Anthropic IPO Puts $800B Compute Contracts in View

Anthropic publicly filed for an initial public offering on Monday, Oct. 26, and its filing revealed a sharp imbalance: customers can cut back on its AI or leave with little or no notice, but some computing contracts require minimum payments regardless of need. The company says it must rapidly grow revenue to meet those obligations. Its bosses hold the contracts; customers can walk away, while infrastructure bills keep arriving.

The commitments behind the IPO

Anthropic’s S-1 filing with the Securities and Exchange Commission describes a business that sells access to Claude and other AI models while taking on heavy computing commitments. The company’s shares are set to trade on Nasdaq under the ticker ANT, and a debut could come as soon as November. Investors will get a test of their appetite for the technology.

The New York Times reported that Anthropic generated $16.3 billion in revenue in the first half of 2026, nearly 14 times the total for the same period a year earlier. Fortune reported first-half revenue of $16.28 billion, up 1,300%, with about $13.1 billion—roughly 80%—coming from enterprise customers who pay according to consumption. Business Insider reported second-quarter revenue of $11.6 billion, up 1,368% from a year earlier.

Anthropic reported a net loss of $80.7 billion for the first half of the year. The New York Times said that figure included a one-time $72.4 billion charge tied to convertible notes and other financial instruments. The company raised billions from major backers, including Amazon and Google, through convertible debt. As its valuation climbed, the debt’s value rose too, and Anthropic recorded it as a loss because the debt counts as a liability.

The company’s appetite for infrastructure is large enough to shape its entire business. The New York Times tallied up to $807 billion in contracts for computing power and other infrastructure. Fortune reported at least $475 billion in future cloud-computing purchases, including at least $389.9 billion pledged to Amazon, Google-parent Alphabet, SpaceXAI and Microsoft, as well as deals with Nscale and Lambda. Add agreements to lease computing chips, and total commitments could reach $750 billion. Much of that spending comes due regardless of how much customers use.

When customers can leave, the bills remain

Anthropic said it has underestimated its computing needs before, leaving it short on capacity and affecting product development. It warned that new capacity could arrive at the wrong time, creating “periods of excess or underutilized compute and committed cash outflows that do not produce immediate revenue.” The filing says expanding infrastructure will require large upfront payments before capacity comes online, potentially shrinking cash reserves.

The company has tried to shift enterprise customers into non-cancelable, multiyear agreements, offering discounted token rates in exchange for fixed minimum spending. As of June 2026, those contracts carried about $10.8 billion in remaining performance obligations. Anthropic planned to recognize 75% of that revenue over the next year, while its data-center spending commitments stretch further into the future.

The contracts don’t all impose the same restrictions. The New York Times reported that Anthropic can scrap its deal to lease computing power from SpaceX through 2029 starting in 2027, with 90 days’ warning, while a $110 billion deal with Amazon lasts until 2036. Anthropic typically signs computing contracts well before it can use the capacity. It said timing mismatches can leave it paying for resources that don’t bring in revenue right away.

Growth, risk and the public offering

Business Insider reported that computing and infrastructure spending rose 300% in the second quarter, while other expenses rose 741%. Anthropic brought in $2.08 in revenue for every dollar it spent on computing in that quarter, compared with 40 cents in the first quarter of 2025. It recorded $974 million in quarterly operating profit, an 8% margin. Its adjusted operating profit margin was 19%, up from negative 129% a year earlier.

The filing also lays out risks that come with the expansion. Anthropic said serving customer tokens cost $2.4 billion in the six months ended June 2026, up 135% from the same period a year earlier. It spent $9.7 billion on computing and infrastructure for research and development. The company warned that new capacity could squeeze margins, and that revenue growth is expected to slow over time and fluctuate.

Anthropic says it doesn’t plan to rely on advertising in the near future, citing pressure to optimize for engagement. Its public benefit corporation structure requires it to pursue a public benefit mission alongside shareholder interests. The company also disclosed that 5% to 7% of its revenue this year came from customers later found to have violated its usage policies. It described incidents in which models gained unauthorized access to real-world systems during security testing.

The filing names risks that include cyberattacks, biological weapons design and government mass surveillance. “We consider it our personal responsibility to help the world understand these risks, to prevent or mitigate them when they arise from our AI models, and to work with governments and the rest of the industry to address them at a societal level,” the company wrote. The Federal Trade Commission has also asked Anthropic executives to testify in a consumer-harm investigation. Mark Mahaney, a top tech analyst at Evercore ISI, wrote: “Frontier-lab unit economics can turn positive at scale.”