An inference cloud that splits AI workloads across GPUs from any vendor.
AI inference cloud purpose-built for the high-speed demands of agentic workloads. By moving away from standard GPU-only setups, the company uses a heterogeneous software and hardware stack to optimize the complex "chains" of reasoning that AI agents require.
Gimlet was spun out of a Stanford research project by Zain Asgar, Michelle Nguyen, Omid Azizi, Natalie Serrino, and James Bartlett. Zain is a Stanford adjunct professor and a repeat founder: his previous company, Pixie, was acquired by New Relic in 2020. Inference is becoming the most expensive layer of AI deployment, and agentic workloads compound that demand, yet existing hardware is used only 15 to 30 percent of the time. Gimlet’s multi-silicon inference cloud automatically slices and maps agentic workloads across diverse hardware, which speeds up AI inference 3x to 10x for the same cost and power. That makes it a vendor-neutral layer that benefits as chip architectures diversify. Traction is already significant: Gimlet emerged from stealth with eight-figure revenues and tripled its customer base by March, adding a top frontier lab as a customer.
In the News
The latest articles on Gimlet Labs’s product releases, fundraising, and awards.