Anthropic Boosts AI Compute Deal with Google and Broadcom to 3.5 GW Capacity in 2027

Anthropic's expanded partnership with Google and Broadcom reflects a significant increase in AI training infrastructure, driven by soaring demand and revenue growth to $30 billion. This escalation underscores the rapid development and scaling in AI research and deployment, potentially influencing market dynamics. Further developments in contract scope or capacity could reshape compute resource availability for AI projects.

Anthropic has expanded its AI compute capacity agreements with Google and Broadcom, securing 3.5 gigawatts of tensor processing unit (TPU) power starting in 2027, according to reports from TechCrunch and Quartz. This deal reflects Anthropic's growing demand for AI infrastructure as its annual revenue run rate surpasses $30 billion, more than tripling since late 2025.

Broadcom, as detailed in filings highlighted by Tom’s Hardware, will supply the TPU capacity to Anthropic, marking a strategic alliance between leading chipmaker Broadcom, Google’s AI hardware division, and the AI research group. The expanded deal underlines rising needs for specialized AI chips and datacenter networking equipment to advance large-scale AI operations.

The Register adds that Broadcom is developing next-generation AI acceleration chips specifically tailored for Google, and Anthropic’s planned use of 3.5GW of these accelerators signals increased collaboration aimed at meeting future AI compute demands. The partnership illustrates the intertwining of hardware innovation with AI software development on a massive scale.

Hacker News underscores that this ramp-up is intended to support Anthropic’s rapid growth in AI model training and deployment, reinforcing the company’s position in the competitive AI market. The agreement represents one of the largest announced AI compute capacity arrangements to date.

Looking ahead, stakeholders will monitor how this expanded compute capacity influences Anthropic’s AI advancements and market positioning, especially as infrastructure scaling remains critical for next-generation AI capabilities. The 2027 start date indicates ongoing investment in long-term AI hardware partnerships.