Microsoft CEO Satya Nadella warns that an AI market dominated by a few frontier models faces societal rejection

"Our priority has to be building a frontier ecosystem, not just a frontier model"

Great article! and timely, after the takedown of Fable and the launch of Fusion

@ClementDelangue yes! we need open source models to win

@ClementDelangue nobody wants a future where every company is just an api wrapper paying rent to three sets of weights. decentralizing the stack isn't just a vibe it's a survival tactic

fair, but these are two different claims. satya's pitch is firms owning their learning loop, not open models. you can have a thousand companies building private IP on top of three closed models, and the model layer is still concentrated. open weights fixes that part. the loop doesn't.

@ClementDelangue yet the very same Satya Nadella is already looking for new ways of profitmaxxing

@ClementDelangue quoted economist type energy but the models are already eating everything and society is just watching

where does the line actually get drawn?

@ClementDelangue Absolutely. Just for fun, Microsoft Office Revenue circa $35bn.

Everyone is still talking with one eye on their incentive structure.

@ClementDelangue He probably agrees because wealth does look like concentrating under him, but yeah, makes sense

@ClementDelangue Centralised intelligence of any kind just doesn't compute, same lessons as the early Internet

@ClementDelangue Oui enfin bon, Microsoft découvre le problème après avoir passé 2 ans à verrouiller l’infra et la distrib de l’IA. La captation de valeur les dérange d’un coup, quel hasard.

@ClementDelangue Clem, sometimes you have to fight the same monsters you once were!

Very real Karma points time for @Microsoft

@ClementDelangue I reacted to the same post by Mr. Nadella this weekend! He/we only need to look at the revolution, Linus Torvalds started a few decades ago! The power of Linux is immeasurable! That is how the models should be!

@ClementDelangue Decentralized AI is the only way forward

@ClementDelangue AI in the future has a lot to be thought about tbh. The risks short term and the risks long term. I hope we can manage it’s impact impact in the long run.

@ClementDelangue do you agree with the approach he suggests? like data factories or such in each company

@ClementDelangue competition will solve this before politics does.

open source keeps closing the gap, switching providers is trivial, and when costs collapse, the monopoly value just evaporates.

@ClementDelangue Clem, But other way to look at the article is even Big Tech CEO is not sure how exactly the AI future and its impact on the society will be

Would like to hear your views

@ClementDelangue End of the day , model is just a runtime. No one can or should create a moat around it

@ClementDelangue societal permission is a polite word for antitrust. the incentive to concentrate hasn't changed.

The pricing data shows a wider field than the concentration worry suggests. We benchmark 51 vendors weekly: open weights run about 80% below proprietary for comparable capability, and the spread across vendors for the same task runs past 80%. That is a competitive market, not a captured one, at least on price. Whether it stays that way is measurable, which is why we track it.