SAN FRANCISCO - Anthropic disclosed nearly $895.4 billion in computing and infrastructure commitments in an amended initial public offering filing Monday, revealing the extraordinary long-term cost of competing at the frontier of artificial intelligence.

The maker of the Claude AI system reported $17 billion in revenue for the three months ended Sept. 30, up from $1.4 billion in the same quarter last year. Its operating loss narrowed to $567 million from $3.1 billion, while an adjusted measure showed operating income of $2.5 billion.

Despite that operating improvement, Anthropic recorded a quarterly net loss of $23.3 billion, according to the filing as reported by The Associated Press. For the first nine months of 2026, revenue reached $33.2 billion, roughly 13 times the $2.5 billion reported for the comparable period a year earlier.

The nine-month net loss was $104 billion, but that figure cannot be read as equivalent to cash consumed by the business. It included a $94.3 billion change in the fair value of convertible notes, derivative liabilities and agreements for future equity. Anthropic said the accounting charge was driven mainly by the rising valuation of its outstanding convertible financing.

The compute commitments are more concrete. Anthropic said it had entered purchase agreements and leases totaling about $895.4 billion, substantially all tied to computing and infrastructure. About $447.5 billion is expected to be incurred by the end of 2030, with most of the remainder due from 2031 through 2045.

Approximately $69.8 billion is due in the 12 months following Sept. 30. The contracts include capacity from major cloud providers, and the company warned that forecasting demand years in advance could leave it paying for more computing power than customers ultimately require.

Anthropic nevertheless argued that its efficiency is improving. Revenue grew 1,127% from the year-earlier quarter while compute and infrastructure expense increased 382%. The filing said the company held $95.6 billion in cash and cash equivalents and $36.3 billion in short-term investments at the end of September.

The amended registration statement still leaves the number of shares and expected price range blank. Anthropic has applied to list Class A shares on the Nasdaq Global Select Market under the ticker ANT, but the filing does not establish when the offering will be completed or how much it will raise.

Its governance structure will also receive scrutiny. Anthropic is a Delaware public benefit corporation and says an independent Long-Term Benefit Trust is intended to preserve its safety mission. At the same time, a founder-controlled entity is set to hold 50.1% of voting power on matters submitted to shareholders generally.

AI leadership now requires industrial-scale capital

Anthropic's filing shows why the AI competition can no longer be understood as a contest among clever software laboratories. Training and serving frontier models requires chips, electricity, data centers, cooling systems and multiyear capacity contracts on a scale once associated with national infrastructure.

That investment strengthens an American company competing against state-supported Chinese rivals. Claude's enterprise growth suggests that U.S. innovation is producing products businesses are willing to pay for, and the narrowing operating loss is more informative than the dramatic net-loss headline alone.

But patriotism is not a substitute for financial discipline. An $895.4 billion commitment is enormous even when spread over two decades and protected by contractual termination rights. Public investors need to understand how much capacity is firm, what happens if suppliers fail to deliver and how quickly model efficiency could make older infrastructure less valuable.

The financing arrangements also reveal concentration. A small group of cloud and chip providers supplies the essential machinery for the most advanced AI companies while also investing in them. Those relationships can accelerate construction, but they can bind customers, suppliers and financiers together so tightly that a failed forecast spreads risk across the same ecosystem.

American markets offer an important democratic advantage: a company seeking public money must disclose risks, accounting assumptions, contracts and governance before investors vote with their capital. The filing does not eliminate uncertainty, but it makes the scale visible in a way that state-directed competitors often do not.

Regulators should preserve that transparency without trying to choose a technological winner. Investors should separate rapid sales growth from accounting charges, adjusted metrics from standard results, and committed infrastructure from current cash expense.

Anthropic may prove that demand for capable AI can outrun the cost of supplying it. It may also show that the race for intelligence locked companies into costly infrastructure before the market stabilized. The amended filing places that wager in public view: extraordinary growth on one side, nearly nine hundred billion dollars of commitments on the other.