Anthropic is assembling roughly 10 gigawatts of compute capacity by the end of 2027, seven times the approximately 1.4 gigawatts it had online at the close of 2025. The expansion spans Google and Broadcom’s TPU (specialized AI processing chip) capacity, AWS Trainium, AMD MI450 GPUs and large data-center agreements across several continents. A partnership announced in April 2026 is intended to secure approximately 3.5 GW of next-generation TPU capacity beginning in 2027, while Anthropic’s AWS agreement includes up to 5 GW of Trainium capacity. Its AMD collaboration covers up to 2 GW of MI450 GPUs, with the first 1 GW expected online by the first half of 2027. Anthropic has also leased a 2.16 GW Australian campus for inference (running trained AI models), scheduled to launch in 2027, and agreed with Nscale to obtain 460 megawatts under a contract reportedly worth $45 billion over six years, with operations starting in late 2027. Compute capacity is projected to reach roughly 5 GW by the end of 2026 before doubling to the 10 GW target a year later. The push coincides with annualized revenue exceeding $30 billion by April 2026, up from roughly $9 billion at the end of 2025. Anthropic has more than 1,000 customers spending over $1 million annually on its Claude AI models, while projections suggest annualized revenue could exceed $65 billion. By combining suppliers and placing capacity in Australia alongside North American infrastructure, the company is positioning itself to address data-sovereignty requirements and reduce latency across markets as inference workloads become more geographically distributed than training.