Yacine and Doria spotlight an asymmetry where US firms turn worldwide AI research into paid frontier systems while many of the original contributors, from Europe and beyond, hit paywalls or API limits that keep those same models out of reach.
Modern AI resulted from research made also by many non-US scientists (Hinton, the French folks, Linnainmaa, many others). The pre-training corpus was produced worldwide with massive code contribution from Europe OSS. What is happening with frontier LLMs is unacceptable.
Hinton’s neural-net breakthroughs, Linnainmaa’s backpropagation work, and datasets such as Common Corpus show clear non-US fingerprints in the foundations, yet the precise share of European code inside proprietary US training runs stays unmeasured in public sources.
Recent curbs on European use of models like Anthropic’s frontier releases illustrate the pattern, though routine claims of contributor-level unaffordability lack detailed public pricing breakdowns beyond general subscription costs.
Many users appreciate mentions of European contributors and the role of global OSS research in modern AI with thanks and excitement, while others accuse training data practices of stealing open source code or argue taxes hinder development.
No Digg Deeper questions have been answered for this story yet.
Remember that with the web there was the same story. The WWW was invented in Europe. The open source stack that created a true revolution of startups was massively based in Europe. Yet all the value was mostly captured in the US. Now it is worse than in the web case.
@hepin1989 I think the solution could actually be exporting ASML machines to China and in exchange do a fundamental AI agreement with Chinese folks. GPU shortage fixed much faster, AI problem too.
And yeah it does feel like a third world asymmetric trade: exporting raw material, hardly affording the manufactured output.
Even more recently and to the point; many RL/synth env sources are EU export.
@antirez Europe should impose export controls on lithography machines—ceasing both sales and maintenance in the future.
@volfpeter @antirez @MistralAI I guess you can check this and this
@antirez Who is stopping Europe from building its own frontier LLMs?
@antirez Don't forget about the ARM arch designed by Acorn.
@antirez Europe has been, and continues to be, very naive when it comes to geopolitics and understanding how the world works
@antirez Americans are essentially of European descent, yet this current generation has turned its back on its own heritage. I think you are absolutely right; back during the Ming Dynasty, China was actually quite advanced, but many books were stolen. A great deal of knowledge was taken
@antirez EU has a lot of talents but its leadership and policies are extremely backwards.
@antirez @hepin1989 美国要求25%的零件就不能对中国卖，所以欧洲永远无法卖给中国光刻机，而且中国的模型都是开源的，欧洲直接用就可以了，但是最好别全买英伟达
@antirez Everything is downstream of tax. Reduce it, ie. stop confiscating, and you will see Europeans develop value for society.
I generally don't agree with "give it to China as well". I'm conscious it's fear and evidence from the past shows it's wrong (the chip ban and the raise of Huaweii ...?). Hard to deal with fear though.
You raised a serious, good-faith point worth examining seriously (with a bit of help from AI-lice ;-)
The global, collaborative roots of AI (European researchers, open-source contributions, etc.) deserves respect and US-centric frontier model dominance is suboptimal.
Export ASML EUV lithography machines to China in exchange for a deep AI collaboration agreement... aims to accelerate progress, ease hardware bottlenecks, and avoid zero-sum decoupling.
I respect this.
Pure technological isolation has real downsides.
But on balance, the risks for the West (and especially Europe) are substantial, and maybe your specific idea looks more like a high-stakes gamble than a clear win.
The proposal has appeal.
- **Hardware reality**: Advanced AI depends on cutting-edge chips. ASML’s EUV machines are the only ones capable of high-volume production at the most advanced nodes (roughly 7nm and below, critical for top-tier GPUs and AI accelerators). More machines globally could increase supply, potentially easing shortages and lowering costs over time.
- **Collaboration upside**: China has scale, talent, manufacturing prowess, and different research approaches. A genuine “fundamental AI agreement” could theoretically speed breakthroughs in efficiency, multimodal models, or applications.
- **Pragmatism**: Strict controls have costs. China is already innovating around restrictions (multi-patterning on older DUV tools to reach ~7nm/5nm-class processes). Full decoupling is expensive and imperfect. These are legitimate arguments, especially from a European perspective frustrated by feeling sidelined between US tech giants and Chinese state-driven scale.
@antirez Treating them like weapons requiring export restrictions is indeed ridiculous, but the US government has form: see the Pretty Good Privacy case in the 90s:
@antirez cc @steeve / @zml_ai :)
@antirez @hepin1989 Da ex $Amat [2004-2009+ 2 sabbatici] ho considerato il ban di $Asml verso la Cina il piu' grande errore industriale europeo degli ultimi 30 anni: un ban senza ottenere nulla in cambio se non la pernacchia di sleepy Joe. $Intc ha cancellato il suo investimento a Magdeburgo, 1/3
@xinguankeli @antirez 他说怎么就怎么吗？要把全球人民吃干抹净。
@antirez Thanks for mentioning Linnainmaa, the name is not seen often enough in these contexts 🙏
Why it’s likely a net negative (and potentially very costly).
EUV lithography isn’t just another industrial tool: it is one of the most complex machines humans have ever built (school-bus sized, hundreds of millions of dollars each, requiring constant specialized maintenance). It is the critical chokepoint for advanced semiconductors. Current policy (US-led, with Dutch/EU alignment) deliberately withholds the most advanced tools from China precisely because they are dual-use: the same chips power consumer AI, data centers, *and* military applications (autonomous systems, surveillance, targeting, cyber, hypersonics, etc.).
China’s military-civil fusion strategy means civilian tech gains directly strengthen state capabilities.
No EUV machines have been sold to China under existing controls. China *has* made credible progress despite restrictions (SMIC producing 7nm-class chips via workarounds; Huawei Ascend chips competitive in some AI workloads). But it remains meaningfully behind on leading-edge processes, yields, and efficiency. Controls have slowed them.
Exporting EUV machines would likely **dramatically accelerate** China’s ability to close the gap in advanced logic chips and AI hardware.
@sergiuprt @antirez @MistralAI Really? Is there a blog post or something about it? I'd love to have a look
