French startup ZML opens AI inference to any chip

Nvidia still dominates AI hardware, but the field is opening up — and a small Parisian startup just made a big move. ZML, backed by Yann LeCun, released inference software that lets open-source LLMs run at max speed on AMD, Google TPU, Apple Metal, Intel Arc — and yes, Nvidia too. Founder Steeve Morin says the point isn’t to knock Nvidia down, but to break the vendor lock-in that’s been quietly suffocating the AI inference scene. Inference is now more important than training in daily AI use, so having hardware flexibility matters. ZML’s software lets enterprises mix cheaper or more energy-efficient chips without the usual performance trade-offs. The 20-person team raised $20M thanks to Morin’s track record with Zenly. The move could shake up the ‘inference gold rush’ — and give novel European chipmakers a fighting chance. Why it matters: If AI is going to be truly embedded in everyday work and life, it can’t stay locked into one vendor’s garden — ZML just handed out the keys.