Cerebras Systems will supply its AI computing systems to Gimlet Labs, a cloud computing startup, under an agreement reported by WTAQ. The deal pairs the maker of the industry's only wafer-scale processors with a young cloud provider looking to differentiate its infrastructure offering.

For Cerebras, the arrangement marks another route to market beyond its own cloud service and its direct enterprise sales. The company builds its AI systems around the WSE family of chips — processors fabricated at effectively a full wafer scale rather than being diced into individual dies. That architecture gives a single chip vastly more on-chip memory and core count than conventional GPUs, a proposition Cerebras has pitched as a way to sidestep the interconnect bottlenecks that slow multi-GPU clusters on large models.

Details of the agreement — the number of systems involved, deployment timelines, pricing, or revenue commitments — were not disclosed in the report. What is confirmed is the supply relationship itself: Cerebras hardware will sit inside Gimlet Labs' cloud infrastructure, making the startup one of the independent operators now building services around non-NVIDIA silicon.

That framing matters commercially. Nvidia's CUDA ecosystem still dominates AI cloud capacity, and most hyperscalers remain overwhelmingly GPU-based. Deals like this one give challenger accelerator vendors a foothold: a cloud provider that offers Cerebras systems can target customers who want large-memory, high-throughput inference and training without managing clusters of networked GPUs themselves. Cerebras has pursued this playbook before through its own managed cloud, where it sells per-token and per-slot access to its hardware. Partnering with an outside cloud operator extends that reach without Cerebras having to build out the full commercial stack alone.

For Gimlet Labs, whose public profile remains thin, the supply deal signals an infrastructure strategy built on differentiated silicon rather than on price competition with the major clouds. Startups in this position typically win customers where the underlying hardware delivers a specific advantage — in Cerebras' case, the wafer-scale chip's large fast on-chip memory, which the company argues removes the memory-bandwidth wall that constrains GPU inference at long context lengths.

The agreement also lands amid a shifting supply picture for AI accelerators. GPU lead times and pricing pressures have pushed cloud operators and model developers to qualify alternative architectures, and investors have backed a widening field of accelerator vendors seeking capacity at leading-edge foundries. Cerebras, headquartered in Sunnyvale, California, has spent years qualifying its wafer-scale design for volume manufacturing — an engineering problem distinct from conventional chip production, since the entire wafer must function as one device.

How much volume this deal represents is the open question. A single startup customer does not move a vendor's revenue base on its own, and Cerebras has not published system shipment figures that would anchor an estimate. Investors and analysts tracking the accelerator market will watch whether Gimlet Labs expands its initial deployment, and whether Cerebras converts similar supply agreements with other cloud operators — the pattern that would turn individual deals into a genuine second sales channel alongside its direct cloud business.