Cerebras Systems reported second-quarter 2026 results that beat expectations on guidance, but investors zeroed in on a weaker hardware line. The company posted GAAP revenue of $180.1 million, up 74% from a year earlier, while core revenue came in at $210 million, up 103%. Hardware revenue dropped to $54.1 million, down 23% year over year, and the stock fell as much as 17% in after-hours trading after rising 11.6% in Wednesday’s regular session.
Cloud revenue overtook hardware in the quarter
The biggest change in the quarter was the split between hardware sales and cloud-related revenue.
According to the earnings report, hardware revenue fell from $70.3 million in the same period last year to $54.1 million, a 23% decline. Cloud computing and other services climbed from $33.0 million to $126.0 million, up 281%. On a core basis, cloud and other services revenue reached $127.7 million, up 287%.
That shift shows Cerebras is no longer relying mainly on sales of AI computing systems. As customers including OpenAI expand demand for AI inference compute, the company is generating a larger share of revenue by delivering compute through the cloud.
Chief Executive Officer Andrew Feldman said Cerebras would keep its two-track model of hardware and data center services. 「We want to meet customers where they are growing the fastest,」 he said.
Even so, the hardware decline remained the central issue for the market. Cerebras has long positioned its wafer-scale engine, or WSE, as a challenger to Nvidia GPUs. A drop in hardware revenue suggests that business has yet to settle into a steady, linear growth path. Feldman said volatility is part of the nature of the hardware business.
Guidance moved higher, but investors stayed cautious
Cerebras gave guidance that came in above market expectations despite the drop in hardware sales.
The company said third-quarter core revenue should be about $214 million to $216 million, with a midpoint of $215 million, above the average analyst estimate of $212 million. Core gross margin is expected to be 38% to 40%, compared with market expectations of about 36%.
It also raised full-year guidance. Cerebras now expects 2026 core revenue of $880 million to $890 million, up from prior guidance of $855 million to $865 million. Full-year core gross margin is expected at 41% to 43%, up from 38% to 41% previously. Analysts had been looking for adjusted gross margin of 35.89% for the year.
Other core metrics improved as well. Core total revenue for the second quarter was $209.9 million, up 103% year over year. Core gross margin reached 40.6%, about 9.4 percentage points higher than a year earlier. Core net loss narrowed sharply to $6.908 million from $40.5 million in the same quarter last year.
Still, the quarter’s revenue mix kept pressure on sentiment. Investors appeared to weigh not only whether Cerebras could beat near-term revenue estimates, but also how durable and how high-quality that growth looks if hardware remains uneven.
GAAP and core metrics painted different pictures
The market’s reaction also reflected closer scrutiny of margins and accounting adjustments.
Since its IPO, Cerebras shares have already seen a large run-up, with investors betting the company could use its wafer-scale chip architecture to take AI inference share from Nvidia. Against that backdrop, a 23% decline in hardware revenue was enough to trigger profit-taking even with fast cloud growth and higher full-year guidance.
One figure that stood out in the report was hardware gross margin. On a GAAP basis, second-quarter hardware gross margin was just 1.8%. On a core basis, hardware gross margin was 38.8%.
The company said its core metrics exclude items including customer warrant amortization, stock-based compensation and data center expenses. That created a visible gap between GAAP and core reporting.
AI inference remains the center of the growth story
Cerebras is putting more of its strategy behind AI inference.
Unlike model training, inference happens after users send requests to AI applications such as chatbots. As AI moves from model development into broader commercial deployment, demand is rising for low-latency, high-throughput inference compute.
Cerebras said its WSE product uses a wafer-scale design that integrates large amounts of compute on a single massive chip. The company argues that this reduces data transfer across many chips in conventional GPU systems and improves AI inference speed.
The company said its systems can already support OpenAI’s GPT-5.6 Sol at 750 tokens per second. It is also working with Advanced Micro Devices, or AMD, on a disaggregated inference offering that is expected to enter production in the fourth quarter of 2026. Cerebras said it plans to bring that technology to Amazon Web Services’ Amazon Bedrock in the first quarter of 2027.
Cerebras also said it has signed new cloud capacity agreements with customers including AI coding company Cognition and Lovable. Existing customers include Block, Figma, AlphaSense, GSK and CrowdStrike.
HBM-free design is part of its supply chain pitch
At a time when AI chip makers are dealing with tight high-bandwidth memory, or HBM, supply and rising prices, Cerebras is trying to turn its architecture into a cost advantage.
The company said its wafer-scale design does not rely on HBM, CoWoS advanced packaging or a 3-nanometer process, all of which are constrained parts of the current AI chip supply chain.
Feldman said the sharp rise in Nvidia AI chip prices is tied to higher HBM costs, while Cerebras could have an advantage in an environment of rising component prices because it does not use HBM.
Whether that advantage can translate into sustained hardware sales growth remains unproven. The quarter’s hardware revenue decline showed the company is still working through the challenge of turning technical differentiation into stable commercial revenue.
$25.4 billion in remaining performance obligations puts focus on delivery
Cerebras is leaning on a large order book and a sizable cash position as it moves into the next stage.
As of the end of June, the company had $25.4 billion in remaining performance obligations, indicating a substantial volume of contracted revenue that has not yet been recognized. Cerebras said it plans to more than triple revenue in 2027.
The company also said it raised about $6.4 billion in its IPO this year. As of June 30, it held about $8.6 billion in cash, cash equivalents, restricted cash and short-term investments, along with $850 million in debt financing capacity.
To support those commitments, Cerebras is expanding production and data center infrastructure. The company said manufacturing capacity in 2026 will increase by more than 10 times, with new production lines added at Flex, Sanmina and Rocket EMS. As of the end of June, contracted, under-construction or already operational data center capacity scheduled for delivery by the end of 2027 exceeded 600MW.