Has Europe's Great Hope For AI Missed Its Moment? (ft.com) 39
France's Mistral AI is facing mounting pressure over its future as an independent European AI champion, as competition intensifies from U.S. tech giants and China's emerging players. The Paris-based startup, valued at $6.5 billion and backed by Microsoft and Nvidia, has struggled to keep pace with larger rivals despite delivering advanced AI models with a fraction of their resources.
The pressure increased this week after China's DeepSeek released a cutting-edge open-source model that challenged Mistral's efficiency-focused strategy. Mistral CEO Arthur Mensch dismissed speculation about selling to Big Tech companies, saying the firm hopes to go public eventually. However, one investor told the Financial Times that "they need to sell themselves."
The stakes are high for Europe's tech ambitions. Mistral remains the region's only significant player in large language models, the technology behind ChatGPT, after Germany's Aleph Alpha pivoted away from the field last year. The company has won customers including France's defense ministry and BNP Paribas, but controls just 5% of the enterprise AI market compared to OpenAI's dominant share.
The pressure increased this week after China's DeepSeek released a cutting-edge open-source model that challenged Mistral's efficiency-focused strategy. Mistral CEO Arthur Mensch dismissed speculation about selling to Big Tech companies, saying the firm hopes to go public eventually. However, one investor told the Financial Times that "they need to sell themselves."
The stakes are high for Europe's tech ambitions. Mistral remains the region's only significant player in large language models, the technology behind ChatGPT, after Germany's Aleph Alpha pivoted away from the field last year. The company has won customers including France's defense ministry and BNP Paribas, but controls just 5% of the enterprise AI market compared to OpenAI's dominant share.
ASML (Score:4, Interesting)
Seems like Europe is content selling the golddiggers the tools they need while the miners are burning through endless money to train models that they then have to give away for dimes on the dollar to be in any way competitive.
Mistral models do quite well in certain niches like Pixtral, but I don't think they are much interested in entering the billion dollar model race.
Of course with R1 showing it can be done much cheaper, I guess we will see a ton of companies flooding the market with chatgpt level models. I'm sure some european entries will be there too.
Re:ASML (Score:4, Informative)
Seems like Europe is content selling the golddiggers the tools they need ...
That's true, the US tech-barons with planet sized egos who laugh at Europe for supposedly having no tech industry build their entire empires on chips made with Dutch lithography machines.
Re: (Score:3)
And then they discover that all they get from AI are polished turds.
Re:ASML (Score:5, Interesting)
And then they discover that all they get from AI are polished turds.
I'm not so sure about that. It's just that from what I can tell at the moment DeepSeek seems far more interesting than the US bloated AIs the US tech-barons are currently creating by throwing money at the problem. If DeepSeek pans out and if it isn't a 'CCP psyop' created specially to embarrass Trump and the US's mercurial business geniuses (LoL, echoes of COVID), then DeepSeek is a very impressive feat of mathematics and engineering. The polished turds thing is just how the US works. They inflate these bubbles, obscene amounts of money get thrown at the bubble, everybody drinks the kool-aid dances to the buzz and then everything collapses and eventually people pull useful stuff out of the ruins. The Chinese get hit by sanctions, they have limited GPU supplies so they throw maths, engineering and science at the problem cut off a whole lot of bloat and end up with a massively distilled product. Sometimes I think the world would be better off just isolating the US from the rest of the planet's economy every 20 or 30 years when the Americans descend into one of their bubble-mania sessions. The way I see the US markets and the way they work is something like this: https://si.wsj.net/public/reso... [wsj.net]
Re: (Score:1, Insightful)
Re: (Score:2)
Remember: America innovates, China duplicates and Europe regulates.
As one liners go, that's true enough. But although I've never set foot in Europe, their innovations are absolutely essential to my tech career. They just don't get Wall Street attention because the regulation keeps the billionaires away.
China? Fully worthless and more trouble than its worth.
Re: (Score:2)
Seems like Europe is content selling the golddiggers the tools they need while the miners are burning through endless money to train models that they then have to give away for dimes on the dollar to be in any way competitive.
Except that Europe (well basically only ASML) is selling nails instead of pickaxes. The revenue and profits numbers for ASML, TSMC, and Nvidia (as you go up the vertical product ladder) are
ASML: revenue=$26B profit=$8
TSMC: revenue=$96B profit=$39
NVDA: revenue=$113B profit=$71B
And that's just the trailing 12 months. Nvidia's number are still exploding (despite the DeepSeek news) and will likely be more than revenue=$140B profit=$85B going forward.
Does it matter? (Score:5, Interesting)
Even assuming LLMs aren't abused as much as they are today and are used for legitimate purposes, does it matter if Europe developed it when such a significant amount of the work, including DeepSeek, is being done under open source licenses?
Indeed, maybe they dodged a bullet, given that if DeepSeek is what it claims to be (and as of now there doesn't appear to be any evidence it isn't) then all the investors who shoveled billions into OpenAI or their own LLM departments like Google are about to lose their shirts.
Can Europe compete? (Score:5, Interesting)
The more general question for Europe is whether they can compete in tech. It's not just LLM AI, but also hardware, software infrastructure (e.g. operating systems), applications, etc. There'll be lots of potential in the application of AI beyond the current use cases for LLMs (which too often sound to me like 'software for lazy people" :-) ) I think machine learning for specific tasks, integrated into systems, will be the real Big Thing for AI.
This will require education, training, policy to support startups, culture, etc, etc.
What they will NOT be successful at doing is taking the "tariff approach" by taxing/punishing successful (American) tech companies in the hope that some European company can produce competing products. (At best, tariffs can produce markets for manufacturing, but software and software-intensive systems are not "manufacturing", as compared to automobiles.)
Re:Can Europe compete? (Score:4, Insightful)
No matter how smart you are, working 80 hours a week makes you dumb.
Anyway, Europe may be better off waiting and seeing on this one. The tech will probably become a commodity pretty quickly, and the most important thing will be to have strong protections for citizens.
Re: (Score:2)
Europe tends to compete on quality. AI developed here will be higher quality, more reliable, more respectful of privacy and copyright.
Right now if you want to use AI there is a good chance that you will end up breaking someone's copyright. Likely it was trained on stolen data, and can be made to spit it out word-for-word with a bit of coaxing. It will probably tell your customers that Hitler did nothing wrong too.
There was a case recently where an airline chatbot gave a customer a refund, but the airline sa
Re: (Score:2)
I would agree that 'quality' is a good positioning for Europe. But it will also take business skills and a favorable regulatory environment. Part of that should be the ability to sustain a start-up/new technology without the "quarterly profits/get there first and then make it work" approach that often characterizes US tech. ("Rapid Unscheduled Disassembly"?)
A friend ran the US subsidiary of a French company. Together, they had very good technology. The US subsidiary was profitable, but the French parent
Re:Does it matter? (Score:5, Insightful)
Europe won't make an AI that rivals OpenAI or Deepseek directly. They'll make one that works well for Europe.
For probably 100 years, if you gave a problem to a European and an American, the American would put the biggest, noisiest, most gas guzzling muscle-car engine on it, and would solve the problem with brute force and ignorance. The European would make a far more modest, far less aggressive solution which would likely be very calm and subtle about whatever it does.
Historically, Europeans probably haven't been as quick as the Americans to get to a solution, and so commercially Europe is always seen as being a poor performer. Often times, the big aggressive American solution is seen as the "winner" and so further work on the subject comes to an end in Europe. Europe's "give up" attitude is something of a limiting factor, but it's approach to make things differently to the Americans most definitely is not. Indeed, there are noises in the tech industry at least, that maybe there's a "better" way to do a number of things rather than the Tech Bro' stuff we get from Silicon Valley.
If Europe does anything, it'll be a solution that works in Europe. It won't be loud and brash, it won't be the biggest, or the fastest or the most expensive. It'll be "just right" for something or other. Whether that makes it a commercial success or not remains to be seen. If Europe actually holds the course and continues to invest in its own solutions, then they'll likely do just fine. If they stick to type, they'll cut funding and essentially give up at the first hurdle, and so unsurprisingly, won't achieve very much.
Re: (Score:2)
Many European nations avoided the mistakes made by the US not rushing into everything. I suspect this AI thing will be another as it really seems to be a solution looking for a problem.
Another old joke about Europe:
Heave
Re: (Score:2)
Even assuming LLMs aren't abused as much as they are today and are used for legitimate purposes, does it matter if Europe developed it when such a significant amount of the work, including DeepSeek, is being done under open source licenses?
Indeed, maybe they dodged a bullet, given that if DeepSeek is what it claims to be (and as of now there doesn't appear to be any evidence it isn't) then all the investors who shoveled billions into OpenAI or their own LLM departments like Google are about to lose their shirts.
If AI models are open sourced and can be trained in a dispersed manner rather than by obscenely expensive super computers only mega-corporations can afford then literally nobody will lead AI anymore than anybody leads a FOSS project. The most powerful AIs would be created in the distributed FOSS manner by many volunteers and trained by volunteers running their own training nodes and that will result in another huge democratization of the tech industry. This has happened before with Linux and all the user-la
Re: (Score:2)
Even assuming LLMs aren't abused as much as they are today and are used for legitimate purposes, does it matter if Europe developed it when such a significant amount of the work, including DeepSeek, is being done under open source licenses?
Indeed, maybe they dodged a bullet, given that if DeepSeek is what it claims to be (and as of now there doesn't appear to be any evidence it isn't) then all the investors who shoveled billions into OpenAI or their own LLM departments like Google are about to lose their shirts.
If AI models are open sourced and can be trained in a dispersed manner rather than by obscenely expensive super computers only mega-corporations can afford then literally nobody will lead AI anymore than anybody leads a FOSS project. The most powerful AIs would be created in the distributed FOSS manner by many volunteers and trained by volunteers running their own training nodes and that will result in another huge democratization of the tech industry. This has happened before with Linux and all the user-land software that was built on top of it. Tech-barons like Sam Altman with dreams of huge fortunes as a key player in an corporate AI oligopoly are going to be disappointed [hindustantimes.com]. The most ironic thing about DeepSeek is that if it is legit then DeepSeek will have accomplished with a minimal financial investment, maths and engineering and old fashioned optimization what OpenAI and others threw billions of dollars at. Then DeepSeek crashed US tech stocks by doing the most 'communist/collectivist' thing possible by open sourcing their LLMs. Ayn Rand would be applled, Ronald Reagan is spinning in his grave ... I legitimately find this turn of events hilariously funny.
I hope you're right. I'd love to see the wanna be AI overlords embarrassed into oblivion, AND get actual benefits from open source AI solutions. That would be the best outcome from all of this AI hype. Which is why I'm suspicious it won't go down that way.
Re: (Score:2)
Why is it that everyone keeps not understanding what R1 is. It's not a foundational model. It used all the current models and distills down from that, extracts reasoning with Reinforcement Learning techniques. The cost was smaller because of some shortcuts they made in the CUDA code, as they were doing Crypto mining with the cards they had. Also I'm sure they re not reporting all the costs associated with it.
So if a new bigger better foundational model is made, and you don't open source it, then it can and
My hopes...? (Score:2)
AI is commodity now (Score:2)
Crazy regulations (Score:1)
The technology is still only a few years old (Score:2)
What makes anyone think we're beyond the point where people can no longer jump in? There are a LOT of years to go.
Re: (Score:2)
There are a LOT of years to go.
More like decades to centuries. And the next milestone, which is currently farm far out of reach, is actual usefulness. Oh, wait, LLMs are useful for creating "better crap" and attack code. Such a big win!
EU basics (Score:1)
What, wasting money on artificial stupidity? (Score:2)
Seriously, anybody that is "late" to this game or does not join at all is already a winner.
open source (Score:2)
You don't really need do beat DeepSeek, forking it should be enough.