Cerebras secures 700 MW of AIDC capacity as it expands global footprint

Home AI Infrastructure News Cerebras secures 700 MW of AIDC capacity as it expands global footprint
Cerebras

Cerebras’ CEO Andrew Feldman told a conference call with investors that the firm is in the final stages of contracting for a data center in São Paulo, Brazil

In sum – what to know

700 MW secured – Cerebras has secured approximately 700 gross MW of data center capacity for delivery by the end of 2027, with most expected to come online from mid-year.

Global expansion – More than 75% of the company’s data center capacity expected to come online in 2027 will be outside the U.S., with facilities planned in Canada, France, and Norway.

Capacity constraints – Demand for Cerebras’ cloud services continues to outpace supply, while management says data center availability is limiting revenue growth.

Cerebras Systems has secured approximately 700 gross megawatts (MW) of data center capacity for delivery by the end of 2027, as the AI chipmaker expands its infrastructure footprint across several regions.

Speaking during the company’s third-quarter 2026 earnings call, the company’s chief executive officer Andrew Feldman said more than 75% of the company’s data center capacity expected to come online in 2027 would be outside the U.S. New facilities in Canada, France, and Norway are expected to come online over the coming months. Cerebras also signed contracts during the quarter for additional capacity in Norway and Finland.

The company is also in the final stages of contracting for a data center in São Paulo, which Feldman described as its first in South America.

The expansion reflects the importance of infrastructure availability to the company’s growth plans. Feldman said data center availability remains a constant constraint across the industry, while CFO Robert Komin explained that bringing large new data center capacity online can take between 15 and 18 months.

Cerebras has been building its own cloud infrastructure to serve customers that want to consume AI tokens rather than deploy hardware themselves. The company also sells systems directly to customers for installation in their own data centers. Management sees the two channels as complementary, allowing customers to access Cerebras technology through either hardware purchases or cloud services.

The company’s cloud business has grown rapidly, but available infrastructure is limiting its ability to increase capacity further. Revenue from cloud and other services reached $145 million in the third quarter, up 272% year over year. Sequential cloud revenue growth was 13%.

However, Komin said Cerebras currently expected cloud revenue to be approximately flat sequentially in the fourth quarter. He stressed that the outlook was not driven by pricing or demand weakness: demand for the company’s solutions continued to outpace supply by a wide margin. “Our ramp in cloud revenue is limited entirely by our data center capacity,” Komin said.

Cerebras expects to begin increasing capacity in December, but Komin said the timing was too late to affect fourth-quarter revenue. The company anticipates more substantial increases in data center capacity across 2027, with the bulk of the additional capacity coming online from the middle of the year.

Komin said larger portions of the approximately 700 MW currently under contract would be delivered during that period. He also described 2026 as a year of building capacity, capabilities, and adding customers to support the company’s expected revenue growth in 2027.

Increasing capacity per megawatt

Alongside expanding its physical infrastructure, Cerebras is looking to increase the amount of AI inference capacity it can deliver from each data center.

Feldman identified disaggregated inference as one way to achieve this, saying the approach could increase by up to five times the number of tokens a data center can deliver. He argued that generating more high-speed tokens per megawatt would allow each data center to produce markedly more revenue.

The company’s cloud already serves customers using a heterogeneous architecture that combines AMD GPUs for prefill with Cerebras CS-3 systems for decode. Feldman said Cerebras had developed a software platform to enable production-scale disaggregated inference, designed to work with AMD and other GPUs.

The same solution is expected to be deployed across the Cerebras inference cloud and at selected data center customers. Feldman said disaggregated inference delivers more high-speed tokens for the same amount of power, potentially improving economics for cloud operators and reducing the cost per token for AI users.

This website uses cookies to improve your experience. We'll assume you're ok with this, but you can opt-out if you wish. Accept Read More