Three years ago, Luminal co-founder Joe Fioti was designing chips at Intel when he hit a realization: the real bottleneck in modern compute wasn’t hardware, it was software.
“You can make the best hardware on earth, but if it’s hard for developers to use, they’re just not going to use it,” he said.
That frustration led to Luminal, a startup focused entirely on improving the software tools that sit between developers and GPU hardware. On Monday, the company announced a $5.3 million seed round led by Felicis Ventures, with angel investors including Paul Graham, Guillermo Rauch, and Ben Porterfield. The startup also participated in Y Combinator’s Summer 2025 batch.
Luminal’s core business mirrors new wave “neo-cloud” companies like CoreWeave and Lambda Labs: it sells compute. But instead of competing on sheer GPU supply, Luminal differentiates by squeezing more performance out of existing hardware, using sophisticated optimization techniques. Its main target is the compiler layer, the system that translates written code into GPU instructions.
Today, the industry standard is Nvidia’s CUDA, a crucial but often overlooked reason behind Nvidia’s dominance. While parts of CUDA are open-source, other layers remain highly specialized. Luminal sees an opportunity to build an open, more flexible alternative, especially as companies scramble for compute and look for cheaper, faster inference solutions.
Related: South Korea’s Bone AI Pushes Into “Physical AI” With $12M Seed And Bold Defense Ambitions
The startup joins a rising wave of AI inference-optimization startups, including Baseten, Together AI, Tensormesh and Clarifai, all aiming to reduce the cost of running large AI models. But Luminal faces strong pressure from in-house optimization teams at major AI labs, which can fine-tune for a single family of models, an efficiency advantage Luminal doesn’t have.
Still, Fioti believes the market is growing too quickly for competition to slow them down.
“It’s always possible to spend six months hand-tuning a model architecture on specific hardware,” he said. “You’ll probably beat any compiler. But our bet is that anything short of that will still be incredibly valuable at scale.”
With a founding team from Intel, Apple, and Amazon, and a fast-growing demand for efficient inference, Luminal is aiming to become a critical layer in the AI compute stack,not by selling more GPUs, but by making them smarter.
