Paris‑based AI startup ZML announced today that its new LLMD inference server is available for free, letting developers run open‑source large language models on a wide array of hardware—including Nvidia GPUs, AMD GPUs, Google’s TPUs, Apple’s Metal framework and Intel’s Arc GPUs. The company says the platform eliminates the silos that force customers into single‑vendor ecosystems, delivering peak performance—or even faster speeds—across the board.

Founder and CEO Steeve Morin, a former VP of engineering at Zenly (acquired by Snapchat in 2017), told TechCrunch that optimizing inference has become more critical than model training as AI embeds itself in everyday workflows. “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 said.

ZML’s backing includes AI luminary Yann LeCun, who serves as an advisor, and a $20 million Series A round led by 20VC, AALVC, Drysdale Ventures, Kima Ventures, Kindred Capital, LocalGlobe, Puzzle Ventures and others. The startup’s lean team of about 20 engineers has been able to move quickly, Morin added, citing the firm’s Paris headquarters as a strategic advantage.

LLMD is not open source, but the free launch is designed to collect real‑world usage data. Morin explained that the approach lets ZML measure demand before deciding on a pricing model, avoiding premature monetization that could hinder adoption. “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,” he said.

The product arrives amid what industry observers call an “inference gold rush.” Competitors such as Baseten, Inferact (the team behind vLLM) and RadixArk (the commercial arm of SGLang) are also targeting the fast‑growing market for efficient prompt processing. While vLLM and SGLang overlap with LLMD on functionality, ZML’s broader hardware compatibility differentiates it, especially for enterprises looking to mix and match chips for cost or energy considerations.

Morin emphasized that ZML maintains a good relationship with Nvidia, the current market leader, even as the startup works to broaden chip options. “We have reached the point where we are co‑designing silicon,” he noted, hinting at future collaborations that could include emerging European chip makers such as Axelera, Fractile, Kalray, OLIX, Q.ANT, SiPearl, SpiNNcloud and VSORA.

Analysts see the move as a potential disruptor for the AI hardware market, where vendor lock‑in has driven pricing pressures. By enabling the same model to run on less expensive or more energy‑efficient processors, ZML could help enterprises lower operating costs while maintaining performance.

Looking ahead, Morin said the team has more releases planned and remains confident that Paris is the right place to build Europe’s next generation of AI infrastructure. “I couldn’t do ZML anywhere but in Paris,” he concluded.

Dieser Artikel wurde mit Unterstützung von KI verfasst.
News Factory APP - agentische News für besseres SEO & AEO.