None of this will be a choice made by experts. This will not be a decision made from a standpoint of what is the most rational way to set up everything.
The decisions will be made by politicians. It will be made on the basis of power, control, fear, ignorance, public outrage, etc. It will be made in a moment of intense public uproar. They will not trust AI systems to secure the economy against other AI systems, that approach will take over a decade to deploy fully anyway, we don't have that kind of time. As the attacks rapidly increase, the politicians will reach for a hammer solution in desperation, and that one solution is intensive network isolation.
The politician will say: why are we allowing foreign actors access to our networks to harm us with AI? The answer is: there's no reason to allow it, it is not necessary. Note that China has done just fine with its network isolation, its economy has not languished for it, they will be viewed as proof that it can work well.
Big tech won't be prevented from selling to the rest of the world. They'll follow local laws/rules just as they have to right now. They'll help other nations build their own isolated networks (smaller nations will federate with other nations on this), and they'll offer services within those networks. Canada for example would be very likely to federate with the USAnet and EUnet.
> They will not trust AI systems to secure the economy against other AI systems, that approach will take over a decade to deploy fully anyway
They may not trust it, but the danger cannot realistically grow here faster than the solution, as it is the same tech.
> Big tech won't be prevented from selling to the rest of the world. They'll follow local laws/rules just as they have to right now.
The data centres are big and hard to move. Severing the trans-Atlantic internet connections will force big tech to build in Europe new data centres to support Europe, and take no advantage from time zones having Europe finish it's work day before California starts theirs.