Anthropic had $95.6 billion of cash and cash equivalents and $36.3 billion of short-term investments on 30 September 2026, its amended Form S-1 shows, against approximately $69.8 billion of contractual obligations expected to be paid in the following 12 months. By our arithmetic, that is about 1.9 times; counting the $14.8 billion available under its credit facility, about 2.1 times.
Those 12 months are the near end of an aggregate committed spend of approximately $895.4 billion, substantially all of it for compute and infrastructure. This page sets out how the money comes in, where it is committed to go, why the loss line is so large, and what the filing still leaves blank before the IPO.
Business Pill · Run-Rate Revenue
A one-minute Business Engineer explainer of run-rate revenue. It explains the general idea only and does not discuss Anthropic or its IPO.
The key insight: As we read the S-1/A, Anthropic’s near-term position and its long-term commitments are different questions: $131.9 billion of cash and short-term investments covers about 1.9 times the $69.8 billion due within a year, by our arithmetic, while the full committed spend is approximately $895.4 billion.
The S-1/A reports revenue of $33,242 million for the first nine months of 2026, including $16,956 million in the third quarter. The second quarter was the only one in the filing with operating income: $974 million on $11,555 million of revenue. The third quarter had an operating loss of $567 million, and the nine months an operating loss of $1,260 million.
Operating activities provided $6,569 million of cash in the nine months, investing activities used $36,546 million and financing activities provided $110,735 million, according to the filing. It says Anthropic had raised an aggregate of $141.1 billion of capital by 30 September 2026.
The filing defines Run-Rate Revenue as consumption-based fees over a trailing 28-day period multiplied by 13, plus monthly subscription fees multiplied by 12. The amended filing leaves the latest figure blank: “our Run-Rate Revenue was $ as of October 31, 2026”. As a simpler yardstick, third-quarter revenue times four is about $67.8 billion, by our arithmetic; that is not Anthropic’s run-rate measure.
Anthropic’s cash and cash equivalents, short-term investments and available credit facility at 30 September 2026, against contractual obligations expected to be paid in the following 12 months and nine-month 2026 revenue. Source: Anthropic Form S-1/A (SEC EDGAR, 2 November 2026). The 1.9x ratio is our arithmetic.
Where the Money Is Committed to Go
The filing describes its most significant compute arrangements one by one. Rent under five-year equipment leases with Broadcom and financing entities totals approximately $161.2 billion excluding interest. Anthropic committed to spend at least approximately $121.3 billion under a hosting agreement with Google between April 2026 and July 2033, and at least $110.0 billion with Amazon between May 2026 and April 2036.
It agreed to spend up to $120.5 billion with SpaceXAI over terms extending into 2029, at least $44.9 billion with Microsoft between November 2026 and June 2033, up to approximately $44.6 billion with Nscale and approximately $41.3 billion with Lambda. Outside the obligations table, it estimates $125.0 billion to $150.0 billion for additional TPU compute and expects AMD spending to exceed $20.0 billion.
Those seven stated amounts add up to about $643.8 billion, by our arithmetic. They mix minimums, up-to amounts and lease rent, so the sum is indicative only. Of the $895.4 billion total, approximately $447.5 billion is expected before 31 December 2030, about half by our arithmetic.
The SpaceXAI fee is the largest monthly line the filing names: “approximately $3.1 billion in the aggregate per month once all capacity has been delivered, which we expect in December 2026”. That is about $37.2 billion a year, and about 55% of the third quarter’s average monthly revenue of about $5.65 billion, by our arithmetic. After initial non-cancelable periods ending in January and March 2027, either party may cancel each agreement on 90 days’ notice, the filing says.
Update, 5 Nov, 19:50Z: two-thirds of the named compute deals are owed regardless of use. The eight compute counterparties the S-1/A names, listed above, total $663.8 billion by our arithmetic, counting AMD at the filing’s expectation that spending will exceed $20.0 billion. Of that, $437.4 billion, or 66%, is the Broadcom lease rent plus the Google, Amazon and Microsoft minimum spends; for Google and Amazon the filing says the spend is owed “regardless of actual usage”.
The other $226.4 billion is capped, billed only as capacity arrives, cancellable on notice or an estimate: up to $120.5 billion with SpaceXAI, up to about $44.6 billion with Nscale, about $41.3 billion with Lambda and AMD’s Helios systems, where “The agreement does not specify a fixed dollar commitment”. Broadcom, Google and SpaceXAI together are $403.0 billion, 61% of the total, by our arithmetic. A post on X by @StockMKTNewz at 18:20 UTC summed the same list as $664 billion, with the same $437.4 billion and $226.4 billion split.
Why the Loss Line Is So Large
The nine-month net loss was $104.0 billion. About 91% of it, $94.3 billion, is the change in fair value of convertible notes, derivative liabilities and SAFEs, by our arithmetic, a non-cash line that moves with the valuation behind those instruments.
The filing says Anthropic “will continue to recognize changes in fair value until these instruments are converted into equity, repaid, or otherwise settled”. The IPO is one such trigger: under Amazon’s note agreements, outstanding principal and accrued interest convert into Class N common stock at the lower of $2.0928 or $5.0478 per share, depending on the agreement, and the IPO price.
What the Filing Still Leaves Blank
Besides the run-rate figure, the S-1/A leaves the offering price blank, referring to an “assumed initial public offering price of $ per share”. Until a range is filed, outside estimates fill the gap. On 4 November, Polymarket traders put a 55% chance, by our arithmetic, on Anthropic shares starting to trade between 16 and 20 November, and 64.8% on an IPO price valuing the company at $1.50 trillion to $1.75 trillion.
Go deeper, 5 Nov, 08:41Z: the year-by-year schedule. The S-1/A’s contractual obligations table schedules $29,165 million of payments for 2027, $44,357 million for 2028, $47,752 million for 2029 and $45,070 million for 2030. Anthropic’s $131.9 billion of cash and short-term investments covers the table’s 2026-2029 columns about 1.04 times, by our arithmetic.
As we read the filing, the operating line and the net loss describe different things: the nine months had an operating loss of $1,260 million, while $94.3 billion of the $104.0 billion net loss is a fair-value change that the filing ties to notes until they convert.
As we read the commitments, they come in different forms: minimums with true-ups for Google and Amazon, up-to amounts with cancellation rights for SpaceXAI, and lease rent for Broadcom. That is why we treat their $643.8 billion sum as indicative only.
Anthropic, PBC, Form S-1/A, 2 Nov 2026
“Because much of the related capacity has not yet been delivered, the timing of our payments, and in some cases the amounts, may differ from those shown.”
Three Implications
FOR LIQUIDITY Cash and short-term investments of $131.9 billion compare with about $69.8 billion due in the 12 months after 30 September, about 1.9 times by our arithmetic.
FOR COMMITMENTS Approximately $447.5 billion of the $895.4 billion total is expected before 31 December 2030, about half by our arithmetic.
FOR THE IPO The S-1/A leaves both the offering price and the latest Run-Rate Revenue figure blank.
The framework says of models, compute, silicon and foundries: “They have to be priced with an embedded discount for the political volatility of the control plane.”
As we read it, Anthropic’s filing shows the compute side of that in contract terms: minimum spends, up-to amounts and leases running to 2036. The framework does not discuss Anthropic’s S-1.
The 1.9 times figure compares balances at one date with payments expected over the next year; it does not include cash Anthropic will earn or raise, or commitments it signs later. The filing notes that because much of the capacity has not yet been delivered, the timing and in some cases the amounts of payments may differ. The offering price, the listing date and the latest run-rate figure are not in the filings we read.
This Week In The AI Supercycle
Anthropic’s Reality Check, The Capital Squeeze, & The New World Order
This week’s issue: for two years the AI build-out was priced as a growthstory. This week it started being priced as a credit story.
Anthropic enters its IPO with $131.9 billion of cash and short-term investments against about $69.8 billion of obligations due within a year, about 1.9 times, by our arithmetic, inside a committed spend of approximately $895.4 billion. Its operating loss for the nine months was $1,260 million; the $104.0 billion net loss is mostly a fair-value change that the filing ties to the notes until they convert.
A note on sourcing. Figures are from Anthropic’s amended Form S-1 filed on SEC EDGAR on 2 November 2026, directly or as reported in our earlier pieces; Polymarket odds are market prices read on 4 November and are not forecasts. Ratios, sums and annualisations are our arithmetic.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
