Startup Gimlet Labs is solving the AI inference bottleneck in a surprisingly elegant way

Gimlet Labs has raised an $80 million Series A led by Menlo Ventures for its 'multi-silicon inference cloud' — software that orchestrates AI workloads across heterogeneous hardware including NVIDIA, AMD, Intel, ARM, Cerebras, and d-Matrix chips simultaneously. The technology slices agentic AI workloads so each step runs on the most suitable chip type, claiming 3x–10x inference speedups at the same cost. The startup targets large AI model labs and data centers, noting that existing hardware sits idle 70–85% of the time. Gimlet launched in October with eight-figure revenues and has since doubled its customer base. The company has raised $92 million total and employs 30 people.