US-based artificial intelligence chip manufacturer Cerebras is launching a multibillion-dollar expansion into the European market, with plans to bring its first data center capacity online by the end of 2026. The company intends to establish a comprehensive network of AI data centers across France and the Nordics by the end of 2027, boasting a combined power capacity of 200 MW. This strategic move aims to address the rapidly growing demand from European businesses, research institutions, and government entities seeking localized, low-latency infrastructure as an alternative to existing capacity in the US and Asia.
Chief Executive Andrew Feldman, speaking at the RAISE Summit in Paris, described the initiative as a massive expansion effort. He noted that the company is responding to extraordinary growth in demand for generative AI computing power, which is currently outpacing their ability to keep up. By deploying these data centers, Cerebras expects to satisfy specific European requirements regarding data sovereignty while providing high-speed AI inference infrastructure for increasingly complex workloads.
Power capacity serves as the primary metric for these facilities, as electricity remains the most significant constraint for AI scaling. While typical enterprise data centers operate between 1 and 20 MW, Cerebras’ planned network aligns more closely with large-scale infrastructure needs. A portion of this new capacity is earmarked to support existing partnership workloads with OpenAI. Furthermore, these new deployments will provide the high-performance inference chips necessary for AI agents. Appetite for inference-specific chips has exploded as more people use AI agents, a new type of interface that can carry out tasks autonomously on behalf of users.
The expansion comes at a time when investment in AI infrastructure is accelerating throughout Europe, a region where Nvidia currently dominates with technology powering over 90% of announced AI factory projects. Despite this, transatlantic tensions have led to increased caution among firms and governments regarding over-reliance on US providers. Cerebras, which focuses on chips dedicated to AI inference rather than intensive training, is well-positioned for this shift, bolstered by its successful $5.5 billion IPO in the US this past May, which remains one of the largest in Wall Street history.
Power capacity is the main yardstick for AI data centres because electricity has become the key constraint on expanding AI computing. For comparison, smaller enterprise data centres typically consume between 1 and 20 MW, whereas hyperscale facilities operated by cloud providers can draw 100 MW or more.
Founded in 2015, Cerebras has focused on chips dedicated to AI inference.
