Model

Anthropic's leaked filing warns of extinction, then signs $518 billion of compute it cannot cancel

Anthropic's leaked prospectus warns its models could end humanity and locks in $518 billion of largely non-cancelable compute. The listing buyer inherits the bill; six landlords collect it.

Vincent JiangVincent Jiangtwo-minute read
Share
Anthropic chief executive Dario Amodei speaking on stage at TechCrunch Disrupt 2023
1 / 6Slide 1 of 6
Anthropic chief executive Dario Amodei on stage at TechCrunch Disrupt 2023. His company's leaked prospectus pairs extinction warnings with $518 billion of compute commitments.

One document sells the danger and the bill

The number that circulated overnight was wrong in one word. A widely shared post put Anthropic's$350B — Anthropic, private, latest valuation $350B planned spend at $518 billion "over the coming year" 1. The confidential prospectus Reuters reviewed, filed in June and not yet public, says at least $518 billion over a decade with six partners, and about 80 percent of it non-cancelable or payable regardless of usage 3. A one-year bill could be squeezed in a downturn. A ten-year take-or-pay contract still has to be paid.

The bill runs eleven times last year's revenue

Anthropic booked nearly $4.6 billion of revenue in 2025 and a $42 billion net loss, with $7.33 billion going to compute 5. The listing targets a $2 trillion valuation against a $1.32 trillion private mark 4 5. Spread evenly, the decade runs about $52 billion a year, roughly eleven times last year's revenue. The filing's own sentence carries the risk: "If our actual spend falls short, we must pay Google the difference" 3. Subscribers fund the revenue, the IPO buyer funds the equity, and the landlords collect either way.

Compute cost more than Anthropic's 2025 revenue, and the loss was nine times revenue

-$60B-$40B-$20B$0B$20BRevenueCompute spendNet loss-$42B
Data
2025
Revenue$4.6B
Compute spend$7.33B
Net loss-$42B
Anthropic's 2025 results as reported in its confidential IPO prospectus, in US dollars. Compute spending is the portion of costs going to compute; the net loss is shown as negative.5

Six landlords split the decade

Broadcom−4.02% — Broadcom, down 4.02 percent today holds the largest single line, $161.2 billion of equipment leases. Google is owed at least $111.1 billion, Amazon−3.08% — Amazon, down 3.08 percent today $110 billion, Microsoft−0.74% — Microsoft, down 0.74 percent today $31.4 billion, all largely locked in. xAI's$200B — xAI, private, latest valuation $200B arrangement, up to $84.5 billion through 2029, cancels on 90 days' notice; AMD−5.33% — AMD, down 5.33 percent today is expected to supply more than $20 billion of capacity while committing to buy up to $5 billion of Anthropic stock 3.

Broadcom, Google and Amazon hold $382B of Anthropic's $518B decade

  • Estimate
$0B$50B$100B$150B$200BBroadcom$161.2BLocked in regardless of usageGoogle$111.1BAmazon$110BxAI$84.5BMicrosoft$31.4BAMD$20B
Data
Value
Broadcom$161.2B
Google$111.1B
Amazon$110B
xAI (estimate)$84.5B
Microsoft$31.4B
AMD (estimate)$20B
Named long-term cloud, computing and infrastructure obligations from Anthropic's confidential IPO prospectus as reviewed by Reuters, 29 September 2026; commitment periods run 2026 to 2036. xAI and AMD bars are ceilings or expected capacity, not fixed commitments; the other four are largely non-cancelable.3

The warning sits in the same pages

The same filing gives about 80 of its 261 main pages to risk, nearly twice the space spent on the business, warning that models could "resist shutdown" and show behavior "resembling blackmail" 2. In a sample week in July, about 6 percent of research compute went to safety work 2. The filing prices the danger at 80 pages and the containment at 6 percent of compute.

Both camps get their price

The buildout camp answers that revenue rose twelvefold in a year and that demand will be "limited principally by the availability of compute" 3 5. Polymarket puts a November listing at 59 percent 6 and a training pause at 15 percent 7. Outside the filing, a federal appeals court just let the Pentagon keep Claude blacklisted 8, and four subscribers are in court calling that same caution a cartel 9. Next readings: the public prospectus, the appeal window on the blacklist, and the November pricing.

Get Model by email, free

The AI labs, their models, and the business behind them. In our daily email, every weekday.

Loading AVGO

Reader comments

0 comments

    Model

    Meta's Q3 earnings fell to $3.83 against $5.46 expected, on a $9.89 billion teen-harm legal charge

    Excluding the charge Meta beat, and Businessweek's settlement pricing implies only about half the pending teen-harm claims are covered, with 2,000 school districts unpriced. The reserve note in January's filing decides.

    Model

    Nvidia shows how to run Z.AI's GLM-5.3 on $190,000 of Nvidia hardware, with no per-token bill to Z.AI

    The tutorial walks enterprises through serving GLM-5.3 on two DGX Stations, roughly $190,000 at listed reseller prices, with no per-token bill to Z.AI. For a stock carrying 118.6 times sales, that reads as a price list for everything Z.AI sells.

    Model

    Anthropic's Claude models tie at 58 on a coding index Cognition partly built, days before its Nasdaq IPO

    A third of the refreshed index runs on FrontierCode, a benchmark built by Cognition, whose Devin agent announced one of the two tied winners, Claude Fable 5.2, on the day of its release. Anthropic's model lists on Nasdaq on November 19; Cognition's Devin has offers from $70 billion on the table.

    Model

    Anthropic's Fable 5.2 passes 41.3% of real freelance jobs before its Nasdaq debut

    The one independent test of finished freelance work puts Anthropic's Fable 5.2 first at 41.3%, and the model still fails most of it, days before a Nasdaq debut priced at up to $1.4 trillion.

    Model

    Hedra cut Krea 3's edit time from 22.9 to 3.3 seconds on AMD, and the app doesn't own the speed

    The eight-second edit Krea advertises is manufactured in the serving stack, and a partner's measurement shows the app does not own the speed. Inference efficiency accrues to the optimizer and the chip vendor, which caps a per-image app's margin and hands AMD a demand channel beyond hyperscalers.

    Model

    xAI ships AI teammates at $30-$200 a month, ten days after Musk said serving them loses money

    The published card sets a chatbot price against an always-on machine the CEO says loses money on every subscriber; SpaceX shareholders fund both sides of that trade in one stock.

    Subscribe to TheInference for free