Anthropic announced on April 6, 2026, a major expansion of its partnership with Google and Broadcom that will provide approximately 3.5 gigawatts of next-generation TPU compute capacity starting in 2027. The agreement, which extends through 2031, represents one of the largest AI infrastructure commitments to date and positions Anthropic to compete directly with OpenAI's compute capabilities.
Deal Combines Google Financing With Broadcom Custom Chip Manufacturing
The three-way partnership leverages Google's financing for Anthropic data centers while Broadcom handles custom TPU chip manufacturing and AI rack component supply through 2031. Broadcom has locked in a long-term supply agreement with Google to manufacture future TPU generations, ensuring consistent hardware availability for Anthropic's scaling needs. The compute capacity represents part of a broader "multiple gigawatts" agreement between the parties.
Vast Majority of Infrastructure Will Be US-Based
The new compute infrastructure will be predominantly sited in the United States, a strategic decision that could make Anthropic eligible for U.S. Department of Defense contracts. Previously, Anthropic lost government contract opportunities due to dependencies on non-US compute resources. This domestic focus aligns with broader U.S. efforts to maintain AI technology leadership while preventing intellectual property transfer to competitors.
Partnership Positions Anthropic for Frontier Model Competition
The massive infrastructure investment signals Anthropic's commitment to scaling frontier AI models and competing with OpenAI, which is also racing toward a potentially record-breaking IPO by the end of 2026. The 3.5 gigawatt allocation starting in 2027 provides the computational foundation needed for training increasingly large language models and multimodal systems. Bloomberg reported that Broadcom will ship Google TPU chips directly to Anthropic as part of the arrangement.
The announcement reached the Hacker News front page with 88 points and 28 comments, indicating strong interest from the developer community in AI infrastructure developments.
Key Takeaways
- Anthropic will access approximately 3.5 gigawatts of next-generation TPU compute capacity starting in 2027 through a partnership with Google and Broadcom extending to 2031
- Broadcom secured a long-term deal to manufacture Google's future TPU chips and supply AI rack components through 2031
- The vast majority of new compute infrastructure will be located in the United States, potentially making Anthropic eligible for Department of Defense contracts
- The partnership combines Google financing for data centers with Broadcom's custom chip manufacturing capabilities
- The announcement comes as both Anthropic and OpenAI prepare for potentially record-breaking IPOs by the end of 2026