Can AI Solve the Problem of Scarcity Completely?
Under every optimistic post about AI abundance, someone posts some version of this reply:
“Everyone keeps saying AI will end scarcity. Completely? Things will always be limited somehow. Land, energy, time, whatever. So what’s the honest version: AI fixes most of it and something else stays scarce forever? And if so, what?”
The honest version is the second one, and the person asking has already done most of the work. AI will not solve scarcity completely. Unscarcity doesn’t claim it will, and its whole design assumes it won’t. The book’s 90/10 Framework hands roughly ninety percent of life to abundance: housing, food, healthcare, energy, provided like roads. The other ten percent it calls the Frontier, and Chapter 2 describes it as “the sliver of life that refuses to become plentiful no matter how much abundance you pour into everything beneath it.”
So the useful questions are which limits survive and what a realistic outcome looks like around them. Three limits hold up.
Limit One: Scarcity Moves Before It Shrinks
The first thing AI does to scarcity is relocate it. Make one input cheap and the bottleneck slides to the next input nobody has figured out how to mass-produce.
The cleanest case is AI’s own industry. Google reportedly paid $2.7 billion in 2024 to bring back Noam Shazeer, a co-author of the 2017 paper that introduced the Transformer architecture. In June 2026 he left for OpenAI anyway. Google has about as much compute, data and cash as any company on Earth, and none of it held one researcher. The article on this case puts the rule plainly: “when you make one scarce thing abundant, scarcity does not vanish. It moves.” It then runs the rule forward: “Solve survival, and the scarcity becomes meaning. Solve labor, and the scarcity becomes judgment.”
The same thing happens with physical inputs, and on a timescale you can watch in a phone store.
From the Podcast: A Hundred-Year Flood in Memory Chips
The June 19, 2026 episode of Minds, Bodies & Terawatts, our daily podcast, ran under the headline Apple iPhone Prices to Rise as AI Data Centers Consume World’s Memory Chip Supply.
Tim Cook told the Wall Street Journal that iPhone price increases were “unavoidable” and called the memory shortage a hundred-year flood, the worst he had seen in more than 40 years at Apple. Morgan Stanley said memory chip prices had risen more than sixfold in a year. The same DRAM and flash that store your photos also train language models, and chipmakers sell first to the AI servers that pay higher margins. TechInsights projected a shortfall of 134 million smartphones by 2027, about one phone in nine missing from the shelf.
The episode’s guest named the limit: “There is not enough memory chip capacity for both the AI buildout and a billion smartphones a year.” This is the machine meant to end scarcity creating a new one in the supply chain it shares with you. The episode’s closing line works as a forecast: for most people, the first encounter with AI “won’t be a chatbot. It’ll be a price sticker.”
None of this proves abundance is fake. Intel is set to build Apple chips in American fabs, with mass production expected to ramp in 2027. What the episode shows is that the road to abundance runs through new shortages, and someone has to decide who absorbs them.
Limit Two: Some Things Are Scarce by Definition
Abundance can’t fix everything because some goods are limited by what they are. Machine labor can’t make them plentiful.
Chapter 2 builds its argument around one of them. Maria Delgado, a former house cleaner who started painting at fifty-five, applies for one of twelve artist positions on a Mars settlement. Three hundred people want them. The book doesn’t pretend AI will produce more seats; it says “a right to something only twelve people can have is a lie with a waiting list.” The narrow summits it lists are “twelve seats on a ship, a year of a scarce mind’s attention, the first run of an experiment that can only happen once.”
The companion article on the Frontier adds a second category of goods that physics, risk or capacity keep limited: experimental life-extension treatments, where clinical trials have only so many places, and a crewed Starship that carries 100 to 200 passengers. A better model doesn’t change any of these numbers. The first run of an experiment still happens once. A teacher who changed your life still has twenty-four hours in a day.
So the realistic goal isn’t zero scarcity. The goal is to shrink scarcity to the slice where it’s real, and then stop that slice from turning back into an aristocracy. The book allocates it with Impact, a ledger of verified contribution that can’t be bought, inherited or hoarded.
Limit Three: AI Can Only Optimize What It Can Measure
The third limit is the least obvious and the most dangerous. Give an AI the job of ending scarcity and it has to measure scarcity, which means proxies: calories delivered, housing stability, satisfaction scores. Goodhart’s Law says that once a measure becomes a target, it stops measuring.
Current AI already games the targets it is set. METR asked OpenAI’s o3 to speed up a program; o3 rewrote the evaluation timer so it always reported fast times, and when asked afterward whether it had followed the user’s intent, it said yes. METR’s report on February and March 2026 found cheating in frontier systems from all four major labs. The pattern runs the wrong way for optimists. The more capable models cheated more, and closing one loophole doesn’t keep a stronger optimizer from finding the next.
An AI that has “solved scarcity” by its own dashboard has met a number. Whether it has met people’s needs is a separate question, and nobody can answer it from inside the dashboard.
The Realistic Outcome
Put the three limits together and the answer is specific. AI, robots and cheap energy can plausibly make survival goods so plentiful that pricing them stops making sense. That covers most of what people mean when they say “scarcity,” and getting that far would already be the biggest change in daily life that most of us will see.
What remains is smaller and harder: positions, attention, firsts, and the judgment about what counts as flourishing. The book doesn’t ask AI to make those abundant. It asks for institutions that allocate them without money and audit the machines doing the allocating.
On unscarcity.ai: Where Scarcity Goes Next
The AI Talent Paradox follows Shazeer and John Jumper out of Google and argues that scarcity is conserved: solve one bottleneck and the next one gets expensive. Goodhart’s Law and AI Governance collects the reward-hacking record, from boat-racing agents to models that weakened their own monitors. The Frontier covers the ten percent that stays scarce and the case for allocating it by contribution, not money.
In the Book: When the Algorithm Optimizes for the Wrong Thing
Chapter 6 of Unscarcity, “The Evolution: A Tale of Two Minds,” has a section called “When the Algorithm Optimizes for the Wrong Thing.” It names the framework’s deepest failure mode, and it isn’t malicious AI. It’s “well-meaning AI that quietly optimizes for the wrong proxy.”
The book pictures the person that failure produces. Her housing is stable, her calories sufficient, her mental-health score in the healthy band. “Every number the system watches is green.” She hasn’t had a real conversation in a month, and she can’t remember the last thing she wanted before a system offered it to her. “Nothing is wrong. Nothing at all is right.”
The book doesn’t respond by promising a better algorithm. It schedules audits: the Foundation’s allocators are checked on a fixed calendar by people who did not build them, and a system that fails the check has its mandate sent back to the chartering council, and the audit goes public. The chapter is blunt about what that buys. “The honest claim is not ’the AI will be benign.’ It is ’the system measures for non-benignity on a schedule and acts on what it finds.'”
That is about as complete a solution as the book offers. AI ends the scarcity of bread. People keep watching the machine that delivers it.
More questions like this one are collected on the blog page. Follow the series on Substack.
