Table of Contents
Cerebras told investors that the firm is in early discussions regarding potential data center deployments in Israel, the UAE, Australia, Singapore, India, and Indonesia
In sum – what to know:
Global infrastructure push – Cerebras is expanding data center capacity across North America and Europe while exploring potential deployments in additional international markets.
Supply chain advantage – The company said its wafer-scale architecture avoids industry bottlenecks tied to HBM memory, CoWoS packaging, and advanced semiconductor manufacturing nodes.
Strategic partnerships deepen – Cerebras highlighted progress with OpenAI and AWS as it seeks to capitalize on growing demand for high-speed AI inference.
Cerebras Systems is accelerating its global data center expansion as demand for AI inference infrastructure continues to grow, with the company adding capacity across North America and Europe while exploring additional international deployments.
During the company’s first-quarter 2026 earnings call, co-founder, chief executive officer and president Andrew Feldman said Cerebras has added new data centers across the United States, Canada, and Europe, including deployments in France and the Nordic region. The company is also in early discussions regarding potential data center deployments in Israel, the UAE, Australia, Singapore, India, and Indonesia.
“It’s a dogfight out there,” Feldman said, referring to competition for data center capacity. “We’re expanding the capacity we need to serve customers, and we’re doing it with urgency.”
The expansion comes as Cerebras reported core revenue of $191.3 million for the first quarter, up 92% year-over-year. Core hardware revenue reached $111.6 million, while core cloud and services revenue increased 167% year-over-year to $79.8 million.
Feldman also said the company is positioning itself for what he described as decades of growing demand for AI compute.
A key element of Cerebras’ strategy is its focus on AI inference speed. Feldman argued that faster inference directly translates into higher productivity and broader AI adoption. “Fast tokens are the most valuable tokens because they get more work done in less time,” he said.
The executive said the company’s wafer-scale architecture enables an order-of-magnitude performance advantage while also helping it avoid several supply chain constraints affecting the broader AI infrastructure market.
According to Feldman, current industry bottlenecks include shortages of high-bandwidth memory (HBM), limited CoWoS advanced packaging capacity at TSMC, and tight supply of leading-edge 3-nanometer manufacturing capacity.
Cerebras avoids many of those constraints because its systems use SRAM integrated directly onto its logic wafer rather than separate HBM chips, do not rely on CoWoS packaging, and operate at the 5-nanometer node rather than depending on 3-nanometer manufacturing capacity.
“The binding constraint in the market right now is HBM memory. It’s in short supply, it’s expensive, and we don’t use it,” Feldman said.
To support growing demand, Cerebras has expanded manufacturing capacity in the United States, adding hundreds of thousands of square feet of manufacturing and clean-room space. The company has also broadened its manufacturing ecosystem by expanding its relationship with Flex and adding Sanmina as a second major contract manufacturing partner.
On the customer front, Cerebras highlighted recent progress with OpenAI and Amazon Web Services (AWS).
Feldman said Cerebras began production deployments for OpenAI roughly five weeks after signing an agreement in late 2025 and confirmed that GPT-5.4 is currently running on Cerebras infrastructure for OpenAI engineers and select customers.
The company also recently finalized a definitive agreement with AWS to begin technical collaboration and prepare for deployments of Cerebras systems in AWS data centers. The planned architecture will combine AWS Trainium3 chips with Cerebras’ CS-3 systems in a disaggregated inference architecture in which Trainium handles prefill workloads and Cerebras performs decode operations.
“The demand environment is strong, but this is not just about demand,” Feldman said. “It’s about building the infrastructure required for the next phase of AI.”
RCRTech recently interviewed Jean-Philippe Fricker, co-founder of Cerebras. The first part of the interview can be viewed by clicking this link, while the second part can be accessed through the following link.
