MONDAY, NOVEMBER 16, 2026 Tech news, minus the noise.
N News Cast
LATEST
Home  /  Cloud
Cloud

Cerebras Signed Microsoft as Its Third Hyperscaler, and Its OpenAI Deal Is Running Late

Cerebras signed Microsoft as its third hyperscaler customer and posted record Q3 revenue of $230.9 million. At the same time, its biggest OpenAI deal is running late on power and buildings.

Cerebras Signed Microsoft as Its Third Hyperscaler, and Its OpenAI Deal Is Running Late
Source: Retro-Computing Society of Rhode Island, Wikimedia Commons (CC BY-SA 4.0)

Cerebras Systems reported record third-quarter revenue on November 12, 2026, and buried the bigger news inside the same announcement: the Cerebras Microsoft deal, with Microsoft signing on as a cloud customer. It is the third hyperscaler to commit to Cerebras hardware, after Amazon Web Services and Oracle, and it is the strongest signal yet that the AI chip challenger is becoming part of the mainstream cloud stack.

The numbers

Cerebras reported core revenue of $230.9 million for the quarter ended September 30, up 70 percent year over year and ahead of the $214.7 million analysts expected. The company guided fourth-quarter core revenue to $310 million to $330 million, against $271 million expected, and raised its full-year outlook. The stock surged after the announcement, closing up more than 11 percent on November 13.

The revenue mix tells its own story. AI cloud accounted for $145 million of the quarter, and the company expects that line to stay relatively flat in the fourth quarter, constrained by data center capacity that only starts ramping in December. Core hardware revenue was $86 million, excluding the recently signed Oracle deal. Cerebras now earns most of its money renting out its chips, not selling them, which is a cloud business wearing a chip company's clothes.

The Cerebras Microsoft deal, and what is not in it

Cerebras said in its results release that it had signed a cloud deal with a top-three hyperscaler, and CEO Andrew Feldman told CNBC the customer is Microsoft, which signed in October. Revenue from the deal should begin to show up in the first quarter of 2027, with more details coming in the first half of 2027. Microsoft has not commented, and Cerebras disclosed no size, term, or contract value.

The Microsoft deal is not yet reflected in Cerebras's $26.1 billion order book, which also excludes the Oracle agreement. That means the backlog figure, large as it is, understates the contracted demand. The company also said it has secured 700 gross megawatts of data center capacity for delivery by the end of 2027, which is the capacity it needs to serve these customers.

The OpenAI problem

The same earnings call carried a warning. Cerebras is running behind on its biggest contract, a take-or-pay deal with OpenAI for 750 MW of capacity. Feldman said the problem is land, power, and data center buildings, not chips. Delays give OpenAI credits, which Cerebras has already taken off its backlog, and a delay of 90 days or more lets either side cancel that slice of capacity. The company estimates less than 5 percent of committed megawatts could be open to cancellation this year, and no capacity has been cancelled.

That framing deserves a careful read. The bottleneck for AI infrastructure is no longer silicon. It is the physical world, and the numbers show it: a gigawatt of AI data center now costs about $49 billion, according to Morgan Stanley, as covered in our Vera Rubin data center analysis.: getting power to a site, getting a building finished, getting a grid connection. Cerebras can manufacture its wafer-scale chips. It cannot manufacture a substation.

The next chip, and what to watch

The next generation is already in motion. The Wafer Scale Engine 5 taped out on June 15, came back from the fab ahead of schedule, and is running models in system testing. The CS-5 system built on it is targeted to enter production in early summer 2027, built on TSMC's 5-nanometer node, and is expected to more than double the speed of the CS-4.

Three things will decide whether this quarter was a turning point or a peak. First, whether the Microsoft revenue actually arrives in Q1 2027 on the schedule Feldman described. Second, whether the OpenAI delays stay small and the credits stay manageable. Third, whether data center capacity ramps in December as promised, because flat AI cloud revenue in Q4 is only acceptable if Q1 makes up for it.

The capacity race is global. AWS just opened its first region in Chile with $4 billion behind it, and Google committed $17 billion to finish the Crusoe data center site in Wyoming. Cerebras has now signed all three of the largest cloud providers it could plausibly sign, and it did so while posting record revenue and raising guidance. The question is no longer whether anyone wants its chips. It is whether the company can build the power and buildings fast enough to deliver them.

AS
News Cast Editorial Team

We cover AI, gadgets, startups and security — original reporting and explainers, written by humans, for humans. Got a tip? Reach us via the contact page.