Anthropic Locks In $200B Google Cloud Deal Through 2031

Anthropic agreed to spend $200 billion with Google Cloud over five years, the largest cloud-services commitment ever signed by an AI company. The Information reported the deal Tuesday. The contract pulls Google’s tensor processing units, hosting capacity, and Google Cloud software through 2031. The figure represents more than 40% of Google Cloud’s $462 billion contractual backlog disclosed at the end of Q1 2026.

Stack that against what Anthropic actually earns today. Annualized run rate sits near $30 billion as of March, up from roughly $9 billion at the close of 2025. The new Google bill averages $40 billion a year for the next five.

That’s the easy part of the story. The harder part is the math underneath.

The $200 Billion Number, And Why Google’s Backlog Just Got Personal

Alphabet shares climbed about 2% in after-hours trading after the report broke on May 5. The move was small for a number this big, suggesting investors had already priced in most of the dependency. What hasn’t been priced in is who pays whom.

Two of every five dollars Google has booked for future cloud delivery now come from a single customer building a single family of models. Reuters could not independently verify the figure, and Anthropic and Google declined to comment. What is on the public record is the April 6 disclosure of multi-gigawatt TPU capacity through Google and Broadcom, plus Alphabet’s separate plan for an equity tranche of up to $40 billion.

The contract sits on top of a partnership that has been deepening since 2023. Google holds about a 14% stake in Anthropic going into the latest equity round, and the new TPU capacity is co-designed with Broadcom on TSMC’s 2nm process node.

The Money Loop Nobody At Either Company Wants To Spell Out

Days before the $200 billion figure leaked, Alphabet announced an investment of up to $40 billion in Anthropic itself. The structure is staged. Ten billion in cash now at a $350 billion valuation, with another $30 billion released as Anthropic hits performance milestones.

Read the two contracts side by side and the loop is hard to miss. Google writes Anthropic a $40 billion equity check. Anthropic writes Google a $200 billion services check. Net of equity, Anthropic owes its part-owner $160 billion in cloud and chip spend.

This pattern isn’t isolated. Amazon committed up to $25 billion of equity to Anthropic in April, on top of $8 billion in prior rounds, disclosed in Anthropic’s expanded Amazon compute partnership announcement. In return, Anthropic agreed to spend more than $100 billion over ten years on AWS infrastructure, with nearly 1 gigawatt of Trainium2 and Trainium3 capacity coming online by year-end 2026.

Stack it together and the math is plain.

• $200 billion: Anthropic’s five-year commitment to Google Cloud through 2031
• $100 billion+: Anthropic’s ten-year commitment to AWS for Trainium and related capacity
• $30 billion: Anthropic’s annualized revenue run rate as of March 2026
• $380 billion: post-money valuation set by the February Series G recorded in Anthropic’s $30 billion Series G announcement

Roughly $300 billion in named compute commitments, plus undisclosed CoreWeave dollars, sit against a revenue line that has only just crossed $30 billion. The bet is the curve, not the current ledger.

Three Clouds, Four Chip Families, One Customer

Anthropic isn’t picking a side. It’s renting all of them. The company splits its training and inference workload across Google’s TPU silicon, Amazon’s Trainium accelerators, and Nvidia GPUs hosted by AWS, Google Cloud, and CoreWeave.

The CoreWeave piece is the newest leg. CoreWeave disclosed a multi-year agreement with Anthropic on April 10, with capacity scheduled to come online later this year on Nvidia’s Vera Rubin generation. Specifics on dollar value were withheld, but the deal helped push CoreWeave’s contracted backlog past $66.8 billion. Details sit in CoreWeave’s investor disclosure of the multi-year Anthropic agreement.

The diversification has a clear logic. No single foundry, no single fab line, no single accelerator generation can absorb this demand alone. Splitting the bill across four chip families also gives Anthropic pricing leverage at every renewal, even with Google now holding a meaningful equity stake.

How A $30 Billion Run Rate Funds A $40 Billion-A-Year Compute Bill

It doesn’t, on today’s numbers. Anthropic is forward-spending against a revenue curve that has compounded above 10x annually for three straight years, with Sacra estimating roughly 1,400% year-over-year growth into March.

The bet is that a 2026 Claude run rate of $30 billion stretches past $100 billion within the contract window, with margin expansion as TPUs and Trainium silicon undercut Nvidia hardware on cost per token. Anthropic’s CFO said as much on April 6 alongside Anthropic’s expanded Google and Broadcom partnership announcement.

Krishna Rao framed the spend as deliberate.

“This groundbreaking partnership with Google and Broadcom is a continuation of our disciplined approach to scaling infrastructure. We are building the capacity necessary to serve the exponential growth we have seen in our customer base while also enabling Claude to define the frontier of AI development.”

Independent analyst Ben Thompson read the same partnership through Google’s lens. “Anthropic needs compute, and Google has the most: it’s a natural partnership, particularly for Google,” he wrote in Stratechery’s analysis of the Anthropic-Google alliance. Thompson’s added concern is opportunity cost. Anthropic now has to allocate fixed compute across Claude consumer, Claude API, Claude Code, and frontier research at a margin that will reflect what it pays hyperscalers and neoclouds for the privilege.

Broadcom, Nvidia, And The Chip Stack Behind The Bill

Behind every dollar of cloud spend sits a piece of silicon. The $200 billion contract leans on Google’s seventh-generation Ironwood TPU and the eighth-generation parts Google previewed at TSMC’s 2nm node, with Broadcom co-designing the accelerators. Broadcom disclosed a 3.5-gigawatt compute commitment tied to the agreement in an SEC filing.

The capacity timeline maps the spend.

1. End of 2026: nearly 1 gigawatt of Trainium2 and Trainium3 capacity online for Anthropic via AWS
2. 2027: first wave of dedicated Google TPU capacity comes online under the $200 billion deal, per Google Cloud’s April 6 announcement of the expanded Anthropic TPU partnership
3. Through 2028: CoreWeave begins phased rollout of Nvidia Vera Rubin capacity
4. By 2031: full $200 billion in Google services and chips delivered

Energy is the other constraint behind the bill. A gigawatt of AI compute draws roughly the steady-state demand of a mid-sized U.S. city. Anthropic’s combined commitments add up to more than 5 gigawatts on the AWS side, multiple gigawatts on the Google side, and additional load on CoreWeave’s grid footprint.

Long-context training pulls especially hard on this kind of capacity, the same pressure shaping rivals chasing larger windows. A look at the model side of that race appears in our coverage of Subquadratic’s 12-million-token context window launch.

The chip diversity also addresses a single-supplier risk Anthropic learned the hard way through 2024. Heavy concentration on Nvidia silicon would leave the company exposed to allocation cuts during shortages and to whatever margin Nvidia chooses to charge. Splitting work across Trainium, TPU, and GPU forces every chip vendor in the stack to compete on cost per useful token.

That competition is exactly what Google has wanted for a decade. Custom TPU silicon, designed in-house and fabricated by TSMC, gives Google a pricing lever its competitors don’t have at the same scale.

The Risk Sitting Inside Google’s $462 Billion Backlog

A backlog this concentrated cuts both ways. If Anthropic delivers on the revenue curve its investors are paying for, Google Cloud’s anchor-tenant exposure compounds into a multi-year asset. If Anthropic stalls, two-fifths of Google Cloud’s booked future revenue stalls with it.

Across the three majors, contracts tied to Anthropic and OpenAI now account for more than half of the roughly $2 trillion in combined contracted cloud backlog at AWS, Microsoft Azure, and Google Cloud. Two AI labs. Half the future cloud revenue of three of the biggest companies on Earth. The dependency is structural, and it isn’t getting more diverse.

Some of that pressure is already showing up on benchmark scoreboards rivals use to grade frontier models. Recent results across cyber and reasoning tests appear in our breakdown of GPT-5.5, Mythos, and the latest AISI cyber evaluations.

Frequently Asked Questions

Is The $200 Billion Anthropic-Google Deal Officially Confirmed?

Not by either company. The Information published the figure on May 5, 2026, citing sources familiar with the agreement. Reuters could not independently verify the dollar amount, and both Anthropic and Google declined to comment. What is public is Anthropic’s April 6 announcement of multi-gigawatt TPU capacity through Google and Broadcom, plus Alphabet’s separate disclosure of up to $40 billion in equity investment at a $350 billion valuation.

Why Is Anthropic Spending More Than Its Current Revenue On Compute?

Because the bet is on the revenue curve, not the current run rate. Anthropic’s annualized revenue grew roughly 1,400% year-over-year to about $30 billion by March 2026. The company is signing five and ten-year capacity contracts that price out at a fraction of what spot GPU rentals would cost in 2030. Investors funded the strategy with a $30 billion Series G in February at a $380 billion post-money valuation.

Will This Deal Change Pricing For Claude Users?

Not directly, and not soon. The new Google capacity comes online starting in 2027, and the cost per token of TPU and Trainium silicon is meant to come in below current Nvidia GPU economics, not above. The bigger consumer-side variable is inference volume. Heavier Claude usage at the API and product level is what fills the capacity Anthropic is buying, which is also what funds the $40 billion-a-year cloud bill.

Could Anthropic Walk Away From Any Of These Contracts?

Very unlikely without major financial penalty. Multi-year cloud commitments at this scale carry minimum-spend clauses, and the equity tied to Google’s $40 billion plan and Amazon’s $25 billion stake creates further alignment. The contracts are structured so Anthropic’s compute demand and its biggest investors’ cloud revenue move in the same direction. A meaningful breach would also reset the equity terms across both partnerships.

How Does This Compare To Microsoft’s Deal With OpenAI?

It’s bigger on the cloud side. Microsoft’s Azure partnership with OpenAI underpins Azure’s AI revenue base, but a single $200 billion five-year commitment to one cloud is, on public reporting, larger than any single tranche Microsoft has booked from OpenAI. Anthropic has also spread its bets across three cloud vendors and four chip families, where OpenAI has only recently begun to diversify beyond Microsoft.

The next few quarters of Alphabet, Amazon, and Anthropic earnings will tell the harder story. Compute commitments at this scale have to reverse out somewhere, in either model revenue, gross margin, or the value of Google’s stake.

The first place to watch is whether Google Cloud’s $462 billion backlog grows or thins around its newest anchor tenant. The second is whether Anthropic’s revenue line keeps doubling fast enough to make the math work before 2031.