French startup ZML challenges NVIDIA's hegemony in the AI chip market
In an era of rapid AI development, NVIDIA's absolute market dominance is facing a serious challenge. French startup ZML has introduced new software backed by Turing Award winner Yann LeCun. This product allows AI models to run at maximum speed on chips from various manufacturers. This was reported by Techcrunch.com reports .
According to TechCrunch, the new inference server called ZML/LLMD is designed to run open-source large language models (LLM) efficiently on various processors, not just NVIDIA, but also AMD, Google TPU, Apple Metal, and Intel Arc. This technology allows companies to avoid vendor lock-in and fully utilize existing hardware resources.
A revolution in the inference process
Currently, using ready-made AI models, i.e., the inference process, is becoming more important than training AI. ZML founder Steeve Morin explains that many companies are dependent on specific chip manufacturers due to software and architectural barriers. The new platform removes these barriers and unlocks the maximum potential of the chips.This solution is particularly relevant given the rising costs and energy consumption of AI. The ZML startup enables enterprises to use combinations of cheaper or more energy-efficient chips. This, in turn, makes the implementation of AI systems more accessible and economically efficient.
New opportunities for European chips
The software introduced by ZML is expected to be a major boost not only for giants but also for emerging chip manufacturers. In his interview, Morin listed several promising companies:- Axelera and Fractile;
- Kalray and OLIX;
- SiPearl and SpiNNcloud;
- Q.ANT and VSORA.
This news is also important for specialists and local IT companies. Open-source models running on various chips allow for the optimization of computing power and cost reduction. This step by the ZML startup signals the beginning of an 'inference gold rush' in the AI market.
Comments 0
…