Pomegra
Gimlet Labs raises $300M in a Series B led by a16z, with Arm and Microsoft's M12 joining in, as the chip-agnostic inference startup reaches a $3B valuation six months after its $80M Series A.
- Andreessen Horowitz led the $300M Series B, with Arm, M12, Sapphire Ventures, Menlo Ventures, and Factory also participating.
- The round brings Gimlet's total funding to $392M and triples its implied valuation from just six months ago.
- Gimlet claims its multi-silicon inference software delivers up to 10x gains in throughput by routing AI workloads to their best-fit hardware.
Lead
Gimlet Labs, a San Francisco-based AI infrastructure startup, closed a $300 million Series B on September 4, 2026, led by Andreessen Horowitz, with new strategic investors Arm and M12 - Microsoft's corporate venture fund - joining returning backers Sapphire Ventures, Menlo Ventures, and Factory. The round sets Gimlet's post-money valuation at $3 billion. It follows an $80 million Series A that closed roughly six months prior, meaning the company's paper value nearly quadrupled in half a year.
What Does Gimlet Actually Build?
Gimlet builds inference orchestration software that disaggregates AI workloads and routes them across heterogeneous accelerators - GPUs, purpose-built AI chips, SRAM-based architectures, and CPUs - rather than locking customers to a single silicon vendor. The company calls this a "multi-silicon inference cloud." In practice, it means a customer's inference stack can pull from whatever mix of hardware delivers the best price-performance for each phase of a given workload.
The pitch is not novel in concept. Workload orchestration across heterogeneous compute has been a data-center staple for years. What Gimlet is selling is the AI-specific version: software smart enough to know which part of an agentic AI pipeline runs cheapest on a Nvidia GPU versus an Arm-based accelerator versus a specialized inference chip. The company claims throughput improvements of up to 10x versus single-vendor deployments, though independent benchmarks validating that figure are not public.
Why Are Strategic Investors - Not Just VCs - Writing Checks?
The presence of Arm and M12 alongside a16z is the round's most telling signal. Both are strategic, not financial, investors. Arm's participation makes sense on its face - Gimlet's software is designed to run on Arm-based chips, and the two companies have an active compatibility agreement. An Arm-compatible inference layer that routes workloads toward Arm silicon is, effectively, a sales tool for Arm's chip licensees.
Microsoft's M12 adds a different dimension. Microsoft Azure is a major consumer of inference capacity and has been expanding its accelerator portfolio beyond Nvidia. A portfolio company that helps enterprise clients run inference across mixed-chip environments fits Microsoft's interest in reducing customer dependency on any single GPU vendor, including Nvidia.
For a16z, this is a follow-on to a thesis already placed. The firm has backed the company since its seed stage. The Series B size - nearly four times the Series A - suggests internal confidence, though it also raises the question of burn: a startup needing $300M twelve months into commercial operations is either growing extremely fast or spending at a rate that warrants scrutiny.
What Does This Valuation Imply About the Series A?
At an $80M Series A and a $3B Series B valuation, Gimlet's implied round-over-round step-up is steep by any measure. If the Series A was priced at even a modest $300M to $500M valuation - a reasonable assumption for an early-stage infrastructure play - the Series B represents a 6x to 10x markup in six months. That is a venture outcome compressed into pre-revenue or very early revenue territory. Whether the underlying business metrics justify the valuation is unknown; Gimlet has not disclosed revenue or customer counts.
Competitive Context
Gimlet is not alone in the heterogeneous inference space. Hyperscalers including AWS, Google Cloud, and Microsoft Azure all offer multi-accelerator inference options natively. Independent infrastructure startups have targeted the same problem from the software layer. Gimlet's differentiation, per its positioning, is the independence from any single cloud or silicon vendor. Whether enterprise buyers value that independence enough to route inference through a third-party abstraction layer - rather than the native tooling from their primary cloud - remains an open commercial question.
The Arm partnership is concrete evidence of at least one form of hardware-vendor validation. The NVIDIA ecosystem conspicuously absent from the investor list; Nvidia's own inference software remains the default for GPU-heavy workloads, and any erosion of that default is a slow process.
Outlook
Gimlet enters the next phase well-capitalized and backed by investors with direct hardware and cloud skin in the game. The $3B valuation puts pressure on the company to show commercial scale, not just technical credibility. The multi-silicon inference bet pays off only if enterprise buyers actually diversify away from single-vendor GPU stacks - a trend that is moving, but slowly. The strategic alignment of Arm and M12 gives Gimlet distribution leverage that pure financial rounds do not. Whether that translates into revenue at a pace consistent with the valuation will determine whether this round looks prescient or premature by the time the next one is due.
