You are about to be made richer as a customer and poorer as an owner. The first half feels so good you won’t notice the second.

Sometime soon you’ll pay twenty dollars for something that used to cost two thousand. A legal opinion. A design. A diagnosis. A quarter’s worth of analyst work, done before your coffee cools. It will feel like a gift, and you will tell people about it.

Here is what you won’t feel: that two-thousand-dollar bill didn’t vanish. It moved.

It used to be somebody’s salary, and salaries get spent at the shops, taxed by the state, spread around a town. Now it is somebody’s margin, and margin gets booked as equity, owned by a few, and parked. You will celebrate the cheaper service and never connect it to the thinner paycheque, because the discount lands in your lap today and the displacement lands on someone else next year.

That is the trick of this whole transition. Almost no one is angry about it, because the bribe arrives first.

We are arguing about the wrong number

The public fight over AI is stuck on jobs. Will they go, how many, how fast, and will “reskilling” save us. It’s the wrong number. The argument was never really about wages. It’s about the cap table.

For two centuries, inequality was tolerable because of one promise: growth spreads. The founder got rich, but the factory hired. Investors won, but workers were paid, and the wage was the pipe through which the gains reached everyone else.

Strip out the worker and you don’t just remove a job. You remove the pipe. The whole pitch of AI is to grow output without growing headcount, and when that works, the gains don’t trickle down through salaries. They shoot up through ownership, to founders, shareholders, chipmakers, the cloud, whoever holds the machine.

This is why “it’s just the internet again” is wrong. The internet created middlemen; AI deletes them. One gave us platforms, the other offers replacement. Politically, those are different animals.

The money is already pooling at the top

This isn’t a forecast. The scale is already on the books:

  • McKinsey puts generative AI’s potential at $2.6-4.4 trillion a year.
  • The IMF reckons 40% of the world’s jobs are exposed, around 60% in rich economies.
  • US private AI investment hit $285.9 billion in 2025, per Stanford’s AI Index.
  • Nvidia’s data-centre arm alone booked $75.2 billion in a single quarter, up 92% on the year.

Look at where that last number lands. Not in a million pay packets, but in the accounts of the handful of firms that sell the shovels. That is the whole story in miniature: enormous value, narrow ownership.

Now be honest about the asymmetry, because it’s worse than “the rich get richer.” The consumer windfall is at least real and widely shared, cheaper tools for everyone with a phone. But not everyone gets even that. The radiographer whose scan-read is now automated doesn’t get a cheaper radiographer. She gets a pink slip. The gift is broad and the loss is concentrated, and a broad, mild gift almost always out-votes a narrow, severe loss. That is precisely why the anger never organises.

You can’t tax your way back in

The reflex fix is to let them grow and tax the profit. It’s built for the wrong economy. Profit moves, deferred, restructured, booked offshore, and AI’s crown jewels aren’t coal seams inside a border. They’re models, weights, compute contracts and IP that live everywhere and nowhere.

Worse, the real money doesn’t show up as income at all. It shows up as valuation. A founder’s net worth can climb by billions, and a model can erase a profession, long before the tax system sees a single rupee.

We built the tax machine for a world where humans earned wages. AI is building a world where machines earn the margin. You cannot tax your way onto a cap table you were never on.

The honest answer is a share, not a handout

Which is why the cash-handout solutions, UBI, transition cheques, “AI dividends,” solve the wrong problem. A handout keeps you fed while confirming your new role: a dependent of the machine, not a part-owner of it. It buys quiet. It changes nothing about who owns the future.

The harder answer is ownership, and the precedent is dull and proven. Norway turned its oil into one of the largest sovereign funds on earth. Alaska still cuts its residents a cheque from oil revenue. When a national resource concentrates wealth, the public takes a stake, not as charity, as equity.

The AI version is not the state seizing the labs. A government dictating how models get built would politicise the frontier and shield the incumbents it’s meant to check. The point isn’t control, it’s a claim: a public fund holding small, diversified, professionally run stakes across the stack, so that when AI compounds, citizens compound with it. Get back on the cap table, or stay a customer forever.

This is a legitimacy problem, not a charity one

The industry is asking for staggering trust, your data, your work, your kids’ schooling, the grid, national security, while minting the largest private fortunes in history. People will wear that while their own lives visibly improve. Build the other world, where a few own the machine and everyone else rents access to intelligence by the month, and the reckoning won’t arrive as a polite essay. It’ll arrive as a ballot.

Every technological age writes a new social contract. Industry got unions and public schooling, oil got sovereign funds, the internet got data empires. AI will write one too, and the only choice is whether it’s drafted before the anger or after it.

The paradox is the cruel part. The more spectacularly AI delivers everything its builders promise, the harder it becomes to justify a world where so few own the thing now doing the work.

The next fight won’t really be left against right, or even labour against capital in the old sense. It’ll be plainer than that.

Those who own the machine, and those who rent their lives from it.