Anthropic announced a large expansion of its cloud computing deal with Google Cloud. The company said it may use as many as one million of Google’s Tensor Processing Units over the next year. This capacity is expected to bring more than one gigawatt of AI compute online by 2026.
Anthropic described the arrangement as worth tens of billions of dollars. The company said the extra TPU capacity will support training and serving future versions of its Claude models. The deal will add raw compute and Google Cloud services to Anthropic’s existing infrastructure mix.
Why Anthropic chose TPUs
Anthropic said it selected Google TPUs for their efficiency and price performance. The company has already run models on Google hardware and on other clouds. Anthropic leaders said TPUs give an efficient path to scale without building massive data centers.
Google is also improving its TPU line. The newest generation, called Ironwood, aims to raise throughput and lower cost. That helps companies that need large-scale training and inference.
What this means for Claude
Anthropic said the added compute will let it train and run larger and faster Claude models. The company targets enterprise features and tools that need persistent capacity. The expansion aims to meet what Anthropic called rapidly growing demand from business customers.
Anthropic also noted that Claude Code and other developer-facing tools have driven substantial usage. The extra capacity is meant to keep response times low and to cut operating costs per query.
Scale and industry context
The scale of this deal is notable when compared with prior arrangements across the industry. Companies are racing to secure large blocks of compute. Anthropic’s deal follows a string of massive supply agreements in the AI sector. Reuters and other outlets noted the broader trend toward multibillion-dollar compute commitments.
Anthropic’s move also highlights how cloud providers compete on custom AI hardware as an alternative to vendor GPUs. Firms that need steady and efficient training capacity are weighing TPUs against GPUs and other accelerators.
Business and customer signals
Anthropic serves hundreds of thousands of business customers. The company has said enterprise demand forms a large share of its revenue. The extra TPU capacity supports larger enterprise contracts and global deployments. Industry reporting notes that Anthropic aims to scale revenue rapidly in the next year.
Google also has a commercial interest in the deal. The company has invested in Anthropic in recent years and is positioning Google Cloud as a place to run industry-scale AI workloads. The agreement deepens that commercial link.
Technical and operational notes
A one gigawatt of compute does not mean one gigawatt of electrical draw for a single data center. It describes an industry measure of peak AI processing capacity. The deal will bring capacity across Google Cloud regions in stages through 2026.
Anthropic will combine TPUs with other chips in its deployments. Anthropic will retain multi-cloud relationships. The company has used several clouds and chip types in the past. This deal does not remove other supplier options, but it does give Anthropic a large, long-term source of TPU capacity.