Gimlet Labs, a startup that has made solving AI inference bottlenecks its main focus, has raised $80 million in a Series A round, TechCrunch reported on March 23.
Gimlet Labs has developed a “multi-silicon inference cloud” that can run AI workloads across multiple types of hardware at the same time.
The multi-silicon inference cloud combines traditional CPUs, AI-focused GPUs and high-capacity memory systems to help maximise AI performance.
Tim Tully, a managing partner at Menlo Ventures, said, “A single chip cannot handle all AI tasks, but Gimlet Labs’ software connects different hardware to optimise performance.”
McKinsey said current data centre hardware utilisation is only about 15 percent to 30 percent, and improving it is expected to deliver cost savings of hundreds of billions of dollars. Gimlet Labs stressed that it provides a solution that increases AI inference speed by 3 to 10 times while keeping costs and power consumption unchanged.
Gimlet Labs provides its product as software or through its own cloud API. It is also working with major chipmakers including Nvidia, AMD, Intel and Arm.
Chi-gyu Hwang
delight@d-today.co.kr