Cerebras has landed a deal to supply Gimlet Labs with AI hardware that can consume roughly 100MW, giving fast chatbots a hefty electricity bill.
According to Reuters, the agreement covers CS-4 systems delivered over one to two years, with Gimlet planning to offer the hardware through its cloud in 2027. Neither company disclosed the financial terms. Gimlet will operate and maintain the systems once delivered.
The announcement adds some technical detail to the spending. Gimlet will combine Cerebras’ wafer-scale processors with GPUs, assigning different stages of inference to the hardware best suited to each task.
The companies are targeting speeds of up to 3,000 tokens per second for AI agents and real-time applications. That remains a performance claim, with results dependent on the model and configuration.
Gimlet Labs co-founder and chief executive Zain Asgar said: “Inference speed matters. It determines how productive AI can be.”
Gimlet and Cerebras say they have worked on joint customer projects since last year, with an integrated system already serving tokens in private deployments. Their first Cerebras-powered Gimlet Cloud data centre is expected online later this year. The separate CS-4 availability timetable reported by Reuters points to 2027.
The collaboration will extend across infrastructure design, developer tools, APIs, software optimisation and production operations.
