4 minBusiness
Anthropic’s IPO filing reveals executive security costs and compute spending
Anthropic’s S-1 prospectus shows large compute commitments, a reliance on cloud rivals and plans for its own data centers, along with details on executive pay and security.
Anthropic’s IPO filing, a 318-page S-1 prospectus made public with the U.S. Securities and Exchange Commission, shows how central computing power is to the company’s business. The document states that the ability to develop, train and deploy frontier AI models depends on access to enormous and growing quantities of specialized compute, and that failing to secure enough of it is a significant risk factor.
The filing puts a figure on that dependence. Anthropic has committed to spending between $542 billion and $567 billion on future compute and infrastructure. On December 31, 2025, the company signed a $417 billion compute deal with various providers, with terms running from one to 20 years beginning in 2026 and 2027. It also took on additional capacity starting between 2028 and 2031, totaling $125 billion to $150 billion.
Compute is Anthropic’s largest operating expense, at $5.6 billion in the second quarter of 2026. The company generated $11.6 billion in revenue in the same quarter, which the filing treats as a healthy revenue-to-compute ratio, a metric Anthropic says it intends to keep improving. By that measure, it brought in a little over two dollars for every dollar spent on compute.
Anthropic sources its compute from Amazon, Google and Microsoft, which are also its competitors and, in the filing’s words, have made significant investments in the company while acting as customers for its models. The S-1 warns that these suppliers could prioritize their own compute needs and could gain insight into Anthropic’s business that might inform their competitive strategies. The company has also built a proprietary orchestration layer that routes workloads across AWS Trainium, Google TPUs, Nvidia GPUs and AMD GPUs, which it says reduces costs and dependence on any single supplier.
To further reduce that reliance, Anthropic is pursuing a strategy it calls Anthropic Direct, which involves setting up its own data centers through third-party partners. Starting around September 30, 2026, the company began entering purchase commitments and lease agreements for the initiative. Those partners will secure the real estate, power and facilities and manage the data centers, while Anthropic leases the compute it needs. Its rent payments over the five-year lease total $125.2 billion, excluding interest. Its largest infrastructure partner in 2026 is Broadcom, with $42 billion in potential financing.
The filing also includes a deal with SpaceX, owned by Elon Musk, under which Anthropic entered a series of agreements from May to October 2026 to access up to $120.5 billion in compute through 2029, powered by Nvidia GPUs that SpaceX owns. Other disclosures in the document cover executive compensation, with co-founders Dario Amodei and Daniela Amodei identified as the two most highly paid executives, and spending on personal security for top leaders. Anthropic did not immediately respond to requests for comment on the compute figures.
7
