AI labs pivot to smaller deals

Anthropic and OpenAI are now hunting for smaller data center deals of 20-30 megawatts, shifting from their earlier focus on massive multi-gigawatt projects.

Exploring UK, Nordic, US sites

Anthropic has explored smaller capacity agreements across the UK and Nordic countries, while OpenAI has looked at similar deals in the Nordics and US.

Speed beats scale

Smaller deals offer faster access to usable capacity. Companies can deploy workloads quicker by securing a few megawatts at existing powered sites rather than waiting for massive new builds.

“Securing a few megawatts at an existing powered site can be more practical than waiting for a much larger block in one location.”

Training versus inference

Training AI models needs massive computing power, but running them day-to-day through inference can use smaller clusters spread across multiple locations.

“Training a large model typically requires many chips working closely together.”

Inference workloads surging

Inference is expected to overtake training as the dominant data center workload by 2027. In 2025, inference used 9% of global data center capacity versus 14% for training.

Crusoe invests in smaller sites

Crusoe, which built a major Texas data center for OpenAI, is now investing in smaller facilities that are faster and cheaper than large builds facing delays.

Crusoe raises $3.9 billion

The data center company announced a $3.9 billion funding round at a $30.9 billion valuation, reflecting booming demand from the AI buildout.