OpenAI and Anthropic scramble for smaller data centers as massive gigawatt projects lag — 20-30 MW facilities to provide capacity as mega structures undergo construction

Anthropic and OpenAI are reportedly scrambling to secure smaller AI data center deployments as they race to bring more compute capacity online while gigawatt-scale projects undergo construction. According to a September 18 CNBC report, the two AI giants are exploring agreements for existing facilities with roughly 20-30 MW of capacity, despite already committing tens of billions of dollars to much larger AI infrastructure projects.

Anthropic has reportedly sounded out potential deals of that size across the UK and Nordic countries, according to CNBC, citing four people familiar with the discussions. OpenAI has also explored smaller deployments in the Nordics, while sources said both companies have discussed similar opportunities in the United States.

“We’re building a diversified compute portfolio to meet growing demand for AI around the world,” an OpenAI spokesperson told CNBC. The company added that different workloads require different infrastructure and that it evaluates potential deployments based on performance, reliability, timing, and cost. Anthropic declined to comment.

The smaller facilities would complement the enormous AI campuses both companies are already pursuing. Anthropic, for example, reportedly signed a roughly $45 billion agreement with Nscale for around 460 MW of compute capacity at a West Virginia development. OpenAI, meanwhile, said earlier this year that its Stargate infrastructure commitments had already surpassed the project's original 10 GW target, before announcing another 3 GW in Georgia and 8 GW in Ohio.

These projects provide the large clusters of accelerators needed to train increasingly large AI models, in which thousands of GPUs must work together via high-bandwidth interconnects. However, they also take years to develop, requiring land, grid connections, substations, cooling infrastructure, and huge amounts of power before the accelerators can begin operations.

Meanwhile, both OpenAI and Anthropic need immediate capacity to meet the inference demand for their models. As it turns out, inference can be distributed across multiple smaller clusters, making smaller data centers appropriate for these tasks and increasing the demand for such facilities.

Smaller deployments also circumvent some of the issues plaguing large data center buildouts. In the U.S. alone, local opposition blocked 45 data center projects worth $68 billion in the second quarter of 2026 over land usage, noise pollution, and electricity and water consumption. Existing, small facilities do not face the same pressure.

“Securing a few megawatts at an existing powered site can be more practical than waiting for a much larger block in one location,” Jabez Tan, head of research at Structure Research, told CNBC. “For workloads that can operate across separate sites, a collection of smaller deployments can add up to substantial capacity.”

Renting capacity is already standard practice for AI companies, as AI labs lease capacity from hyperscalers, dedicated data center operators, and rapidly expanding GPU-focused neoclouds. This allows them to spread workloads across multiple providers.

Follow Tom's Hardware on Google News, or add us as a preferred source, to get our latest news, analysis, & reviews in your feeds.