Vendors & EquipmentCerebrasMicrosoftAI ChipsAI Cloud
Cerebras Lands Microsoft as Third Hyperscaler Customer
Microsoft becomes Cerebras' third hyperscaler customer after AWS and Oracle, as Q3 revenue hits $231M and remaining obligations reach $26.1B ahead of a 700MW capacity build.
Vendors & EquipmentWhy it matters
- Microsoft is Cerebras' third hyperscaler customer, after AWS and Oracle; deal details due in 2027.
- Q3 2026 revenue was $231 million, up 70 percent year-on-year; core net income turned positive at $30 million.
- Remaining performance obligations reached $26.1 billion as of September 30, 2026, excluding Oracle and Microsoft deals.
- Over 75 percent of Cerebras' planned 2027 data center capacity is outside the US, across Canada, France, Norway and Finland.
- About 5 percent of the OpenAI contract value may be subject to a 90-day delay cancellation clause; no cancellations so far.
The story
Microsoft has signed on to deploy and offer Cerebras' AI chips, making it the third hyperscaler customer for the wafer-scale silicon specialist after Amazon Web Services and Oracle. Cerebras CEO Andrew Feldman revealed the contract on the company's Q3 2026 earnings call and told investors that more details will arrive in 2027.
The deal lands in a quarter when Cerebras posted $231 million in revenue, up 10 percent sequentially and 70 percent year-on-year, beating expectations. AI cloud contributed $145 million of that total. Core hardware revenue was $86 million and excludes the recently signed Oracle contract. Gross margins stood at 41 percent; the company expects 48 percent next quarter, 50 percent in the near term and 60 percent as a long-term target.
Cerebras turned core net income positive at $30 million, with year-to-date adjusted core EBITDA also positive at $2 million. Remaining performance obligations climbed to $26.1 billion as of September 30, 2026 — a figure that still excludes both the Oracle and Microsoft deals. The company raised full-year 2026 core revenue guidance to a range of $940 million to $960 million.
Why did hyperscalers sign?
CFO Bob Komin credited the company's own cloud rollout, launched in early 2025, as the decisive factor in winning large deployments. He said: "In order to win the large deployments that foundation labs and hyperscalers require, Cerebras needed to demonstrate that our systems could deliver our industry-leading inference performance at scale. After having the opportunity to thoroughly evaluate our solution in our private cloud, these customers can then choose to buy our hardware or purchase our cloud offering."
The strategy reflects a structural reality: hyperscalers plan data center compute allocations years in advance and often lack near-term availability for new hardware. AI natives, by contrast, want to scale compute quickly without owning data centers. The Cerebras cloud bridged that gap and became the evaluation path to hardware contracts.
Where is the capacity going?
Data center constraints dominate the roadmap. AI cloud revenue is expected to stay flat in Q4, limited by capacity that only begins ramping in December. Larger portions of the planned 700MW footprint are under contract for deployment from mid-2027.
Feldman said the constraint comes down to land, power and data center shells. "Data center availability remains a constant constraint across the industry, and we are no exception," he told investors, adding that the evolving US policy landscape could further complicate domestic development.
As a result, more than 75 percent of planned 2027 data center capacity sits outside the US:
- Facilities in Canada, France and Norway expected online in coming months
- New capacity contracts signed in Norway and Finland during Q3
- A São Paulo data center in final contracting stages — Cerebras' first in South America
What about the OpenAI delays?
Feldman devoted part of the call to countering speculation. He confirmed the OpenAI partnership expanded in Q3 from Q2 and called the response to OpenAI's Ultrafast GPT-5.6 Sol preview "overwhelming." He denied that Cerebras hardware could not technically support OpenAI's Astra model but conceded that Cerebras "didn't have sufficient capacity to support Ultrafast in a broad rollout, including for Astra."
He acknowledged the partnership is running behind schedule on initial tranches because the company has taken longer than it wanted to contract and bring online the required capacity. If delays exceed 90 days, either party can cancel that portion of the contract; Cerebras estimates roughly 5 percent of full contract value may fall under that condition, and no cancellations have occurred.
On yield concerns, Feldman rejected the claim that low hardware margins signal manufacturing problems, attributing them to the manufacturing ramp and final shipments under special pricing programs for an early customer. "We expect our hardware margins to improve markedly over the next several quarters, eventually achieving our long-term target of 50 percent," he said.
The rebuttal follows short-seller Culper Research disclosing a short position and claiming Cerebras wafers break down after roughly two years of use — a lifespan Culper said "undermines the entire business."
On product, CS-5 production starts in summer 2025, with CS-3-to-CS-4 crossover expected in Q1 2027. Cerebras expects to exit 2026 with ten times its start-of-year manufacturing capacity, and it forecasts tripled revenue in 2027. Shares stood at $97.75 at the time of writing.
Also reported
Source: datacenterdynamics.com
More from Daniel Okafor
Show full bio
Senior reporter covering regulation, policy and their impact on Carrier business, 5G and 6G, fiber and broadband at Telecom Gazette.
536 articles