I think the question today is not who will use AI. That part is largely settled: everyone who can, already is. The harder and more revealing question is the opposite one. Who cannot use AI, and why can they not use it? And to be clear, this is not a technology question. It is a question of availability, and of the ability to pay the foundational cost of getting started.
IT used to be the great equalizer
I have said this often during my sessions: for a long time, people from middle class and lower class backgrounds did remarkably well in IT. The reason was simple. The economic barrier to learning was low. A modest second hand computer, a cheap internet connection, and free and open source tools were enough to teach yourself to code, to build real things, and to compete on merit rather than on capital. The field rewarded curiosity and persistence far more than it rewarded money. That openness is a large part of why the industry produced so much talent from places and families that no one expected.
AI has raised the floor
Today that is no longer the case. If you want to run your own experiments in machine learning or AI, you face a very different situation. Serious local work needs expensive hardware: capable GPUs, large amounts of memory, and the power and cooling to keep them running. The alternative is to rent capacity from third party providers, which means working inside controlled environments that you do not own. You get metered usage, rate limits, shifting terms of service, content restrictions, and the constant possibility that the model or feature you depend on is changed, priced up, or withdrawn without notice. Either way, the price of entry has moved up, and the person without capital is either priced out or left at the mercy of a handful of providers.
This is the quiet inversion that bothers me most. The very thing that once let people climb without money now increasingly asks for money first.
Regulation can shrink access even further
There is a second force, and it worries me as much as cost. The fear mongering around AI gives governments and controlling agencies a ready justification to write laws that reduce the reach of these tools for a large part of the world’s population. Once fear becomes the framing, restriction becomes easy to sell, and it is rarely the powerful who lose access first.
The recent Fable case is a small but important example of how quickly access can be curtailed once a reason is found. It does not take much. You never know when an agency decides that you cannot use this model, or that model, or the other one, due to this reason, that reason, and some third reason. Each restriction sounds defensible on its own. Put together, they can quietly wall off whole categories of people from tools that are fast becoming basic infrastructure.
Who actually gets left out
Put the two forces together, cost and regulation, and it becomes clear who cannot use AI. It is the student without a capable machine. It is the self taught developer in a region where cloud access is expensive, throttled, or restricted. It is the small business that cannot absorb enterprise pricing. It is, in short, almost exactly the population that IT once lifted. If we are not careful, AI will undo the one property that made computing such a powerful ladder in the first place.
Why this matters, and what helps
None of this is an argument against AI. It is an argument for keeping the door open. That is why open weight models, efficient models that run on modest hardware, and private, on premise deployments matter far beyond their technical merits. They are what decides whether the next generation can teach itself and build the way many of us did, or whether it will have to ask permission and pay rent for the privilege.
So the question of who cannot use AI is really a question about what kind of field we want this to be. We should answer it deliberately, and soon, before cost and law answer it for us.
