Anthropic has committed at least $518 billion over the next decade for computing and infrastructure, with about 80% of the obligations either non-cancellable or payable even if the company does not use the contracted capacity.
Artificial intelligence company Anthropic has committed at least $518 billion over the next decade to build computing and other infrastructure, and roughly 80% of those obligations cannot be cancelled or must be paid regardless of whether the company uses the capacity, according to a confidential IPO prospectus reviewed by Reuters.
The commitments illustrate the enormous infrastructure requirements facing companies developing advanced AI systems and the financial risks involved in securing computing capacity years in advance.
Anthropic, the company behind the Claude AI models, has told investors that access to computing power is becoming a major constraint on the development of increasingly sophisticated AI systems. The company expects demand for advanced AI computing to exceed available supply and says future AI development will be “limited principally by the availability of compute.”
The scale of the commitments also highlights the increasingly close financial relationships between Anthropic and major technology companies that simultaneously serve as its investors, cloud providers, distributors and competitors.
$518 billion in infrastructure commitments
Anthropic's prospectus says the company expects to spend at least $518 billion over 10 years with six infrastructure partners.
About 80% of that amount is either non-cancellable or requires Anthropic to make payments even if it does not use all of the computing capacity it has contracted for.
Among the largest commitments are:
• Google: at least $111.1 billion through 2033.
• Amazon: about $110 billion through 2036.
• Microsoft: $31.4 billion through 2033.
• Broadcom-related equipment leases: about $161.2 billion, largely non-cancellable.
• xAI: up to $84.5 billion for Nvidia-based computing capacity through 2029, although those commitments are largely cancellable with 90 days' notice.
The figures represent contractual obligations and potential spending over multiple years rather than money Anthropic is paying immediately.
That distinction is important because the $518 billion figure describes the company's expected infrastructure commitments over the coming decade.
Anthropic could have to pay even without using the capacity
The filing provides an unusual insight into the economics of AI infrastructure.
Anthropic's agreements with Google and Amazon include minimum-spending commitments. If the company does not use enough of the contracted capacity, it can still be required to pay the difference.
Anthropic said its Google commitment runs from April 2026 through July 2033, while its Amazon commitment runs from May 2026 through April 2036.
The Microsoft agreement, worth $31.4 billion, runs from November 2026 through May 2033 and is described as non-cancellable except in the event of an uncured material breach by Microsoft.
Anthropic's Broadcom-related equipment leases are also largely non-cancellable, except under specified circumstances such as default.
These arrangements provide Anthropic with access to computing infrastructure but also create significant fixed financial obligations.
Why Anthropic is locking in so much computing power
Anthropic's explanation is straightforward: it expects demand for AI computing to continue growing rapidly.
Training and operating advanced AI models require enormous amounts of processing power. Companies therefore compete for access to chips, data centres, electricity and cloud infrastructure.
Anthropic believes securing computing capacity ahead of time is necessary because available supply could become a bottleneck as AI systems become more capable.
The company has consequently been moving beyond a traditional cloud-only model and increasingly securing dedicated infrastructure and directly leased computing resources.
The strategy is intended to give Anthropic greater certainty that it will have enough computing capacity to train and deploy future generations of Claude.
Amazon commitment exceeds $100 billion
Anthropic has already publicly disclosed part of its expanding relationship with Amazon.
In April, Anthropic announced an agreement committing more than $100 billion over 10 years to AWS technologies and securing up to 5 gigawatts of new computing capacity.
The company said the infrastructure would be used to train and deploy Claude and would include Amazon's Trainium chips.
Amazon has also invested heavily in Anthropic.
Under the April agreement, Amazon announced a $5 billion investment in Anthropic, with the potential for another $20 billion in the future. That followed roughly $8 billion in previous investment.
The relationship demonstrates how AI companies and infrastructure providers are increasingly becoming financially interconnected.
Amazon benefits from Anthropic's demand for AWS computing, while Anthropic receives access to the infrastructure it needs to expand Claude.
Google and Microsoft are also major partners
Anthropic's infrastructure commitments extend beyond Amazon.
The company expects to spend at least $111.1 billion with Google's cloud business and $31.4 billion with Microsoft under long-term infrastructure arrangements.
The relationships are notable because Google and Microsoft are also developing their own AI products.
Microsoft has invested heavily in OpenAI while also providing cloud infrastructure for AI companies, whereas Google operates its own Gemini AI models and has invested in Anthropic.
Anthropic acknowledged the potential tension in these relationships in its prospectus.
The company said Amazon, Google and Microsoft occupy several roles simultaneously: they can be investors, customers, cloud providers, distributors and competitors.
Anthropic warned that those interests may not always be aligned with its own.
Anthropic is also working with xAI
The filing revealed another major computing arrangement involving xAI, Elon Musk's artificial intelligence company.
Anthropic could spend as much as $84.5 billion through 2029 on Nvidia-based computing capacity supplied by xAI.
Unlike several of Anthropic's other infrastructure commitments, the xAI arrangements are largely cancellable with 90 days' notice.
The agreement illustrates the increasingly complex network of relationships developing around AI infrastructure.
Rather than building every component itself, Anthropic is securing computing capacity from several companies while also developing more dedicated infrastructure of its own.
AMD plans to supply additional computing capacity
Anthropic's filing also showed a deeper relationship with Advanced Micro Devices.
AMD has committed to purchasing up to $5 billion of Anthropic stock and is expected to supply AI computing capacity worth more than $20 billion.
The arrangement gives Anthropic another major source of computing infrastructure as it attempts to reduce the risk of depending too heavily on any single provider.
The development also reflects the growing competition between semiconductor companies to supply the AI industry.
Nvidia remains a dominant provider of AI accelerators, while AMD has been expanding its own AI-chip business.
Anthropic's financial performance
The infrastructure commitments come as Anthropic experiences rapid revenue growth but continues to spend heavily on computing.
According to its IPO prospectus, Anthropic's revenue increased roughly 12-fold in 2025 to $4.59 billion.
However, the company reported an operating loss of about $8.06 billion, with computing and infrastructure costs representing a major expense.
Anthropic's reported net loss was approximately $42 billion, although a substantial portion of that figure resulted from accounting adjustments connected to financing instruments rather than ordinary operating expenses.
At the end of 2025, Anthropic held about $20.28 billion in cash and short-term investments, according to reporting based on the prospectus.
The numbers underline the financial challenge of operating at the frontier of AI development.
Even with billions of dollars in annual revenue, the cost of computing, research, data centres and other infrastructure remains extremely high.
Revenue is growing, but customer concentration remains a risk
Anthropic's prospectus also shows that its rapidly growing revenue is concentrated among a relatively small number of customers.
Its two largest direct customers each accounted for about 12% of 2025 revenue, meaning the two together represented roughly 24% of sales.
That concentration creates a different kind of risk from Anthropic's infrastructure obligations.
While the company has committed to pay large amounts to infrastructure providers, some customers are not locked into similarly long-term agreements and could reduce their spending.
This creates a situation in which some of Anthropic's costs are relatively inflexible while parts of its revenue base may be more variable.
The AI infrastructure race is becoming more expensive
Anthropic's $518 billion commitment is comparable in scale with other enormous AI infrastructure plans.
OpenAI's Stargate project, for example, was announced as a planned $500 billion AI infrastructure initiative, with costs expected to be shared among OpenAI, SoftBank, Oracle and MGX.
Anthropic's commitments therefore provide another indication of the extraordinary amount of capital being directed toward AI infrastructure.
The spending is not limited to computer chips.
AI infrastructure requires data centres, electricity, cooling systems, networking equipment, cloud capacity and long-term facilities.
As companies compete for limited capacity, they are increasingly entering multiyear agreements to guarantee access.
The commitments create both opportunity and risk
For Anthropic, locking in infrastructure provides greater certainty that it can continue training and deploying increasingly powerful AI systems.
But it also means the company could face substantial financial obligations if demand for Claude grows more slowly than expected, if AI technology changes rapidly or if newer computing systems make existing capacity less attractive.
Anthropic itself warned investors that if access to third-party computing is curtailed, repriced or terminated, its business and financial condition could be adversely affected.
The company's dependence on major technology providers is therefore both a source of computing capacity and a potential vulnerability.
Anthropic's IPO puts the spending under scrutiny
The disclosures come as Anthropic prepares for a potential public listing.
The company filed confidentially with the U.S. Securities and Exchange Commission in June, meaning its complete IPO paperwork has not yet been publicly disclosed.
The infrastructure commitments are likely to be closely examined by prospective investors because they represent substantial future financial obligations.
The company's rapid revenue growth provides one side of the investment story, while its enormous computing expenses and long-term commitments provide another.
The filing offers investors a clearer picture of the scale of spending required to remain competitive in frontier AI.
A bet on sustained AI demand
Ultimately, Anthropic's infrastructure strategy reflects a major bet: that demand for advanced AI will continue growing rapidly enough to justify the enormous cost of securing computing power.
The company says the availability of compute will be one of the principal limits on future AI development.
Its contracts show how seriously Anthropic is acting on that assessment.
With at least $518 billion in long-term infrastructure commitments, most of which cannot simply be cancelled, Anthropic has effectively committed itself to a decade-long expansion of its computing capacity.
Whether the investment produces sufficient returns will depend on the growth of Claude, enterprise AI adoption, the cost and availability of computing, and the pace at which AI technology evolves.
For now, the filing provides one of the clearest indications yet of the financial scale behind the race to build increasingly powerful AI systems.
