This Hot French Startup Just Released a Free Tool to Break Nvidia’s Hardware Lock-In

Hot French Startup ZML Releases Free Inference Engine to Break Nvidia's Chip Lock-In:

ZML/LLMD lets open-source LLMs run at peak speed across Nvidia, AMD, Google TPU, Apple Metal, and Intel Arc — and the Paris startup is giving it away for free, for now.

$20M: Seed funding raised, led by backers including 20VC and Kima Ventures

5+: Chip architectures supported: Nvidia, AMD, Google TPU, Apple Metal, Intel Arc

20: People on ZML's Paris-based team

Paris Startup Backed by Yann LeCun Releases Free Peak-Performance AI Inference Server:

1: Breaking Nvidia's Silo, One Chip at a Time:

Nvidia's dominance in AI hardware has never been in serious doubt — but the field around it is getting crowded. Paris-based startup ZML, backed publicly by Turing Award winner Yann LeCun, has released ZML/LLMD, an inference-performance server designed to run open-source large language models at peak speed across a wide range of chips, including Nvidia, AMD, Google's TPU, Apple Metal, and Intel Arc. The goal, according to founder Steeve Morin, is to break the software and architecture barriers that currently lock enterprises into a single hardware vendor.

Inference, not training, is now the priority. As AI moves from experimentation into daily workflows, the processing of prompts, or inference, has been outpacing model training in importance. Morin argues that inference infrastructure has remained fragmented behind the scenes, forcing companies into unnecessary trade-offs between performance and vendor flexibility.

The idea is to give people back the power to create their own system and achieve real efficiency gains that allow AI to be disseminated. — Steeve Morin, Founder, ZML

2: A Boost for Europe's Emerging Chipmakers:

Cross-chip compatibility could be a lifeline for challenger silicon. By making it easier for enterprises to mix and match hardware, ZML's approach may particularly benefit newer, less-established AI chipmakers competing for attention against Nvidia's dominant supply chain. Morin pointed to a number of chipmakers ZML is working with, several of which happen to be European, including Axelera, Fractile, Kalray, OLIX, Q.ANT, SiPearl, SpiNNcloud, and VSORA.

Region matters less than ambition, Morin said. What he cares about most is doing engineering work that hasn't been attempted anywhere else, describing ZML's collaboration with chipmakers as reaching the point of co-designing silicon itself, rather than simply optimizing software for existing hardware.

3: Free for Now, in a Crowded Inference Market:

ZML/LLMD is launching free, unlike ZML's earlier open-source framework. The company's first public project, an inference-focused ML framework released in 2024 and updated this past March, was fully open source. ZML/LLMD is not, but Morin chose to release it for free anyway, prioritizing usage data and adoption over near-term revenue.

He said he'd rather measure where monetization is most effective than slow early growth by charging too soon.

Competition in inference is intense and well-funded. ZML is entering what's been called the inference gold rush, competing against Baseten, recently valued at $13 billion, Inferact, the commercial venture behind vLLM, and RadixArk, the company built around SGLang. Both vLLM and SGLang partially overlap with what LLMD does, but Morin frames ZML's ambitions as broader, spanning software and silicon co-design rather than software alone.

4: Lean Team, Strong Cap Table, Paris Roots:

A 20-person team is fueling a fast release cadence. Morin credits ZML's small size as a reason the company has been able to move quickly, with additional releases already planned. The startup raised $20 million from venture firms including Harry Stebbings' 20VC, >commit, AALVC, Drysdale Ventures, Xavier Niel's Kima Ventures, Kindred Capital, LocalGlobe, and Puzzle Ventures — funding built partly on Morin's track record as VP of engineering at Zenly, acquired by Snapchat for nine figures in 2017.

Notable founders are backing the company too. ZML's cap table includes Dagger and Docker founder Solomon Hykes, Hugging Face co-founders Clément Delangue and Julien Chaumond, and Yann LeCun, who now leads AMI Labs. Morin sees the backing as validation that Europe's AI startups can build ambitious infrastructure without relocating.

I couldn't do ZML anywhere but in Paris. — Steeve Morin, Founder, ZML

The Bigger Picture:Infrastructure Flexibility Is the New Competitive Edge:

ZML's bet is that enterprises no longer want to be locked into a single chip vendor's roadmap or pricing. That same principle — flexibility without sacrificing performance — is what drives Otherworlds AI's approach to enterprise AI deployment.

Whether it's optimizing inference costs or orchestrating agents across your existing tech stack, our Agent+ Business AI Platform, powered by Google Opal, gives your team the freedom to build without lock-in.

Ready to see what a flexible, cost-efficient AI stack looks like for your business? Explore Agent+ starting at $297/month at otherworldsai.com.