tightly integrated CPU-GPU architecture for large-scale AI workloads. Expected to deliver 3.3x the inference performance of Blackwell Ultra.

Kyber prototype shown. Nvidia's next rack architecture after Rubin. 144 GPUs in vertical compute trays for higher density and lower latency. Shipping 2027 as Vera Rubin Ultra.

Feynman architecture previewed. The 2028 roadmap. Potentially built on TSMC's 1.6nm process. Jensen teased "chips the world has never seen before."

Uber autonomous fleet. Nvidia Drive AV software powering Uber across 28 cities on four continents by 2028, starting with LA and San Francisco next year. Huang called it "the ChatGPT moment for autonomous driving." New partners: Nissan, BYD, Geely, Isuzu, Hyundai.

NemoClaw launched. Nvidia's enterprise-grade implementation of OpenClaw, built on Nvidia Agent Toolkit. Jensen urged every company to "adopt an OpenClaw strategy" and called AI agents "the next computing platform."

AI token budgets. Jensen proposed giving Nvidia engineers annual AI token budgets alongside salaries, with compute credits worth nearly half base pay. "AI compute access is emerging as a key recruitment lever."

Sources: CNBC | Tom's Hardware | TechRadar | Nvidia Blog | ASV Insider

Rebel Fund's analysis: 35% of W26 startups score in the top 20% of all YC companies ever evaluated. No previous batch has come close. The entire distribution curve has shifted right, not just the top companies. More startups predicted to succeed. Fewer predicted to become zombies.

One company walked into Demo Day at $27M ARR.

The composition: 64% B2B. Only ~5% consumer. Heavily weighted toward infrastructure and hard technical problems. Healthcare is ~10% of the batch. Legal tech ~4% (following Harvey and Legora's fast unicorn runs). Physical-world AI is back: mines, drones, satellites, battery packs, chip design, radar for autonomous vehicles, cattle-mustering drones, space solar arrays.

Why it's stronger: Younger founders (historically correlates with better outcomes in YC data). More Bay Area concentration (geographic signal in Rebel's model). Far fewer consumer companies. More technical depth.

The YC asset class context (from Garry Tan): For investors who backed 3+ companies per Demo Day from 2018-2020, the bottom quartile returned 3.3x TVPI. Median: 5x. Top quartile: 8x. Top decile: 15x. The bottom quartile of Demo Day investing outperforms the top quartile of broader venture.

Multiple observers project this batch produces ~20 unicorns from 196 companies. A 10% hit rate against YC's historical 4.5%.
Sources: Rebel Fund / Jared Heyman | New Economies | TLDL | Y Combinator

GPT-5.4 launched March 5. The first general-purpose model with native computer use. It can operate your desktop, browser, and applications autonomously.

OSWorld-Verified: 75.0% success on desktop navigation. Human performance: 72.4%. GPT-5.2: 47.3%.

GDPval: Matches or beats industry professionals in 83% of knowledge work comparisons across 44 occupations.

Hallucinations down: 33% fewer false claims vs GPT-5.2. Full responses 18% less likely to contain errors.

Context window: Up to 1M tokens in the API. 47% fewer tokens on some tasks.

New Excel and Google Sheets plugins. GPT-5.4 Thinking for Plus/Team/Pro subscribers. GPT-5.4 Pro for maximum performance.

Sources: OpenAI | TechCrunch | Fortune | VentureBeat

Two data sets that look contradictory until you understand what's actually happening.

Lenny Rachitsky's State of the Product Job Market (March 24): 7,300+ open PM roles globally (highest in 3+ years, up 75% from early 2023 lows). 67,000+ open engineering roles (26,000 in the US). AI-specific roles are hockey-sticking. Design roles have plateaued (possibly the first AI casualty). Bay Area concentration increasing. Remote declining. Tech recruiter openings approaching 2022 peak levels, a leading indicator of sustained hiring.

Oracle: Cutting 20,000-30,000 employees to free $8-10B for AI data centres. $100B+ in debt. $156B in infrastructure commitments. US banks pulling back. Considering selling Cerner ($28.3B acquisition) to fund the buildout.

The paradox resolved: AI is creating jobs at AI companies and AI-adjacent roles while eliminating jobs at companies that need to fund AI infrastructure or whose work AI can now do. It's not "AI creates jobs" or "AI kills jobs." It's both, simultaneously, in different parts of the economy. The winners are people who can build with AI. The losers are people whose work AI can replace or whose employers need their salaries to buy GPUs.

Chris Tottman's GTM framework is worth reading alongside this data. His analysis shows B2B GTM effectiveness has fallen from 78% in 2018 to 47% in 2025. More than half of every sales and marketing dollar is structural waste. The core insight: your ACV is your GTM compass. If your go-to-market motion costs more than your price point can sustain, no amount of hiring fixes it. If you're a founder figuring out your sales motion, his piece is essential reading."

Sources: Lenny's Newsletter | Bloomberg | CIO

Prediction markets went from $50M weekly volume pre-2024 election to over $6B today. That's 100x in 24 months. Polymarket and Kalshi saw $4.2B+ on the 2024 election alone.

Now accessible through Robinhood, Coinbase, DraftKings, and FanDuel, unlocking 100M+ retail accounts already linked to bank accounts. Research projects the category could hit $1T annual trading volume by 2030.

Chamath's 60+ page deep dive covers the architecture, the competitive landscape, regulatory clarity post-Kalshi v. CFTC, and how prediction markets are being used in the private sector to surface intelligence. If you're in fintech or building anything adjacent to markets, this is the category to watch.

Source: Social Capital

VI. Anthropic Court Update: Judge Says Ban "Looks Like Punishment"

Yesterday's hearing. U.S. District Judge Rita Lin told the government the Pentagon's actions were "troubling" and appeared to be "an attempt to cripple" Anthropic rather than address a genuine security concern. She questioned whether the designation was proportionate: "That seems a pretty low bar."

The government's argument shifted notably. Their lawyer said the Pentagon worries Anthropic could "in the future take action to sabotage or subvert IT systems" during military operations. Anthropic noted this concern never came up during negotiations and only appeared in court filings.

150 retired federal and state judges filed an amicus brief supporting Anthropic. Microsoft filed a brief. OpenAI and Google DeepMind researchers filed briefs. Senator Elizabeth Warren wrote to Hegseth expressing concern.

Judge Lin expects to rule in the next few days. If the injunction is granted, Anthropic can continue doing business with government contractors while the case plays out. If not, the company has said it could lose billions.

Sources: CNBC | NPR | Fortune | Al Jazeera

1. Claude Certified Architect Curriculum (Anthropic, via AI Corner) Anthropic built a certification covering Claude Code, Agent SDK, API, and MCP. The exam is locked to partners, but the full curriculum is public. Five domains: Agentic Architecture (27% weight), Claude Code Config (20%), Prompt Engineering (20%), Tool Design/MCP (18%), Context Management (15%). The key production failure pattern across all domains: architecture problems, not model problems. If you're building on Claude, this is the technical benchmark. (AI Corner)

2. Stratechery: "Agents Over Bubbles" (Ben Thompson) Thompson's GTC analysis argues agentic AI is fundamentally changing the shape of compute demand. He no longer believes we're in a bubble because agents create sustained, recurring inference demand that justifies current infrastructure spending. His key insight: OpenClaw proves his thesis that OpenAI and Anthropic are sustainably differentiated through their integration of harness and model was wrong. Nvidia's inference pivot (Groq LPU) reflects genuine concern about a world dominated by just two companies. (Stratechery)

3. Nvidia State of AI Report (3,200+ respondents) 87% of enterprises say AI reduced annual costs. 25% report cost decreases greater than 10%. 86% say AI budgets increase in 2026. 44% deploying or assessing AI agents. Retail/CPG leads with 37% reporting 10%+ cost reduction. (Nvidia Blog)

We sit at the center of 550k+ founders, operators, and LPs. Our goal is to eliminate the "random walk" of fundraising by using our data to make the one warm connection that actually matters.

Thanks for reading this edition. If you got value from it, share it with a founder, teammate, or friend who’d use it too it keeps this community sharp and growing. Not for you? You can always update your preferences anytime.

Want more practical ideas on how to raise smarter and build better?
Join the crew below and follow me on LinkedIn to never miss an update.

If you’d like tailored help you can book time with me on hubble here.

— Kenneth Kelly
Founder @Tech Venture Navigator

^{What do you think about Tech Venture Navigator?}
💡 ^{Love it |} 🔥 ^{Useful | 👍 Solid | 🤷‍♂️ Okay |} 🚫 ^{Skip it
or Leave feedback here.}

^{Got ideas, topics, or frameworks you’d love to see covered? Just hit reply or drop me a message, we read every note.}