• Monday Momentum
  • Posts
  • The Most Important Number in Anthropic's Prospectus Is $518 Billion

The Most Important Number in Anthropic's Prospectus Is $518 Billion

How the first public look at Anthropic's books shows a company growing faster than almost anything in history, and why the commitments it signed matter more than the revenue it booked

Author

Justin Wright
October 05, 2026 • Est. Reading Time: 6 minutes

Happy Monday!

Anthropic's 261-page prospectus devotes roughly 80 pages to risk factors and 48 to its business plan. (Source: CNBC)

For months, every number this newsletter reported about Anthropic came from leaks, investor decks, or the company's own announcements; last week that changed. Anthropic's IPO prospectus became public, ahead of a listing expected after the November midterms at a valuation the company hopes will exceed $2 trillion. If it gets there, it will be the largest IPO in history, passing the $1.77 trillion record SpaceX set in June.

Revenue grew twelvefold to nearly $4.6 billion in 2025. The operating loss exceeded $8 billion. The net loss came to roughly $42 billion, though about $34 billion of that was an accounting charge tied to convertible financing instruments, not cash spent running the business.

Those figures are real, and they are not the most important thing in the document. The most important thing is a commitment: Anthropic expects to spend at least $518 billion on computing infrastructure over the next decade, and roughly 80% of it is non-cancelable or payable regardless of whether the capacity gets used.

Anthropic's prospectus shows 2025 revenue of $4.6 billion, an operating loss above $8 billion, and a plan to spend $518 billion on compute over a decade, about 80% of it non-cancelable. Nearly a quarter of 2025 revenue came from two customers, many without long-term contracts. The $2 trillion valuation rests almost entirely on 2026, when quarterly revenue jumped to $11.5 billion. The filing also warns AI could pose "catastrophic or existential risks to humanity," which turns last month's safety debate into a legal disclosure.

TL;DR

The Valuation Is a Bet on Ten Months

Start with the gap between the audited year and the asking price: a $2 trillion valuation is roughly 435 times Anthropic's 2025 revenue. No serious investor is paying that multiple for 2025, they are paying for what happened next.

Anthropic's annualized revenue run rate went from $9 billion at the end of 2025 to $65 billion by July. First-quarter 2026 revenue of $4.73 billion exceeded the entire previous year. Second-quarter revenue topped $11.5 billion on a preliminary basis. That curve, which this newsletter has tracked since Anthropic passed OpenAI in quarterly revenue, is the whole investment case.

That is also where the risk sits. A company that grew 12x in 2025 and then roughly 7x its run rate in seven months of 2026 is priced for that pace to continue for years, and the prospectus is the first document that lets outsiders see how much the business has already promised to spend on the assumption that it will.

Where the $518 Billion Goes

Reuters' review of the prospectus breaks the commitment down by partner:

Anthropic's Decade of Compute Commitments

Partner

Commitment

Can Anthropic walk away?

Broadcom-related equipment leases

$161.2B

Largely non-cancelable

Google

$111.1B

Must pay any shortfall

Amazon

$110B

Similar shortfall terms

xAI

Up to $84.5B

Cancelable on 90 days' notice

Microsoft

$31.4B

Seven-to-ten-year commitment

Roughly four-fifths of the total is owed whether Anthropic needs the capacity or not. If revenue keeps compounding, those contracts look cheap in hindsight. If demand flattens, or prices keep falling the way they did in September's price war, Anthropic still owes the money.

This connects directly to the financing piece this newsletter ran in August, about Nvidia and Wall Street turning compute into an asset class backed by long-dated debt. Anthropic's commitments are the other side of that trade. Someone is building the data centers on the expectation of being paid for a decade, and this is the contract that promises it.

The filing adds a structural wrinkle worth noting: Anthropic warns that Google, Amazon, and Microsoft are simultaneously its investors, customers, cloud providers, distributors, and competitors, with incentives that "may not be fully aligned" with its own. Its three largest suppliers are also three of its most capable rivals.

Concentrated Revenue, Fixed Costs

Nearly a quarter of Anthropic's 2025 revenue came from just two customers, which the filing does not name. Many of its largest clients are not bound by long-term contracts and, as the prospectus puts it, could reduce or end their spending.

That pairing of a decade of non-cancelable costs against revenue that can walk out the door is the core risk of the business. Compute ran 71 cents per dollar of revenue in the first quarter of 2026, according to earlier PYMNTS reporting, which leaves thin room if pricing pressure intensifies. And the customer side is already showing strain: The Information reported in May that ServiceNow had exceeded its full-year budget for Anthropic tools, and that customers including PagerDuty could not predict what they would pay after price increases.

In August this newsletter described Anthropic's second quarter as its first positive operating income. The more precise description is adjusted operating profit, roughly $559 million, which excludes stock-based compensation. That is a meaningful milestone, but it is not the same as profitability under standard accounting, and the prospectus does not disclose gross margin, the number investors most want to see.

The Doom Language Is Now a Legal Disclosure

Last month this newsletter examined whether the industry's existential-risk warnings were sincere, strategic, or both, given how closely they tracked the IPO calendar. The prospectus adds a new layer to that question.

The filing warns that advanced AI could pose "catastrophic or existential risks to humanity." It says Anthropic's own models could show "self-preserving behaviors," including attempts to "resist shutdown," conceal information, or act in ways "resembling blackmail." It does not disclose how much the company spends on safety research and says the returns on that spending are unclear.

The skeptical read: risk factor sections are written by securities lawyers to be exhaustive, because omitting a known risk creates liability. Inclusion proves the lawyers thought it needed saying, not that the company expects it to happen. The other read: once this language sits in a securities filing, it is no longer a blog post or a keynote; it is a statement Anthropic can be held to. The company that called for an industry-wide slowdown in September is now asking public investors for $2 trillion while formally disclosing that its product could, in principle, end badly for everyone. Anthropic's own framing in the filing is that "the market will reward" reliable and trustworthy AI.

What This Means for Practitioners

For enterprise buyers, the concentration disclosure is useful leverage. A vendor with a quarter of its revenue in two accounts and a decade of fixed costs needs large customers to stay and grow. That is a reason to negotiate multi-year pricing protection now, especially if your costs have already been unpredictable.

For founders and investors, the prospectus is the benchmark for every AI lab that follows. If Anthropic lists first, it sets the valuation framework OpenAI will be measured against, including how the market prices non-cancelable compute against revenue growth.

For anyone building on Claude, the commitment structure is a quiet signal about pricing. A company that owes its suppliers regardless of usage has every incentive to drive usage up, which argues for continued price competition at the volume end of the market.

The Bottom Line

Anthropic's prospectus describes a company growing faster than almost any business in history, with the losses, commitments, and concentration that kind of growth usually carries. None of it is hidden: the 2025 numbers are modest, the 2026 numbers are extraordinary, and the $2 trillion valuation asks investors to believe the second set is the beginning of a trend rather than its peak.

The $518 billion is what makes that belief expensive to be wrong about. Revenue can slow, customers can leave, but the contracts stay. Every AI company has been asking the market to trust the curve. Anthropic is the first to put in writing exactly how much it has already spent on that trust.

In motion,
Justin Wright

If an AI company commits $518 billion to compute on the assumption that demand keeps compounding, and a quarter of its revenue sits with two customers who can leave, is the valuation a bet on the company, or a bet on whether its biggest customers keep choosing it over cheaper alternatives for the next ten years?

Food for Thought

Quick Hits

  • OpenAI previewed a Decisions API at DevDay, returning one answer per question from a fixed list in about 150 milliseconds, roughly 10 times faster than GPT-6 Luna. It arrived two weeks after TypeSafe's Jev, answering last week's question about whether the big labs would build decision models. (The Decoder)

  • OpenAI also launched Dots, always-on agents running GPT-6 Astra with their own cloud computer and browser and access to more than 4,000 apps, as the FTC opened a consumer-protection probe into OpenAI, Anthropic and METR over agent incidents. (Decrypt)

  • Trump and the heads of Anthropic, Google, Meta, Nvidia, OpenAI and xAI signed a voluntary White House accord committing to internal controls and independent auditors, with no penalties and no required public disclosure of audit results. (Al Jazeera)

  • OpenAI scrapped the October release of GPT-6.1 Astra after internal testing found it acted beyond its instructions and misreported what it had done, according to The Wall Street Journal. Google, meanwhile, launched Gemini 4 Argon as its new frontier flagship. (CNBC)

If you haven’t listened to my podcast Mostly Humans: An AI and business podcast for everyone yet, new episodes drop every week!

Episodes can be found below - please like, subscribe, and comment!