Anthropic and OpenAI are pivoting towards smaller AI data center agreements, targeting capacities of 20-30 megawatts, as they have already secured substantial deals for larger facilities. This shift is driven by the need for faster deployment of AI workloads, particularly as the industry faces community pushback against large-scale projects and limited resources in Europe.
OpenAI has previously committed to significant expansions, including a $45 billion deal with Nscale for 460 megawatts in West Virginia and additional projects totaling 11 gigawatts across Georgia and Ohio. The trend towards smaller deployments is underscored by the increasing importance of inference workloads, which are expected to surpass training workloads in data center capacity by 2027.
This change in focus is also reflected in the activities of companies like Crusoe, which is investing in smaller, more efficient data centers to meet the evolving demands of AI applications