AI Acceleration Without NVIDIA: A Free Tool for Any Chip
French startup ZML, backed by Turing Award winner Yann LeCun, has released ZML/LLMD, a free tool that speeds up language model inference on a wide range of AI chips. The software cuts the operating cost of running AI in production and reduces lock-in to a single vendor, above all NVIDIA. It is most relevant for teams running AMD, Intel and other accelerators.
AI-processed from TechCrunch; edited by Hamidun News
On July 8, 2026, French startup ZML released a free tool for accelerating AI on any chip: ZML/LLMD, a software layer that speeds up AI model inference across a wide range of hardware accelerators. The project is backed by Yann LeCun, a Turing Award winner and Meta's chief AI scientist.
What ZML/LLMD does
ZML/LLMD is a software layer that lets teams run language models faster and more cheaply, regardless of chip vendor. Cross-platform support is its main advantage: today's market for inference tooling is heavily skewed toward NVIDIA. Popular libraries such as TensorRT-LLM and CUDA-oriented builds of vLLM are optimized by default for the "green" vendor's GPUs. Teams on AMD, Intel Gaudi, AWS Trainium and other accelerators must either accept lower performance or spend engineering resources adapting the stack themselves.
ZML/LLMD positions itself as a single tool that works equally well on different hardware and lowers the cost of running AI products in production.
- Product: ZML/LLMD, an optimization layer for LLM inference
- License: free to use
- Coverage: a wide range of AI chips from different vendors
- Goal: reduce the operating cost of inference in production
Why LeCun's name changes how the startup is perceived
Yann LeCun is one of the three 2018 Turing Award winners (together with Geoffrey Hinton and Yoshua Bengio) and has been Meta's chief AI scientist since 2018. His work on convolutional neural networks became the foundation of modern computer vision and laid the groundwork for the current wave of deep learning.
LeCun's public backing is a signal to the market and to investors: ZML has at least caught the attention of top-tier researchers. For developers who have not yet heard of the startup, having such a name among its supporters noticeably lowers the barrier of distrust toward a new tool.
Context: why inference costs more than training
For most AI products in mature production, inference spending, literally the cost of answering millions of requests, has long exceeded training spending. By industry estimates, the cost split is shifting toward 10 to 90 in favor of inference. That is why optimizing the speed and cost of running models is one of the most competitive niches in AI infrastructure in 2026.
ZML/LLMD's cross-chip compatibility could give companies leverage in negotiations with vendors and reduce the risk of dependence on a single supplier.
What this means
ZML/LLMD targets a real pain point: inference operating costs grow with the scale of AI products. A free distribution model lowers the barrier to entry and may attract a broad engineering community. How well the tool beats specialized solutions on specific hardware will become clear from independent benchmarks as its user base grows.
*Meta has been designated an extremist organization and is banned in Russia.
- Nov 1, 2026Local AI models agreed to pay scammers, while simple rules blocked all 42 schemes
- Oct 31, 2026Open-source LLM gateway KTMS 1.0 plugs your own GPUs into the cloud and survives outages
- Oct 31, 2026A 22B-Parameter Video Model Runs on an RTX 3080 with 10 GB of VRAM
- Oct 29, 2026T-Tech Introduces LIMe, a Method That Trains Language Models With 15% Less Compute
Key Questions Answered
What ZML/LLMD does?
ZML/LLMD is a software layer that lets teams run language models faster and more cheaply, regardless of chip vendor. Cross-platform support is its main advantage: today's market for inference tooling is heavily skewed toward NVIDIA. Popular libraries such as TensorRT-LLM and CUDA-oriented builds of vLLM are optimized by default for the "green" vendor's GPUs. Teams on AMD, Intel Gaudi, AWS Trainium and other accelerators must either accept lower performance or spend engineering resources adapting the stack themselves. ZML/LLMD positions itself as a single tool that works equ
Why LeCun's name changes how the startup is perceived?
Yann LeCun is one of the three 2018 Turing Award winners (together with Geoffrey Hinton and Yoshua Bengio) and has been Meta's chief AI scientist since 2018. His work on convolutional neural networks became the foundation of modern computer vision and laid the groundwork for the current wave of deep learning. LeCun's public backing is a signal to the market and to investors: ZML has at least caught the attention of top-tier researchers. For developers who have not yet heard of the startup, having such a name among its supporters noticeably lowers the barrier of distrust tow
What are the key details on Context: why inference costs more than training?
For most AI products in mature production, inference spending, literally the cost of answering millions of requests, has long exceeded training spending. By industry estimates, the cost split is shifting toward 10 to 90 in favor of inference. That is why optimizing the speed and cost of running models is one of the most competitive niches in AI infrastructure in 2026. ZML/LLMD's cross-chip compatibility could give companies leverage in negotiations with vendors and reduce the risk of dependence on a single supplier.
Ship AI Faster with Founders, Engineers & Weekly Architecture Reviews
Inside Vibe Club (hamidun.com/club) — weekly live masterminds with Zhemal Khamidun, Tech Radar, full LMS access, and 370+ production Claude Code skills.
The AI world, distilled — once a week
Seven stories that actually mattered, hand-picked. No noise, no reposts, no press releases.
Done! Check your inbox for a confirmation.