Founder Steeve Morin argues that software and architecture barriers have forced enterprises into inefficient silos, making inference optimization more critical than model training. While Nvidia remains a key player, ZML’s platform enables companies to mix and match hardware, potentially utilizing more cost-effective or energy-efficient alternatives. Morin believes this flexibility is essential for the next wave of AI dissemination, particularly for emerging European chipmakers like Axelera and SiPearl, which are eager to integrate with novel software stacks.
In section Startups & Technology
French startup ZML launches free tool to break Nvidia chip monopoly
Paris-based ZML is challenging the dominance of Nvidia by releasing LLMD, a new inference server designed to run open-source language models across diverse hardware. By supporting chips from AMD, Google, Intel, and Apple, the startup aims to slash AI costs and bypass the vendor lock-in currently stifling the industry.
The startup faces a crowded field, competing against well-funded rivals like Baseten and projects like vLLM and SGLang. However, Morin claims ZML’s advantage lies in its ability to co-design silicon, a feat made possible by a lean team of 20. Backed by $20 million from investors including Kima Ventures and LocalGlobe, the company is prioritizing adoption over immediate revenue. LLMD is currently free to use, a strategic choice Morin hopes will provide the usage data necessary to build a sustainable business model without stifling growth. High-profile supporters, including Hugging Face executives and Yann LeCun, suggest the project has already gained significant traction within the developer community.
Comments (0)
No comments yet. Be the first!