Gimlet Labs Raises $300M at $3B Valuation for Its AI Inference Cloud

Gimlet Labs has raised $300 million in a Series B round at a $3 billion valuation, according to a company announcement on September 4, 2026. Andreessen Horowitz led the round, with Sapphire Ventures, M12, and chip designer Arm joining as new investors alongside existing backers Menlo Ventures and Factory. The raise brings Gimlet's total funding to $392 million and roughly triples its valuation from where it stood after emerging from stealth less than a year ago.

What Gimlet actually sells

Gimlet's pitch is narrow and, if it works, genuinely useful: don't run every AI inference workload on the same type of chip. The San Francisco company built software that splits an AI model's inference process into phases and routes each phase to whichever hardware handles it best, whether that's a GPU, a custom accelerator, or a CPU from a different vendor entirely. The company calls it a "multi-silicon inference cloud," and it's aimed squarely at agentic AI workloads, the multi-step, tool-calling systems that are far heavier and more variable than a single chatbot reply.

Co-founder and CEO Zain Asgar put it plainly in the announcement: "We're able to deliver unprecedented performance because Gimlet software intelligently slices and orchestrates their workloads across different types of hardware." The company claims up to 10x gains in throughput and interactivity from that approach, though those are Gimlet's own figures rather than an independent benchmark.

The traction behind the price tag

A $3 billion valuation for a company that only left stealth in October 2025 sounds aggressive until you look at what Gimlet says it's landed since. By March 2026 it had tripled its customer base and signed one of the top three frontier AI labs and one of the top three hyperscalers as customers. It also says it has since secured billions of dollars in contracted revenue for its Gimlet Cloud product and is scaling infrastructure toward hundreds of megawatts. Gimlet joined MLCommons, the industry benchmarking consortium, in June 2026, a move that reads like a company trying to get its performance claims independently measured rather than just self-reported.

Why this matters beyond one funding round

The bigger signal here isn't the dollar figure. It's who's writing checks. Arm doesn't typically invest in AI infrastructure startups unless it sees its own chip designs benefiting from wider adoption of heterogeneous compute. A hyperscaler and a frontier lab both becoming paying customers, not just pilot partners, suggests the biggest AI spenders are actively looking for ways to stop over-relying on a single chip supplier for inference, even as that same supplier keeps winning most of the training market.

That's a real shift. For most of the current AI boom, the story has been simple: buy more GPUs from one company. Gimlet's raise is a bet that the inference side of AI, which is where the actual day-to-day cost of running these systems lives, splits differently than the training side did.

The verdict

$300 million is a big check for a company with no independently audited performance numbers yet. But the investor list and the customer list both say something that matters more than the price: serious money believes inference workloads won't stay locked to one type of chip. If Gimlet's real-world numbers hold up to outside scrutiny once MLCommons results are public, this round will look cheap in hindsight. If they don't, it joins a long list of infrastructure startups that raised big on a good pitch and a thin track record.