AI orchestration startup Gimlet Labs has raised $300 million at a $3 billion post-money valuation, only six months after closing an $80 million Series A. Andreessen Horowitz led the round, with new backers Arm Holdings and M12, Microsoft's venture fund, joining alongside existing investors Menlo Ventures, Eclipse, Factory, Prosperity7 and Triatomic.
A multi-silicon inference cloud
Gimlet builds what it calls a multi-silicon inference cloud, software that distributes AI workloads across GPUs, CPUs and specialised accelerators rather than assuming everything runs on the same chip. Its platform can slice a single model across different architectures and route each stage to the hardware best suited to it, delivering what the company says are three-to-ten-times inference speed improvements on the same cost and power envelope.
The company was founded by former Pixie Labs engineers led by Zain Asgar, whose previous startup was acquired by New Relic. Gimlet says its customer base has tripled since its March financing, adding a top-three frontier lab and a top-three hyperscaler.
Building beyond software
Deploying multi-chip inference exposed a physical constraint: data centres themselves were largely designed around homogeneous fleets. Asgar told Bloomberg the company is now helping customers configure their own facilities and building data centres of its own. Different chips require different cooling systems, temperatures and power topologies, which turns a software problem into an infrastructure problem.
Where a16z is placing the bets
The round fits a broader Andreessen Horowitz push into the physical layer of AI. The firm recently expanded its Growth fund to $8.5 billion and launched a separate $1.1 billion Machine Age Fund targeting chips, networking, storage, data centres and robotics. Gimlet slots into that thesis: rather than backing another Nvidia challenger, a16z is financing the coordination layer that sits above every accelerator.
Compute economics as an investment theme
The round also lands the same day Crusoe finalised its own $3 billion raise at a $30 billion valuation for its AI data-centre buildout, and days after Nscale hit $14.6B on the back of Europe's sovereign AI push. Together the deals show investors financing both raw compute capacity and the software that squeezes more useful work out of it — a market that increasingly resembles infrastructure finance rather than conventional venture. Gimlet's speed of markup, from $80M Series A to $3B valuation in six months, captures how quickly capital is repricing the orchestration layer.
Reporting based on coverage from Bloomberg, Tech Funding News and TechStartups.