Luminal raised a $25 million Series A led by Standard Capital to expand its software for running AI models across different chips. The San Francisco company aims to give developers more options beyond Nvidia’s hardware through its compiler and inference cloud.
Its compiler searches for efficient ways to perform a model’s calculations on a particular processor, then generates code for that chip. The aim is to reduce the specialist engineering required to move models between hardware platforms, a task Luminal’s September technical post describes as a barrier to switching suppliers.
Customers can pay for computing through Luminal’s managed cloud or license its software for their own infrastructure. The company targets developers of coding assistants, voice agents and image tools, whose applications repeatedly run AI models to produce responses.
The Series A follows a $5.3 million seed round disclosed in November 2025 and led by Felicis Ventures. Luminal says every investment firm from that round invested again. The new money will fund engineering, broader model and hardware support, and sales, according to a company statement reported by TNGlobal.
In a September 17 technical post, Luminal reported image-editing costs up to 47% below an optimized Nvidia H200 setup when it ran Flux.2 Klein 9B on AMD’s MI300X. The cheaper setup took 16%–31% longer per image across three tested sizes. The comparison used different hourly chip rental prices and excluded model loading, compilation and warmup from timing.