OpenAI and Anthropic are shifting their infrastructure strategy toward smaller data centers. While previously focused on massive projects like the $500 billion Stargate initiative, the companies are now seeking facilities with capacities of 20–30 MW to accelerate scaling.
Anthropic is exploring partnerships in the UK and Scandinavia, while OpenAI is also looking for partners in the Nordic region. Both companies are considering the construction of smaller data centers within the United States to bypass the delays and local opposition associated with gigawatt-scale sites.
The industry shift is driven by a move from model training to inference. According to JLL, the share of data centers dedicated to inference is expected to grow from 9% in 2025 to 37% by 2030, while training-focused capacity will remain lower.
Market players are adapting to this trend. Nvidia is focusing its support initiatives on smaller projects, and the company Crusoe has shifted its strategy toward smaller facilities. Crusoe recently secured $3.9 billion in funding, reaching a valuation of $30.9 billion.
