Model Competition vs Infrastructure Consolidation

One of the more interesting things in today’s AI news isn’t about which company may have the best model.

It’s what is happening underneath the models.

OpenAI launched GPT-6 Astra. Anthropic, Meta and Google are all pushing competing models. At the model layer, competition is increasing and leadership looks increasingly temporary.

At the same time, though, Nvidia is moving to acquire Hugging Face for almost $13 billion, potentially bringing one of the most important open-model distribution platforms into the ecosystem of the company already dominant in AI compute.

Microsoft is pushing even further upstream, where AI competition increasingly includes cloud infrastructure and data centres -- and now long-term energy supply.

The continuing parallels with the dot-com boom provide an interesting insight here.

A huge number of internet companies came and went. Even infrastructure giants such as Cisco were badly hit when the bubble burst.

But the infrastructure being built underneath the boom didn't disappear.

The networks, fibre, data centres and computing infrastructure became part of the foundation on which the next generation of internet businesses was built.

Perhaps AI is beginning to show us something similar.

Because we may be watching two different things happen at once.

AI is becoming more competitive at the model layer, while potentially becoming more concentrated at the infrastructure and platform layers.

That’s quite significant, because the long-term winners in AI may not necessarily be the companies that temporarily produce the smartest model.

The winners may instead be the companies that control the compute, cloud, energy, data platforms and distribution environments through which those models become economically useful.

Simply put: Model leadership may be temporary. Control of the AI stack is much harder to displace.