Cerebras announces six new AI inference data centers in North America and Europe

Cerebras/WSE-3 / CS-3

Cerebras announced six new inference data centers in North America and Europe, including Oklahoma City (300+ CS-3 systems) and Montreal, raising its capacity 20-fold to over 40 million Llama 70B tokens per second. About 85% of the capacity is in the US; customers included Mistral, Perplexity and Hugging Face.

• Six new facilities: Santa Clara, Stockton, Dallas (online); Minneapolis Q2, Oklahoma City Q3, Montreal Q3, Midwest/East and Europe Q4 2025.
• Capacity: 40 million Llama 70B tokens/sec aggregate, a 20-fold increase; Oklahoma City alone has 300+ CS-3 systems.
• 85% of capacity U.S.-based; customers: Mistral (Le Chat), Perplexity (AI search), Hugging Face (model platform).