Cerebras Systems plans to supply its CS-4 AI systems to cloud-computing startup Gimlet Labs for a deployment expected to consume about 100 megawatts of power.
Reuters reported that the rollout will focus on high-speed AI inference and is expected to take place over roughly one to two years, with Gimlet targeting cloud availability in 2027. Financial terms were not disclosed.
Why 100MW is significant
Power has become one of the defining constraints of AI infrastructure. A 100MW deployment is large enough to underline how quickly cloud providers are moving from small experimental clusters toward dedicated facilities built around AI workloads.
The figure should be understood as the expected power capacity of the planned deployment, not a benchmark of computing performance.
What Cerebras is supplying
The deal centres on Cerebras CS-4 systems. Cerebras has built its strategy around wafer-scale processors designed to handle large AI workloads with a different architecture from conventional GPU clusters.
Why inference is becoming a battleground
Training large models attracts much of the attention, but inference is the work performed every time a model answers a prompt, runs an agent or processes a request in production. As usage grows, inference becomes a recurring cost.
Cloud providers therefore have strong incentives to offer faster and more efficient inference. That creates opportunities for alternative hardware architectures and specialised accelerators.
What the timeline means
The reported one-to-two-year deployment window indicates that this is infrastructure being built out over time rather than capacity available immediately. Gimlet is targeting 2027 for cloud availability.
What is not yet known
Financial terms have not been disclosed. The companies also have not publicly provided a complete schedule showing how capacity will be phased, which regions will receive the service first or what pricing model Gimlet will use.
Bottom line
The Cerebras-Gimlet agreement is another sign that AI infrastructure is diversifying. The next phase of cloud competition will involve specialised systems, power capacity and the ability to deliver low-latency inference as a service.
