Hot French startup ZML is releasing a free product to speed up predictions on all AI chips

The days of Nvidia’s unrivaled market dominance are not over, but rivals and choices are coming from all directions.

ZML, the hot French AI startup founded by Turing Award winner Yann LeCun, has released inference-performance software that allows a variety of major open source languages ​​to run on a variety of chips — including Nvidia’s, AMD’s, Google’s TPU, Apple’s Metal and Intel’s Arc.

With ZML/LLMD, the newly launched LLM server, the company’s desire is to break down existing silos and make different chips available for AI use cases at their highest available speed, and sometimes faster, ZML founder Steve Morin told TechCrunch.

As AI integrates into our work and daily lives, improving prediction — aka, data processing — has surpassed model training in importance, but often feels invisible behind the scenes, with software and architectural barriers leading to vendor lock-in, Morin said.

The promise of higher performance across all types of chips is a technological feat, but it could also be a market disruptor, amid growing fears about AI-related costs.

ZML hopes to provide enterprises and clouds with the option of using a mix of chips, some of which may be less expensive or use less power. “The idea is to give back to people the power to make their own plan and get the benefits that efficiency allows them [AI] to be distributed,” said Morin.

Such software assistance could help the creators of novel AI incubators, many of them from Europe, Morin noted, citing Axelera, Fractile, Kalray, OLIX, Q.ANT, SiPearl, SpiNNcloud, and VSORA. But more than where they come from, what is important to him is that ZML can work with them on “things that have never been done before anywhere in the world.”

That doesn’t mean Morin is still strong at Nvidia. He is not, in part because of its existing provision. He told TechCrunch that ZML has a good relationship with the AI ​​chip giant, which has been preparing for the rise of thinking.

Speculation has been the site of massive investment, so much so that the trend has been hailed as “forecasting gold.” So ZML has competition like Baseten, which was recently valued at $13 billion; Inferact, from the creators of the open source project vLLM; and RadixArk, the commercial company behind SGlang.

Both vLLM and SGlang compete slightly with LLMD, but Morin’s ambitions for ZML cover a wide spectrum. “We have reached the point where we are integrating silicon,” he said. He also credited ZML’s 20-person team as the reason why the Paris-based startup has been able to move faster, with more releases on plans.

It also helped that this small group was well supported for its size. Thanks to his track record as VP of engineering at Zenly, which Snapchat acquired for nine figures in 2017, Morin raised $20 million from companies including Harry Stebbings’ 20VC, >commit, AALVC, Drysdale Ventures, Xavier Niel’s Kima Ventures, Kindred Capital, LocalGlobe, and Puzzle.

Unlike the original ZML community project, an understanding-oriented ML framework released in 2024 and updated in March, ZML/LLMD is not open source. But it is presented as a free product for the purpose of learning how to use it. “I prefer to measure again [then generate revenue] where it works a lot without stunting my growth because I’ve been very greedy since I left,” said Morin.

It is too early to say whether ZML/LLMD may become a premium product, and what its adoption will look like. But the startup’s cap table is making sure other founders are paying attention, including Dagger and Docker founder Solomon Hykes, Clément Delangue and Julien Chaumond from Hugging Face, and LeCun, now with AMI Labs. This also makes the case that European AI startups can now build at home. “I couldn’t do ZML anywhere but Paris,” said Morin.