Anthropic deepens its Google TPU bet with a multi-gigawatt compute deal

Anthropic is doubling down on the infrastructure race behind artificial intelligence. On April 6, 2026, the company said it had expanded its partnership with Google and Broadcom to secure multiple gigawatts of next-generation TPU capacity, with the new supply expected to begin coming online in 2027.

The deal is a reminder that the biggest AI stories are no longer only about new model releases. They are also about who can secure enough chips, power, and data-center capacity to train and run those models at scale.

• Anthropic says the expanded capacity is tied to future Claude development.
• Google and Broadcom are both involved in the supply arrangement.
• The capacity is expected to start coming online in 2027.

A bigger bet on infrastructure

Anthropic said the agreement builds on its existing relationship with Google and Broadcom and will provide access to multiple gigawatts of TPU capacity. The company framed the move as part of a disciplined push to scale infrastructure in step with customer demand and future model development.

Broadcom separately disclosed that it had also signed a long-term agreement with Google to develop and supply future generations of custom AI chips and related components through 2031. Reuters reported that Anthropic’s portion of the arrangement equates to about 3.5 gigawatts of Google TPU compute starting in 2027.

Why the deal matters

For Anthropic, the logic is straightforward: more compute means more room to train larger models, serve more enterprise customers, and keep pace with rivals. For the broader AI industry, the announcement reinforces a growing reality that access to specialized hardware has become a strategic advantage in its own right.

That shift is reshaping competition among major AI labs. Product launches still matter, but the companies that can secure long-term compute supply may be better positioned to ship faster, train more ambitious systems, and support rising demand without bottlenecks.

The compute race is now public

The April 6 announcement arrives as AI companies continue to sign increasingly large infrastructure deals, often with chipmakers, cloud providers, and custom accelerator partners. Those agreements are now a major signal to investors and customers alike, not just a back-end detail.

Anthropic did not say how much the expanded arrangement will cost, and the companies did not publicly detail every technical term of the agreement. But the scale alone makes the message clear: frontier AI is increasingly a race to secure power, silicon, and long-term capacity before the next wave of demand hits.

What to Watch

Watch for whether rivals respond with similar compute-heavy partnerships, and whether Anthropic begins to translate this expanded capacity into faster model releases or new enterprise capabilities. The other question is whether supply, power, or regulatory constraints slow the pace at which these enormous AI infrastructure plans can actually be deployed.