SpaceX's $2T Case, Nvidia's Shock Selloff, America Turns on AI, Trump Pulls AI Order, Bond Crisis?

Key points

• **Andrej Karpathy joins Anthropic to lead recursive self-improvement on the pre-training team.** Chamath frames the value: 'recursive self-learning puts these models on overdrive AND autopilot. You can potentially live out the order-of-magnitude-per-year improvement Moore's Law of AI.' **Gavin's response makes the bigger investment-strategy move: 'Anthropic was EBIT-positive per the WSJ in the most recent quarter. The returns-circular-funding bear case collapses if OpenAI + Anthropic + Gemini + Cursor + xAI + open source aggregate to $200-300-400B of ARR at high margins this year.'** **Direct triangulation with [Gavin's $11B-of-ARR-in-one-month framing on ILTB this week](/issues/2026-05-24)** — Karpathy hire is the talent capstone underneath the financial story.
• **SpaceX S1 filed — the largest IPO ever proposed (>2x Saudi Aramco).** Headline data: **$1.75T valuation, $75B raise, June 12 IPO (ticker SPCX).** Three business units: **Starlink** $11.4B revenue (+50%), $4.4B op income, 10M subscribers; **Space business** $4B revenue (+17%), -$650M op losses; **xAI** $3.2B revenue (+2x), -$6.4B op losses. $20B capex in 2025, 60% AI compute. **Polymarket: 71% chance SpaceX closes day one >$2T market cap.** Confirms [Brad Gerstner's June IPO date prediction from Issue 04 Milken hit](/issues/2026-05-10) — date held within a week, ticker confirmed.
• **The Elon Web Services breakout — Anthropic paying SpaceX $1.25B/month = $15B/year for Colossus 1 + parts of 2 = $45B over 3 years.** Plus Cursor acquisition adds $2-3B more. **'Both parties have a 90-day cancellation provision'** (Gavin's important footnote). **Gavin's structural point on why this can scale faster than expected**: SpaceX builds data centres in 122 days → 91 days → 66 days — *'they build dramatically faster and at lower cost than anyone else. With a clear offtake partner, there is no reason they can't stamp these data centres out really fast. Jensen will allocate GPUs to whoever can plug them in, turn them on, and convert electrons to tokens.'* **The xAI + SpaceX revenue confluence overnight is the most underpriced revenue line in the S1.**
• **The Cursor Composer 2.5 — 3-4 weeks of RL on Colossus 2 with Cursor's coding-data corpus → Pareto-dominant.** Gavin's strongest data-point of the week: 'Cursor allegedly has more tokens of coding data than exist on the public internet. After 3-4 weeks of appropriate RL using their data, Composer 2.5 is Pareto-dominant on the [intelligence vs cost](/issues/2026-05-24) frontier. Same base model as 2 (Kimmy K.25). This is amazing.' **Implication**: xAI/Cursor combo (post Elon-lets-them-on-Colossus) just became a credible frontier lab in 3-4 weeks. **Direct continuation of [Issue 04 Anthropic-monopoly thesis](/issues/2026-05-10)** — but now there's a fourth player on the frontier curve, not three.
• **Nvidia Q1 — the print everyone underestimated.** $81.6B revenue (+85% YoY, +20% QoQ), $58B net income, $48B FCF, 75% gross margins, $5.3T market cap. Capital return: **+$80B buyback, dividend 25x'd (1¢ → 25¢/share), 50% of FCF returned**. **Most-under-discussed disclosure: Nvidia's CPU business is $20B this year — 'overnight, one of the world's largest CPU manufacturers.'** Chamath: 'the domain-specific-architecture market evolution is happening *inside* Nvidia. This is what's so insane.' Cross-reference to [Reiner Pope on chip-design tradeoffs from this week's Dwarkesh](/issues/2026-05-24) — Nvidia is now competing with AMD, ARM, Intel on CPUs while extending the GPU lead.
• **DC-to-DC power architecture redesign as the next Jensen + Elon collaboration.** Chamath's deep-cut: *'Forget all this DC-to-AC-to-DC nonsense. If you could just do DC-to-DC — comes in as direct current, goes right to the rack as DC — it requires a fundamental rearchitecture. Jensen needs a design partner and Elon becomes a natural partner.'* **This is the structural read on why TerraFab + Colossus + xAI + Nvidia is the integrated stack — DC-to-DC requires a willing rack designer, a willing fab partner, and a willing data-centre operator.** Companion to [Gavin's TerraFab disclosure on ILTB this week](/issues/2026-05-24).
• **The AI PR crisis: Matthew Prince's 'measurers' memo + Zuckerberg's 'dystopian' layoffs.** Cloudflare laid off 20% of workforce despite record revenue/growth — labelled them 'the measurers' (people who manage people / measure data). Chamath: *'This was from the PR school of [redacted]. You could not have written a worse memo. You reduce humans to a label and now they have a scarlet letter when they try to get a different job.'* Zuckerberg paired layoffs with 'we're putting recording software on every employee's computer to train our models.' Plus: **3 commencement speeches booed AI this week (including Eric Schmidt's)**. **Trump's planned AI executive order was scrubbed an hour before announcement** — would have required some federal supervision of frontier models. Chamath/Friedberg link this to a CCP-funded anti-AI campaign hypothesis (echoing the Pax Silica AI-supply-chain framing from [Helberg in Issue 05](/issues/2026-05-17)).
• **Sham Sankar's reframe (Palantir CTO, friend of pod):** *'Stop breathlessly asking the model makers what they think. Go to the end user — ask the factory worker using AI saying I added a third shift. Ask the ICU nurse who has more time with patients. Tell THOSE stories.'* Gavin's example: hedge-fund manager's daughter with rare genetic mutation — used LLMs + existing safe drug to take her neuron firing rate from 30-40% to 80-90%, now spinning up a startup to formalise the cure. **The strongest counter-doom narrative on the tape this week — and the one Anthropic and OpenAI should be amplifying instead of dystopian roadmaps.**
• **Trump-Xi summit follow-up (continues from [Issue 05 All-In coverage](/issues/2026-05-17)).** No grand deal materialised. Putin sitting with Xi same day as the US visit. Chamath: *'There's what you see on the surface and what happens behind closed doors. I think there's a way to divide up the gameboard in a way that helps them and helps us.'* **Gavin's contrarian take on selling Nvidia chips to China**: *'Selling deprecated Nvidia GPUs to China LOWERS the odds they develop their own alternative ecosystem. It's the highest-probability path for keeping America ahead in AI.'* Friedberg's geopolitical framing: 'After 1945 the US had the nuclear monopoly and could have controlled the world forever. Instead we rebuilt Germany + Japan. That's testament to American spirit. AI dominance probably leads to another Pax Americana — informed by AI.'
• **Macro flashing red — but 'best house in worst neighbourhood' frame holds.** 10-year hit 4.6% (Bessent's goal was sub-4%). **Japan 30-year at 5.1% — highest ever recorded.** UK yields highest since GFC. Germany highest since 2011. **Korea retail investors borrowing record amounts for AI chip stocks** (echoes 2024 crypto peak). Survey of professional forecasters projects May CPI at 6%. **Friedberg activates 'Dr Doom':** 'Global debt-to-GDP 310%. The spending problem at federal/state/local plus every country creates a cascading effect — eventually massive inflation, money printing, asset inflation to service the debt, spiral takes off.' **Gavin's counter**: 'rates going up + 2026 AI fundamentals can both be true. The tech bubble happened with 10y/30y much higher than today. Nvidia at low-teens multiple of real earnings vs Cisco at 100x forward earnings.' **The dollar-crisis-or-not framing is the macro question hanging over the whole AI trade — file under 'watch carefully through September.'**
• **'AI is seasonal' — a sub-thread worth flagging.** Gavin's framing: 'AI fundamentals appear a little seasonal too. In the past, college students use less ChatGPT/Claude in summer, people work less when weather's nice. With agentic AI, will the fundamentals still be seasonal? We will see.' **Operating implication**: if AI demand follows a sin-wave with summer trough, the Q3 prints from OpenAI/Anthropic/Cursor might disappoint sentiment despite the underlying trajectory. **Worth tracking as the bear-case-trigger candidate for September-October 2026.**

Notable quotes

Themes

• Karpathy joins Anthropic + Anthropic EBIT-positive = ROI bear-case collapse
• SpaceX S1 $1.75T June IPO + Elon Web Services $15B/yr Anthropic deal
• Cursor Composer 2.5 Pareto-dominant after 3wks on Colossus = xAI/Cursor frontier comeback
• Nvidia $20B CPU business overnight + DC-to-DC redesign as Jensen+Elon next move
• AI PR crisis (Cloudflare 'measurers' memo + Zuckerberg dystopia) + Trump EO scrubbed