Cerebras triples cloud revenue to $141.9M and lifts its 2026 outlook to $940M+

Cerebras Systems reported record third-quarter 2026 results on November 12, with GAAP revenue of $210.8M, up 55% from a year earlier, and raised its full-year core revenue outlook to $940M-$960M.

Cloud inference now outsells the chips

The growth came almost entirely from the cloud. Cloud and other services revenue reached $141.9M, a record and up 265% from $38.9M in Q3 2025. Hardware revenue moved the other way, falling to $68.9M from $96.8M. On Cerebras’ adjusted “core” basis, which strips out items such as pass-through revenue and customer warrant amortization, total revenue was $230.9M, up 70%.

For the nine months, GAAP revenue was $584.3M against $338.5M in the same period of 2025. Core gross margin in the quarter was 41.0%, while the GAAP figure was 20.0%, against 40.6% a year earlier.

Guidance is given on the core basis. For Q4, Cerebras expects core revenue of roughly $310M to $330M with a core gross margin of 47% to 49%. For the full year it now expects $940M to $960M, a core gross margin of 42% to 44% and a core operating margin between negative 10% and negative 8%. CEO Andrew Feldman said the company expects to triple revenue in 2027.

New customers and a capacity race

Cerebras named several customer moves. Oracle plans to integrate Cerebras inference into Oracle Cloud Infrastructure, with availability in 2027. The company also announced a cloud deal with what it calls a top-three hyperscaler, with details promised next year, and said it expanded through several neocloud partners. Bilt will use Cerebras inference for its Neighborhood OS platform and Bilt Concierge assistant. Cerebras was a day-zero launch partner for Moonshot AI’s Kimi K3.1, a 2.8 trillion-parameter open-weight model, which it says runs at up to 2,000 tokens per second.

On supply, Cerebras says it has secured 700 gross megawatts of data center capacity for delivery by the end of 2027, with additional contracts in Norway and Finland. It is scaling manufacturing with Flex and Sanmina and expects to exit 2026 with more than ten times the manufacturing capacity it had at the start of the year. Remaining performance obligations stood at $26.1B, and cash, restricted cash and investments totaled $7.8B.

On the product side, the company says its CS-4 system is close to full production, its WSE-5 chip is back from the fab and in system testing, and the CS-5 is targeted for early summer 2027. CFO Bob Komin said the pipeline for high-speed inference grew by more than 10x in the quarter.

Beating Wall Street, but still losing money

Set against analyst expectations, the guidance is a clear step up. According to Yahoo Finance, the Q4 range has a midpoint of $320M against a consensus of $274.9M, and the full-year midpoint of $950M compares with a consensus of $887.5M. Yahoo reported that shares rose more than 16% in after-hours trading on the news.

The headline growth also hides a gap between the two ways of counting. The GAAP net loss was $161.2M for the quarter, or $0.67 per share, compared with a loss of $22.2M a year earlier. The nine-month GAAP net loss was $625.8M, against net income of $263.4M in 2025. The core figures look far better, with core net income of $29.8M in the quarter, and the difference is the set of items Cerebras excludes. Cash spent on property and equipment reached $1,046.1M in nine months, which shows how capital-hungry the capacity build is.

The shift in mix matters as much as the total. A year ago, hardware sales made up most of Cerebras’ revenue. Now cloud inference is about two thirds of GAAP revenue, which ties the company’s results to how long customers keep running workloads and to how fast it can bring data center power online. The company’s own pipeline claim, the 10x increase, is not backed by a dollar figure in the release, and the hyperscaler deal is unnamed, so neither can be checked yet.

Cerebras’ statement that its technology is “the undisputed champion of speed” is a company claim, not an independent benchmark. For readers who use Cerebras today, the practical effect is more capacity: through its own cloud now, and through Oracle in 2027. Nothing in the release gives pricing changes.