ZML, a French AI startup endorsed by the renowned Turing Award winner Yann LeCun, has recently released an inference-performance software that paves the way for a new era of AI chip diversity. This software allows various open-source large language models to run on a wide range of chips, including Nvidia's, AMD's, Google's TPU, Apple Metal, and Intel Arc. The company's ambition is to break existing silos and make different chips available for AI use cases at their maximum speed, sometimes even faster. This is a significant development, as AI integration into our daily lives and work has made inference optimization more crucial than model training. However, the software and architecture barriers often lead to vendor lock-in, which is where ZML comes in. By providing enterprises and clouds with the option to use a mix of chips, some of which might be less costly or consume less energy, ZML hopes to democratize AI and achieve real efficiency gains. This is particularly exciting for novel AI chipmakers, many of which are based in Europe, as it opens up new possibilities for innovation and collaboration. Personally, I think this is a game-changer for the AI industry, as it challenges the dominance of a few major players and encourages a more diverse and competitive market. What makes this particularly fascinating is the potential for Europe to become a hub for AI innovation, as it has been historically underrepresented in the field. However, it's important to note that ZML is not bearish on Nvidia, as it has a good relationship with the AI chip giant and has been gearing up for the rise of inference. In my opinion, this is a strategic move by ZML, as it allows the company to leverage Nvidia's existing supply while also challenging its dominance. One thing that immediately stands out is the fact that ZML has a lean team of 20 people, which has allowed them to move fast and release more products. This is a testament to the power of a small, well-funded team, and it's interesting to see how this compares to other AI startups. What many people don't realize is that ZML's first public project, the inference-focused ML framework, was released in 2024 and updated in March. However, ZML/LLMD is not open-source, but rather a free product with the goal of learning about usage. This is a smart move, as it allows ZML to measure and generate revenue where it is most effective without hindering its growth. If you take a step back and think about it, this is a common strategy for startups, as it allows them to learn and adapt quickly while also generating revenue. This raises a deeper question: how will the AI industry evolve in the coming years, and what role will startups like ZML play in shaping it? A detail that I find especially interesting is the fact that ZML has attracted attention from other founders, including Dagger and Docker founder Solomon Hykes, Clément Delangue and Julien Chaumond from Hugging Face, and LeCun himself. This builds the case that Europe's AI startups can now build from home, and it's exciting to see the potential for collaboration and innovation in the region. In conclusion, ZML's release of the inference-performance software is a significant development for the AI industry, and it has the potential to democratize AI and challenge the dominance of a few major players. It's a testament to the power of innovation and collaboration, and it's exciting to see what the future holds for startups like ZML.
