One OpenAI Deal Is Behind Most of Cerebras's $25.4 Billion Backlog

Cerebras, the AI chipmaker whose wafer-scale processors are pitched as an Nvidia alternative for training and inference, reported $25.4 billion in remaining performance obligations at the end of June, the contracted revenue the company has signed for but not yet recognized as earned. The figure is roughly 29 times Cerebras' expected 2026 revenue, and on its face it reads like a runaway backlog. The quarterly filing, as summarized by The Motley Fool, shows almost all of that weight sits on one agreement, and management itself says less than a quarter of it converts in the next two years.

The deal is the master relationship agreement Cerebras signed with OpenAI in December 2025 for 750 megawatts of AI inference capacity. Cerebras valued that arrangement at more than $20 billion. OpenAI also holds an option to buy an additional 1.25 gigawatts by the end of 2030. In the second quarter alone, $56.8 million of revenue, about 32% of Cerebras' $180.1 million GAAP top line, came from that single customer, according to the same filing summary.

The conversion curve is the part the headline number obscures. Management told investors it expects roughly 22% of the $25.4 billion, or about $5.6 billion, to convert to revenue over the 24 months ending June 30, 2028. Another 43% lands in months 25 through 48. The remainder sits past that window. That is a slow-moving balance sheet item with a long tail, not a near-term cash event.

The line moved from $24.6 billion at the end of 2025 to $25.0 billion at the end of March and $25.4 billion at the end of June, roughly 3% growth over the first half of 2026. A $25.4 billion headline that barely budges over six months reads differently from a $25.4 billion headline that doubled in a quarter.

There is a second change in the curve worth tracking. At the end of 2025, management expected about 15% of the backlog to convert in the trailing 24-month view. The current filing pushes that to 22%. Cerebras is pulling future revenue into the near-term window. The shift reduces the risk that the OpenAI arrangement stalls, and it concentrates the next two years of reported growth on a single counterparty.

Q2 itself was strong. Non-GAAP revenue grew 103% year over year to $209.9 million. GAAP revenue rose 74% to $180.1 million. The inference cloud business, the part that touches the OpenAI deal, nearly quadrupled in the quarter. Cerebras raised its full-year 2026 adjusted revenue outlook to $880 million–$890 million. The backlog is the other number in the same release, and the two are best read together.

A 750-megawatt inference commitment valued at more than $20 billion, with a 1.25-gigawatt option layered on top, means the $25.4 billion figure moves with one customer's data center build-out pace, one customer's power siting decisions, and one customer's appetite for more Cerebras silicon. The filing notes that the conversion timing is at the customer's request, so OpenAI can push deliveries out or pull them forward, and Cerebras reports against that schedule.

Cerebras' next 10-Q will show whether the 22% conversion window holds, whether the second 24-month bucket stays at 43%, and whether OpenAI moves on the 1.25-gigawatt option. The company has not disclosed a public deadline for that option beyond the end of 2030. The $25.4 billion number is large. The part worth holding onto is smaller: one customer, one pacing decision, and a backlog whose shape management itself is still redrawing.