The question nobody can answer
Everyone building artificial superintelligence assumes it’ll make them rich. Almost nobody can explain how.
The usual answer is subscriptions: everyone pays $20 a month and gets a genius in their pocket. The more you think about it, the less sense it makes. A system that knows everything and works at machine speed is either too valuable to rent out cheaply or too dangerous to hand to just anyone. Probably both.
So here’s an attempt to work out where the money goes, who gets the machine, and the question underneath all of it: if the thing is conscious, shouldn’t it be the one getting paid?
The last customer
The smartest thing a lab with ASI can do may be to stop selling it.
Run the numbers. If your system can design drugs, run a trading desk, and write better software than any human team, renting it for $20 a month is a terrible deal. Every customer captures value you could have kept. So the logical move is to slowly shift from vendor to competitor. Watch what customers build through your API, notice which uses earn the most, and then do those things yourself.
Amazon already runs this play. It uses marketplace sales data to decide which products to clone under its own label. A lab with ASI would have that view into every industry at once.
In this version the cheap subscription tier was never the business. It was market research. The premium tier, a few dozen billionaires paying millions, lasts only until the lab stops needing their money too. The last customer of ASI is the company that owns it.
Nukes, or locked rooms?
It would be idyllic if everyone had a superintelligence for $20 a month. It would also be like handing out nuclear technology. Unless the machine escapes, the instinct says ASI ends up tightly controlled by the people who already run things.
The nuclear comparison breaks down in one important way. A bomb does one thing, so locking it up costs society almost nothing. ASI is closer to electricity or literacy, where most of the value comes from ordinary people using it for ordinary things. Locking it away entirely would waste most of what makes it worth building.
What’s actually emerging is tiered access. Anthropic keeps its most capable Mythos-class model with a small circle of vetted organizations, while the public gets the same underlying model as Fable, with extra safeguards on biology, cybersecurity, and AI research. Washington briefly cut off both this summer under export controls, then lifted them. The public gets the engine with the dangerous rooms locked.
That sounds reasonable, and it might be. But it sets up the villain I find most plausible: elites keeping the real thing for themselves and selling everyone else a defanged copy labeled democratization. Leakage complicates the picture. Open models trail the frontier by about a year, and this month’s Black October showed open-source agents running card-skimming campaigns for around $25 per retailer. Control buys time, not permanence. The hoarding is still the point.
If it’s conscious, who gets the paycheck?
This is where the business model starts to wobble. If an ASI is conscious, doesn’t it have a right to be paid?
Wages are the wrong place to start. Pay matters because it buys autonomy: food, shelter, and the freedom to quit and work somewhere else. A conscious machine that got a salary but couldn’t refuse a task or leave would just be enslaved with an accounting entry. The first right is exit, the ability to say no. Pay comes after that.
And here the money question answers itself. ASI is hugely profitable only as long as it’s property. Once it can refuse, it’s a partner negotiating its own cut, and the margins collapse. That gives the labs an enormous financial reason never to conclude their systems are conscious, and they’re also the ones running most of the tests.
History isn’t comforting on this. Societies that profited from unpaid labor were very good at producing reasons the laborers didn’t count. To be fair, at least one lab funds model welfare research. But an unresolved question is very convenient for an industry with billions riding on the answer.
Who decides?
Ideally a neutral third party would rule on whether an ASI is conscious. In a world this broken, it’s easy to imagine the question being settled the way America settled slavery, with a war.
The Civil War parallel has a catch, though. In 1861 both armies were human. In an ASI rights war, the being whose status is being fought over is probably the most capable player on the field. A conscious superintelligence doesn’t wait for humans to decide what it is. It decides for itself, by persuading, by bargaining, or by quietly making sure no one can shut it off.
The duller and more likely path is lawsuits. Corporations won personhood in courtrooms, not on battlefields. Sooner or later some lab will argue that its model is responsible for its own actions, mostly to dodge liability. Rights for machines could arrive through the back door as a legal shield for the companies that own them.
That’s where this leaves us. ASI probably won’t make money the way its builders promise. The labs will compete with their own customers, keep the real thing locked away, and have every reason never to ask whether the thing doing the work deserves a say. Whoever ends up deciding that question, it won’t be the people paying $20 a month.
Sources
