What Cerebras Systems (CBRS) Said on Its Q3 Earnings Call

Key Points

• Record Q3 performance lifted Cerebras’ outlook: Core revenue rose 70% year over year to $231 million, and the company reported its first positive non-GAAP net income as a public company. It raised full-year 2026 core revenue guidance to $940 million–$960 million.
• Data center delays are holding back OpenAI deliveries: Cerebras said it fell behind on initial capacity commitments because of land, power and construction constraints, not chip supply. Less than 5% of the 750-megawatt commitment could face cancellation this year, and none had been canceled at the time of the call.
• New partnerships and capacity support future growth: A Microsoft cloud agreement brings Cerebras’ hyperscaler customers to three, with revenue expected to begin in 2027. The company expects to return to its OpenAI delivery schedule in the second half of 2027 and forecasts revenue to triple in 2027.

Cerebras Systems NASDAQ: CBRS raised its 2026 financial outlook after reporting record third-quarter core revenue and announcing a cloud agreement with Microsoft, while executives said data center delays have put portions of its OpenAI contract behind schedule.

Chief Executive Officer Andrew Feldman said the Microsoft agreement, signed on the day of the earnings call, brought the company’s hyperscaler customer count to three, alongside Amazon Web Services and Oracle. Cerebras expects to disclose more about the Microsoft deal in the first half of 2027, when revenue from the agreement is also expected to begin.

The company continues to expect revenue to triple in 2027, with improving gross margins, as it expands manufacturing and data center capacity. However, Chief Financial Officer Bob Komin said fourth-quarter cloud revenue would be roughly flat sequentially because new capacity will arrive too late to materially affect the period.

Quarterly Results and Outlook

Third-quarter core revenue rose 70% from a year earlier to $231 million, exceeding the midpoint of the company’s guidance by $16 million. Core cloud and other services revenue increased 272% year over year to $145 million, while core hardware revenue was $86 million, up 5% sequentially.

Core gross margin was 41%, up 40 basis points from the second quarter. Core hardware gross margin fell to 31%, which Komin attributed to the final shipments under an older, lower-priced customer agreement. Core cloud and other services gross margin improved to 46.9% from 41.8% in the prior quarter, helped by higher-priced new business.

Core operating loss was approximately $33 million, while core net income was positive at $30 million. Komin said the company delivered positive non-GAAP net income for the first time as a public company.

For the fourth quarter, Cerebras raised its guidance to core revenue of $310 million to $330 million, core gross margin of 47% to 49%, and core operating margin of negative 1% to positive 1%. For full-year 2026, it raised its core revenue forecast to $940 million to $960 million, with core gross margin of 42% to 44% and core operating margin of negative 10% to negative 8%.

Komin said hardware shipments to new customers, including Oracle, are expected to lift fourth-quarter revenue and continue through at least the first half of 2027. He said hardware could account for closer to 50% of revenue in the fourth quarter and first quarter of 2027, although shipment timing depends on customers having data centers ready.

OpenAI Delays and Capacity Constraints

Feldman said Cerebras lacked sufficient capacity for a broad rollout of OpenAI’s Ultrafast offering and had fallen behind on the initial capacity deliveries under its agreement with OpenAI. He attributed the delays to land, power and data center construction, rather than chip or material availability.

OpenAI is a large customer and remains a close partner, Feldman said, adding that its business grew in the third quarter from the second. He also rejected suggestions that Cerebras systems could not technically support Astra-scale models.

Cerebras reported $26.1 billion in remaining performance obligations at Sept. 30. Komin said OpenAI represents the majority of that balance and that Cerebras deducted estimated potential delay credits from it. The figure did not include the Oracle deal, which was signed after the quarter ended.

Under the OpenAI take-or-pay agreement for 750 megawatts of capacity, delays lasting at least 90 days give either party the option to cancel the affected portion. Komin estimated that less than 5% of the total committed megawatts could become subject to cancellation this year. Feldman confirmed that no delayed capacity had been canceled at the time of the call.

Komin said Cerebras expects to return to the delivery schedule in the second half of 2027. Feldman added that OpenAI has the contractual right to purchase hardware, though he said the customer’s own data centers were also behind schedule.

Expansion and Product Roadmap

Cerebras has secured approximately 700 gross megawatts of data center capacity for delivery by the end of 2027, Feldman said. More than 75% of the capacity coming online in 2027 is outside the United States, including facilities in Canada, France and Norway.

The company expects its Cerebras-powered Oracle Cloud inference service to launch in the summer of 2027. Oracle is purchasing Cerebras hardware for deployment in its U.S. data centers, Feldman said.

Cerebras has begun shipping its CS-4 system and expects it to overtake CS-3 deployments in the first quarter of 2027. Its next-generation CS-5 system is undergoing testing and is expected to enter production in early summer 2027. The company also said it has deployed an inference configuration that combines AMD graphics processors with Cerebras CS-3 systems to increase token output.

Cerebras ended the quarter with $7.8 billion in cash equivalents, restricted cash and marketable securities, and had not drawn on its revolving credit facility of up to $850 million, Komin said.

About Cerebras Systems (NASDAQ:CBRS)

Cerebras Systems develops computing systems and software designed for artificial intelligence and other compute-intensive workloads. Its products are built around the Wafer-Scale Engine, a large processor designed to support AI model training and inference, and include the CS-3 system and related software. The company also offers cloud-based access to its AI computing infrastructure, giving customers a way to run AI workloads without deploying its systems on site. Its offerings are aimed at organizations developing and operating large AI models, including businesses, research institutions and public-sector customers.

This instant news alert was generated by narrative science technology and financial data from MarketBeat in order to provide readers with the fastest reporting and unbiased coverage. Please send any questions or comments about this story to contact@marketbeat.com.