Every prediction about advanced AI says the economy is going to boom. Economists say moderate growth. Tech CEOs say explosive growth. Nobody says negative growth.
But what if they're all wrong? What if the thing that makes AI so powerful — replacing human work — is also the thing that breaks the economy's engine?
That's the question economist Alex Imas decided to take seriously. And the answer is more unsettling than you'd think.
Let's start with a story. Imagine a tiny island economy. There are 100 workers and 10 wealthy capital owners. Everyone eats two things: fish and coconuts. Workers catch the fish and pick the coconuts using equipment the owners provide. Workers get paid wages. They spend all of it. The owners are rich — they eat their fill (10 fish and 10 coconuts a day), and that's plenty for them.
Now, someone invents a robot fishing net. It catches fish all by itself. Workers lose their fishing jobs — but that's actually fine! They all move over to picking coconuts. Fish is now super cheap, so workers can buy more stuff with their wages. The economy grows. Everyone's happy.
But then someone invents a robot coconut harvester. Now both jobs are automated. The workers have nowhere to go. They earn nothing. They spend nothing.
And here's the kicker: the owners? They were already eating their 10 fish and 10 coconuts. They're full. They don't want more.
So you've got machines that could produce 10,000 units of stuff — but only 200 units of demand. Why would anyone build more machines or produce more goods if there's nobody to buy them?
The economy just shrank. Not despite the amazing technology — because of it.
When economists model the impact of AI, they mostly focus on the supply side — how much more stuff can we make, how much cheaper can we make it? And on those terms, AI looks incredible. If you can cut labor costs dramatically, output should jump.
But that logic has a hidden assumption baked in: that someone is buying all that new output.
Most economic models assume the economy smoothly adjusts. And historically, it has! When phone operators got automated away, those workers found other jobs. When farming got efficient, people moved to factories. When factories moved overseas, people moved to services. There was always somewhere for human labor to go.
The worry with truly advanced AI is: what if there isn't somewhere to go? What if AI automates enough sectors, fast enough, that millions of people lose income all at once — and with it, their ability to buy things?
Imas works through two different models. Both arrive at the same dark place, but through different doors.
Door #1: The Demand Collapse
This one is pretty intuitive. Workers spend almost everything they earn. Wealthy capital owners don't — they're already comfortable. When AI shifts income from workers (who spend a lot) to owners (who don't), total spending in the economy drops.
You might think: "But everything is so much cheaper now! Won't people just buy more?" And yes — cheaper prices help. But there's a limit. Think about light bulbs. The cost of lighting a room dropped by a factor of 40,000 between 1800 and the 1990s. Did we buy 40,000 times more light? Of course not. At some point, every room is bright enough.
If prices fall but people are already satisfied — and the people who would spend more don't have any money — you get a nasty gap between what the economy could produce and what it actually does.
In one numerical example, AI causes a nearly 48% drop in GDP. Not because of a failure of technology, but because of a failure of customers.
Door #2: The Slow Drain
The second model, from a paper called "Robots Are Us," tells a slower, creepier story. It's about what happens across generations.
Imagine two kinds of goods: mass-produced stuff (call it "corn") and deeply human services that can't be automated (call them "prayers" — think therapy, spiritual guidance, artisanal crafts). Young workers earn wages and save for retirement. Those savings fund the economy's capital — the factories, the infrastructure, the machines.
When AI arrives, code starts replacing coders. At first there's a boom. But eventually, legacy software piles up and new coders become less valuable. Wages fall. When wages fall, people save less. When people save less, there's less capital investment. The capital stock actually shrinks.
The result is a world where technology is better than ever, but the physical economy — the stuff that actually makes things — is hollowed out. Output falls below where it started. Everyone is worse off.
The researchers call this "immiserating growth." Progress that makes you poorer.
Probably not. And Imas is clear about this.
The conditions you need for the economy to actually shrink are pretty extreme. You need near-total automation of human work. You need no policy response. You need people to stop inventing new things to want. You need institutions to just sit there and do nothing.
Any one of those might be plausible on its own. All of them at once? That's a stretch.
History shows that people are remarkably good at inventing new desires. Air conditioning, international travel, cosmetic surgery, video games — none of these existed a century ago. Advanced AI could create entirely new categories: personalized AI companions, immersive virtual worlds, life extension therapies.
And governments tend to act before things get truly catastrophic. The political pressure from mass unemployment would be enormous.
Even if AI doesn't shrink the economy, these models reveal a real force that could make AI-driven growth much more disappointing than the optimists expect. Think of it less as "the economy will collapse" and more as "the economy will grow — but a lot less than the tech could deliver, because too few people have money to spend."
And here's the thing the GDP number won't capture: even if the economy keeps growing on paper, the welfare consequences of shifting all income to capital owners could be devastating for ordinary people. Growth and human wellbeing are not the same thing.
The most interesting part of Imas's essay might be the policy section, because the solution is surprisingly boring — in a good way.
The idea: if capital is going to earn a bigger and bigger share of the pie, then more people need to own capital. A sovereign wealth fund — where the government invests on behalf of citizens and pays out dividends — would keep money flowing to consumers without taxing innovation.
This isn't science fiction. Alaska has been doing it for decades with its Permanent Fund. The UAE does it. Several countries do it. The twist Imas adds: make the ownership stakes inalienable — you can't sell or trade them. In a world where wages disappear, you can't afford to let people lose their only remaining income stream to bad luck or bad actors.
The elegant thing about broad capital ownership is that it hits both problems at once. It gives families spending money (fixing the demand collapse). And it gives them assets that grow (fixing the savings drain). No innovation-killing taxes required.
Will advanced AI shrink the economy? Almost certainly not. But could it disappoint — delivering a fraction of the growth the technology makes possible, while concentrating wealth in ways that leave most people worse off?
That's not just possible. Without deliberate policy choices, it might be the default.