Gimlet Labs raises $80M to split AI across any chip
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Gimlet Labs raised an $80 million Series A led by Menlo Ventures, with total funding now at $92 million; the startup publicly launched in October with eight-figure revenues and has since doubled its customer base in four months.
- Gimlet's "multi-silicon inference cloud" software splits AI agent workloads across diverse hardware because, as Menlo's Tim Tully wrote, "inference is compute-bound; decode is memory-bound; and tool calls are network-bound" — meaning no single chip handles all steps.
- Founder Zain Asgar, a Stanford adjunct professor and previously exited founder (his prior startup Pixie was acquired by New Relic in 2020), says AI applications use deployed hardware only 15% to 30% of the time, calling it "hundreds of billions of dollars" in wasted idle resources.
- Gimlet claims it speeds AI inference 3x to 10x for the same cost and power, and says it can even slice the underlying model so each portion runs on the best-suited chip architecture.
- The company's hardware partners include NVIDIA, AMD, Intel, ARM, Cerebras, and d-Matrix; the product is aimed at the largest AI model labs and data centers, not rank-and-file developers.
Why it matters: With McKinsey projecting nearly $7 trillion in data center spending by 2030 and Asgar's own claim that 70-85% of deployed AI hardware sits idle, even a fraction of Gimlet's stated 3x-10x efficiency gain could redirect tens of billions in capital spend for the hyperscalers and frontier model labs who are the actual buyers — not the developer market the term sheet volume might imply.
Ask SkimNews
