OpenAI and Anthropic Have Bet Billions on Massive Data Centers. Now They’re Looking at Smaller Ones,
Anthropic and OpenAI are looking beyond massive, multibillion-dollar data center projects and pursuing smaller deployments that could bring artificial intelligence computing capacity online faster.
The two leading AI labs have spent the past year securing enormous amounts of computing infrastructure, including facilities measured in hundreds of megawatts and even gigawatts. But people familiar with the discussions told CNBC that both companies are now exploring deals for significantly smaller sites with roughly 20 to 30 megawatts of capacity.
Anthropic has discussed potential agreements of that size in the United Kingdom and Nordic countries, four people familiar with the talks told the outlet. OpenAI has also explored smaller deployments in the Nordics, according to two of the sources. At least one source was familiar with discussions involving both companies about similar projects in the United States.
Smaller facilities could complement those developments from larger ones by giving AI companies access to computing power sooner as large data center projects become more complicated. “We’re building a diversified compute portfolio to meet growing demand for AI around the world,” an OpenAI spokesperson said.
“Different workloads need different infrastructure, so we have conversations with a range of partners and assess opportunities based on our requirements, performance, reliability, timing and cost,” the spokesperson added.
The smaller-data-center push comes on the heels of extraordinary infrastructure commitments from both companies.
Anthropic struck a roughly $45 billion cloud agreement with Nscale that includes about 460 megawatts of computing capacity at a West Virginia data center development, Bloomberg reported. OpenAI, meanwhile, said in April that commitments connected to its Stargate AI infrastructure initiative had already surpassed its original 10-gigawatt target.
The company has since committed to another 3 gigawatts of development in Georgia and 8 gigawatts in Ohio. But enormous projects can take years to build and increasingly face obstacles. Communities across the United States have pushed back against new data centers over concerns including electricity demand, water use, land development and their potential impact on utility costs.
Europe presents another challenge, with available power and suitable land constrained in several important data center markets. That makes smaller projects attractive because of what Jabez Tan, head of research at Structure Research, described to CNBC as “speed to usable capacity.” “Securing a few megawatts at an existing powered site can be more practical than waiting for a much larger block in one location,” Tan said.
Training cutting-edge AI models generally requires enormous clusters of chips operating closely together. Inference, the process of running trained models and responding to users’ requests, can often be distributed across smaller clusters in multiple locations, making smaller centers more attractive.
As products such as ChatGPT and Claude attract more users and AI becomes embedded across businesses, the infrastructure required to serve those models could become an increasingly important part of data center demand.
Real estate firm JLL expects the share of data center capacity devoted to inference to surpass training in 2027. Its projections show inference rising from 9% of global data center workloads in 2025 to 37% by 2030. Training, by comparison, is projected to account for 13% by the end of the decade.
The trend is already attracting infrastructure companies. Nvidia announced in February that it would work with data center stakeholders to study smaller facilities designed for distributed inference.
