Three years ago, Luminal co-founder Joe Fioti was working on chip design at Intel. He was working on making the best chip possible, but the more important bottleneck was in the software.
“You can make the best hardware on the planet, but if it’s difficult for developers to use, they won’t use it,” he told me.
Now he has founded a company focused entirely on that problem. On Monday, Luminal announced $5.3 million in seed funding in a round led by Felicis Ventures with angel investments from Paul Graham, Guillermo Rauch, and Ben Porterfield.
Fioti co-founders Jake Stevens and Matthew Gunton come from Apple and Amazon, respectively, and the company was part of Y Combinator’s summer 2025 batch.
Luminal’s core business is simple. The company sells computing, similar to neo-cloud companies like Coreweave and Lambda Labs. But while those companies are focused on GPUs, Luminal is focused on optimization techniques that allow it to squeeze more compute out of its infrastructure. In particular, the company focuses on optimizing the compiler, which sits between the written code and the GPU hardware. This is the same developer system that caused Fioti a lot of headaches at his previous job.
Currently, the industry-leading compiler is Nvidia’s CUDA system. This is an underrated element in the company’s phenomenal success. But many elements of CUDA are open source, and while much of the industry is still scrambling for GPUs, Luminal is betting there’s a lot of value to be gained from building out the rest of the stack.
It’s part of a growing number of inference optimization startups that are gaining in value as companies look for ways to run models faster and cheaper. Inference providers like Baseten and Together AI have specialized in optimization for years, but smaller companies like Tensormesh and Clarifai are now emerging with a focus on more specific technical tricks.
Luminal and other members of the cohort will face stiff competition from optimization teams from leading laboratories that have the advantage of optimizing against a single family of models. Luminal, which works for clients, has to adapt to whatever models emerge. But Fioti says that despite the risk of being overtaken by hyperscalers, he’s not worried because the market is growing fast enough.
“You can always spend six months manually tweaking a model architecture on a particular piece of hardware, and you could probably beat the performance of any kind of compiler,” Fioti says. “But our big bet is that aside from that, multipurpose use cases are still very valuable economically.”
Source link
