Gimlet Labs, which has now raised $92 million, introduces the world’s first 'multi-silicon inference cloud.' This innovation splits AI tasks across diverse processors like CPUs, GPUs, and specialized chips, significantly boosting efficiency. Founders say existing hardware operates at 15%-30% capacity, wasting billions. Partners include Nvidia, AMD, Intel, and Cerebras. The startup aims to accelerate AI inference workloads by 3-10x, making it more cost-effective.
With its software layer, Gimlet allows different parts of AI models to run on different chips, reducing idle times and improving overall performance. The company already generates eight-figure revenues and plans to expand its team and inference cloud to meet growing demand.