Cerebras to Supply AI Systems to Cloud Startup Gimlet Labs in Push for Ultrafast Inference

Gimlet Labs plans to use the hardware to power its inference cloud — the systems that generate responses from AI models at speed.

Gimlet chief executive Zain Asgar told Reuters the company will target customers building their entire products around AI, and that speedy inference can make a decisive difference in areas such as cybersecurity, voice applications and financial analysis.

The collaboration pairs Cerebras' wafer-scale compute with the Gimlet Cloud in what the companies describe as a purpose-built disaggregated inference cloud, spanning datacenter infrastructure all the way to developer APIs.

Together they plan to deliver speeds of up to 3,000 tokens per second for demanding agentic and real-time applications, with the first Cerebras-powered Gimlet Cloud datacenter expected to come online later this year.

Gimlet Cloud combines the Cerebras Wafer Scale Engine with GPUs into an integrated inference solution, using advanced inference disaggregation technology to orchestrate model execution so that each phase of inference runs on the silicon best suited to it.

In AI, speed drives user experience and engagement: for real-time applications, from voice and video AI to agents and assistants, latency can be the difference between an interaction that feels seamless and one that feels slow.

"The deal between Gimlet and Cerebras is further validation of how easy it is to deploy, even in heterogeneous environments," Cerebras chief executive Andrew Feldman said, referring to installing the hardware alongside chips and systems from other companies.

The companies declined to disclose financial terms, and Gimlet will be responsible for maintaining and operating the systems once delivered.

Cerebras plans to supply its CS-4 systems — unveiled over the summer — over the next one to two years, with Gimlet planning to make the hardware available in its cloud in 2027.

The deal is the latest sign of surging interest across the AI industry in hardware purpose-built for inference, the step in which a trained model generates a response to a prompt.

Earlier this year, Cerebras signed a deal to supply OpenAI with chips, and last year Nvidia signed a licensing agreement with Groq for the startup's inference chips and hardware — evidence that the giants of the AI industry are treating raw inference speed as a strategic advantage on par with model capability itself.

By Hamza Faisal