Inside the numbers behind Anthropic's biggest infrastructure commitment yet — from 2027 TPU deliveries to a run-rate revenue jump past $30 billion.
Anthropic has signed a new agreement with Google and Broadcom to secure multiple gigawatts of next-generation TPU (Tensor Processing Unit) capacity, with the infrastructure expected to begin coming online in 2027. The company describes the vast majority of that capacity as US-based, extending a commitment announced in November 2025 to invest $50 billion in American computing infrastructure.
The announcement lands alongside a sharp acceleration in Anthropic’s commercial performance. Run-rate revenue has surpassed $30 billion, up from roughly $9 billion at the end of 2025 — more than a tripling in eight months. The deal is the clearest signal yet that the constraint on frontier AI growth is no longer customer interest but the availability of computing power.
What Anthropic Announced
The agreement deepens two existing relationships rather than creating entirely new ones. It expands Anthropic’s work with Google Cloud, building on the increased TPU capacity announced in October 2025, and strengthens its ties with Broadcom, which designs and supplies custom silicon used in large-scale AI systems.
Anthropic has not published a full deployment schedule. The company says capacity will start arriving in 2027, but an exact timeline for bringing all of the contracted gigawatts into service is Not specified.
Gigawatts, in this context, describe electrical power draw rather than chip counts — a shorthand the industry now uses because data centre power availability has become the practical ceiling on AI expansion.
Key Details at a Glance
| Item | Detail |
|---|---|
| Partners named | Google and Broadcom |
| Technology | Next-generation TPUs (Tensor Processing Units) |
| Scale | Multiple gigawatts of capacity |
| Timeline | Begins coming online in 2027; full deployment schedule Not specified |
| Primary location | Vast majority sited in the United States |
| Related commitment | $50 billion US computing infrastructure investment announced November 2025 |
| Run-rate revenue | Surpassed $30 billion, up from approximately $9 billion at end of 2025 |
| Large enterprise customers | Over 1,000 spending more than $1 million annually as of April 2026 |
| Primary cloud and training partner | Amazon Web Services, including collaboration on Project Rainier |
| Cloud availability of Claude | AWS (Bedrock), Google Cloud (Vertex AI), Microsoft Azure (Foundry) |
| Deal value | Not specified |
| Official website | anthropic.com |
Revenue and Enterprise Demand Behind the Expansion
The commercial backdrop explains the scale of the commitment. Anthropic’s run-rate revenue — an annualised figure based on current recurring performance rather than audited full-year results — has moved past $30 billion from roughly $9 billion eight months earlier.
The composition of that growth matters as much as the headline. When Anthropic announced its Series G fundraising in February 2026, it reported more than 500 business customers each spending over $1 million a year on Claude services. By April 2026, fewer than two months later, that figure had doubled to exceed 1,000 customers.
Doubling the count of seven-figure accounts in under two months points to deployments moving out of pilot stage and into production workloads. Organisations spending at that level are typically embedding a model into customer-facing products, internal engineering workflows, or document-heavy operations rather than running experiments.
Why Anthropic Runs Claude on Multiple Chip Platforms
Anthropic trains and serves Claude across a range of AI hardware, including AWS Trainium chips, Google TPUs and NVIDIA GPUs. The company frames this as a way to match individual workloads to the silicon best suited to them, improving performance while reducing exposure to any single supplier.
Performance matching. Training runs, fine-tuning and high-volume inference have different characteristics, and different accelerators handle them with different efficiency.
Supply resilience. Advanced AI chips have faced periodic supply pressure across the industry. Spreading demand across vendors limits the impact of any one shortage or delay.
Continuity for customers. Enterprises running Claude in critical systems need confidence that capacity constraints on one platform will not interrupt service.
Amazon Web Services remains Anthropic’s primary cloud provider and training partner, and the two companies continue to collaborate on Project Rainier. The Google and Broadcom agreement adds capacity rather than replacing that relationship.
Where Claude Sits Across the Cloud Market
Claude is available to customers across all three of the largest cloud platforms: Amazon Web Services through Bedrock, Google Cloud through Vertex AI, and Microsoft Azure through Foundry. Anthropic describes Claude as the only frontier model with that breadth of availability.
For buyers, multi-cloud distribution reduces the pressure to migrate an entire technology stack in order to adopt a particular model. For Anthropic, it broadens the addressable market without requiring the company to own every layer of delivery.
Why This Matters
The clearest read-through is that enterprise AI spending has moved past the experimental phase. More than 1,000 organisations committing over $1 million a year each to one vendor suggests frontier models are being treated as operating infrastructure with budget lines to match.
It also reframes where competitive advantage sits. If demand continues to outpace available compute, the ability to secure power, chips and data centre capacity years in advance becomes as decisive as model quality itself.
There is a risk dimension too. Committing to capacity that arrives in 2027 requires a bet that demand will still be climbing when it does. Anthropic is signalling confidence that it will be — but capacity contracted years ahead is, by nature, a forecast rather than a certainty.
Industry Context: Compute as a Strategic Asset
Anthropic is not alone in this. Major AI developers and cloud operators have broadly moved toward long-dated hardware and data centre commitments as a way of guaranteeing future supply. Publicly announced terms, timelines and volumes vary widely between companies and are frequently disclosed only in part.
The structure of Anthropic’s deal is also typical of how the AI supply chain has matured. Rather than building everything in-house, leading model developers increasingly assemble capacity through partnerships with specialised silicon designers and hyperscale cloud providers, spreading capital requirements across several balance sheets.
Readers should treat cross-company comparisons cautiously. Metrics such as run-rate revenue, contracted gigawatts and customer counts are defined differently by different organisations and are not always directly comparable.
What Happens Next
The immediate milestone to watch is 2027, when the new capacity is expected to start coming online. Until then, Anthropic will continue serving demand through existing infrastructure and partnerships, including its work with AWS.
Other open questions include the financial terms of the agreement, the specific sites involved, and how the capacity will be split between training and inference. None of those details have been disclosed, and further specifics are Not specified at this stage.
Anthropic has said its focus remains on scaling Claude’s capabilities while maintaining the safety and alignment standards it has emphasised in recent months. Company announcements are published on its official website at anthropic.com.
Editorial Note
This article reports on a corporate announcement and publicly stated figures. It is general news coverage, not investment, procurement or technical advice. Figures such as run-rate revenue are company-reported and are not audited annual results. Readers making business decisions should verify current details directly with the companies involved.
Frequently Asked Questions
What exactly did Anthropic announce? A new agreement with Google and Broadcom to secure multiple gigawatts of next-generation TPU capacity, with infrastructure expected to begin coming online in 2027.
What is a TPU, and why are gigawatts used to measure the deal? TPUs are Tensor Processing Units, chips designed by Google to accelerate AI workloads. Gigawatts measure electrical power draw, which has become the practical way to describe the scale of large AI data centre capacity.
When will the new capacity be available? Anthropic expects it to start coming online in 2027. A timeline for full deployment of all the contracted capacity is Not specified.
How much is the deal worth? Not specified. Anthropic did not disclose financial terms in the announcement.
Where will the infrastructure be located? The vast majority will be sited in the United States, extending the $50 billion American computing infrastructure commitment Anthropic announced in November 2025.
How fast is Anthropic’s revenue growing? Run-rate revenue has surpassed $30 billion, up from approximately $9 billion at the end of 2025 — more than a tripling in eight months. Direct comparisons with other AI companies are not published by the company.
How many enterprises are spending heavily on Claude? More than 1,000 business customers were each spending over $1 million annually as of April 2026, doubling from the 500-plus reported at the time of the February 2026 Series G announcement.
Does this replace Anthropic’s relationship with AWS? No. Amazon Web Services remains Anthropic’s primary cloud provider and training partner, and the companies continue to collaborate on Project Rainier.
Why does Anthropic use several hardware platforms? Running across AWS Trainium, Google TPUs and NVIDIA GPUs lets the company match workloads to suitable hardware, improve resilience and reduce dependence on a single vendor.
Does the deal change where customers can access Claude? No. Claude remains available through AWS Bedrock, Google Cloud Vertex AI and Microsoft Azure Foundry.
