France’s ZML wants to break Nvidia lock-in with free cross-chip AI software

Paris startup ZML, backed by AI pioneer Yann LeCun, released free software called ZML/LLMD that runs open-source language models across Nvidia, AMD, Google, Intel, and Apple chips, aiming to break Nvidia's lock-in on AI hardware. The tool is an inference server designed to let enterprises mix and match cheaper or more efficient silicon, potentially boosting European chipmakers and reducing AI costs.

A Paris startup wants to loosen Nvidia’s grip on AI, not with a new chip, but with software. ZML has released a free tool that runs open-source models fast across Nvidia, AMD, Google, Apple and Intel silicon alike. Nvidia still rules AI hardware, but its walls keep thinning. ZML, a Paris startup backed by AI pioneer Yann LeCun, has released free software that runs open-source language models across a mix of chips, TechCrunch reports https://techcrunch.com/2026/07/08/hot-french-startup-zml-releases-free-product-to-speed-inference-across-lots-of-ai-chips/ . The list spans five targets: Nvidia, AMD, Google’s TPUs, Intel and Apple. The tool, ZML/LLMD https://zml.ai/llmd/ , is an inference server. Inference means running a trained model to answer prompts, the part of AI that now eats most of the compute. Founder Steeve Morin says the goal is to break the silos that lock users to one vendor, and to squeeze each chip to its top speed. Why a mix of chips matters Cost is the driver. As AI bills climb, enterprises and clouds want the freedom to pick cheaper or less power-hungry silicon for a given job. “The idea is to give people back the power to create their own system,” Morin said. Do that well, and it reads less like a feature and more like a wedge under Nvidia’s moat. It could also lift a wave of novel chipmakers https://thenextweb.com/news/furiosaai-rngd-europe-equinix-lisbon , many of them European. Morin name-checked Axelera, Fractile, Kalray, SiPearl, VSORA and others. Software that treats their chips as first-class, not second-best, gives buyers a real reason to try them. A crowded, costly race Morin does not write off Nvidia, and says ZML has a good relationship with the chip giant. But the field is crowded. The “ inference gold rush https://thenextweb.com/news/baseten-1-5bn-round-13bn-valuation-ai-inference ” has minted rivals like Baseten, recently valued at $13bn, plus the teams behind the open-source projects vLLM and SGLang. All chase the same prize: making AI cheaper to run https://thenextweb.com/news/sail-research-80m-ai-agent-inference . Morin thinks ZML reaches further. “We have reached the point where we are co-designing silicon,” he said. His lean team of 20 has shipped fast, with more releases to come. Why it matters LLMD ships free for now to gather usage, not yet a paid product. Its unusual root is the bigger signal. A tool built to loosen Nvidia’s grip https://thenextweb.com/news/google-nvidia-playbook-tpu-circular-financing-anthropic and to back Europe’s own AI stack https://thenextweb.com/news/tensorx-8-million-sovereign-ai-infrastructure landed from Paris, not Silicon Valley. Morin, who raised $20m from investors including Xavier Niel’s Kima Ventures, put it plainly. “I couldn’t do ZML anywhere but in Paris,” he said. Get the TNW newsletter Get the most important tech news in your inbox each week.