French AI startup ZML launches free multi-chip LLM inference server
Original: Hot French startup ZML releases free product to speed inference across lots of AI chips
Why This Matters
ZML's multi-chip inference software addresses rising AI infrastructure costs and hardware fragmentation across the industry.
French AI startup ZML, backed by Yann LeCun and $20M in funding, released ZML/LLMD, a free open LLM inference server supporting Nvidia, AMD, Google TPU, Apple Metal, and Intel Arc chips, aiming to eliminate vendor lock-in and maximize inference performance across hardware.
ZML, a Paris-based AI startup founded by former Zenly VP of Engineering Steeve Morin, has launched ZML/LLMD, a free LLM inference server designed to run open-source large language models across a wide range of AI chips — including Nvidia, AMD, Google TPU, Apple Metal, and Intel Arc. The product targets the growing inference market, which Morin says has overtaken model training in strategic importance but remains fragmented due to software and hardware silos that create vendor lock-in. ZML's goal is to enable enterprises and cloud providers to mix chip vendors, potentially reducing costs and energy consumption. The startup also highlighted partnerships with European chipmakers such as Axelera, Fractile, Kalray, and SiPearl, noting it is now 'co-designing silicon' with some partners. ZML competes partially with vLLM-based Inferact and SGLang-backed RadixArk, as well as Baseten, recently valued at $13 billion. The 20-person team has raised $20 million from investors including 20VC, Kima Ventures, LocalGlobe, and Kindred Capital. Morin previously led engineering at Zenly, acquired by Snapchat for nine figures in 2017. ZML/LLMD is the company's second public release, with more planned.