Related Articles
Meta Quietly Enters AI Gaming With Pocket App
AI News
Jul 3, 2026
French AI startup ZML has launched a free product designed to improve how large language models run across different types of AI chips.
The new product, called ZML/LLMD, is an inference server built to help open-source large language models operate efficiently on a wide range of hardware. The platform supports chips from Nvidia, AMD, Google TPU, Apple Metal, and Intel Arc, giving developers and companies more flexibility in how they deploy AI systems.
AI inference, which refers to the process of generating responses from AI models after they have been trained, has become increasingly important as businesses integrate AI tools into daily workflows. As demand grows, companies are looking for ways to reduce costs, improve performance, and avoid being locked into a single chip provider.
ZML founder Steeve Morin said the company wants to break down the barriers that currently separate different chip ecosystems. Many AI deployments rely heavily on specific hardware and software stacks, which can limit choice and increase costs.
With ZML/LLMD, the startup aims to give enterprises and cloud providers the ability to use a mix of chips while still achieving strong performance. This could be especially valuable for companies trying to manage the rising expense of running AI applications at scale.
The product may also benefit newer AI chipmakers that are trying to compete in a market still dominated by Nvidia. By making it easier to run AI models across different hardware platforms, ZML could help smaller or emerging chip companies become more useful to developers and enterprise customers.
The launch comes as the AI inference market becomes one of the most competitive areas in artificial intelligence. While model training has received much of the attention in recent years, inference is now a major focus because it directly affects the cost and speed of real-world AI applications.
ZML faces competition from companies and projects such as Baseten, Inferact, vLLM, RadixArk, and SGLang. However, ZML is positioning itself as more than just another inference software company. Morin has said the startup is also working closely with hardware companies and is moving toward deeper chip-level collaboration.
ZML has attracted attention from notable investors and technology figures. The Paris-based startup has raised $20 million from firms including 20VC, Kima Ventures, Kindred Capital, LocalGlobe, Puzzle Ventures, and others.
The company has also drawn support from well-known figures in the AI and developer ecosystem, including Yann LeCun, Solomon Hykes, Clément Delangue, and Julien Chaumond.
Morin previously served as vice president of engineering at Zenly, the social mapping app acquired by Snap in 2017. His background has helped ZML gain early credibility despite having a relatively small team of around 20 people.
Unlike ZML’s earlier public machine learning framework, ZML/LLMD is not open source. However, the company is making the product free at launch in order to understand how developers and businesses use it.
The move suggests that ZML is prioritizing adoption before deciding how to monetize the product. By lowering the barrier to entry, the startup may be able to gather valuable feedback and build a user base before introducing paid features or enterprise pricing.
ZML’s launch also reflects the growing strength of Europe’s AI startup ecosystem. The company is based in Paris, a city that has become increasingly important in the AI sector thanks to strong technical talent, investor interest, and a growing number of ambitious startups.
As AI infrastructure becomes more important, ZML is betting that companies will want more freedom in how they run models. If ZML/LLMD delivers strong performance across many types of chips, it could become a useful tool for businesses looking to build faster, cheaper, and more flexible AI systems.
Share your thoughts about this article.
Be the first to post a comment!