Detailed Analysis
Anthropic announced on April 6, 2026, a landmark infrastructure partnership with Google and Broadcom to secure multiple gigawatts of next-generation Google TPU compute capacity — its largest infrastructure commitment to date. The deal, reported by CNBC to represent approximately 3.5 GW of capacity, is expected to come online starting in 2027, with the vast majority of infrastructure sited within the United States. The TPUs involved are co-designed by Broadcom and manufactured by TSMC, with an emphasis on high efficiency measured in teraflops per watt. Anthropic CFO Krishna Rao framed the deal as a continuation of what the company describes as a "disciplined approach to scaling infrastructure," undertaken in direct response to surging demand for its Claude family of models.
The business context underlying this commitment is striking. Anthropic's annualized revenue run-rate has exceeded $30 billion in 2026, a dramatic acceleration from approximately $9 billion at the close of 2025. The company recently reported that the number of enterprise customers spending more than $1 million annually has doubled to over 1,000, signaling that Claude has achieved deep penetration in high-value commercial deployments. This revenue trajectory explains the urgency of securing compute at gigawatt scale, a threshold that would have seemed extraordinary even two years prior but now reflects the raw power requirements of frontier model training and large-scale inference workloads. The new Google-Broadcom deal follows earlier infrastructure expansions, including a commitment to access up to one million TPU chips announced in October 2025 and a broader $50 billion pledge to U.S. AI infrastructure made in November 2025.
The partnership also illuminates Anthropic's deliberate multi-platform hardware strategy. Rather than consolidating compute on a single provider, the company trains and runs Claude across a diversified portfolio that includes Amazon Web Services Trainium chips via Project Rainier — currently a million-chip cluster — as well as NVIDIA GPUs and Google TPUs. Claude itself is made available across all three major cloud platforms: AWS Bedrock, Google Cloud Vertex AI, and Microsoft Azure Foundry. This redundancy is not merely opportunistic; it provides both operational resilience and negotiating leverage, ensuring that no single hardware or cloud dependency can constrain deployment capacity. The deepened relationship with Google and Broadcom through this deal nonetheless represents a meaningful strategic alignment, given that Google has also been a significant equity investor in Anthropic.
At a broader industry level, the announcement reflects an accelerating pattern among frontier AI developers of treating compute acquisition as a primary competitive variable. The use of gigawatts as a unit of measurement for AI infrastructure has moved from the speculative to the operational, underscoring how dramatically the energy and capital intensity of the sector has scaled. Anthropic's willingness to commit to multi-year, multi-gigawatt contracts years before capacity comes online signals confidence in sustained demand growth and positions the company alongside hyperscalers in the scale of infrastructure planning required. The emphasis on U.S.-sited capacity also aligns with broader geopolitical currents around domestic AI sovereignty, suggesting that regulatory and national-security considerations are now embedded in infrastructure strategy at the highest levels of the industry.
Read original article →