Gimlet plans to add Cerebras systems to its inference cloud
Gimlet Labs announced a partnership to add Cerebras systems to the cloud where it runs AI models. The integrated system is serving private deployments, while the first Cerebras-powered Gimlet data center remains planned for later this year. Gimlet expects customer access to newer CS-4 systems in 2027. The companies' speed and capacity targets are not independent performance measurements for a broadly available service. The agreement centers on combining different processors for inference work.
Artificial Intelligence··Evening
Gimlet and Cerebras partner on inference cloud
Gimlet Labs announced a partnership on September 28 to add Cerebras processors to Gimlet Cloud, where AI models produce answers. Superpower Daily reported the collaboration and the target of opening the first Cerebras-powered data center later this year. Inference is the stage when a trained model answers a request; delays accumulate when an agent makes repeated model calls. The companies aim to reduce that delay. The data-center date is a deployment target, not evidence of a broadly available service already running for customers.[1], [2]
Model phases are assigned to different chips
Gimlet plans to place the preparation and token-generation phases of a model run on different processors. Cerebras' Wafer Scale Engine, with substantial on-chip memory, is intended especially for generation, while graphics processors handle other phases. Gimlet expects developers to use one application interface without managing that split themselves. The partners target speeds of up to 3,000 tokens per second for demanding agent and real-time uses. That figure is a company target, not an independently verified speed across a generally available service.[1]
From private deployments toward a new data center
Gimlet says the companies began joint customer work last year and that an integrated setup already generates tokens in private deployments. That early use has a different reach from the first Cerebras-powered Gimlet data center planned for later this year. Gimlet also expects to become an early partner for Cerebras' newer CS-4 system, with customer access anticipated in 2027. Deployment and speed targets can be assessed as installations arrive. The agreement now links the hardware supplier and Gimlet's cloud software in one planned service.[1]
