The days of Nvidia’s unparalleled market dominance aren’t over, but challengers and choices are arising from all directions.
ZML, a hot French AI startup endorsed by Turing Award winner Yann LeCun, has released inference-performance software that allows a variety of open source large language models to run on a variety of chips — including Nvidia’s, AMD’s, Google’s TPU, Apple Metal, and Intel Arc.
With ZML/LLMD, the newly launched LLM inference server, the company’s ambition is to break existing silos and make different chips available for AI use cases at their maximum available speed, and sometimes faster, ZML founder Steeve Morin told TechCrunch.
As AI becomes integrated into our work and everyday lives, optimizing inference — aka, the processing of prompts — has been outpacing model training in importance, but often feels patchy behind the scenes, with software and architecture barriers that lead to vendor lock-in, Morin said.
The promise of achieving peak performance across a variety of chips is a technological feat, but it could also be a market disruptor, amid mounting fears over AI-related costs.
ZML hopes to provide enterprises and clouds with the option to use a mix of chips, some of which might be less costly or consume less energy. “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 said.
Such a software assist may help novel AI chipmakers, many of which happen to be from Europe, Morin observed, citing Axelera, Fractile, Kalray, OLIX, Q.ANT, SiPearl, SpiNNcloud, and VSORA. But more than their region of origin, what matters to him is that ZML can work with them on “things that haven’t been done before anywhere in the world.”
That doesn’t mean Morin is bearish on Nvidia. He’s not, in part because of its existing supply. He told TechCrunch that ZML has a good relationship with the AI chip giant, which has been gearing up for the rise of inference.
Inference has been an area of such intense investment that the trend has been hailed the “inference gold rush.” So ZML has competition such as Baseten, recently valued at $13 billion; Inferact, from the creators of open source project vLLM; as well as RadixArk, the commercial company behind SGLang.
Both vLLM and SGLang partially compete with LLMD, but Morin’s ambitions for ZML cover a broader spectrum. “We have reached the point where we are co-designing silicon,” he said. He further credited ZML’s lean team of 20 people as the reason why the Paris-based startup has been able to move fast, with more releases in the plans.
It also helped that this small team is well funded for its size. Thanks to his track record as VP of engineering of Zenly, which Snapchat acquired for nine figures in 2017, Morin raised $20 million from venture firms including 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. But it is launching as a free product with the goal of 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 is too early to tell when ZML/LLMD might become a paid product, and what its adoption will look like. But the startup’s cap table confirms that other founders are paying attention, including Dagger and Docker founder Solomon Hykes, Clément Delangue and Julien Chaumond from Hugging Face, as well LeCun, now with AMI Labs. This also builds the case that Europe’s AI startups can now build from home. “I couldn’t do ZML anywhere but in Paris,” Morin said.