Anthropic, Google, and Broadcom expand AI infrastructure ties

Anthropic has entered into expanded chip and computing agreements with Google and Broadcom that the companies describe as worth hundreds of billions of dollars, covering multiple gigawatts of next-generation compute capacity, according to reporting by the Financial Times and the Wall Street Journal.

Broadcom will supply future versions of Google's custom AI chips under the expanded collaboration, while also providing Anthropic with access to computing capacity, CNBC reported. The deal represents a deepening of relationships that span chip design, manufacturing commitments, and cloud infrastructure provisioning.

The agreements reinforce Anthropic's position as one of the most heavily resourced independent AI developers. The company separately disclosed that it has topped $30 billion in annualised revenue, according to The Information, a figure that would place it well ahead of earlier public estimates of its commercial traction.

For Broadcom, the deals extend its role as a designer of custom AI accelerators for hyperscale clients. The company's stock gained following a regulatory filing that disclosed details of the Google arrangement, according to Barron's. Google has long relied on Broadcom for the design of its Tensor Processing Units, and the new agreement appears to extend that relationship into forthcoming chip generations.

The scale of compute being contracted reflects the capital intensity now required to train and run frontier AI models. Anthropic, backed by Google among others, is locking in infrastructure supply at a moment when access to AI-grade silicon remains a binding constraint across the industry.