Nvidia’s grip on the AI market remains strong, but competition is emerging from multiple fronts. ZML, a fast-growing French AI startup backed by Turing Award winner Yann LeCun, has launched inference-performance software capable of running various open-source large language models across different chips—including those from Nvidia, AMD, Google’s TPU, Apple Metal, and Intel Arc. As artificial intelligence becomes more embedded in daily life and work, optimizing inference (the process of handling prompts) has grown more critical than training models. Yet, behind the scenes, this optimization often remains inconsistent due to software and architecture barriers that create vendor lock-in, according to ZML’s founder, Morin. Achieving peak performance across diverse chip types is a technical milestone, but it could also disrupt the market amid rising concerns over AI costs. ZML aims to give enterprises and cloud providers the flexibility to mix chips, some of which may be cheaper or more energy-efficient. “The idea is to give people back the power to create their own system and achieve real efficiency gains that allow [AI] to be disseminated,” Morin explained. Such software support could also help novel AI chipmakers, many of which are based in Europe, he noted, citing companies like Axelera, Fractile, Kalray, OLIX, Q. ANT, SiPearl, SpiNNcloud, and VSORA. More than their geographic origins, what matters to Morin is that ZML can collaborate with them on “things that haven’t been done before anywhere in the world.”

EMBED_PLACEHOLDER_0

This doesn’t mean Morin is pessimistic about Nvidia, partly due to its established supply chain. Inference has attracted so much investment that the trend is now called the “inference gold rush.” Consequently, ZML faces competitors like Baseten (recently valued at $13 billion), Inferact (from the creators of the open-source project vLLM), and RadixArk (the commercial entity behind SGLang). Both vLLM and SGLang partially overlap with LLMD, but Morin’s vision for ZML is broader. “We have reached the point where we are co-designing silicon,” he said. He credits ZML’s lean team of 20 people for enabling the Paris-based startup to move quickly, with more releases planned. The small team also benefits from solid funding, thanks to Morin’s track record as VP of engineering at Zenly (acquired by Snapchat for a nine-figure sum in 2017). He raised $20 million from venture firms such as Harry Stebbings’ 20VC, >commit, AALVC, Drysdale Ventures, Xavier Niel’s Kima Ventures, Kindred Capital, LocalGlobe, and Puzzle Ventures. Unlike ZML’s first public project—the inference-focused ML framework released in 2024 and updated in March—ZML/LLMD is not open source. However, it launches as a free product aimed at learning about usage. “I’d rather measure and [then generate revenue] where it is most effective without hindering my growth stupidly because I have been too greedy from the get-go,” Morin said. It’s too early to predict when ZML/LLMD might become a paid product or what its adoption will look like. But the startup’s cap table shows that other founders are paying attention, including Dagger and Docker founder Solomon Hykes, Hugging Face’s Clément Delangue and Julien Chaumond, and LeCun, now with AMI Labs. This also reinforces the idea that Europe’s AI startups can now build locally. “I couldn’t do ZML anywhere but in Paris,” Morin concluded.