Cerebras Systems has delivered its second quarterly earnings report as a publicly traded company, posting figures that underscore the AI chipmaker's rapid expansion as it positions itself against Nvidia's dominance in the AI accelerator market.

The company reported Q2 2026 core revenue of approximately $210 million, reflecting more than 100% year-over-year growth. CEO Andrew Feldman told investors that core revenue is expected to more than triple in 2027, supported by $25.4 billion in performance obligations already recorded on the company's books.

Financial Results and Guidance

GAAP revenue for the quarter came in at $180.1 million. The gap between GAAP and core revenue stems from long-term contract accounting rules that defer recognition of certain contracted amounts.

Building on these results, Cerebras raised its full-year 2026 revenue guidance to a range of $880 million to $890 million.

The company's initial public offering was a notable milestone in the semiconductor sector. Cerebras raised $6.4 billion in gross proceeds when it went public in May 2026, ranking among the largest semiconductor IPOs on record. The company also expects its manufacturing capacity to increase more than tenfold by the end of 2026.

Wafer-Scale Technology and the CS-4 Roadmap

Cerebras distinguishes itself from conventional chipmakers by building processors that occupy an entire silicon wafer — the dinner-plate-sized disc that traditional fabs normally slice into hundreds of individual chips. Cerebras bypasses that slicing step entirely. The approach addresses a fundamental bottleneck in large-scale AI compute: in conventional clusters, moving data between separate chips across a network introduces latency and energy costs that grow with model size. By keeping an entire model's computation on a single wafer, Cerebras reduces inter-chip communication overhead.

The company's current flagship system is the CS-3, powered by the WSE-3 architecture. That processor packs 4 trillion transistors onto a single wafer. By comparison, Nvidia's H100 contains approximately 80 billion transistors.

Feldman and CFO Bob Komin are scheduled to unveil the CS-4, Cerebras' next-generation system, next week. The announcement will be closely watched as a signal of whether Cerebras can sustain its transistor-density advantage against Nvidia's accelerated product cadence, which includes the Blackwell generation now shipping.

Strategic Partnerships

The most prominent partnership on Cerebras' roster is a multi-year agreement with OpenAI valued at more than $20 billion. The deal is structured around delivering 750 megawatts of inference compute capacity. Inference — the process of running trained AI models to generate outputs — is the fastest-growing segment of AI compute demand as companies shift from training models to deploying them in production at scale.

Cerebras has also established a collaboration with Amazon Web Services to provide inference capabilities.

These partnerships underpin the $25.4 billion in performance obligations — a figure representing contracted future revenue that Cerebras expects to recognize over the coming years. The scale of those obligations reflects the broader industry trend of hyperscalers and AI labs locking in multi-year compute supply agreements amid persistent constraints on advanced chip availability.