Luminal raises $25 million to run AI inference on chips beyond Nvidia's

Luminal, a Y Combinator startup that builds software to run AI models across different chips, has raised $25 million in a round from Standard Capital, Tech Funding News reported. It has now raised about $30.3 million in total.

Luminal's search compiler tests mathematically equivalent ways of executing a model on a particular chip and picks the fastest, then reuses that result for repeated workloads. It pairs the compiler with an inference cloud so developers can deploy models without managing hardware, aimed at high-volume uses such as coding agents, voice agents and image editing. The company says it reached about 90 trillion tokens a month within four months of opening its public cloud, and has committed a large share of that work to Amazon's Trainium chips through a partnership with Amazon.

"Luminal's ability to get the most out of GPUs that have already been purchased and powered is incredibly high leverage," said Dalton Caldwell of Standard Capital. CEO Joe Fioti is a former Intel chip designer. The money goes into engineering, sales and support for more models and hardware.