French startup ZML launches free software to speed AI inference across Nvidia, AMD, and rival chips

ZML Releases Free Cross-Chip AI Inference Server

ZML, a French AI startup backed by Turing Award winner Yann LeCun, has launched ZML/LLMD, a free inference-performance software designed to run open-source large language models across multiple AI chips 1

1

. The newly released LLM inference server supports Nvidia, AMD, Google TPU, Intel Arc, and Apple Metal, aiming to break existing silos and deliver maximum available speed across diverse hardware platforms 1

1

.

Founder Steeve Morin told TechCrunch that the software addresses a critical gap in AI infrastructure: as AI becomes embedded in daily work and life, the need to optimize large language model inference has outpaced model training in importance, yet software and architecture barriers often lead to vendor lock-in 1

1

. The tool runs on any major chip, treating each as a first-class target rather than forcing users into a single vendor's ecosystem 2

2

.

Tackling AI Compute Costs Through Hardware Flexibility

The promise of achieving peak performance across various AI chips could prove to be a market disruptor amid mounting concerns over AI compute costs 1

1

. ZML hopes to provide enterprises and cloud providers the option to use a mix of chips, some of which might be less costly or consume less energy. "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," Morin explained 1

1

.

As AI bills climb, this flexibility reads less like a feature and more like a wedge under Nvidia's moat 2

2

. The ability to break vendor lock-in gives buyers real reasons to experiment with alternative silicon, potentially reshaping procurement strategies across the industry. Cost now drives the push for multi-chip strategies, as enterprises seek cheaper or less power-hungry options for specific workloads 2

2

.

Supporting European Chip Innovation and Co-Designing Silicon

The software assist may help novel AI chipmakers, many of which happen to be from Europe, Morin observed, citing Axelera, Fractile, Kalray, OLIX, Q.ANT, SiPearl, SpiNNcloud, and VSORA 1

1

. Software that treats these chips as first-class gives buyers a genuine alternative to established players 2

2

. What matters most to Morin is that ZML can work with these companies on "things that haven't been done before anywhere in the world" 1

1

.

Morin's ambitions extend beyond software optimization. "We have reached the point where we are co-designing silicon," he said, suggesting ZML's influence may shape future hardware development 1

1

2

2

. This positions the Paris-based startup at the intersection of software and hardware innovation, potentially accelerating AI infrastructure democratization.

Navigating the Inference Gold Rush

AI inference has become an area of intense investment, a trend hailed as the "inference gold rush" 1

1

. ZML faces competition from Baseten, recently valued at $13 billion, Inferact from the creators of open-source project vLLM, and RadixArk, the commercial company behind SGLang 1

1

. Both vLLM and SGLang partially compete with LLMD, though Morin's ambitions cover a broader spectrum 1

1

.

Despite the competitive landscape, Morin maintains he's not bearish on Nvidia, partly because of its existing supply, and notes that ZML has a good relationship with the AI chip giant 1

1

2

2

. Nvidia has been gearing up for the rise of inference, recognizing that processing prompts now consumes most AI compute 1

1

.

Funding, Strategy, and Europe's AI Ambitions

ZML's lean team of 20 people has been able to move fast, with more releases planned 1

1

. The small team is well funded for its size: thanks to Morin's track record as VP of engineering at Zenly, which Snapchat acquired for nine figures in 2017, he raised $20 million from venture firms including Harry Stebbings' 20VC, Xavier Niel's Kima Ventures, LocalGlobe, and others 1

1

2

2

.

Unlike ZML's first public project released in 2024 and updated in March, ZML/LLMD is not open source but launches as a free product with the goal of learning about usage 1

1

. "I'd rather measure and [then generate revenue] where it is most effective without hindering my growth stupidly because I have been too greedy from the get-go," Morin said 1

1

. The startup's cap table includes Docker founder Solomon Hykes, Hugging Face's Clément Delangue and Julien Chaumond, and Yann LeCun, now with AMI Labs 1

1

.

The launch builds the case that Europe's AI startups can now build from home. "I couldn't do ZML anywhere but in Paris," Morin said 1

1

2

2

. This signals a shift where critical AI infrastructure innovation emerges from Paris rather than Silicon Valley, potentially reshaping the geographic distribution of AI development 2

2

.