Anthropic is making an enormous bet on the infrastructure behind Claude.
The artificial intelligence company has lined up hundreds of billions of dollars in future cloud, computing and equipment obligations. Many of those commitments cannot simply be switched off if demand changes.
According to an IPO prospectus reviewed by Reuters, Anthropic expects infrastructure-related obligations of roughly $518 billion over the coming years. Around 80% of the commitments are either non-cancelable or require payment regardless of how much capacity Anthropic ultimately uses.
That puts a striking number behind something the AI industry has been talking about for years: frontier models need an extraordinary amount of computing power.
For Anthropic, securing that capacity now means accepting substantial financial commitments stretching well into the next decade.
Google, Amazon and Microsoft Sit at the Center of the Buildout
Anthropic’s future infrastructure bill is spread across several of the world’s largest technology companies. Reuters reported that the Claude developer has committed at least $111.1 billion to Alphabet’s Google, alongside $110 billion to Amazon and another $31.4 billion to Microsoft. These are not ordinary cloud arrangements where spending can always be reduced quickly when usage falls.
The Google commitment reportedly runs from April 2026 through July 2033. If actual spending comes in below the agreed amount, Anthropic could still be required to pay the difference. A similar structure applies to Amazon, where the commitment extends from May 2026 through April 2036.
Anthropic’s Microsoft obligation runs from November 2026 through May 2033. Reuters reported that the $31.4 billion commitment is non-cancelable unless Microsoft has an uncured material breach. Together, the deals show how far ahead frontier AI companies are now planning their compute needs.
Broadcom Adds Another Huge Piece to the Infrastructure Puzzle
Cloud contracts tell only part of the story. Anthropic also has approximately $161.2 billion in Broadcom-related equipment lease obligations, according to the prospectus reviewed by Reuters, with much of that capacity tied to long-term arrangements.
Broadcom has become an increasingly important part of Anthropic’s compute strategy. The company has been working with Google and Broadcom on next-generation AI infrastructure intended to support future model training and inference at much larger scale.
Anthropic has also avoided relying on a single accelerator platform. Claude workloads can run across Google TPUs, AWS Trainium chips and Nvidia GPUs, giving the company more flexibility in how it deploys models across different infrastructure environments.
That hardware diversity does not remove the financial pressure, though. Anthropic still has to support large multi-year commitments while making sure future usage grows enough to justify them.
Anthropic Is Reserving Compute Years Before It Needs All of It
There is a practical reason AI companies are locking in capacity so early. Compute infrastructure cannot always be added on demand. High-end chips, power access, cooling, networking and data center construction all come with long lead times.
Anthropic’s relationship with Amazon reflects that reality. The companies announced plans involving up to five gigawatts of new computing capacity, giving Anthropic access to a major pool of infrastructure for training and operating Claude.
The company has also made extensive use of Amazon’s Trainium chips and other AWS systems. That gives Anthropic a large base of custom AI infrastructure while reducing its dependence on any one chip supplier.
This is a very different model from a startup renting compute only when needed. Anthropic is effectively reserving industrial-scale infrastructure years ahead of projected demand.
Claude’s Growth Is Driving the Spending
Anthropic’s infrastructure expansion is closely tied to rising demand for Claude. The company has reported strong growth across enterprise and consumer use, putting more pressure on the systems required to serve those workloads reliably.
Enterprise adoption has become particularly important. Claude is increasingly used through direct subscriptions, APIs and cloud platforms, while businesses are deploying the models for coding, research, automation, customer service and internal productivity tools.
More users also means more inference demand. Training the next generation of models may attract the most attention, but serving millions of prompts every day can consume vast amounts of compute on its own.
That helps explain why Anthropic is thinking beyond its immediate needs. The company is trying to avoid a future where demand grows faster than the infrastructure available to support it.
The $518 Billion Figure Comes With a Catch
The $518 billion figure is eye-catching, but it should not be treated as a single capital project or a lump-sum payment being made today. The total represents a combination of long-term cloud contracts, equipment leases and infrastructure commitments spread across several years.
What matters more is how those obligations are structured. A large share of them cannot easily be canceled, meaning Anthropic could still be responsible for payments even if usage grows more slowly than expected.
That creates a straightforward risk. If demand for Claude continues to rise sharply, Anthropic may benefit from having secured capacity early. If growth weakens, those same contracts could become expensive obligations.
The broader AI industry is moving in the same direction. Model developers want guaranteed access to compute, while cloud providers and chip companies want long-term commitments before investing in new infrastructure.
Anthropic’s Infrastructure Bet Extends Into the Next Decade
Anthropic’s filings offer a rare glimpse into the economics behind frontier AI. Building advanced models is only one part of the cost. Running them continuously for businesses and consumers requires another layer of infrastructure that is just as demanding.
Future Claude systems will need more chips, more data center space, more power and more networking capacity. Those requirements have to be planned well before the models themselves are released.
Anthropic is therefore making a long-term bet that AI demand will continue expanding through the 2030s. Its agreements with Google, Amazon, Microsoft and Broadcom are designed to make sure the company has enough infrastructure if that growth materializes.
The trade-off is clear. Anthropic gains access to enormous amounts of compute, but it also takes on obligations that may be difficult to unwind.
Sources
Reuters — Anthropic’s $518 billion AI buildout hinges largely on deals that cannot be canceled, filing shows
https://www.reuters.com/business/anthropics-518-billion-ai-buildout-hinges-largely-deals-that-cannot-be-canceled-2026-09-29/
Anthropic
https://www.anthropic.com/
