AI Just Cracked an 80-Year-Old Math Problem (for About $1,000)
A famous math problem stumped the smartest humans alive for the better part of a century. An OpenAI model just solved it — not by following the textbook, but by trying the “improbable” thing no human had the patience to chase. The bill came to roughly $1,000 in compute.
Ben Cohen · The Wall Street Journal · May 30-31, 2026 · Capital Wealth analysis by Sean Anees Saifi
An 80-year-old problem, finally solved.
“You May Want to Make Sure You Are Sitting Down”
That’s how one of the world’s leading mathematicians opened a recent note. The reason: an OpenAI model resolved a version of the “unit-distance problem,” one of the famous Erdős problems that had stumped humans since 1946. Mathematicians, who are professionally allergic to hype and demand proof before they’ll accept that water is wet, lost their collective minds.
Fields Medalist Timothy Gowers said that if a human had written the proof and submitted it to the Annals of Mathematics, he’d have recommended acceptance “without any hesitation.” A former OpenAI researcher estimated the whole thing took less than 32 hours and about $1,000 in tokens — “a bargain for a result of this caliber.”
Why AI Won Where Humans Lost
The researchers gave three reasons, and they’re weirdly profound. First, the solution was deeply counterintuitive — humans kept trying to prove the old conjecture; the AI disproved it by experimenting with strategies that looked improbable. Second, humans specialize while AI synthesizes — the model pulled together two fields of math (algebraic number theory and discrete geometry) that have about as much in common as the marathon and the pole vault. Third, and most human of all: AI doesn’t quit.
“It’s the kind of idea you try for a bit, it doesn’t work, and you think maybe you were just too hopeful,” said a Harvard statistician. “So you give up and move on.” AI doesn’t move on. It just keeps plugging away — no breaks to eat, sleep, answer emails, pick the kids up from school, or watch the Knicks. Its “chain of thought” on this one ran more than 75,000 words. (We’ve all had that one coworker.)
What This Means For Your Money
Stories like this are why the AI infrastructure trade has legs: the capability is improving fast and the cost of using it is collapsing. That’s a powerful combination, and it’s showing up in the earnings of the companies that sell the compute — Nvidia (NVDA), the data-center power names, and the cloud providers.
But the investing principle is one of healthy skepticism. A breakthrough is not a business model, and “this changes everything” has emptied a lot of accounts. We own the AI buildout through diversified positions in the infrastructure layer — power, cooling, chips — rather than betting the farm on whichever model is winning this month. The mathematicians demanded proof before they believed. Smart investors do the same with valuations.
