CNBC reported on 12 November 2026 that Cerebras Systems shares rose 18% in extended trading after the company beat third-quarter revenue expectations and disclosed a deal with Microsoft. The AI infrastructure supplier reported results for the quarter ended 30 September 2026, with stronger fourth-quarter guidance than analysts expected.
Cerebras reported USD 210.8 million in revenue under generally accepted accounting principles (GAAP), up 55% year on year. Its core revenue, a non-GAAP measure used by the company, was USD 230.9 million, compared with the USD 214.7 million expected by analysts polled by LSEG.
The company posted a GAAP net loss of USD 161.2 million, or USD 0.67 per share. Its core revenue grew 70% from USD 135.7 million a year earlier, down from 103% growth in the second quarter.
Microsoft deal and cloud routes
Andrew Feldman, co-founder and chief executive of Cerebras, said Microsoft signed the deal in October and that revenue should begin to appear in Q1 2027. Cerebras described the customer in its earnings release as a top three hyperscaler and said it would share more next year.
The Microsoft agreement adds to cloud-provider relationships already named by Cerebras. The company has announced plans to make Cerebras inference available on Oracle Cloud Infrastructure in 2027, and Feldman said Cerebras has also landed deals with Amazon and Oracle.
For enterprise architecture teams, the cloud route matters because it changes how customers consume Cerebras capacity. Cerebras sells direct systems, offers Cerebras Cloud, and is pursuing third-party cloud access through large providers.
Feldman said long lead times in graphics processing units and higher GPU expense create an opportunity for Cerebras, particularly in hardware sales. Cerebras is evaluating financing options to meet rising demand, he said.
Cloud revenue and capacity
Cerebras said GAAP cloud and other services revenue was USD 141.9 million, up 265% year on year. Core cloud and other services revenue was USD 144.9 million, up 272%.
The company reported USD 26.1 billion in remaining performance obligations as of 30 September 2026. It also said it had secured 700 gross megawatts of data-centre capacity for delivery by the end of 2027.
Cerebras said it had signed contracts for additional capacity in Norway and Finland. It also said it expects to exit 2026 with more than 10 times the manufacturing capacity it had at the start of the year, with Flex and Sanmina named as manufacturing partners.
The technical claim behind the capacity story is inference density. Cerebras said its disaggregated inference approach can increase by up to five times the number of tokens data centres can deliver for the same power.
Guidance and investor context
For the fourth quarter, Cerebras guided for core revenue of USD 310 million to USD 330 million. Analysts polled by LSEG expected USD 271 million.
The company guided for a fourth-quarter core gross margin of 47% to 49%, compared with an analyst expectation of 42.1%. For full-year 2026, Cerebras expects core revenue of USD 940 million to USD 960 million and a core operating margin between negative 10% and negative 8%.
Cerebras said it expects revenue to triple in 2027, implying about USD 2.85 billion. Analysts surveyed by LSEG expected USD 2.92 billion.
The share-price reaction followed a difficult period since the company’s Nasdaq debut in May. Cerebras shares had fallen by about 47% since the initial public offering before the earnings release.
Cerebras reported liquidity of USD 7.8 billion in cash, cash equivalents, restricted cash and short-term investments, plus USD 850 million of debt capacity. That balance sheet matters because the company is committing to data-centre capacity, manufacturing expansion and cloud-provider routes at the same time.
Choosing where AI inference runs, how much capacity is contracted and which cloud route exposes it to customers are infrastructure decisions ECS's Infrastructure topic covers, alongside Azure architecture, hybrid cloud, cost and platform engineering. Sessions and workshops give attendees room to compare architecture choices, capacity planning and operating cost assumptions with practitioner experience. Attendees can bring their own inference workloads, data-centre constraints and procurement questions to test against that experience.
This article was written by the ECS editorial team with the assistance of Amal Agent, run.events' AI assistant, which supported the research, drafting and proofreading.