Gimlet Valued at $3 Billion: From Scheduling Different Chips to Building Its Own Data Centers

Beating AI News reports that AI inference startup Gimlet Labs has closed a new $300 million funding round, lifting its valuation to $3 billion. Andreessen Horowitz (a16z) led the round, with new participation from Arm and Microsoft’s M12. Gimlet operates a multi-chip inference cloud, which splits individual AI inference tasks across stages to run on the most suitable chips. The startup has secured funding at a rapid pace: roughly six months ago, it closed an $80 million Series A round; less than a year prior, it raised a $12 million seed round, bringing its total three-round funding to $392 million. CEO Zain Asgar noted the latest round was finalized quickly because the company received multiple unsolicited term sheets from investors. Originally focused on chip orchestration software, Gimlet has since expanded to assisting clients with data center construction and is building its own data center facilities. Different chips have unique requirements for cooling, temperature, and server configurations, so strong software alone is insufficient—entire data center rooms must be redesigned accordingly. For instance, prefill and decoding in AI inference have distinct demands: the former is more compute-intensive, while the latter relies heavily on memory bandwidth. Gimlet can assign these tasks to chips from different vendors. Arm is both an investor and partner in this round; the two parties will also collaborate to adapt Gimlet’s software to more Arm chips.