Archive — US — techcrunch
Hot French startup ZML releases free product to speed inference across lots ofAI chips
French AI startup ZML has launched ZML/LLMD, a free inference-performance software that allows various large language models to run on multiple AI chips, aiming to optimize AI processing and reduce vendor lock-in. Endorsed by Yann LeCun, ZML seeks to empower enterprises with chip flexibility amidst growing competition in the AI sector, particularly against Nvidia.
Key points
- ZML/LLMD is a newly launched inference server for large language models.
- The software runs on various chips: Nvidia, AMD, Google TPU, Apple Metal, Intel Arc.
- Aim: Break vendor silos, enhance chip utilization, and improve speed.
- ZML seeks to offer cost-effective and energy-efficient AI solutions.
- Steeve Morin emphasizes collaboration with emerging AI chipmakers from Europe.
- ZML is funded with $20 million from notable venture firms.
- The product is free initially to learn about user interactions.
- ZML faces competition from companies like Baseten and Inferact.
Why it matters
- AI inference optimization has become crucial as models are increasingly integrated into daily life.
- The rise of AI chipmakers reflects a broader trend in diversifying AI processing capabilities.
- ZML's development underlines a shift towards democratizing AI technology amidst Nvidia's dominance.
- The 'inference gold rush' indicates increased investment and innovation in AI processing technologies.
ZML's innovative approach to AI inference alongside chip flexibility could significantly disrupt the current market landscape, enabling a new wave of cost-efficient AI technologies and fostering growth in European AI startups.