More or Less · Topics · Episode #168

The St. Petersburg Paradox: Sam's Theory for Why Nobody Can Actually Price the AI Bubble

Clip from the full recording, 25:08–30:27 · download mp4

After a Blackstone exec openly admits "we don't know" who pays for AI data-center debt, Sam breaks out a 300-year-old math puzzle to explain why the AI narrative has no coherent price — and accuses the industry of knowingly selling a story that was always going to evolve.

Jessica shares a live anecdote from The Information’s AI event: at a Blackstone panel on data-center financing, a subscriber asked who’s going to pay for the debt behind the AI infrastructure buildout, and the executive’s answer was simply, “we don’t know.” Sam says that moment captures the whole problem, and reaches for the St. Petersburg Paradox — a classic probability puzzle he’s giving a talk on next week — to explain it: in a coin-flip game where every round doubles the payout, the mathematically correct price to play is infinity, yet no rational person would actually pay anywhere close to that, because the math describing unbounded upside breaks down against real-world risk of total loss.

His point: AI’s economic narrative has been pitched, explicitly or not, as exactly this kind of infinite-payout game — whoever reaches AGI first captures unbounded value — which means markets literally cannot price it using normal tools. Dave connects this to a recent essay by OpenAI’s head of research suggesting AGI isn’t a single point in time but a gradual, decades-long diffusion of capability, which Sam says is a very different (and much less lucrative-sounding) story than what was being sold three years ago, when the pitch was a single moment of winner-take-all arrival. Jessess adds that this narrative shift literally shows up in contract language — AGI clauses were written into deals like Microsoft’s because agreements were being signed faster than anyone could actually define the term. Sam’s closing argument: this isn’t some unknowable mystery evolving in real time — it’s a predictable result of selling a narrative, and the real question is how much credit the industry deserves for being “intellectually honest” about updating that story versus simply having moved on from one era of salesmanship to the next.

Key points

  • At a Blackstone panel, an executive responded to a question about who pays for AI data-center debt with a blunt "we don't know."
  • The St. Petersburg Paradox: in a coin-flip game that doubles payout every round, the mathematically 'correct' price to play is infinity, yet virtually no one would pay close to that in practice — because the math of unbounded expected value breaks down against real catastrophic-loss risk.
  • Sam's argument: AI's winner-take-all AGI narrative was effectively an infinite-payout pitch, which is precisely why markets have struggled to price it using normal financial tools.
  • Dave cites a recent essay from OpenAI's head of research arguing AGI is a gradual, decades-long diffusion rather than a single point in time.
  • Jessica notes that AGI definitions were written directly into financial agreements (e.g., with Microsoft) because deals were moving faster than anyone could pin down the term.
  • Sam's closing point: the shifting AGI narrative isn't some mysterious evolution — it was a predictable outcome of salesmanship, and the real question is how much credit to give an industry for 'discovering' things it should have known all along.

Where they landed

SamThe entire AI investment narrative has functioned as an unpriceable infinite-payout bet, and the shifting AGI definition reflects salesmanship more than genuine discovery.
JessicaCorroborates with on-the-ground reporting (Blackstone, Microsoft contract language) that the industry itself can't answer these financial and definitional questions.
DaveConnects the paradox to OpenAI's own evolving public position that AGI is gradual rather than a single moment.

Quotes

“A subscriber said, who's gonna pay for the debt? And the executive at Blackstone said, we don't know.”— Jessica
“If it's an infinite series, the logical answer is you actually would pay infinity to play the game.”— Sam
“The problem with a lot of the AI narrative is that when you talk about infinite outcomes, there is no price. There is no way to model it.”— Sam
“The idea that there was some magical AGI point in time to be won by one company was obviously ridiculous, but it was a hundred percent what was being sold.”— Sam

Suggested tweets

A Blackstone exec was asked who pays for the AI data-center debt. The answer: "We don't know." Jessica Lessin says that's exactly where the industry is right now.

Tweet this →

Sam Lessin breaks out the 300-year-old St. Petersburg Paradox to explain why nobody can actually price the AI bubble: when the payout is framed as infinite, there is no correct price.

Tweet this →

"The idea that there was some magical AGI point in time to be won by one company was obviously ridiculous, but it was a hundred percent what was being sold." — Sam Lessin

Tweet this →

All topics · Full episode #168 + transcript · Subscribe on YouTube · Spotify