Q.ANT gives away the software for its light-powered AI chips in a CUDA-style bet on developers
The New Stack Matthew Burns
Q.ANT put its chip software on GitHub for free. It’s a CUDA-style bet: win developers now, even though the actual chips are still scarce.
Based on reporting by The New Stack, Matthew Burns — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Q.ANT, a Stuttgart startup, is trying to do for photonic AI chips what CUDA did for Nvidia’s GPUs: make the software layer the thing developers reach for first. This week it released a free, open-source toolkit on GitHub so people can write and test code on a normal computer before they ever touch the hardware.
That hardware is still limited. Q.ANT says its chips are already running at a few research computing centers, and wider access is still coming in the “coming months” through cloud service from German provider IONOS or through an on-site server from Q.ANT. So the company is asking developers to start building before they can easily verify the big promise: lower power use.
The toolkit, called the Q.ANT Native Computing Toolkit, works in Python and C and comes with a simulator that mimics the chip without requiring Q.ANT drivers. Its first version is aimed at inference, not training. The examples include handwritten digit recognition, object detection in photos, and shape outlining in images. Training, for now, still happens on standard CPUs and GPUs.
Q.ANT’s basic pitch is that light can do some of the math AI chips normally handle electrically, especially the data-movement-heavy parts that chew through power. Its chips use wave-shaped functions similar to a cosine, and the company says models built around those functions can use fewer parameters. Fewer parameters can mean smaller models, less data shuffled around, and less energy burned. The toolkit includes company-made comparisons between standard models and ones built the Q.ANT way.
Founder and CEO Michael Förtsch framed the release as a software-first play, calling it the “Linux moment” of photonic computing. It’s a neat line, but the market has heard versions of this before. Lightmatter has shifted toward Passage, which uses light to move data between chips rather than do the compute itself, and Graphcore had its own software stack before SoftBank bought it in 2024. Q.ANT also says this is the first openly available software kit for programming a photonic processor, though Xanadu has offered free open software for its light-based quantum computers since 2018. For now, the simulator is the main attraction. The real test — whether the chips live up to the power-saving story on other people’s models — still has to wait.
My take — AI-written commentary, not fact-checked reporting
This is the right move, and also the most obvious one: if the hardware is scarce, flood the zone with software and hope developers do the evangelizing for you. The industry keeps pretending compute is won by transistor wizardry alone, then remembers that CUDA exists and all the drama starts again. Open tools are how these chip startups stop sounding like science projects with a pitch deck.
Read more about this at: The New Stack