Cerebras plans to expand its European data center capacity to 200 MW by 2027, according to a report from Indiatimes. The American AI hardware company, best known for its wafer-scale processors, is scaling up its European footprint to serve rising demand for AI compute power.

The move puts Cerebras among the AI hardware makers that are no longer only selling silicon. They are building and operating the capacity themselves, close to the customers and the regulators that govern where European data can be processed. The expansion is a response to demand for AI inference in Europe.

That timeline is aggressive, two years is fast for capacity at this scale in Europe. Grid connection queues, permitting, and local power constraints have become the limiting factor on new builds rather than the availability of the hardware itself.

What it means for operators

Wafer-scale AI accelerators concentrate an unusual amount of power into a small physical envelope, and a 200 MW target signals racks that run far denser than a traditional data hall was designed to hold. When load arrives in blocks like this, the constraint moves from the chip to the room. Airflow, inlet temperature, and per-rack power headroom start deciding what can actually be deployed, and hall-level averages stop being a useful answer. Operators sharing a grid with builds of this size should also expect the power around them to tighten, which makes measured capacity, rack by rack, worth more to release stranded capacity.

Source: Indiatimes