Cerebras Systems has agreed to supply cloud startup Gimlet Labs with AI chips amounting to 100 megawatts of compute, according to a report by Quartz. The figure is the strongest concrete number in the deal, and it frames the transaction in the currency that now dominates AI infrastructure procurement: raw power consumption rather than unit counts or dollar value.

Neither company has disclosed a revenue figure, a delivery schedule, or a shipment volume for wafer-scale processors, according to the report. The 100-megawatt commitment is therefore a capacity statement, not a confirmed revenue number, and investors reading the announcement should treat pricing and timing as open questions.

The structure of the deal marks a commercial milestone for Cerebras. The company has spent years positioning its wafer-scale technology as an alternative to GPU-based clusters for large-model training and inference. A multi-megawatt commitment from a cloud operator — rather than a single research deployment or a government-linked order — gives Cerebras a repeatable commercial customer in the fast-growing AI cloud market.

For Gimlet Labs, the bet is on differentiation through silicon. Cloud startups compete against hyperscalers that buy accelerators at enormous volume and negotiate accordingly. Committing to 100 megawatts of Cerebras compute, rather than to conventional GPUs, gives the startup a supply position its larger rivals do not replicate, and a potential performance-per-watt story aimed at AI customers whose deployments are increasingly constrained by grid access rather than capital.

Megawatt-based contracting has become the standard shorthand for AI infrastructure scale because power, not chips, is now the binding constraint on data center expansion. Utilities and grid operators across the United States are struggling to interconnect gigawatt-scale AI campuses, and cloud providers increasingly size and price their offerings in energy terms. A 100-megawatt commitment places Gimlet Labs' planned Cerebras capacity in the same planning language used for hyperscaler buildouts, even though the absolute scale remains far smaller.

The deal also carries supply chain weight. Cerebras designs its chips at wafer scale, an architecture that departs from conventional reticle-limited manufacturing, and its production depends on advanced foundry capacity that remains concentrated at a small number of suppliers. Whether Cerebras can secure enough wafer supply to fulfill a 100-megawatt deployment on a commercial timeline will determine whether the agreement functions as a capacity contract or a roadmap.

Competitive dynamics matter here as well. The AI accelerator market is dominated by Nvidia, and challengers including Cerebras have struggled to convert technical differentiation into sustained multi-customer volume. A publicly stated 100-megawatt order from a cloud provider gives Cerebras a reference deployment it can take to other infrastructure buyers, potentially shifting the competitive conversation from benchmarks to delivered megawatts.

Key details remain undisclosed. The report does not specify when deliveries begin, how the 100 megawatts will be phased across installations, whether the agreement includes minimum purchase commitments or take-or-pay terms, or how Gimlet Labs will price the resulting cloud services against GPU-based rivals.

Those gaps define what to watch next. Confirmation of a delivery schedule and pricing structure would convert the announcement from a headline-scale commitment into a bankable commercial contract, and would sharpen the comparison between Cerebras-based capacity and GPU clusters on performance per watt — the metric that increasingly decides who wins AI cloud deals.