Under the terms announced on April 6, Broadcom will continue to design and supply Google’s Tensor Processing Units (TPUs), the proprietary accelerators that power everything from Google Search to its Gemini AI models alongside networking components for Google’s next-generation AI racks, with commitments running through to 2031. Financial terms were not disclosed.

The deal cements Broadcom’s role as Google’s primary silicon design partner at a time when demand for custom AI chips is accelerating sharply.

Google’s TPUs have increasingly become a growth engine for Google Cloud, offering a credible alternative to Nvidia’s GPUs for AI workloads and attracting growing interest from third-party customers.

Broadcom shares rose around 3% in after-hours trading following the announcement, a market response that reflects just how strategically significant the custom chip business has become for the company.

The chipmaker’s AI-related revenue is projected to reach $46 billion in 2026, up 134% year-on-year, as hyperscalers and AI labs alike move away from general-purpose GPUs towards purpose-built silicon.

Alongside the Google agreement, Broadcom also confirmed a separate deal with Anthropic, the AI safety company behind the Claude model family. Under that arrangement, Anthropic will gain access to approximately 3.5 gigawatts of AI computing capacity drawing on Google’s TPU infrastructure, beginning in 2027, a notable signal that Anthropic is diversifying its hardware strategy beyond Nvidia GPUs and Amazon Web Services.

As model training and inference workloads scale dramatically, the economics of custom silicon are increasingly compelling, delivering better performance per watt and lower long-term costs than off-the-shelf alternatives.

For Broadcom, securing long-term commitments from both Google and Anthropic solidifies its position as an indispensable architect of AI infrastructure, sitting alongside Nvidia rather than in its shadow.

RELATED STORIES

AI chip demand continues to strain big tech supply chains

All about Jericho, Broadcom’s new chip to advance AI data centre networks