Cerebras Stock Sends Massive Expansion Signal at Paris Summit

The announcement lands just weeks after Cerebras completed the largest semiconductor IPO in history, raising $6.4 billion. The fresh capital appears to be going straight into global buildout, and Europe is now first in line.

'Shark Tank' Star Kevin O'Leary 'Can't Stand It' When Young People Spend $28 on Lunch But Only Make $70k — 'I Mean, That's Just Stupid'

Broadcom Lands $30 Billion Chip Deal With Apple. Why It's a Win-Win for AAPL and AVGO.

Intel Future Price Hikes Send Huge Demand Signal to Wall Street

European companies, research labs, and government agencies have grown tired of routing their AI workloads through data centers in the United States and Asia. That distance adds delay, something Cerebras has built its entire business around eliminating.

Feldman has argued for months that speed determines the usefulness of AI. People only keep using AI tools that respond quickly. If an AI is slow, people give up on it, and if it's fast, they use it constantly.

Most AI chips (like Nvidia's (NVDA) GPUs) store data on separate memory chips located next to the processor. Every time the AI needs that data, it has to send a request across a wire to fetch it, which takes time. This is the "memory bottleneck."

Cerebras uses a different process. Instead of using separate memory chips, it prints the memory directly onto the same giant chip as the processor. There's no wire and no round trip, so data moves almost instantly. Basically, Cerebras designs wafer-scale chips that skip the memory bottlenecks that slow down traditional GPU setups, and the same argument is now driving the company overseas.

Cerebras plans to bring its first European data center capacity online by the end of 2026, with facilities spreading across France and the Nordics. By the end of 2027, the company aims to reach a total European capacity of 200 megawatts. Norway and Finland are already on the list for new sites.

"We are contracting significant capacity for 2027, with data centers slated for Norway and Finland as we actively build across Europe," Feldman said. "These deployments will enable us to move decisively on what our customers have been asking for: fast, high performance AI compute located in Europe."

Part of that new European capacity is expected to support OpenAI workloads, extending the two companies' existing partnership. On the company's first earnings call as a public company, held in June, Feldman detailed a definitive agreement signed with OpenAI in December 2025 worth more than $20 billion in compute purchases over several years.

Cerebras says it moved from signed contract to running a live model for OpenAI in just 35 days. GPT 5.4 is now running on Cerebras hardware for select OpenAI customers, and the two companies are working to bring GPT 5.5 onto the platform next.

If a meaningful slice of that OpenAI demand shifts to European data centers, it gives Cerebras a built-in customer for the new capacity rather than a bet that demand will show up later.

Cerebras also struck a deal with Amazon.com's (AMZN) Amazon Web Services earlier this year, pairing AWS' Trainium chips for the initial processing stage of an AI request with Cerebras hardware for generating the response. Executives have said that impact should show up mainly in 2027.

For European businesses, the pitch is simple. Cerebras offers faster AI running on servers physically closer to home, with data that does not have to cross an ocean to get an answer.

"Our customers don't just want AI compute. They want it close to home, powered responsibly, and available fast," Feldman said. "This expansion and capacity plan reflects our confidence in Europe as a long term growth market for Cerebras."

Cerebras reported first quarter core revenue of $191.3 million, up 92% year-over-year (YOY) with cloud and services revenue jumping 167%. Executives have said the current constraint on growth isn't chip supply or customer demand.