So we've been hearing for months from the AI influencers that - "what the hell is happening in Europe??". "There are no frontier labs in Europe!!!". In fact in the link above - Ethan Mollick even claims that it's totally impossible. Not even a shred of the right stuff available in Europe. Until suddenly Mistral dropped Le chonk - Mistral Large 4, which - benchmaxxing be damned - is certainly a model at an appropriate scale and ambition level. Obviously the world overflows with 'meh' comments, they are still months behind - but come on, this is just silly - why should the developments of 2026 be so different from 2026, 2024, 2023?
I've not seen all the influencers retract their stupid claims - but Mistral Large 4 is not even alone; German groups and UK groups are building (smaller) open models also.
And why wouldn't there be? At this point there's maybe 10-30 labs targeting the 1T model size. A lot of these models are very good. If it's more or less trivial to start a new lab in China or the US that can compete what would make it hard in Europe?
Back in 2023 - Sam Altman made remarks that it would be 'hopeless' for anyone to try to catch up to OpenAI. This was before he lost his lead, financially and performance wise, to Anthropic. He backpedalled and said that this as about the size of investment - and that makes more sense - except of course the Chinese labs seem to not require the billions the American labs are spending - alledgedly they are also mainly distilling the big models.
The simple thing - that has been clear for a long time, is this: There are no secrets. There is no moat. There's only data and compute - aka capital. The only moat is money. I'm sure there's some sweet operational secret on inference at scale inside Anthropic and OpenAI - but very hard to believe even these are defensible.
Now, money is obviously not a small problem - but a big problem. And as long as the EU remains too fragmented for its own good so that we don't actually have a truly large homogenous domestick market for services - among them AI services - it will remain a problem. But at the core, AI as a field is an open book.