Luminal raises $25M Series A to run AI models beyond Nvidia chips

Luminal has raised $25 million in Series A funding led by Standard Capital to help AI companies run models on chips outside Nvidia's ecosystem.

The San Francisco-based AI infrastructure company announced the round on Thursday. The money will fund more engineering and go-to-market work as Luminal adds support for more AI models and chip platforms.

Every seed investor doubled down. Y Combinator, Felicis, SagaVC and Liquid 2 all re-upped in the Series A, the company said in a statement.

What does Luminal actually build?

Luminal sells two products: a search compiler and an inference cloud.

The compiler tests mathematically equivalent ways of running a model on a given chip and picks the fastest. Once a model is optimised for a chip, that configuration can be reused. That reuse is what makes the search worthwhile across repeated workloads.

The inference cloud lets developers deploy AI models without managing the infrastructure underneath.

Together, the two products let AI companies use different hardware platforms and cut the cost of running models, Luminal said.

Why does hardware choice matter?

Much of the industry depends on Nvidia's GPU ecosystem. Optimising models for other chips often takes specialist engineering and extensive tuning. Luminal's approach can widen the hardware choices available to AI companies, beyond Nvidia's ecosystem, without forcing them to give up speed or cost.

The company puts the opportunity at $5 trillion in AI infrastructure.

Joe Fioti, co-founder and CEO of Luminal, said the company began by building in the open on GitHub as AI adoption grew. As Luminal sees strong demand for a more efficient way to run models across different hardware, Fioti emphasised that the cost of running AI should not limit who can build it.

How fast is the business growing?

The growth numbers are concrete. Since launching its public cloud after a $5.3 million seed round in November 2025, Luminal has reached 90 trillion tokens processed a month within four months.

A large share of those tokens runs on Amazon's Trainium chips, under a deepening partnership with Amazon. That tie-up gives Luminal a live proof point that its compiler works on hardware outside the Nvidia stack, and a distribution channel into developers already inside Amazon's cloud.

The trajectory matters for the Series A math. A startup that scales from seed launch to 90 trillion monthly tokens in four months can argue its cost advantage compounds as workloads repeat and optimised configurations get reused.

Where does the money go?

Luminal says the Series A will fund:

• Development of its compiler and inference cloud
• Wider model and hardware support
• Expansion of its go-to-market effort

The bet behind the round is straightforward. If AI inference keeps growing and Nvidia supply stays tight and expensive, the value of software that makes other chips competitive rises with every token processed — and Luminal now has $25 million to chase that market.