What happened
Cerebras announced financial results on November 12, 2026, for the quarter ended September 30. The company reported GAAP total revenue of $210.8 million, up 55% year over year, and GAAP cloud and other services revenue of $141.9 million, up 265%. It separately reported “Core” total revenue of $230.9 million and Core cloud and other services revenue of $144.9 million. Cerebras defines Core as a non-GAAP measure that excludes non-cash amortization of customer warrants and stock-based compensation, data-center pass-through revenues and costs, and certain other items; the two measures should not be conflated.
The release gives a mixed picture of profitability. GAAP gross margin was 20%, while Core gross margin was 41%. GAAP operating margin was negative 106%, and Core operating margin was negative 14%. For full-year 2026, Cerebras forecast Core revenue of $940 million to $960 million and Core gross margin of 42% to 44%. Its Q4 forecast was Core revenue of approximately $310 million to $330 million, with Core gross margin of 47% to 49%. The release also reported $26.1 billion in remaining performance obligations as of September 30.
On customer access, Cerebras said Oracle is integrating its fast inference into Oracle Cloud Infrastructure (OCI), and that Cerebras will be available in OCI in 2027. It described prospective users as AI labs, enterprises and developers that rely on Oracle, but did not identify specific models, regions, pricing, service levels or a date within 2027. The release also announced a deal with an unnamed “top three hyperscaler”; Cerebras said further details would come next year. That customer is not identified in the announcement.
Cerebras said it had secured 700 gross megawatts of data-center capacity for delivery by the end of 2027. It also described “heterogeneous disaggregated inference” as a production-scale offering, claiming it can increase the number of tokens data centers deliver by up to five times and produce more high-speed tokens for the same power. These are company claims, not independent measurements. The announcement also mentioned Cerebras inference powering Bilt’s Neighborhood OS platform and Bilt Concierge AI assistant, which serve eight million consumers and 50,000 businesses, according to the company.
What the archive adds
New Runtime's earlier coverage on October 31 described Oracle's accelerator expansion, including chips from Cerebras and several other providers, in Oracle's OCI accelerator expansion. This earnings announcement adds Cerebras's 2027 availability statement and financial results to that known story. An earlier note on October 27 covered NVIDIA's Vera Rubin systems becoming generally available on OCI; that is a separate accelerator offering, not evidence that Cerebras is already available there.
Trend signal
This supports the pattern that agent-ready software exposes capabilities only weakly: Oracle is set to expose another inference capability through its cloud, but the announcement does not yet specify the interface or operational terms that developers could compose. It is more directly a cloud-supply signal than evidence of a new agent integration boundary.
Why now
The release ties the OCI announcement to Cerebras's Q3 results and its stated effort to offer customers more ways to access its systems, including direct sales, Cerebras Cloud and third-party cloud providers. Its stated availability timing is 2027, alongside a plan to deliver 700 gross megawatts of data-center capacity by the end of that year.
Connections to other events
New Runtime's August 16 earlier coverage said OpenAI products had appeared in Oracle Cloud Marketplace for eligible US customers, with other regions planned later. That was a procurement route for OpenAI products; the new Cerebras announcement concerns inference capacity integrated into OCI. Separately, New Runtime covered NVIDIA Vera Rubin systems becoming generally available on OCI on October 27. Together, these announcements show Oracle adding distinct AI compute options, not a single interchangeable service.
Why it matters
A cloud integration can lower the organizational friction of trying specialized inference: teams may be able to keep procurement, identity and workloads within an existing cloud environment. But “available in OCI” does not by itself establish that a workload can move unchanged. Model support, APIs, batching behavior, latency, throughput, regional placement, quotas and price will determine whether the option is useful for a given product.
The revenue figures matter as a supplier signal, not as proof that Cerebras will meet a future service commitment or that its economics work for every workload. The gap between GAAP and Core results is substantial, and both operating-margin measures remain negative. Builders should treat the 2027 date as a roadmap marker and evaluate actual service terms and measured performance once the integration is available.
Limits and uncertainty
The OCI availability is forward-looking and scheduled for 2027; the release does not establish a launch date or independently verify delivery. It provides no OCI pricing, regions, quotas, model list or service-level commitments. Cerebras's fivefold capacity claim for disaggregated inference is not accompanied here by a workload-by-workload methodology or independent validation. The unnamed hyperscaler deal cannot be assessed for scale or deployment, and the company's outlook and capacity plans are subject to execution risk.
Reader takeaway
Cerebras is set to join Oracle's cloud inference options in 2027, while reporting rapid growth in cloud revenue. For builders, this is a reason to watch for concrete OCI terms—not yet a basis for changing infrastructure. Compare deployed latency, model fit and cost once the service can be tested.