Note: This is a research note supplementing the book Unscarcity, now available for purchase. These notes expand on concepts from the main text. Start here or get the book.
The AI Talent Paradox: When Genius Becomes the Last Scarcity
In June 2026, Google lost the two people it could least afford to lose, three days apart.
First, Noam Shazeer walked out the door to OpenAI. Shazeer is not a normal employee. He co-wrote “Attention Is All You Need,” the 2017 paper that invented the Transformer, the architecture underneath every chatbot you have ever used. Google had paid a reported $2.7 billion in 2024 to pull him back from his startup Character.AI, an acqui-hire so expensive it was essentially a ransom. Eighteen months later, the ransom didn’t hold. He left for a direct competitor.
Three days after that, John Jumper left Google DeepMind for Anthropic. Jumper won the 2024 Nobel Prize in Chemistry for AlphaFold, the system that predicted the structure of 200 million proteins and is now used by two million researchers across 190 countries. He had been at DeepMind for almost nine years. Alphabet’s stock dipped on the news, because the market understood what had happened: the most valuable player in the AI race had just demonstrated that its crown jewels have legs.
Here is the part that should keep every AI executive awake. Google has more compute than almost anyone on Earth. It has more data than almost anyone on Earth. It has more cash than the GDP of most countries. It deployed all of that, plus $2.7 billion in cold money, to hold onto one man, and it failed. The single input to frontier AI that money cannot reliably buy turns out to be the input that matters most: a few hundred irreplaceable minds.
Welcome to the talent paradox. It is the clearest real-world demonstration of the one idea the Unscarcity framework is built on.
Scarcity Is Conserved
The whole argument of the book starts with a deceptively simple observation: when you make one scarce thing abundant, scarcity does not vanish. It moves. The bottleneck migrates to whatever you haven’t solved yet.
We are watching this happen in real time at the AI frontier. Compute, the binding constraint five years ago, is being poured into the world by the trainload. OpenAI alone raised $122 billion, and Musk wired up 200,000 GPUs in a Memphis warehouse in 122 days. Capital is so plentiful it has lost the power to shock anyone. Training data is being scraped, licensed, and synthetically generated at planetary scale.
So the bottleneck slid down the line and parked itself on the last input nobody has figured out how to mass-produce: the small cohort of researchers who can actually push the frontier. Abundant compute plus abundant capital plus abundant data does not add up to abundant breakthroughs. It adds up to a bidding war for the handful of people who know what to do with all of it.
This is not a bug in the AI economy. It is what the book predicts will happen to every economy as it climbs. Solve survival, and the scarcity becomes meaning. Solve labor, and the scarcity becomes judgment. The hydra always grows a new head.
Why Genius Hasn’t Commoditized (Yet)
Skeptics have a fair objection. AI is supposed to be the thing that automates cognitive work, so why hasn’t it automated its own R&D? AI already writes 51% of the code on GitHub and roughly 75% of the code at Google. If machines can write the software, why are humans still the bottleneck?
Because writing the code and deciding what code to write are different jobs, and only one of them has been automated.
The abundant thing is execution. A model will happily generate ten thousand lines of a transformer variant, run the training loop, and plot the loss curve. The scarce thing is taste: knowing that attention might replace recurrence in the first place, sensing which of a hundred plausible research directions is worth a six-month bet, recognizing a result that looks like noise but is actually the future. Nobody handed Shazeer the Transformer. He had the judgment to see it before there was evidence it would work.
That kind of judgment is exactly what current AI is worst at. Models are superhuman at interpolation, at producing the average of everything they have already seen. Frontier research is extrapolation, the act of being right about something that is not yet in the training data because it hasn’t happened yet. Until a model can do that reliably, the human who can stays scarce, and scarce things in a high-stakes market get expensive. Two-point-seven-billion-dollars expensive.
The Aristocracy Problem
This is where the paradox stops being a fun business-page story and starts being the book’s problem.
When a tiny group becomes the binding constraint on the most valuable industry in history, the market does what markets do: it bids their price toward infinity. Nine-figure pay packages for individual researchers stopped being rumors in 2026 and started being recruiting strategy. A few hundred people are on track to personally capture a meaningful slice of the wealth generated by the technology that will reshape every other job on the planet.
Left alone, this is how you build a new aristocracy. Not an aristocracy of land or oil, but of irreplaceable cognition. And cognitive aristocracy is stickier than the old kind, because you cannot tax it away or nationalize it. It walks out the door and joins your rival, as Google just learned twice in one week.
This is the fork the book opens with: the choice between a Star Wars future of elite capture and a Star Trek future of abundance as infrastructure. A world where AI’s gains flow to the few hundred people who built it while everyone else rents the output is the Star Wars branch wearing a hoodie. The Iron Law of Oligarchy does not care whether your elite inherited castles or wrote really good CUDA kernels. Concentrated, self-perpetuating power corrodes a society the same way regardless of how it was first earned.
Honor the Genius, Retire the Throne
The framework’s answer is not to pretend talent isn’t scarce. Pretending scarcity away is how you get shortages and resentment. The answer is to separate two things the current system fuses together: the reward for rare contribution, and permanent power over everyone else.
The Foundation handles the first half. The output of all this genius, the abundance it produces, gets treated as public infrastructure rather than a private toll road. You don’t have to be Noam Shazeer, or employed by him, to get food, housing, healthcare, energy, and compute. The genius makes the pie bigger. The pie is not his to ration.
Impact, the framework’s contribution currency, handles the second half, and it does something the dollar refuses to do: it decays. Invent the Transformer and you earn an enormous spike of recognition and standing. Sit on that one invention for fifty years and watch it bleed away at a few percent annually, because Axiom IV of the framework says power must decay. You are honored lavishly for what you did. You are not crowned for it.
We have already sketched what this looks like in practice. Founder Status, the deal that converts yesterday’s billionaires into stakeholders, hands them a high starting reserve of standing that then drains faster than everyone else’s, by design, so the head start is gone within a generation. Apply the same logic to the cognitive elite. The researcher who cracks the next architecture should get the spike. The spike should not become a hereditary throne. A civilization that lets its smartest people get rich is healthy. A civilization that lets its smartest people get permanent is just feudalism with better GPUs.
Is Talent Even the Last Scarcity?
The podcast conversation that surfaced this topic called genius “the last scarcity.” It is a great headline. It is also probably wrong, and the reason it is wrong is the most important part of the story.
Talent is the current bottleneck, not the final one. The entire AI project is a bet that machines will eventually do frontier research too. Anthropic spent 2026 warning that its own models were approaching recursive self-improvement and calling for a global pause, which is another way of saying the talent bottleneck has an expiration date. The day a model can reliably originate the next Transformer, elite human researchers commoditize the way human translators and human coders already have, and the hydra grows its next head.
Where does the bottleneck go then? Probably to energy, the wall the AI buildout keeps slamming into, where every breakthrough is gated by how many gigawatts you can plug in. After energy, maybe coordination. After coordination, maybe wisdom, the genuinely unautomatable question of what all this abundance is even good for. The Labor Cliff was never a single event. It is a wave that breaks across one category of human work after another, and the scarce input keeps relocating one rung up the ladder of abstraction.
That migration is precisely why you cannot fix this by winning the current round. Whoever owns the bottleneck today (compute barons yesterday, talent barons now, energy barons tomorrow) will always be tempted to freeze the music while they are on top. The only durable design is one that assumes the bottleneck will keep moving and refuses to let whoever holds it ossify into a permanent ruling class. Honor the holder. Decay the throne. Repeat each time the hydra regrows a head.
That is the whole trick, and the AI talent war of 2026 is the dress rehearsal.
The Takeaway
Google paid $2.7 billion for one researcher and still lost him. That number is not a curiosity. It is a measurement of how concentrated the value of frontier AI has become, and a warning about who is positioned to capture it. The most valuable resource of the abundance age is the one resource abundance can’t yet manufacture: the judgment to build the next thing.
The book’s wager is that we can have the genius without the aristocracy. Pay the rare mind generously, refuse to let the payment harden into a dynasty, and keep the abundance it produces in the commons where it belongs. Get that design right, and the talent paradox is a phase we pass through. Get it wrong, and the people who automated your job will own the machine that did it, permanently.
Unscarcity is the blueprint for getting it right. Start with the crossroad we are standing on, or read how the Foundation turns private genius into shared infrastructure.
