Paris-based startup ZML, supported by AI luminary Yann LeCun, has unveiled ZML/LLMD, a free inference performance software that enables efficient AI model deployment across multiple chip platforms including Nvidia, AMD, Google TPU, Apple Metal, and Intel Arc.
- ZML/LLMD supports numerous AI chips for faster inference
- Free launch aims to disrupt vendor lock-in and cut costs
- Paris startup benefits from prestigious backing and $20M funding
What happened
ZML, a French AI startup founded by Steeve Morin, released a new product named ZML/LLMD, an inference server designed to accelerate large language model processing across various AI hardware. This software supports chips from leading manufacturers such as Nvidia, AMD, Google, Apple, and Intel, aiming to maximize or even exceed the chips' native performance.
Unlike traditional solutions that often lock users into specific vendor ecosystems, ZML/LLMD enables a flexible approach where enterprises and cloud providers can mix and match hardware based on cost, energy usage, or availability. While the software is not open source, it is offered for free initially to build a user base and gather usage insights for future monetization.
Why it matters
AI inference—the real-time processing of prompts and data through trained models—has become a critical bottleneck for scaling AI applications affordably and sustainably. ZML’s product addresses the common challenges of vendor lock-in and suboptimal utilization of hardware by allowing cross-platform compatibility and optimization.
This innovation could disrupt the current dominance of major chip players by empowering emerging and alternative AI chip manufacturers, especially notable in Europe. This diversification may lower operational costs and energy consumption for AI workloads, accelerating the wider adoption of AI technologies in business and everyday life.
What to watch next
ZML’s approach to co-designing software with silicon manufacturers and its strategy of launching as a free product to foster adoption are significant to watch. The market will be observing how quickly enterprises and cloud providers embrace this multi-chip compatibility and how effectively ZML can convert users into paying customers in the future.
Competition remains intense with other companies developing inference software and frameworks. ZML’s ability to maintain rapid innovation with its lean team, supported by notable investors and AI figures, positions it well for further releases and potentially pivotal contributions in AI hardware-software integration.