OpenAI and Anthropic Seek 20-30MW Data Center Deployments for Inference Workloads

AI labs OpenAI and Anthropic are pursuing smaller-scale data center deployments of 20-30MW as they look to expand their inference footprints beyond the massive training clusters that have dominated AI infrastructure investment in recent years. The two companies are in talks to secure capacity deals at data centers, with the workloads likely tied to inference rather than training, according to CNBC.

The move signals a broadening of AI infrastructure strategy across the industry, as leading labs shift from an exclusive focus on gigawatt-scale training campuses toward a more diversified compute portfolio that includes smaller, distributed facilities capable of serving models in production. Inference — the process of running trained AI models to generate outputs — requires infrastructure with different performance, latency, and reliability characteristics than training, and is expected to grow substantially as AI applications reach mainstream adoption.

OpenAI has held discussions for such deployments with companies in the United States and the Nordic region, while Anthropic has pursued similar conversations in the United Kingdom and the Nordics. The geographic spread reflects both companies' ambitions to serve a global user base with lower-latency infrastructure positioned closer to end users.

"We're building a diversified compute portfolio to meet growing demand for AI around the world," OpenAI said in a statement. "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. We don't comment on specific commercial discussions."

Anthropic's infrastructure commitments have been particularly aggressive. The company has signed at least $517 billion in compute capacity leases over the past 11 months as it scales capacity to train and serve its Claude family of models, underscoring the enormous capital intensity of the frontier AI sector.

The shift toward 20-30MW inference deployments marks a notable departure from the industry's recent emphasis on multi-gigawatt training campuses. While those large-scale projects remain critical for developing next-generation models, the emerging demand for inference-optimized capacity could reshape how data center operators and developers plan their pipelines, creating new opportunities for facilities in regions with favorable power access, connectivity, and proximity to major population centers. For operators, the trend suggests growing demand for mid-sized deployments that can be brought online faster than the sprawling campuses built for training, potentially easing some of the supply chain and energy constraints that have slowed AI infrastructure buildout globally.