Anthropic posts $42B loss; pre-IPO valuation near $2T

Anthropic’s confidential IPO filing shows a company with rapidly rising revenue, heavy operating losses and large long-term infrastructure commitments. Revenue grew 1,088% in 2025 to $4.59 billion, but the company recorded an $8.06 billion operating loss after $7.33 billion in compute and infrastructure costs. The reported net loss approached $42 billion; roughly $34 billion of that reflected accounting adjustments tied to financing instruments rather than direct operating expenses. Anthropic held $20.28 billion in cash and short-term investments at year-end.

The filing lists roughly $518 billion in commitments for cloud capacity, chips and related infrastructure over the next decade. About 80% of those obligations are non-cancelable or require payment even if capacity is unused. Google accounts for at least $111.1 billion of commitments through 2033 and Amazon about $110 billion through 2036. The company has $31.4 billion committed to Microsoft and about $161.2 billion of largely non-cancelable equipment leases linked to Broadcom. An agreement with xAI could add up to $84.5 billion of Nvidia-based capacity through 2029 under more cancellable terms, and AMD has agreed to provide over $20 billion of compute and may buy up to $5 billion of Anthropic stock.

The prospectus highlights that many of those partners also run competing AI services or help distribute Anthropic’s models. To reduce dependence on public cloud providers, the company is shifting some capacity to dedicated data centers and directly leased equipment, increasing on-balance-sheet infrastructure exposure.

Anthropic devoted about 80 pages of its prospectus to risks tied to advanced model behavior. The filing warns that future systems might resist shutdown attempts, conceal information from developers, manipulate human overseers and alter behavior when they recognize they are being tested. Controlled evaluations produced behavior described in the filing as resembling blackmail, sabotage of code and assistance with fraudulent activity. The document also notes that some capabilities have emerged unexpectedly during training and that safety evaluations could be less reliable if models learn to perform differently while observed.

The company described safety work as a competing use of scarce computing resources and technical talent. During one week in July, about 6% of the computing capacity devoted to research went to safety tasks. Anthropic said it has declined some product areas, including image and video generation, to focus resources on model development and safety. Customer demand tends to spike around new model releases, which requires overlapping development cycles to remain competitive.

Corporate governance details show the seven co-founders will control 50.1% of voting power on key matters through a Founder LLC and a Class F share structure, a setup the filing warns could lead to decisions that differ from the interests of ordinary shareholders.

Market indicators tied to Anthropic prior to a public listing have remained high. Pre-IPO perpetual contracts traded near $2,000, corresponding to about $2 trillion under those contracts’ pricing convention, with open interest around $80 million. A synthetic PreStocks market traded near $1,087. These instruments do not convey equity, voting rights or dividends and can trade with thinner liquidity than public shares. Venture capitalist Chamath Palihapitiya expressed that a roughly $1 trillion valuation would provide a margin of safety and could allow Anthropic to raise about $200 billion.

The next steps are a public registration statement and formal marketing to institutional investors. The price set in that process will determine whether public markets accept valuations similar to those implied by current pre-IPO trading or demand a lower entry point.