ZML launches free software to accelerate AI inference across diverse chips

French startup ZML has released ZML/LLMD, a free inference server designed to run open source large language models across a wide array of hardware, including chips from Nvidia, AMD, Intel, and Google.

Startups, Policy & World Desk
Updated Nov 12, 2026 2 min read
ZML launches free software to accelerate AI inference across diverse chips

Paris-based startup ZML has launched ZML/LLMD, a new inference server designed to allow open source large language models to run across a variety of hardware platforms. According to TechCrunch, the software supports chips from Nvidia, AMD, Intel, and Google, as well as Apple Metal. By providing a unified software layer, the company aims to dismantle existing hardware silos, allowing enterprises to deploy AI models on diverse chipsets at maximum speed. The product is currently free, as the startup seeks to understand usage patterns before determining a future monetization strategy.

The release addresses the growing industry challenge of inference optimization. As AI integration increases, the process of running models—rather than training them—has become a critical bottleneck. Founder Steeve Morin told TechCrunch that software and architecture barriers currently force organizations into vendor lock-in, limiting their ability to choose hardware based on cost or energy efficiency. ZML/LLMD is intended to provide the flexibility to mix and match chips, potentially aiding novel AI chipmakers in Europe, such as Axelera, Fractile, and SiPearl, by ensuring their hardware can handle complex AI workloads.

While the inference market is crowded with competitors like Baseten, Inferact, and RadixArk, ZML is attempting to differentiate itself through a focus on silicon co-design. Morin noted that the company’s lean team of 20 people has enabled rapid development cycles. Although ZML/LLMD is not open source—unlike the company’s previous framework released in 2024—the decision to offer it for free is a strategic move to lower adoption barriers. Morin stated he prefers to measure effectiveness and growth before introducing revenue models that might hinder the product's initial trajectory.

The startup has garnered significant attention from industry figures, including Turing Award laureate Yann LeCun and leaders from Hugging Face and Docker, who are listed on the company's cap table. This support underscores a broader narrative regarding the maturation of the European AI ecosystem, with Morin asserting that the company’s development was only possible in Paris. Despite the competitive landscape, ZML maintains a relationship with Nvidia, which Morin described as positive, noting that the chip giant is also preparing for the shift toward inference-heavy workloads.

Ultimately, ZML/LLMD represents an attempt to commoditize the hardware layer of AI deployment. By enabling performance across a heterogeneous mix of processors, the startup hopes to give developers the power to create their own systems, rather than relying solely on the dominant market players. Whether this approach can successfully disrupt the established inference market remains to be seen, but the company’s focus on hardware-agnostic software reflects a growing industry trend toward optimizing for cost and efficiency as AI deployment scales.

Sources

  1. 01 Hot French startup ZML releases free product to speed inference across lots of AI chips — TechCrunch — Startups