What is the future of society after AI dominates most of the tasks of the middle class?
Foundation models—handling text, audio and image and responding with text (and in some cases image)—are where technology stands today.
The model's interface, and its whole "brain," is language for now: we talk about Large Language Models, but we're already seeing a push toward multimodal models. The future of AI will likely unify all branches of machine learning under a single architecture that may not exist yet or is being built, but it will happen.
Likewise, we'll find ways to train models faster or more cheaply, as we've already seen with cases like DeepSeek v3.
So imagine a society, not many years from now, where we've found a way to build these foundation models—fast to train and use (though prohibitively expensive for companies without GPU farms). Remember that training them means using petabytes of data (or maybe exabytes); when I say "cheap," I mean one of these models might cost on the order of $100 million—a very low price for building silicon superhuman intelligence. Running it would be very cheap, as with ChatGPT or Claude today.
Under these assumptions, one question remains:
Okay, but if all those jobs disappear, who will buy what companies produce if there's no employment and no income? Some propose universal basic income (UBI) to support most people hit by this revolution.
But who pays? Big tech through heavy taxes, and their customers. So a handful of companies, likely in the US or China, would have to pay the rest of the world the wages they've saved—both as firms and as customers who replaced workers with AI to cut costs.
In this setup we assume the employer replaces the worker with AI. There are other options—e.g. the worker delegates their work to AI, which also boosts productivity with less labour and again leads to fewer jobs.
In an ideal world we'd clearly define which companies pay those AI taxes: those that use AI, whether they or their workers do the using.
So universal basic income would sit underneath the companies—above all the big foreign ones. The basic livelihood of millions or billions would depend on those companies paying their taxes. That power could match or exceed the government's and would make us heavily dependent on this sector; the government could become a tool of big tech—lobbying at scale.
First, countries without big tech would be at the mercy of a foreign company and its government. Second, the foreign government would be at the mercy of those companies.
What if a country has no home-grown tech company to capture the capital that once went to workers?
Yes, the work of CEOs and boards could be automated—not just programmers, engineers, HR, etc. Top roles can be replaced, and that's not bad news for the real winners of this game: the shareholders of those companies.
It doesn't matter whether the one steering the ship is flesh or silicon; what matters is who takes the treasure. And the best part? Shareholders won't have to share with anyone if there's no one left to share with. They'll have minimal staff supported by AI tools. Basically a machine for becoming a billionaire.
So those who reach the top won't get there by meritocracy—they'll get there by owning the profits of the company. And their children's merit will be being their children, like a royal family. Data centres and the fate of the world will be run by elites who own data centres, not by scientists or engineers.