Anthropic's S-1: $808 Billion in Compute Bets and 19% Margins

Anthropic has locked in as much as $808 billion in compute and chip commitments ahead of its IPO, according to the S-1 prospectus it published this week. That figure includes $45 billion for data center capacity from Microsoft, $110 billion with Amazon, $121 billion with SpaceX, and $115 billion with Google, plus another $125 billion to $150 billion in additional commitments, $45 billion with neocloud Nscale, and $41 billion with rival Lambda. Add at least $20 billion in AMD chip purchases and a colossal $161 billion Broadcom leasing deal, and the total lands between roughly $602 billion and $808 billion — with some contracts cancelable and some not.

The filing gives investors the clearest look yet at the economics of a frontier AI lab. And by extension, it sketches the financial physics of the entire industry.

What do the operating numbers actually show?

As of June, Anthropic generated more than two dollars in revenue for every dollar spent on compute. That's better unit economics than many AI skeptics assumed. Once other operating costs were included, the company posted a 19% adjusted operating margin in its most recent reported quarter — though that figure excludes stock-based compensation, non-cash charitable contributions, and settlements of copyright claims from book authors.

Unit economics are also improving fast: revenue per dollar of compute cost rose 50% between the first and second quarters of 2026. Operations generated $3.9 billion in free cash flow in the first half of 2026, and the company spent $5.6 billion on compute and infrastructure over the same six months.

Anthropic warns the trends may not hold. "As significant incremental compute is brought online in future periods, the associated expense will be recognized in advance of the full revenue that the compute will ultimately support, and we expect operating margin and Adjusted Operating Margin may compress in those periods," the S-1 states. The company also cautions that price competition could "adversely affect our ability to achieve or sustain profitability." The Information reported earlier this week, under the headline "Anthropic's margin is OpenAI's opportunity," that OpenAI is already winning share by undercutting Anthropic on price.

Revenue growth tells a two-sided story. Revenue grew 1,300% between the first half of 2025 and the first half of 2026 — a rate that would comfortably cover the compute bill, but one that almost certainly won't last.

Who actually controls the company?

Anthropic's governance concentrates power tightly. A Long-Term Benefit Trust (LTBT) of five members can appoint the majority of the board, while seven cofounders, through a Founders LLC, hold 50.1% of the voting power to elect the rest. Employees holding special Class E stock can help break certain deadlocks and vote on some LTBT changes. Ordinary investors get Class A shares and will, in practice, always be outvoted — their votes count only when the cofounders split. Backers Amazon and Google hold Class N shares with no voting power at all.

For a company that warns in its own prospectus about authoritarian governments using AI to "create high-tech surveillance states, deepen repression, and make themselves much more difficult to dislodge" — and one that crowdsourced input on the constitution guiding its Claude models — the lack of trust in shareholders looks jarring.

What risks does the filing flag?

The risk factors run from cyberattacks by bad actors to the possibility that the U.S. government nationalizes the company. The biggest is existential. Bloomberg's Matt Levine flagged Anthropic's use of passive voice on this point: "Concerns have been raised both within our organization and by other researchers, regulators, policymakers, and industry participants that sufficiently advanced AI systems could, if not developed, deployed, and operated with appropriate safeguards, oversight, and security measures, pose catastrophic or existential risks to humanity, particularly if such systems were misused, replicated, weaponized, or allowed to operate with a high degree of autonomy." Among those raising those concerns: Anthropic CEO Dario Amodei himself.

What did the S-1 leave out?

Three gaps stand out.

• No Q3 results. Critics on social media noted the omission, even though Anthropic briefed existing investors on those figures at its October 14 investor day. Imputed from S-1 disclosures, Q3 revenue ran 86 times the same period in 2024 — roughly $16.8 billion, matching the leaked investor-day figure. That's about 45% quarterly growth, well below the 147% seen between Q1 and Q2.
• Convertible notes. Some critics compared Anthropic's use of convertible notes to pay compute costs — mostly issued to Amazon and Google, possibly at discounted rates — to dot-com-era tactics of paying suppliers in warrants before an IPO. What happens once the public company has to pay cash?
• No payment schedule. Obligations come due between 2026 and 2033, in some cases 2036, or over one to 20 years. Analysts want a precise timetable to judge whether revenue can cover the bills.

Answers may arrive before Anthropic's shares begin trading, expected between November 9 and Thanksgiving. Until then, investors are pricing an $808 billion bet on growth that's already decelerating.