Musical Chairs

$1.25 billion — Anthropic's monthly rent line to xAI for compute, disclosed today inside SpaceX's S-1 filing. Per month. Roughly $40 billion total through May 2029. Either side can terminate on 90 days' notice. It is, by some margin, the largest recurring infrastructure payment in software history — and it landed on the same day OpenAI's confidential IPO filing started circulating, SpaceX submitted its S-1, and Anthropic's private secondaries traded to $1 trillion. The public-market trillion-dollar club has eight seats. Three lobsters are circling. Two of them sit down.

Stop signing multi-year AI vendor contracts at IPO-window prices. Tomasz Tunguz published a chart this morning showing Google's API prices have tripled in a year and OpenAI's are up 40%. Every lab is hiking. Pichai told I/O yesterday that customers are "blowing through their annual token budgets, and it's only May." The vendor leading any AI category today may not be leading twelve months from now. Trial. Renew on demonstrated performance. Do not underwrite the lab's duration risk on your team's behalf.

Re-run your AI ROI math against the change-management clock — not the capex clock. Acrisure cut 2,250 jobs today citing AI directly. KPMG embedded Claude in 276,000 seats today. Different clocks. The layoff happens in eighteen months. The Claude rollout into global tax and PE workflows is a multi-year program. The Bessemer math the IPO market is pricing assumes both clocks run together. They do not. Plan your 2027 AI revenue on the slow clock and your 2026 AI savings on the fast one.

If you hold Anthropic, OpenAI, or SpaceX through a fund, ask your GP about exit strategy this week. The IPO is not the public's entry point. It's the private fund's exit point. The easy multiple ran in private secondaries between 2024 and now. Anthropic moved from $20B to $900B in eighteen months. OpenAI did multiple secondaries at successively higher marks. The retail buyer and the public-market index are being asked to take supply at peak optimism. If you're on the early side of the cap table, the bell at the NYSE is the bell on your trade.

Keeping up with AI is hard. We know — we do this daily. If any of these action items strike a chord, or you simply want an Outsider perspective on how you can improve your business with AI, we'd love to hear from you.

Outsider Labs Advisory →

Wednesday, May 20

1. The trillion-dollar IPO race has two chairs and three labs. OpenAI's confidential S-1 is reportedly filing this week at a $1 trillion target. SpaceX filed today for a June 12 IPO claiming a $28.5 trillion total addressable market — "the largest TAM in human history," by their own words — most of it in AI compute. Anthropic raised primary at $900 billion with private secondaries hitting $1T. The eight existing trillion-dollar public companies all earned the seat over decades. The math doesn't clear all three. (Full analysis below.)

2. Nvidia went parabolic. $81.62 billion quarter, up 85% year over year, $91B Q2 guide against a $86.8B consensus, AWS adding more than a million Blackwell and Rubin GPUs this year alone. Jensen Huang's call quote: "demand has gone parabolic." The picks-and-shovels trade is still the trade. Michael Dell, on stage at Dell Technologies World today, said the supply chain crisis now extends into 2028 and possibly 2030. If you're a CIO sitting on procurement orders, his exact advice was: buy now.

3. OpenAI solved a famous open math problem with a general-purpose reasoning model. The planar unit distance problem, first posed by Paul Erdős in 1946 — 80 years open. The proof came from a model that wasn't built for math. The OpenAI researcher who saw the output: "I had trouble sleeping for the first couple of nights." On the same day CNBC ran a piece arguing that cheap Chinese inference is eating OpenAI's pricing power, OpenAI handed itself an S-1 talking point about capability the cheap models can't yet replicate. The Erdős result is the bull case for premium pricing, served on a silver tray.

4. Acrisure cut 2,250 jobs and named AI. Eleven percent of the Grand Rapids insurance firm's global workforce, on top of 400 cut in October. CEO Greg Williams in the internal letter: "It's about changing how work gets done." Same week, the company opened a $184 million amphitheater downtown. The numerator on every AI ROI deck just got a face. The denominator — actual incremental revenue from automated work — is much harder to find on the 10-Q.

5. AI-discovered vulnerability-to-exploit windows have collapsed to 20 hours. Per the Cloud Security Alliance report cited in IAPP today: in 2018, the gap from vulnerability discovery to confirmed exploit was 2.3 years. In 2025 it was 23.2 days. In 2026 it is roughly 20 hours. The Cybersecurity Information Sharing Act sunsets in September without reauthorization. If you run a security operation, you have eighteen months to redesign for an attack tempo that no human team can match.

Two Chairs, Three Lobsters

The public-market trillion-dollar club has eight members. Eight. Apple founded it in 2018, Microsoft followed, then Nvidia (now $5.5T after today's blowout quarter), Alphabet, Amazon, Meta, Saudi Aramco, and Berkshire Hathaway. Every single member earned its seat with decades of receipts. Every single one generates extraordinary free cash flow. Not one of them is in a capex emergency relative to its operating earnings. The club is exclusive by construction. The public market doesn't admit new members because a company shows up with a story. It admits them when a company shows up with a balance sheet.

Three labs are crashing the gate this month. OpenAI, filing confidential S-1 this week at $1 trillion. Anthropic, at $900B primary and $1T in secondaries. SpaceX, S-1 filed today, June 12 IPO, valuation expected at or above $1T. Two chairs. Three lobsters. The music is playing.

The pair trade the smartest allocators are running, even without naming it: long SpaceX (or Anthropic), short OpenAI. Run the chain. SpaceX's S-1 targets $185/kg to orbit on Starship, eight times cheaper than Falcon 9's current internal cost. The vehicle has flown nine test flights. Falcon 9 has 450+ successful landings. But if Starship clears, Cursor — which signed a $10B call option with SpaceX in April — gets subsidized compute and commoditizes the coding-agent layer that is OpenAI's most credible enterprise revenue line. Long SpaceX is structurally short OpenAI through Cursor.

Anthropic wins on every dimension that matters — models, talent, enterprise share (34.4% on Ramp vs OpenAI's 32.3%), product velocity (KPMG 276K seats + Bristol Myers Squibb across drug discovery, today), and moral posture. The wrinkle is the $1.25B/month. They are renting compute from a competitor on a 90-day-terminable contract. Krishna Rao's own math explains why: "If you buy too much, you go out of business. If you buy too little, you can't serve your customers." The math favors renting today. The math also has a quarterly reset button held by Elon Musk.

There is a feature of the IPO transition that the recent boom in private secondaries has obscured. Private capex is funded by patient money — venture LPs, sovereign funds, multi-year theses. Public capex is funded by operating cash flow, debt against hard assets, or stock issuance at a price the market is willing to pay. A freshly-public lab that misses an enterprise revenue quarter watches its stock fall. A falling stock can't be used to fund operations through issuance without diluting existing shareholders into the next leg down. Anthropic's $15 billion in annual rent doesn't pause when the stock does. The 2022-2023 SaaSpocalypse showed exactly this — public software companies pre-emptively marked down because institutional investors saw the macro change before any earnings miss. They survived. They spent two years in the wilderness explaining their capex math at half the valuation.

There is an implicit Microsoft put under OpenAI. Microsoft owns ~49% of the for-profit subsidiary's economic interest, has board influence, and every reason to absorb the company at distress before a competitor does. That put exists at a price. That price is some discount to wherever OpenAI's market cap clears. Microsoft's fiduciary duty runs to Microsoft's shareholders, not OpenAI's IPO buyers. The investor who pays $1 trillion at the bell does not get the put at $1 trillion.

When the music stops, you want to know which lobster you bet on. Long the operator. Long the optionality. Short the brand. Find your chair before the song ends.

At COAI today: Full analysis of the Musical Chairs trade, the Starship economics gate, the post-IPO capex squeeze, and the Microsoft put pricing problem at getcoai.com.

— Harry and Anthony

Sources: