The Real AI Question Isn't ‘Is It Powerful?’ — It's ‘Who Owns What It Produces?’

Almost every loud argument about artificial intelligence is an argument about capability. How smart is it? How fast is it improving? Will it take our jobs, cure our diseases, or end the world by Tuesday? These are the questions that fill the headlines, the conference stages, and the funding rounds. But if you are trying to work out who benefits from AI — and whether that group includes you — the capability debate is mostly a distraction. It is loud, it is genuinely fascinating, and it tells you almost nothing about how the technology will land in your actual life.

I want to make a simple case here: the question that decides outcomes is not how powerful is this thing? It is who owns what it produces, and who is allowed to keep the gains? Same capability, two completely different futures — depending entirely on the answer.

Capability is real. It is also beside the point.

Let me be fair, because the capability story is not nonsense. Better models genuinely do useful things. A doctor with a good diagnostic assistant may catch what she would have missed. A small business owner drafts contracts she once paid a lawyer for. A student in a town with no good teachers gets a patient tutor at midnight. These are real, broad benefits, and I am not going to pretend they vanish. Millions of people will get something out of this. That much is true.

But notice how the capability framing quietly changes the subject. It invites you to argue about the tool as if the tool were the outcome. Whether a hammer is well-made tells you nothing about who ends up owning the house. A more powerful engine can pull a family across the country or it can pull coal out of a mine faster for someone who will never share the profit. The horsepower is the same. What differs is the arrangement around it — who holds the keys, who takes the surplus, who absorbs the cost.

So when someone tells you AI is getting more capable, the honest response is: yes, and? Capability is an input. Benefit is an outcome. The wire between them is ownership, and that wire is where the whole story actually happens.

Whether a hammer is well-made tells you nothing about who ends up owning the house.

Who benefits from AI today

Look at where value is pooling right now, and a pattern shows up that I have seen with almost every major technology. The largest, most durable gains are concentrating among the people who own the layers the rest of us have to rent.

Start with the obvious owners. Whoever controls the frontier models captures a toll on everything built on top of them. Whoever owns the compute — the chips, the data centres, the power contracts — collects rent whether a given AI company succeeds or fails, the way the people selling shovels did better than most of the miners. Whoever holds the proprietary data that makes a model useful in a specific domain has something no amount of cleverness can route around. And whoever owns the platform — the place where users actually show up — can bundle AI into an existing moat and make it nearly impossible to compete with. These are not four separate winners. They are increasingly the same handful of firms, stacked. I explore that concentration more directly in my piece on who owns AI, and it is worth reading alongside this one, because ownership is the machinery underneath everything I am arguing here.

Then there is the second group of beneficiaries, less discussed but just as important: firms that use AI to cut costs. When a company replaces forty support agents with three people supervising a model, the work still gets done — but the wage bill that used to flow to forty households now flows, as profit, to shareholders. The productivity is real. The question of who keeps it has a very clear early answer. It is not the forty.

I am not claiming precise ownership stakes or throwing out percentages I cannot verify. I do not need to. The pattern is legible without the exact figures: value is accruing fastest to whoever sits closest to the models, the compute, the data, and the distribution — and to the balance sheets that convert automation into margin.

Who could benefit — and why they mostly don't, by default

Here is the part that should bother you, because the same technology could deliver enormous gains to ordinary people, and there is nothing in the machine itself stopping it.

Workers could benefit. An AI that makes a nurse, a paralegal, or a mechanic twice as effective could mean shorter hours for the same pay, or the same hours for better pay, or more people able to do skilled work that was previously gate-kept behind years of training. Consumers could benefit through genuinely cheaper services, not just cheaper-to-produce services whose savings quietly stay upstream. Whole regions locked out of expertise — legal, medical, educational — could leapfrog.

None of that is fantasy. All of it is technically available today. So why does the default drift the other way?

Because gains flow along the lines of ownership, not the lines of contribution. When a tool raises output, the surplus goes to whoever has the power to claim it — and in most arrangements that is not the worker whose labour was augmented or the consumer whose data trained the thing. This is the capture pattern I keep coming back to, and it is not unique to AI. I have traced it before in how technology gets captured: a technology arrives with broadly shared promise, and then, absent deliberate rules, the value quietly consolidates upward until the many are renting from the few. Some writers now call the endgame of that consolidation technofeudalism — a world where you no longer own the platforms and models you depend on, you merely pay to access them, forever.

The word default is doing heavy lifting here, and I want to be honest about it. Nothing about AI forces the extractive outcome. But nothing about it prevents that outcome either. Left alone, capital-heavy technology concentrates, because concentration is the path of least resistance. If you want the other future, someone has to choose it, on purpose, against the grain.

The same capability can raise your wage or cut it

This is the hinge of the whole argument, so let me make it as sharp as I can. Take one fixed capability — say, a model that automates 30% of the tasks in a particular job. That single, unchanged capability can produce two opposite results.

• In one arrangement, the firm keeps the worker, uses the freed-up 30% for higher-value work, and shares the productivity gain as higher wages or shorter weeks. The worker is better off. The capability lifted her.
• In another arrangement, the firm cuts headcount, banks the savings as profit, and uses the remaining leverage — a labour market now looser because the model can do part of the job — to hold everyone else's wages down. The same worker is worse off, or gone. The capability cut her.

Nothing changed about the AI between those two stories. What changed was ownership, bargaining power, and rules — whether workers have a claim on the gains, whether markets are competitive enough that savings get passed to consumers, whether anyone can even build an alternative or is locked into renting from a single provider. When one company can quietly set the terms for a whole sector, that is the AI monopoly problem, and it is the difference between a tool that lifts you and one that is used on you.

So the debate about whether AI is "good for jobs" is malformed. It has no answer at the level of the technology. It only has answers at the level of arrangements. Ask a better question and the fog clears immediately.

Nothing changed about the AI between those two stories. What changed was ownership.

How AI could actually benefit ordinary people

None of this is a counsel of despair. I am not arguing the outcome is fixed — I am arguing the opposite. The outcome is a choice, and choices can go the other way if enough people fight for the right things. Here is roughly what fighting for the right things looks like.

• Open models and open weights. When capable models can be run, inspected, and adapted by anyone, the toll booth on the frontier gets harder to defend. Openness is not charity; it is competition, and competition is what forces gains outward instead of letting them pool.
• Public and shared infrastructure. Compute is becoming as foundational as roads or electricity. When only a few private owners hold it, everyone else builds on rented land. Public options — university clusters, shared national compute, cooperatively owned resources — change who can participate at all.
• Real data rights. The models are built on our collective output — our writing, our images, our behaviour. If the people who generated that data have no claim on what it produces, the value flows one way by design. Data rights are how you attach a claim.
• Competition, enforced. Bundling AI into existing monopolies is the fastest route to a rent-extraction economy. Antitrust that treats models, compute, and distribution as the choke points they are is not anti-technology; it is what keeps the benefits contestable.
• Sharing productivity gains. When automation raises output, the surplus does not have to become pure profit. It can be shared through wages, hours, ownership stakes, or public dividends. That is a policy and bargaining choice, not a law of physics.

Notice that not one of these items is about making AI more capable. Every one of them is about ownership and rules — about redirecting where the gains flow. That is not an accident. It is the whole point.

Change the question, change what you fight for

If you spend your attention on how powerful is AI?, you will end up rooting for or against a capability, cheering benchmarks, dreading or worshipping the next release — and none of it will move a single dollar toward your household or a single hour off your week. You will be a spectator to a fight about a tool.

If instead you ask who owns what this produces, and who keeps the gains?, the whole landscape rearranges. Suddenly the important battles are not about model size at all. They are about open weights versus closed, public compute versus private toll booths, data rights versus data extraction, competition versus consolidation, shared gains versus captured ones. These are winnable fights. They have precedents. And they are decided by what people demand, not by what the models can do.

The capability debate asks you to marvel. The ownership question asks you to choose. Marvelling is comfortable and it changes nothing. Choosing is harder and it changes everything. So the next time someone tells you how astonishing the latest AI is, believe them — and then ask the only question that determines whether it will ever be astonishing for you: who benefits, and who decided that they should?

Change the question, and you change what you fight for. That is not a small shift. It is the difference between watching this technology happen to you and having a hand in how it turns out.