FreshSynapse

AI trends & daily updates, in plain English

AI Business

Cerebras Beats Q3, Raises Its Outlook — and Signs Microsoft as Its Third Hyperscaler

By FreshSynapse Editorial · · 4 min read

A close-up of a computer motherboard with gold heatsinks and red, yellow and blue memory slots

Photo: Hippopx (public domain). Illustrative image.

What happened. Cerebras Systems reported third-quarter 2026 results on 12 November that beat expectations and confirmed a strategic shift: the chipmaker now earns most of its money renting out computing power rather than selling hardware. Core revenue was $230.9 million, up 70% from a year earlier and ahead of the $214.7 million analysts expected in LSEG data cited by CNBC. Cloud and other services brought in $144.9 million — up 272% year on year and about 63% of core revenue — while core hardware revenue slipped to $86 million from $96.8 million a year earlier.

Microsoft is the newest customer. Cerebras said in its results release that it had signed “an important cloud deal with a top three hyperscaler”, and chief executive Andrew Feldman named Microsoft as the customer. The software giant signed in October, Feldman told CNBC, and “we believe revenue begins to show up in Q1 ‘27”. Microsoft joins Amazon Web Services and Oracle, which plans to offer Cerebras inference inside Oracle Cloud Infrastructure in 2027. Cerebras gave no value or terms for the Microsoft contract and said more would come next year.

The numbers behind the beat. Core net income came to $29.8 million, or $0.13 per share, against a loss of $10.9 million a year earlier — and against an expected loss. But profitability is not there yet on the strictest measure: on a GAAP basis, which counts stock compensation and the warrants Cerebras has issued to big customers, the company lost $161.2 million. Core gross margin was 41% and the core operating margin was negative 14%. Cerebras raised its full-year 2026 core revenue guidance to $940–960 million, up from $880–890 million, and guided fourth-quarter core revenue of $310–330 million — both above analyst estimates. Shares jumped about 17% in extended trading, though they had fallen roughly 47% since the company’s May listing on Nasdaq.

Capacity is the constraint. Cerebras reported $26.1 billion in remaining performance obligations, and said it has 700 megawatts of data centre capacity under contract for delivery by the end of 2027, including new sites in Norway and Finland. Reuters reported that the backlog is underpinned by Cerebras’ deal to supply computing power to OpenAI, and that Feldman said on the analyst call that Cerebras lacked the capacity to support a broad rollout of OpenAI’s Ultrafast service tier. The company ended September with $7.8 billion in cash and investments plus $850 million in debt capacity, and Feldman told CNBC it is weighing financing options to keep up with demand. Its next-generation CS-5 system, built on the Wafer Scale Engine 5 processor now in system testing, is targeted for early summer 2027, and Cerebras expects to triple revenue in 2027.

Why it matters. Cerebras is the clearest challenger to Nvidia’s grip on AI inference, and this quarter shows the challenge is real but capacity-limited: demand for fast inference is outrunning its ability to build data centres. Microsoft’s signature matters because hyperscalers do not sign chip suppliers lightly — it is a vote that wafer-scale chips have a durable place beside GPUs. The results landed the same day Jensen Huang resigned as Nvidia CEO, which makes the week a turning point for the whole AI hardware story. The chip race is also reshaping the supply chain: TSMC has approved up to $1 billion in Anthropic and OpenAI, and Google is adding new TPUs for confidential AI and huge context windows.

Sources

  • Cerebras
  • Microsoft
  • earnings
  • AI chips
  • AI infrastructure

Related articles