Cerebras Raises Full-Year Outlook as Cloud Revenue Jumps 265%

Cerebras Systems raised its full-year core revenue forecast to $940 million to $960 million on November 12, 2026, after third-quarter cloud and services revenue jumped 265% year over year to $141.9 million. Total GAAP (generally accepted accounting principles) revenue rose 55% to a record $210.8 million, but the GAAP net loss widened to $161.2 million from $22.2 million a year earlier.

The new full-year target replaces the $880 million to $890 million range the company set in its August 12, 2026 second-quarter release. The numbers frame a company whose growth now depends on inference capacity delivered as a service rather than chip sales alone — and whose path to sustainable economics depends on execution, not demand.

Why does the raised outlook matter?

Cerebras sits at the center of a race to supply high-speed AI inference, the process of running trained models to generate outputs such as text tokens. Its third-quarter mix shows the shift: cloud and other services supplied most of the quarter's GAAP revenue, while hardware revenue declined from a year earlier. Growth increasingly comes from selling inference capacity directly, complementing system sales and access through third-party clouds.

The company's revenue backlog reinforces the demand signal. Cerebras reported $26.1 billion in remaining performance obligations (contracted revenue not yet recognized) as of September 30, 2026, up from $25.4 billion at June 30, 2026. The scale of those commitments makes deployment schedules and available capacity central to converting customer demand into reported revenue.

How close is Cerebras to profitability?

Its fourth-quarter outlook calls for a core operating margin between negative 1% and positive 1%, putting the adjusted measure near breakeven. Core results exclude stock-based compensation and data-center pass-through items and add back noncash customer-warrant amortization recorded against revenue, among other adjustments.

The gap between core and GAAP results is large. Stock-based compensation represented $155.1 million of the adjustments in the third-quarter operating reconciliation. That figure helps explain why the core operating result sits close to breakeven while the GAAP net loss reached $161.2 million.

Expansion is also cash-intensive. During the first nine months of 2026, Cerebras used $388.9 million in operating activities and purchased $1.046 billion of property and equipment. The company ended September with $7.8 billion in cash, cash equivalents, restricted cash and short-term investments, providing substantial resources to finance the buildout.

Which new cloud deals did Cerebras announce?

Two distribution agreements accompanied the earnings release.

• Oracle: Oracle is integrating Cerebras fast inference into Oracle Cloud Infrastructure. Availability is planned for 2027.
• Top-three hyperscaler: Cerebras signed a cloud agreement with an unnamed customer it described as a top-three hyperscaler. The company said more information about the agreement is expected in 2027.

Both deals extend Cerebras's reach beyond direct system sales, embedding its inference technology in third-party cloud platforms where most enterprise AI workloads already run.

Can Cerebras deliver the capacity?

The company has secured 700 gross megawatts of data-center capacity for delivery by the end of 2027, including new contracts in Norway and Finland. Its August release separately described more than 600 megawatts live or under contract for delivery by the same deadline; that earlier figure also encompassed capacity already operating.

Manufacturing is scaling in parallel. Cerebras manufactures with Flex and Sanmina, and it expects to exit 2026 with more than 10 times the manufacturing capacity it had at the beginning of the year. Plans call for additional factories and capacity in 2027.

What does the architecture change about the economics?

Cerebras says its disaggregated inference architecture — a design that separates compute and memory functions rather than combining them on a single server — can increase the number of tokens a data center delivers by up to five times. That improvement raises the potential revenue produced per megawatt of powered infrastructure.

The capability is central to the company's effort to supply high-speed inference without a proportional increase in power-hungry data-center capacity, a constraint that increasingly shapes the economics of AI compute.

What comes next?

The raised outlook ties Cerebras's next phase to execution as much as demand. Converting $26.1 billion in customer commitments into deployed inference capacity while managing cash use and GAAP losses will determine whether 265% cloud growth produces sustainable compute economics.