Anthropic has disclosed $895.4 billion in cumulative purchase and lease commitments, placing one of the AI industry’s fastest-growing companies under an obligation load that far exceeds its available cash and highlights how aggressively leading model developers are moving to secure computing capacity, power and data center space.
The commitments, detailed in an updated prospectus filed with the U.S. Securities and Exchange Commission on Nov. 2, cover arrangements tied mainly to computing infrastructure. Anthropic reported that $69.8 billion is due within the next 12 months, while $447.5 billion is scheduled to be settled by Dec. 31, 2030. The remainder runs from 2031 through 2045.
At the end of September, Anthropic held $131.845 billion in cash, cash equivalents and short-term investments. Including available credit facilities, its liquidity totaled about $146.6 billion. The disclosed commitments are therefore roughly 6.8 times its cash and short-term investment balance, underscoring the company’s dependence on sustained revenue growth, outside financing and the ability to convert infrastructure reservations into profitable model usage.
Revenue surged as infrastructure spending accelerated
Anthropic’s third-quarter revenue reached $16.96 billion, up from $11.56 billion in the second quarter and roughly $1.38 billion a year earlier, according to the filing. Revenue for the first nine months of 2026 totaled $33.242 billion, compared with $2.545 billion during the same period in 2025.
That growth has been accompanied by a rapid increase in the cost of operating and expanding AI systems. Computing and infrastructure expenses rose to approximately $9.9 billion in the third quarter from about $5.6 billion in the previous quarter. Those expenses consumed about 58% of quarterly revenue.
For the first three quarters of 2026, computing and infrastructure costs increased by about $14.4 billion from a year earlier, a rise of 329%, Anthropic said. Short-term compute purchases alone accounted for about $5 billion during the third quarter.
Operating expenses climbed to $17.5 billion, from $10.6 billion in the second quarter, pushing Anthropic to an operating loss of $567 million. The company had reported operating income of $974 million in the prior quarter.
Anthropic’s adjusted operating margin, which excludes stock-based compensation and a non-cash employee donation-matching expense, fell to about 15% from 19% in the second quarter. The company said pricing changes and contract discounts reduced the revenue benefit of higher usage by approximately $600 million between the two quarters.
The third-quarter net loss was $23.29 billion, improving from a $66.3 billion net loss in the second quarter. Over the first nine months, net losses totaled $104.039 billion, including roughly $94.3 billion in fair-value changes connected to convertible notes, derivative liabilities and other financial instruments. The third quarter also included a non-cash expense of about $2 billion related to the employee donation-matching program.
Long-term commitments span chips, cloud and power
The prospectus provides an unusually detailed picture of how AI companies are shifting from renting cloud services as needed toward reserving physical capacity years ahead of demand.
Anthropic listed about $321.4 billion in lease commitments that had not yet commenced, alongside approximately $265.6 billion in fixed minimum purchase commitments. The filing said certain arrangements may require payments even if Anthropic’s eventual computing needs fall below expectations. That structure shifts part of the risk from infrastructure providers back to the model developer: unused capacity can become a fixed financial burden rather than a flexible operating expense.
Supplier exposure is concentrated among a relatively small group of technology and infrastructure companies. Anthropic disclosed contract amounts of approximately $121.3 billion with Google, $110 billion with Amazon and $44.9 billion with Microsoft.
The filing also cited SpaceX-related agreements worth up to $120.5 billion, with terms extending to 2036, and Broadcom-related equipment lease commitments totaling about $161.2 billion. The document did not include every anticipated chip purchase in the $895.4 billion figure. It referred to planned purchases of roughly $125 billion to $150 billion in Google chips and more than $20 billion in AMD chips where fixed contractual amounts had not been established.
Anthropic said it is sourcing computing capacity across Amazon Web Services Trainium chips, Google TPUs, Nvidia GPUs and AMD GPUs, with scheduling coordinated across providers. Such a multi-supplier approach could reduce reliance on any single chip platform, but it also adds operational complexity as workloads must be allocated across different hardware and cloud environments.
Anthropic moves closer to data center development
The company’s infrastructure plans extend beyond established cloud providers. Anthropic described relationships with newer compute companies including Fluidstack and Lambda, as well as data center development efforts involving TeraWulf and Hut 8.
For cryptocurrency readers, those names underline the growing overlap between AI infrastructure demand and sites traditionally associated with Bitcoin mining. Mining operators often control land, electrical interconnections and power contracts that have become valuable for high-density computing projects. AI tenants generally seek long-term access to large blocks of electricity and facilities capable of housing specialized hardware, while former mining sites can offer existing grid links or industrial locations suitable for conversion.
Anthropic said it has evaluated the reuse of older coal plants and industrial facilities for data center projects, while working through grid access and permitting requirements. Its early project searches focused on sites of around 200 megawatts or more before expanding toward projects around 50 MW. The company also described plans to supplement supply through battery storage, natural-gas generation and the aggregation of multiple smaller sites.
The prospectus refers to these dedicated infrastructure efforts as Anthropic Direct. Available computing capacity could exceed 3.5 gigawatts in 2026, Anthropic said, with plans to double that level in 2027.
A gigawatt-scale target puts Anthropic’s expansion in the same practical category as major industrial power users. Securing chips alone no longer determines how quickly an AI company can scale; transmission access, substations, permits, generation capacity and data center construction can all become constraints.
Fixed capacity raises execution pressure
Anthropic’s commitments give suppliers and data center partners clearer visibility into future demand, particularly for projects requiring years of construction and power planning. They also raise the cost of any slowdown in enterprise AI spending or any delay in bringing new facilities online.
The company’s revenue trajectory offers one justification for locking in capacity early: third-quarter revenue was more than 12 times the level recorded a year earlier. Yet computing expenses are rising quickly, and discounts have already offset a portion of the revenue gained from higher usage.
For data center developers, power-rich industrial landlords and companies with established grid connections, Anthropic’s disclosures reinforce a changing competitive landscape. Facilities once valued primarily for crypto mining operations can now attract AI workloads with far larger contracted infrastructure budgets, provided they can meet the demanding requirements for reliable electricity, cooling, network capacity and buildout speed.
Want deeper context on AI and finance? Explore Toobit Academy’s guide on Web3, AI, and crypto breaking the internet next.
Disclaimer: The content on this page is provided for general informational purposes only and does not represent the views or financial advice of Toobit. We make no guarantees regarding the accuracy or completeness of this information and shall not be held liable for any errors, omissions, or outcomes resulting from its use. Investing in digital assets involves risk; users should independently evaluate their financial situation and the risks involved. For further details, please consult our Terms of Service and Risk Disclosure.
