Anthropic at $900 Billion, or Thereabouts

Anthropic is reportedly on track to close its current fundraising round by the end of May at a post-money valuation north of $900 billion. The reporting is single-sourced and uses the words "on track," which is the language people use when a round is not yet closed. There is no term sheet, no lead investor named, no round size disclosed. Anthropic has said nothing.¹

I want to take the number seriously anyway, because the interesting question is not whether the round closes at exactly $900 billion. The interesting question is what kind of object a $900 billion private valuation is, given what Anthropic has actually disclosed about itself.

The arithmetic, briefly. The last credible ARR figure for Anthropic is over $400 million, reported by The Information in July 2024.² That is the most recent publicly anchored number. There has been no official update for 2025 or 2026, though it is reasonable to assume the figure has grown materially. Pick a generous extrapolation, say $5 billion ARR, and a $900 billion valuation is a 180x revenue multiple. At the last confirmed figure, it is something closer to 2,250x. Peak-bubble SaaS comps ran 40 to 60x.

This is not a revenue multiple in any sense that a public-markets analyst would recognise. It is something else.

What it actually is. A $900 billion private valuation on a company with single-digit-billions ARR is a bet on who controls frontier AI infrastructure in ten years, priced as if the answer is mostly known. The capital is not buying cash flows; it is buying a position in a duopoly that the capital itself is helping to construct. OpenAI raised $40 billion in April 2025 at $300 billion post-money.³ If the Anthropic round closes near the reported figure, two San Francisco labs will have absorbed roughly $1.2 trillion in private valuation within fourteen months, neither of them public, neither of them disclosing audited financials.

This is the AI performativity frame doing real work. Enterprise procurement decisions in 2026 are anchored to a small number of providers partly because the capital concentration has made those providers unavoidable: they have the compute, the talent floor, the regulatory engagement, the press surface, the developer ecosystem. A CIO choosing between Anthropic and OpenAI for a regulated workload is not, on the margin, choosing between two equally-resourced vendors and two equally-likely futures. The valuation has already pre-selected the choice set.

Amazon's position is the most legible part of this. Amazon's $4 billion commitment to Anthropic, structured partly as AWS compute credits, was announced in September 2023 and expanded in March 2024.⁴ At a $900 billion valuation, Amazon's economic stake, call it 4 to 5 percent, the precise figure is not public, would be worth roughly $36 to $45 billion on paper, against a $4 billion cost basis and the structural benefit of locking Anthropic's inference workloads onto AWS. That is not a venture investment. That is a vertical integration paid for with compute credits, dressed in a SAFE.

The Amazon stake is a vertical integration paid for with compute credits, dressed in a SAFE.

The interesting question for the current round is whether the new capital looks anything like this. A clean cash round at $900 billion from financial investors is one kind of object. A round in which some meaningful tranche is sovereign-wealth money, or further compute-credit arrangements with hyperscalers, or structured paper with liquidation preferences that protect the headline number, is a different kind of object. OpenAI's SoftBank round closed with conditions tied to its for-profit conversion;³ frontier lab rounds at this scale are negotiated objects with terms that the headline number does not capture. "On track to close" tells us nothing about which kind of round this is.

The safety-as-market-position frame is awkward here. Anthropic's Responsible Scaling Policy and Constitutional AI methodology are genuine product differentiators in the enterprise compliance conversation, particularly for regulated-industry customers. The RSP framework, updated in October 2024, ties model deployment to safety evaluations in a way that is legibly defensible to a procurement committee at a bank or a hospital system.⁵ This matters, and it shows up in deal flow.

But the frame has a pricing problem. If Anthropic is worth $900 billion and OpenAI is worth $300 billion, the implicit safety premium is $600 billion, which is roughly two SpaceXes. That is a great deal of money to attribute to a posture that does not currently appear in disclosed ARR as a separable line. The more honest reading is that the safety positioning is a moat-component, not the moat itself; the moat is model capability, distribution, and compute access, with safety as the procurement-friendly wrapper. Dario Amodei said on Lex Fridman in November 2024 that Anthropic is "not trying to be the biggest company in the world."⁶ At $900 billion that statement is going to require some maintenance.

What this is a case of. It is the same case as OpenAI's $300 billion round in April 2025, and the same case as the SpaceX tender at $350 billion in December 2024.⁷ Late-stage private markets are now where the largest enterprise-value bets in the economy get priced, because the businesses being priced are not yet legible to public markets — no audited financials, no quarterly cadence, no comparable comps, capital structures that include compute credits and sovereign vehicles. The private round has become the venue for the kind of valuation that used to require an IPO and a decade of operating history. The thing being purchased is optionality on a future market structure, and the size of the cheque is itself part of the mechanism by which that market structure gets built.

What to watch. Three things. First, whether the round actually closes by end of May, and at what number — "on track" is doing real work in the reporting and a slip to June at a revised valuation would be informative. Second, the composition: who leads, how much is cash versus structured, whether hyperscaler compute arrangements are part of the package, and whether any sovereign vehicle appears on the cap table. Third, whether Anthropic discloses any updated revenue figure in connection with or shortly after the close. A round of this size without an accompanying ARR disclosure is itself a data point: it tells you the round is not being priced off the income statement.

I would not bet on the disclosure.

Footnotes

¶