Vicor’s OpenAI Exposure Through Cerebras

Thesis

OpenAI’s demand for Cerebras creates a credible but indirect hardware-demand path to Vicor:OpenAI capacity commitment → Cerebras systems deployed → Vicor vertical-power modules shipped.Vicor is not exposed to OpenAI token usage directly. It is exposed to the number and design of Cerebras systems built to deliver OpenAI’s contracted capacity.

Scale of the path

Item	Current evidence
OpenAI committed Cerebras capacity	750 MW through 2028
Additional OpenAI option	1.25 GW through 2030
Cerebras remaining performance obligations at June 30, 2026	$25.4B, significantly related to OpenAI
Cerebras revenue associated with OpenAI in H1 2026	$74.4M
Approximate WSE-3 engine power	25 kW
Reported Vicor modules per current WSE-3 system	84
Vicor Q2 2026 revenue / backlog / net income	$143.4M / $380M / $49.8M

As a scale illustration only, 750 MW divided by 25 kW equals roughly 30,000 current-WSE power equivalents. At 84 Vicor module positions per current system, that would imply about 2.5 million module positions. This is not a shipment forecast: the 750 MW may include facility overhead and supporting equipment, deployment will use future WSE generations, and Vicor’s future module count and content per system are undisclosed.Cerebras describes its engine block as delivering power directly into the wafer because traditional packaging cannot achieve the required density. Third-party teardown work identifies 84 Vicor power bricks aligned with 84 WSE regions and converting power near the wafer to roughly 1V. Cerebras CS-3, SemiAnalysis teardown

Why the exposure is credible

• OpenAI is already offering Cerebras-backed inference and lists Cerebras as a cloud-infrastructure subprocessor. OpenAI subprocessor list
• Cerebras has begun recognizing OpenAI-related revenue, so the relationship has moved beyond benchmarking and development.
• Vicor says its lead computing customer is undergoing a steep wafer-scale-engine ramp and is transitioning to second-generation vertical power delivery.
• On Vicor’s Q2 call, an analyst asked whether Cerebras could represent 25%–30% of Vicor’s business in 2027. Management declined to quantify the customer but said the relationship was strong; the percentage is therefore an analyst premise, not company guidance. Vicor Q2 call

The important economic distinction

Cerebras earns from contracted capacity and inference services. Its returns depend on utilization, token pricing, data-center costs, financing, and cloud margins.Vicor primarily earns when Cerebras builds the hardware. That can let Vicor benefit during the capacity buildout before Cerebras proves attractive per-token economics. But Vicor does not receive recurring revenue for every token served after installation. Take-or-pay protection for Cerebras also does not automatically create additional Vicor revenue unless more hardware is deployed.The OpenAI agreement therefore supports Vicor’s deployment volume, not a direct percentage of Cerebras backlog.

What remains unproven

The missing variables are Vicor’s dollar content per system, WSE-4/WSE-5 module counts, pricing, deployment timing, and whether Cerebras adds a second source or changes its power architecture. OpenAI materially de-risks demand for Cerebras; continued Vicor content in the next system generations determines how much of that demand reaches VICR earnings.Bottom line: OpenAI provides a large, multi-year demand anchor for Cerebras, and Cerebras provides a credible lead-customer ramp for Vicor. The exposure is real, but the 750 MW and $25.4B backlog cannot be translated directly into Vicor revenue without disclosed per-system content and deployment data.Underlying Cerebras/OpenAI research