Skip to content
Advertisement
Google deepens Anthropic TPU partnership as Broadcom locks in chip supply through 2031

Google deepens Anthropic TPU partnership as Broadcom locks in chip supply through 2031

Google 2015 logo (Wikimedia Commons, public domain)

Anthropic gets access to roughly 3.5 gigawatts of next-generation TPU capacity starting in 2027, with Broadcom designing the silicon

Google Cloud has expanded its partnership with Anthropic and Broadcom. Under the deal, Anthropic gets access to around 3.5 gigawatts of next-generation Tensor Processing Unit capacity, starting in 2027.

The agreement was announced on April 6, 2026. It builds on commitments the companies first made in October 2025.

Who does what in this deal

The deal has three players, and each has a distinct job.

Google owns the TPU, its custom AI chip. TPUs power the company’s internal machine-learning work, including its Gemini models and various consumer services.

Broadcom co-designs those chips with Google, a working relationship that dates back to around 2016. As part of this arrangement, Broadcom finalized a long-term agreement to develop and supply future Google TPUs, plus networking and rack components, through 2031.

Anthropic is the customer, and an increasingly large one. The AI startup behind the Claude models is committing to a huge amount of Google compute.

The money behind the machines

Broadcom executed a $63 billion structured financing agreement, known as Project North Star, around late October 2026.

Advertisement

The package is backed by major financial institutions, including Bank of America, Citigroup and Blackstone. Its purpose is to finance Anthropic’s purchases of TPUs.

Anthropic’s own filings suggest the appetite is real. According to its S-1 filing, the company projects spending of more than $114 billion with Google.

The broader picture is even larger. As of December 2025, the S-1 highlighted $417 billion in total compute commitments for the future.

Anthropic’s revenue is climbing fast

Anthropic reported a revenue run-rate exceeding $30 billion in April 2026.

At the end of 2025, that number was approximately $9 billion.

A run-rate projects current revenue over a full year. It is a snapshot of momentum, not a guarantee of annual results.

Google wants TPUs everywhere

Google plans to expand TPU sales beyond its own cloud. The company intends to make them available on platforms such as Oracle Cloud Infrastructure starting next year.

Not putting all the chips in one basket

Anthropic is not exclusive to Google. The company runs on a range of hardware, from Google TPUs to NVIDIA chips and AWS solutions.

What this means for the AI compute race

For Broadcom, the agreement through 2031 provides long-term visibility. Designing chips, networking and rack components for Google over that span gives it a steady role in the AI buildout.

For Anthropic, the arrangement secures the capacity it needs to keep training and serving models. The tradeoff is enormous long-term obligations, with $417 billion in compute commitments outlined in its S-1.

The financing structure is the part worth watching most closely. Project North Star ties banks and private capital directly to Anthropic’s chip purchases.

There is also a competitive ripple effect. Commitments this large may push other AI developers and cloud providers to accelerate their own infrastructure spending to keep pace.

Google spent roughly a decade building TPUs mostly for itself, and it now appears ready to sell them to everyone else.

Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our Editorial Policy.