As artificial intelligence adoption continues to grow, the competition is shifting beyond training powerful models to making them run faster and more efficiently. French AI startup ZML is entering that race with the launch of ZML/LLMD, a free inference server designed to help AI models perform at high speed across multiple hardware platforms.
Backed by renowned AI researcher Yann LeCun, the Paris-based startup aims to remove hardware limitations that often force businesses to rely on a single chip vendor.
Built for Multiple AI Chips
Unlike many existing AI inference solutions that are optimized for a specific ecosystem, ZML/LLMD supports a broad range of AI hardware, including:
- NVIDIA GPUs
- AMD accelerators
- Google TPUs
- Apple Metal
- Intel Arc graphics
The platform is designed to run open-source large language models efficiently regardless of the underlying hardware, giving enterprises greater flexibility when deploying AI applications.
Solving the AI Inference Challenge
According to ZML founder Steve Morin, AI inference the process of generating responses from trained models is becoming more important than model training itself.
However, today's AI infrastructure is often fragmented, with software limitations and hardware-specific optimizations creating vendor lock-in. ZML wants to eliminate those barriers by allowing organizations to maximize performance across different chips without sacrificing speed.
The company believes this approach can significantly reduce operational costs while improving energy efficiency.
More Choice for Enterprises
One of ZML's biggest goals is to give cloud providers and enterprise customers the freedom to combine different AI processors based on cost, availability, and performance.
Instead of relying solely on premium GPUs, businesses could deploy workloads across multiple hardware vendors while maintaining competitive inference speeds.
This flexibility may become increasingly valuable as AI demand continues to grow and infrastructure expenses rise.
Supporting the Next Generation of AI Hardware
Morin also sees ZML as an important partner for emerging AI chip manufacturers.
The startup is already exploring collaborations with companies such as:
- Axelera AI
- Fractile
- Kalray
- OLIX
- Q.ANT
- SiPearl
- SpiNNcloud
- VSORA
Rather than focusing only on established hardware vendors, ZML wants to help newer chipmakers unlock performance levels that haven't previously been achieved.
NVIDIA Remains an Important Partner
Although ZML is working to reduce hardware dependency, the company is not positioning itself against NVIDIA.
Morin says ZML maintains a positive relationship with the GPU giant, which has also been investing heavily in inference technologies as the AI market evolves.
Instead of replacing NVIDIA, ZML hopes to give organizations additional deployment options alongside existing infrastructure.
Entering a Crowded AI Inference Market
The inference software market has become one of AI's fastest-growing sectors, often described as the industry's new "gold rush."
ZML joins competitors including:
- Baseten
- Inferact
- RadixArk
While open-source projects like vLLM and SGLang overlap in some areas, Morin says ZML's ambitions extend beyond inference software into deeper hardware and silicon co-design.
Small Team, Strong Backing
Despite having a team of only around 20 employees, ZML has secured $20 million in funding.
Founder Steve Morin previously served as VP of Engineering at Zenly before its acquisition by Snapchat in 2017, helping establish credibility with investors.
The startup has attracted backing from leading venture capital firms, including:
- 20VC
- Kima Ventures
- Kindred Capital
- LocalGlobe
- Puzzle Ventures
Its investor list also includes prominent AI leaders such as Solomon Hykes (Docker founder), Clément Delangue, Julien Chaumond of Hugging Face, and Yann LeCun.
Free Today, Monetization Later
Unlike ZML's earlier open-source machine learning framework, ZML/LLMD is proprietary software.
However, the company is making it available free of charge during its early rollout to better understand customer needs and product usage before introducing a commercial pricing model.
Morin says the current priority is adoption rather than immediate revenue, believing rapid growth will create stronger long-term opportunities.
Final Thoughts
With AI inference becoming one of the industry's most important battlegrounds, ZML is positioning itself as a platform that removes hardware barriers and gives organizations more deployment freedom.
By supporting multiple chip architectures, improving efficiency, and offering its software free during launch, the French startup hopes to accelerate enterprise AI adoption while reducing dependence on a single hardware ecosystem.
As demand for faster and more affordable AI infrastructure continues to rise, ZML could become an influential player in the evolving AI inference landscape.
.jpg)
.jpg)
0 Comments