Cerebras Systems Q2 Earnings Call Highlights

Key Points

• Record results and raised outlook: Q2 core revenue rose 103% year over year to $209.9 million, while margins improved significantly. Cerebras raised its full-year core revenue forecast to $880 million–$890 million and expects more than $25 billion in remaining performance obligations.
• Major capacity expansion underway: The company secured more than 600 megawatts of data-center capacity through 2027 and plans manufacturing output to exceed ten times its 2025 level during 2026. Management identified data-center availability as the primary bottleneck to growth.
• Technology and customer pipeline expanding: Cerebras is developing disaggregated inference partnerships with AMD and AWS, plans AWS Bedrock availability in early 2027, and signed six deals worth more than $30 million each in Q2. It also plans to unveil its fourth-generation CS-4 system and remains on track for CS-5 in the second half of 2027.

Cerebras Systems NASDAQ: CBRS reported record second-quarter core revenue and raised its full-year outlook, as the AI infrastructure company said it is expanding data-center capacity, manufacturing output and customer deployments to support anticipated growth beginning in 2027.

Chief Executive Officer Andrew Feldman said the company beat its guidance for core revenue, core gross margin and core operating margin during the quarter, which also included the completion of Cerebras’ public offering. Management characterized 2026 as a “foundation-building year” as it prepares to serve more than $25 billion in remaining performance obligations, or RPO.

“We delivered record core revenue and beat guidance on all metrics,” Feldman said. He added that the company expects to more than triple core revenue in 2027 and continue growing at multiples in subsequent years, based on progress in capacity, technology and customer additions.

Revenue Growth and Updated Outlook

Chief Financial Officer Bob Komin said second-quarter core revenue was $209.9 million, up 103% from a year earlier. Core cloud and other services revenue rose 287% year over year to $127.7 million, while core hardware revenue increased 17% to $82.1 million.

Komin said cloud and services growth reflected the ramp of the company’s OpenAI deployment, greater usage from other cloud customers and timing related to hardware customers’ data-center expansions. He said the company has several late-stage hardware opportunities representing hundreds of millions of dollars, along with new cloud opportunities for 2027.

• Core gross margin was 40.6%, up about 940 basis points from the prior-year quarter.
• Core cloud and other services gross margin was 41.8%, improving 1,600 basis points year over year.
• Core hardware gross margin was 38.8%, up 510 basis points from a year earlier.
• Core operating loss was $33.6 million, while core operating margin improved to negative 16% from negative 42% a year earlier.

Management said sequential gross margin declined from 46.5% in the first quarter because Cerebras has temporarily rented back some of its systems from cloud customers to meet demand through its private cloud. Komin said this higher-cost rented capacity reduced second-quarter core gross margin by approximately 500 basis points.

The company expects the third quarter to be the low point for core gross margin before an anticipated improvement in the fourth quarter as it brings online more data centers using company-owned systems. Cerebras expects core gross margin to trend toward its target of more than 60% over time.

For the third quarter, Cerebras forecast core revenue of $214 million to $216 million, core gross margin of 38% to 40%, and core operating margin of negative 25% to negative 23%. For the full year, it raised its core revenue outlook to $880 million to $890 million, core gross margin guidance to 41% to 43%, and core operating margin guidance to negative 19% to negative 17%.

Capacity Build-Out Targets Data-Center Bottleneck

Feldman said data-center space remains a bottleneck for both Cerebras and the wider AI industry. Over the past seven months, the company secured more than 600 megawatts of data-center capacity that is either operating now or contracted for delivery by the end of 2027. Its pipeline for further expansion is measured in gigawatts, he said.

The company listed capacity in Alabama, Dallas, Denver, Minneapolis, Santa Clara and Stockton, as well as sites in France, Finland, Manitoba, Montreal, Norway, Saskatchewan and Toronto.

Cerebras also said it is expanding manufacturing through Flex and Sanmina. Komin said manufacturing capacity was already four times above its level in the first half of 2025 and is expected to exceed a tenfold increase during 2026. The company has contracted facilities supporting an additional three to four times growth in 2027, Feldman said in response to an analyst question.

Management said its supply relationship with TSMC has secured wafers needed for planned growth. Feldman noted that Cerebras uses TSMC’s 5-nanometer node and does not use high-bandwidth memory, CoWoS packaging or 3-nanometer fabrication capacity, which he said reduces exposure to some industry supply constraints.

Inference Technology and Disaggregation Plans

Feldman said Cerebras added support during the quarter for OpenAI’s GPT-5.6 Sol model and said the company can serve the model at ten times faster speed. He said supporting frontier models has given Cerebras additional insight that can inform its hardware and software roadmap.

The company also highlighted disaggregated inference partnerships with AMD and AWS. In a disaggregated setup, GPUs or other processors handle the prefill stage of inference, while Cerebras systems handle decode, the process of generating output tokens.

According to Feldman, the combined AMD Helios and Cerebras configuration is designed to maintain Cerebras’ speed while increasing throughput fivefold. He said higher throughput can increase tokens produced per system and per watt, potentially improving data-center economics and gross margins. Cerebras expects disaggregated inference using GPUs to be deployed and available in the fourth quarter, he said during the question-and-answer session.

The company plans to unveil its fourth-generation CS-4 system at its Supernova conference, while remaining on track to launch the CS-5 in the second half of 2027. Management expects new systems to double speed annually for the next several years and plans to increase throughput by more than 20 times over the next 18 months.

Customer Expansion and AWS Availability

Feldman said Cerebras expects its offering to become generally available through AWS’s Amazon Bedrock platform in the first quarter of 2027. He said the company expects first hyperscaler revenue beginning in mid-2027, with a ramp through 2028 and beyond. The company’s $25.4 billion RPO at June 30 did not include backlog from AWS or other hyperscalers, management said.

Outside of OpenAI and hyperscalers, Cerebras signed six deals exceeding $30 million in the second quarter. The company cited new agreements with Figma, Cognition, Lovable, Block, AlphaSense, GSK and CrowdStrike.

Feldman said OpenAI is expected to remain a meaningful share of revenue next year, though AWS, coding companies and security applications are expected to become larger portions of the business over time. He also said the company sees emerging “neo-cloud” providers as a potentially important part of its business in 2027.

Cerebras ended the quarter with more than $8.6 billion in cash equivalents, restricted cash and marketable securities, along with an unused revolving credit facility of up to $850 million.

About Cerebras Systems (NASDAQ:CBRS)

Cerebras Systems is a technology company focused on building artificial intelligence infrastructure, including hardware and software designed to accelerate deep learning and large-scale AI workloads. The company is best known for its wafer-scale processor architecture, which is intended to provide high-performance compute for training and inference applications.

In addition to its AI chips, Cerebras offers systems and related software tools that support researchers and enterprises working with machine learning models.

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.

Continue following MarketBeat

Add MarketBeat as your preferred source on Google to see our latest stories in your feed.

Add As Preferred Source