Cerebras Systems recently announced the construction of six new data centers across North America and Europe to boost its artificial intelligence (AI) inference capabilities. This expansion will significantly enhance the company's computing power, supporting the development of various AI applications.
The plan allocates 85% of computing capacity to the United States. Three facilities are already operational in Santa Clara and Stockton, California, and Dallas, Texas. New centers will open in Minneapolis (Q2 2025), Oklahoma City, and Montreal (Q3 2025), as well as Atlanta and France (Q4 2025).
Image Source Note: Image generated by AI, licensed through Midjourney.
At the heart of these new data centers is Cerebras's Wafer Scale Engine (WSE), a specialized chip architecture optimized for AI applications. The company states its CS-3 system can process 40 million tokens per second for the Llama-70B model, dramatically increasing inference speed. The Oklahoma City facility, expected to house over 300 CS-3 systems, is built to a triple-plus-three standard, withstanding tornadoes and earthquakes and featuring triple-redundant power supplies. It is projected to begin operations in June 2025.
Several prominent AI companies have partnered with Cerebras, including French startup Mistral and its Le Chat assistant, and AI question-answering engine Perplexity. Hugging Face and AlphaSense are also utilizing the Cerebras platform. This technology is particularly well-suited for inference models requiring extensive computation and large token generation, such as Deepseek-R1 and OpenAI o3.
This expansion is part of Cerebras's overall 2025 growth strategy. Some facilities will be operated in partnership with UAE-based G42. In Montreal, a new center managed by Enovum, a Bit Digital subsidiary, is expected to go live in July 2025, boasting inference speeds ten times faster than current GPUs.
Cerebras Systems, a US company specializing in AI chips, employs a unique design philosophy using an entire wafer as a single chip. They have released their third-generation Wafer Scale Engine, the WSE-3. This system is already in use at institutions such as Argonne National Laboratory, the Pittsburgh Supercomputing Center, and GlaxoSmithKline. While advantageous, the technology has limitations, such as lacking native CUDA support (Nvidia's standard) and exhibiting less server compatibility than Nvidia solutions.
Key Highlights:
🌍 Cerebras plans to build six new data centers in North America and Europe, primarily in the US, with full operation expected in 2025.
⚡ The data centers will utilize unique wafer-scale chips capable of processing 40 million tokens per second.
🤝 Several leading AI companies have partnered with Cerebras to leverage its high-speed inference capabilities.
