Gimlet Labs, a startup building software that distributes AI inference workloads across different chip architectures, has raised $300 million in a Series B round led by Andreessen Horowitz, the company announced Thursday. The round values Gimlet at $3 billion — a steep jump from the $80 million Series A it closed just six months ago.
Arm Holdings and Microsoft's venture fund M12 also participated. The company plans to use the funding to expand compatibility with more chip architectures, particularly Arm-based processors, and to deepen its partnership with Arm on optimizing inference across heterogeneous hardware.
Founded in 2023 by Zain Asgar, Michelle Nguyen, Omid Azizi, and Natalie Serrino, Gimlet has developed a software layer that acts as an orchestration engine for AI workloads. Instead of running all inference on Nvidia GPUs — the default for most AI companies — Gimlet's platform analyzes a workload and routes tasks to the most efficient hardware available, whether that's a GPU, a CPU, or a specialized accelerator.
The pitch directly challenges Nvidia's dominance in AI infrastructure. While Nvidia's CUDA ecosystem has made its GPUs the de facto standard for AI training and inference, companies are increasingly looking for alternatives as hardware costs soar and supply constraints persist. Gimlet's multi-silicon approach lets organizations use whatever chips they already own or can acquire more cheaply.
The investment underscores a growing belief in the AI industry that inference — running trained models to generate outputs — will become the dominant cost center as AI applications scale. Companies are racing to find more efficient ways to serve AI models, and chip-agnostic orchestration software represents one of the most promising approaches.
Gimlet's rapid fundraising pace — $80 million to $300 million in half a year — reflects both the urgency of the AI efficiency problem and the confidence investors have in the team's technical approach. The involvement of Arm and Microsoft also signals strategic alignment: both companies benefit from a world where AI workloads aren't locked into a single chip vendor.
