Anthropic and OpenAI Pursue Smaller Data Center Deals to Accelerate AI Deployment

Anthropic and OpenAI Target Smaller Capacity Agreements

Anthropic and OpenAI are actively pursuing smaller data center deals in a strategic shift from their recent focus on massive infrastructure projects1

1

. Sources familiar with the discussions told CNBC that both AI labs are now exploring compute capacity deployments of 20-30 MW capacity across the U.K. and the Nordics, a significant step down from the multi-hundred-megawatt and gigawatt deals they've announced over the past year2

2

. Four people with knowledge of the conversations confirmed that Anthropic has been sounding out agreements within that range across these European regions, while two sources indicated OpenAI had been exploring similar opportunities in the Nordics1

1

. One source also revealed familiarity with talks involving both companies about U.S. capacity deployments at that scale1

1

.

Speed to Usable Capacity Drives Strategic Pivot

The appeal of smaller data center deals lies primarily in what Jabez Tan, head of research at Structure Research, describes as "speed to usable capacity"1

1

. Securing a few megawatts at an existing powered site proves more practical than waiting for a much larger block in one location, according to Tan1

1

. For workloads that can operate across separate sites, a collection of smaller deployments can accumulate into substantial capacity2

2

. An OpenAI spokesperson confirmed this diversified approach, stating: "We're building a diversified compute portfolio to meet growing demand for AI around the world. 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"1

1

. Both AI labs typically rent compute capacity from data center operators and neoclouds and have sought large-scale, long-term agreements in the past1

1

.

Massive Infrastructure Commitments Continue

Despite the pivot toward smaller deals, both companies maintain substantial commitments to large-scale AI infrastructure projects. Anthropic inked a roughly $45 billion cloud deal with Nscale, which will see the AI lab rent around 460 MW of compute capacity at a data center development in West Virginia1

1

. OpenAI's Stargate project has surpassed its original commitment of 10 GW in April and has since committed to developing a further 3 GW in Georgia and 8 GW in Ohio1

1

. However, huge data center projects in the U.S. and further afield are increasingly facing pushback from local communities, with community opposition mounting1

1

. The sector is also under pressure in much of Europe, where available land and power are in short supply, creating significant power constraints1

1

.

Inference Workloads Reshape AI Deployment Strategies

The shift toward smaller deployments reflects fundamental changes in how AI compute is being utilized. Training a large model typically requires many chips working closely together, but many inference workloads can instead serve separate requests across multiple smaller clusters, opening up more locations1

1

. This distinction matters significantly as more AI compute moves from training workloads to serving models in production through inference. The proportion of total data center capacity used for inference workloads is expected to overtake training workloads in 2027, according to a report by real estate company JLL1

1

. In 2025, inference made up 9% of global workloads in data centers compared to 14% for training, but by 2030, inference is projected to use 37% of that capacity, compared to just 13% for training1

1

. In February, Nvidia announced collaboration with several data center stakeholders to study smaller-scale distributed data centers designed specifically for distributed inference1

1

. U.S. company Crusoe, which built a huge data center complex in Texas used by OpenAI, is now investing in smaller data centers that will be faster and cheaper than larger builds, which are facing delays across the U.S., according to the Wall Street Journal1

1

.