ZML, a French AI startup, has released free inference software that lets organizations run AI language models across multiple chip types—Nvidia, AMD, Google, Apple, Intel—instead of being locked into a single vendor. The software is designed to lower AI costs and energy consumption by giving enterprises flexibility in chip choice, and could help smaller European chipmakers compete with Nvidia's market dominance.
Summaries like this, in your inbox every morning.
Sign up free →What happened
ZML, a Paris-based AI startup backed by Turing Award winner Yann LeCun, has launched ZML/LLMD, a free inference server that allows open-source large language models to run on multiple chip types—including Nvidia, AMD, Google TPU, Apple Metal, and Intel Arc. The startup raised $20 million（約32億円） from venture firms including 20VC, >commit, AALVC, Drysdale Ventures, Kima Ventures, Kindred Capital, LocalGlobe, and Puzzle Ventures.
Why it matters
The software aims to break vendor lock-in and reduce AI costs by letting enterprises and cloud providers mix chips—some of which may be cheaper or more energy-efficient. This matters for businesses grappling with rising inference costs, since optimizing how AI systems process prompts has become more important than model training. The move could help smaller European chipmakers gain ground against Nvidia's dominance.
What to watch
ZML/LLMD is launching as a free product with no announced timeline for paid pricing; the company plans to learn about usage patterns before monetizing. The startup has a lean team of 20 people and founder Steeve Morin has flagged more releases to come.
Inference—the step where an AI system produces an answer to a prompt—has become the dominant cost and performance bottleneck in AI deployment, even surpassing model training in importance. ZML is entering a crowded market that includes competitors like Baseten (valued at $13 billion（約2.1兆円）), Inferact (from the creators of open-source project vLLM), and RadixArk (the commercial entity behind SGLang). However, Morin's ambition extends beyond competing with vLLM and SGLang on narrow tasks; he frames ZML's mission as co-designing silicon with chipmakers and giving enterprises genuine optionality in their hardware mix.
The startup's ability to move quickly despite a small team of 20 people reflects both its founder's track record and healthy venture backing—a validation that angel investors and prominent founders see value in addressing vendor lock-in. Notably, the cap table includes Docker and Dagger founder Solomon Hykes, executives from Hugging Face (Clément Delangue and Julien Chaumond), and Yann LeCun, suggesting confidence that Europe can develop competitive AI infrastructure from home. By launching as a free product rather than immediately charging, ZML is prioritizing market adoption and understanding usage before monetizing—a deliberate strategy Morin articulated to avoid alienating early users.
No comments yet. Be the first to share your thoughts!
Log in to join the discussion
Get curated AI news from 200+ sources delivered daily to your inbox. Free to use.
Get Started FreeFree · takes 30 seconds · unsubscribe anytime
1 minute a day. The AI essentials.
200+ sources · Email / LINE / Slack
