Cerebras Systems plans to supply Gimlet Labs with artificial-intelligence hardware representing roughly 100 megawatts of computing capacity, expanding the infrastructure race around high-speed AI inference.
The companies said Monday that Gimlet will integrate Cerebras technology into its cloud-computing platform, with Cerebras supplying its CS-4 systems over approximately one to two years. Gimlet expects to make CS-4 capacity available through its cloud in 2027.
Financial terms were not disclosed.
Inference is the computing stage in which a trained AI model generates responses to users. As AI applications move toward voice interaction, cybersecurity, financial analysis and autonomous agents, response speed is becoming an increasingly important competitive variable.
Gimlet says the combined architecture is intended to use different types of computing hardware for different portions of AI workloads rather than relying on a single processor architecture for every task.
The companies are targeting performance of as much as 3,000 tokens per second for certain agentic and real-time workloads, although that figure is a company performance target rather than an independently verified benchmark.
The first Cerebras-powered Gimlet Cloud data center is expected to begin operating later this year, according to the companies.
The agreement illustrates how the AI infrastructure market is expanding beyond training massive models. As businesses put AI systems into production, inference capacity is becoming a major source of computing demand in its own right.
That creates an opportunity for alternative accelerator architectures competing with conventional GPU-based infrastructure.
Investors should watch the deployment schedule, actual production performance and customer adoption as Gimlet begins making the capacity commercially available.
