Listen to this post: Anthropic’s Revised S-1 Filing: $895bn in Compute Deals, a $104bn Paper Loss

Last updated: 3 November 2026. The figures below come from Anthropic’s amended SEC filing as reviewed and reported by multiple financial-press outlets; where a number is self-reported by the company, that is noted, and primary and press sources are linked inline throughout.
The 60-second version
- Anthropic filed Amendment No. 1 to its S-1 registration statement with the US Securities and Exchange Commission on 2 November 2026, ahead of a planned Nasdaq listing under the ticker “ANT”.
- Third-quarter 2026 revenue came in at roughly $17 billion, up from about $11.6 billion in the second quarter and around twelve times the same quarter a year earlier.
- Nine-month 2026 revenue reached $33.2 billion — but the same nine months produced a net loss of $104 billion, a figure dominated by non-cash markups on convertible notes and SAFEs rather than operating cash burn.
- Disclosed compute and infrastructure purchase commitments now total roughly $895.4 billion, spread across Google, Broadcom, Amazon, SpaceX’s AI unit, Microsoft, Nscale and Lambda, among others.
- Investors are reportedly pushing for a lower valuation of roughly $1.5 trillion to $1.6 trillion for the listing — down from the $2 trillion figure floated in October — according to Reuters and the Wall Street Journal; the amended filing itself still leaves the price range and share count blank.
Key numbers from the amended filing
| Metric | Figure | Context |
|---|---|---|
| Q3 2026 revenue | ~$17.0bn | vs ~$11.6bn in Q2 2026; ~12x Q3 2025 (~$1.4bn) |
| Nine-month 2026 revenue | $33.2bn | January–September 2026 |
| Q3 2026 net loss | $23.3bn | vs $12.1bn in Q3 2025; mostly non-cash revaluation |
| Nine-month 2026 net loss | $104.0bn | driven by convertible-note and SAFE markups |
| Q3 2026 operating result | $567m operating loss | vs $974m operating profit in Q2 2026 |
| Total disclosed purchase commitments | ~$895.4bn | as of 30 September 2026, almost all compute/infrastructure |
| Cash + short-term investments | $131.7bn | as of 30 September 2026 |
| Reported IPO timing | Before end of November 2026 | Nasdaq, ticker “ANT” (unconfirmed date) |
What actually changed in this amendment
Anthropic has been filing and amending IPO paperwork for months, and CurratedBrief has tracked each revision: the 27 October filing disclosed a $2 trillion valuation ask alongside an $80.7 billion paper loss. This amendment does two things that matter more than the headline numbers suggest. First, it folds in a complete third quarter, which is the first full quarter in which Anthropic’s revenue growth rate decelerated on a year-over-year basis even as it accelerated sequentially — 47% quarter-on-quarter growth against roughly 1,127% year-on-year growth is a sign of a business scaling from a small base, not evidence of a new growth phase. Second, it revises several named compute commitments upward: Google’s disclosed commitment rose from $114.5 billion in the original filing to $121.3 billion here, according to ZeroHedge’s review of the amended prospectus. Filings like this are amended, not replaced, precisely because the underlying commercial relationships keep growing in size even between quarterly filings — which is itself a data point about how fast the compute market is moving.
Where the $895 billion is actually going
The $895.4 billion figure is the one every outlet led with, and it is Anthropic’s own disclosure, not a press estimate. According to a breakdown reported by Yahoo Finance’s review of the filing, the seven largest named counterparties are: Broadcom equipment leases at $161.2 billion, Google at $121.3 billion, SpaceX’s AI unit at up to $120.5 billion, Amazon at $110.0 billion or more, Microsoft at $44.9 billion, Nscale at up to $44.6 billion, and Lambda at $41.3 billion. Those seven alone add up to roughly $644 billion — about 72% of the headline figure — with the remainder spread across shorter-term and smaller commitments. CurratedBrief has covered two of these deals as they were struck: the Lambda agreement, which turned Nvidia into something closer to a landlord than a chip supplier, and the Nscale deal that followed Google and Microsoft’s own exit from that arrangement. Both of those original dollar figures have since drifted from what appears in this filing, a reminder that compute contracts are renegotiated and restructured more often than headline deal announcements suggest.
It is worth being precise about what is excluded from the $895.4 billion: deals without a fixed dollar amount, which ZeroHedge estimates at a further $150–175 billion — including a prospective Google TPU tranche for 2028–2031 and AMD’s “Helios” arrangement. That estimate is ZeroHedge’s own arithmetic, not a figure Anthropic discloses, and it should be read as an informed approximation rather than a confirmed number. On the timing side, the filing reportedly shows about $69.8 billion due within twelve months of 30 September 2026, roughly $447.5 billion due before the end of 2030, and the remainder — the majority of the total — stretched out across 2031 to 2045. Spreading commitments over two decades is standard practice for infrastructure-heavy balance sheets, and it matters for how alarmed a reader should be: the near-term cash exposure is a small fraction of the headline number, even if the long-run exposure is real.
Why the net loss ballooned to $104 billion — and why that’s not quite the number to worry about
Most of the day’s coverage led with the $104 billion nine-month net loss without explaining where it comes from, which is the single biggest thing this round of reporting gets wrong. Convertible notes and SAFEs — the instruments many of Anthropic’s earlier investors hold — are carried on the balance sheet at fair value, and that value rises as the company’s implied valuation rises. As investor discussions have moved from the roughly $350 billion range earlier in 2026 toward the $1.5–1.6 trillion range now being discussed for the IPO — down from an earlier $2 trillion ask — the liability associated with those convertible instruments has grown sharply, and that growth shows up as a non-cash loss on the income statement. Reporting attributes roughly $21.9 billion of the $23.3 billion third-quarter net loss to exactly this kind of revaluation. It is a real accounting loss — it affects reported net income and dilution math — but it is not cash leaving the business, and it is mechanically a byproduct of Anthropic’s valuation going up, not down.
The more operationally meaningful number is the swing from a $974 million operating profit in the second quarter to a $567 million operating loss in the third. Even that modest swing is partly explained by one-off items: roughly $5.0 billion of compute recognised under short-term commitments and a $2.0 billion non-cash charge tied to a donation-matching programme, according to figures reported from the filing. Strip those out and the underlying operating trend looks closer to flat than collapsing — a meaningfully different story than “Anthropic lost $104 billion.”
The circularity question
This filing lands in the middle of a live argument about whether AI infrastructure spending is self-reinforcing in a way that overstates real economic demand. CurratedBrief examined one side of that argument in Microsoft’s disclosure that OpenAI generated $10 billion of its cloud revenue alongside a $7.2 billion receivable — a structure where the hyperscaler is simultaneously supplier, landlord and, in several cases, investor. Anthropic’s filing shows the same pattern from the other side: Google, Amazon and Microsoft are all named as compute suppliers here, and all three have separately taken minority stakes in Anthropic. None of this makes the revenue or the commitments fictitious — the compute is real and take-or-pay terms in several deals mean Anthropic owes the money regardless of its own growth rate. But a meaningful share of the sector’s reported growth is large companies doing business with each other’s balance sheets, exactly the dynamic Apollo’s economists and Goldman Sachs have been publicly disagreeing about. This filing doesn’t resolve that argument, but it hands both sides new numbers.
Practical takeaways for people who build, publish or run software on this infrastructure
For teams building on the Claude API or competing against it, a few things stand out beyond the headline figures. The filing discloses a roughly $600 million revenue drag from price reductions already enacted, which suggests API pricing has real downward pressure behind it even as usage grows — plan for continued price competition rather than assuming current rates are a floor. The scale and duration of the named compute commitments (many running to 2045) suggest Anthropic’s capacity planning is now closer to a utility’s than a software company’s, which should ease — not eliminate — concern about sudden capacity shortfalls, since multi-year take-or-pay contracts exist precisely to guarantee supply. Anyone evaluating Anthropic, or a competitor, as a vendor should separate the non-cash net loss from the operating and cash position: with $131.7 billion in cash and short-term investments against $69.8 billion due in the next twelve months, near-term insolvency isn’t the risk here — concentration risk among a handful of infrastructure counterparties is the more credible one.
What we still don’t know
- The actual IPO price range and share count, both of which remain blank in the amended filing as of 2 November 2026.
- Whether the $1.5–1.6 trillion valuation Reuters reports investors are now pushing for will hold, since it is sourced to unnamed investors rather than disclosed by Anthropic itself.
- How the roughly $150–175 billion of commitments without a fixed dollar figure — including the prospective Google TPU tranche and the AMD “Helios” arrangement — will eventually be recognised in future disclosed totals.
- Whether the reported listing date of “before the end of November 2026” holds, given that price range and share count are still undetermined.
- Whether the deceleration in year-on-year growth (even alongside strong sequential growth) is an early signal of market saturation among large enterprise customers, or simply a function of 2025’s unusually small comparison base.
FAQ
Is Anthropic profitable? Not on a net basis, and not consistently on an operating basis either — it posted a small operating profit in the second quarter of 2026 and a small operating loss in the third. Its large net losses are overwhelmingly a non-cash accounting effect of its own rising valuation, not a sign of cash-flow distress.
What is the $895 billion commitment actually for? Almost entirely compute and data-centre infrastructure: multi-year contracts with cloud and chip-adjacent suppliers including Google, Broadcom, Amazon, Microsoft, SpaceX’s AI unit, Nscale and Lambda.
When is Anthropic’s IPO expected to happen? Financial-press reporting on the filing points to a Nasdaq listing under the ticker “ANT” before the end of November 2026, though the amendment itself does not yet set a price range or share count, so that timeline is not guaranteed.
Is the $104 billion nine-month net loss a sign of financial trouble? Not in the way the word “loss” usually implies. The great majority of it comes from marking convertible notes and SAFEs to their now much higher implied value as Anthropic’s valuation has climbed — a paper loss driven by the company getting more valuable, not less.
Sources
- Amendment No. 1 to Anthropic PBC’s Form S-1, filed with the SEC (EDGAR) — primary filing.
- The Next Web: “Anthropic’s amended IPO filing shows $17bn revenue, $895bn commitments”
- Yahoo Finance: “Anthropic’s new IPO filing puts its computing bill near $900 billion”
- Forbes: “Anthropic Hit $17 Billion In Third Quarter Revenue Ahead Of IPO—But AI Spending Commitments Hit $900 Billion”
- ZeroHedge: “Anthropic Files Updated S1: Committed Spend Jumps To $900BN, Only Google Has More”
- Reuters (via Investing.com): “Anthropic’s loss shrinks, revenue jumps ahead of blockbuster IPO” — reporting on investors pushing for a lower $1.5–1.6 trillion valuation.
