Twenty people in Paris threw down the gauntlet to the most expensive monopoly in the tech world. The French startup ZML, vouched for by Turing Award winner Yann LeCun, released a free server for running AI models called LLMD - software that lets open language models run at maximum speed on almost any chip: Nvidia, AMD, Google TPU, Apple Metal, Intel Arc.

Why does this matter? Because the current practice is the opposite: you buy chips from one manufacturer, get tied to its software, and for years afterward pay whatever it tells you. Founder Steve Morin argues that this very lock-in to a single supplier is the industry's most expensive habit - and that customers should be able to mix chips, cheaper or more efficient, without a performance penalty. At a time when AI bills grow faster than the revenue it brings, the argument sells itself.

Morin is no unknown: his previous startup Zenly was bought by Snapchat in 2017 for a nine-figure sum, and for ZML he raised 20 million dollars from investors including Xavier Niel's fund and London's 20VC. The competition is serious - Baseten was recently valued at 13 billion dollars - but the Paris team is playing a different card: neutrality between chips, instead of yet another closed platform. LLMD is not open source, but it's free, with simple logic: growth first, monetization later.

There's a European note to the story too. Morin openly says software like this could open the way for new European chipmakers - from SiPearl to Axelera - which without it have no chance of breaking into a market tailored to Nvidia's measure. Can 20 engineers from Paris really move the market? So far no one has managed it. But every monopoly looks eternal - until the day someone shows the wall was made of paper.