Podcast Summary: "Why OpenAI Won’t Survive an AI Crash"
Podcast: Solutions with Henry Blodget
Host: Henry Blodget (Vox Media Podcast Network)
Guest: Paul Kedrosky, Venture Capitalist, Writer, and Researcher
Date: January 26, 2026
Main Theme
This episode examines the current AI investment boom, drawing parallels with past economic bubbles (dot-com, railroads, telecom) and predicting a looming “AI crash.” Guest Paul Kedrosky argues that OpenAI and other “frontier model” AI companies are unlikely to survive, while the seismic infrastructure investment powering today’s tech optimism will result in cheap, ubiquitous AI—fundamentally transforming the future.
Key Discussion Points & Insights
1. The Nature of the Current AI Boom
- Parallels with Past Bubbles: Kedrosky compares today's AI/data center build-out to past boom-bust cycles, highlighting similarities and differences.
- “We've got this lunge towards capital expenditure to create this... fabric underneath society. Much like what happened with canals... or like the railroads... or like telecom and the fiber build out around 2000.” (02:15, Paul)
- All Four Bubble Elements Intersect: Real estate, technology, loose credit, and government policy are colliding in this cycle, making it unprecedented in American history.
- “This is the first bubble in U.S. history that's at the intersection of all four... Each person thinks they're acting in their own interest. And the combination of them all is kind of crap.” (03:54, Paul)
2. Winners and Losers in the AI Cycle
- Frontier Model Companies (e.g., OpenAI) Likely to Lose: Due to high capital requirements, slowing innovation, and rapid commoditization.
- “I'm on the record as saying I don't think there'll be a ChatGPT 8. I don't think OpenAI will be around long enough to do that...these companies are going to be commoditized. It's already happening.” (05:30, Paul)
- Big Tech’s Role Misunderstood: Current tech giants (“Mag 7”: Nvidia, OpenAI, etc.) are unlikely to be long-term winners; winners will emerge later, as with previous tech build-outs.
- “Almost no one in the current Mag 7 strikes me as a likely winner here…that’s the mistake people make.” (06:47, Paul)
- Infrastructure Providers Take the First Hit: Data centers and supporting vendors are heavily exposed; many rely on debt, compounding risk when token prices collapse.
3. Economics of AI: The Token Problem
- Token Price Collapse: The cost structure underpinning AI companies is structurally unstable.
- “It's horrific. It's good for the consumer...but it's a lousy business from the standpoint of being the producer of these things because they're facing this exponential declining cost.” (10:35, Paul)
- Data Center Economics: As token prices fall, revenues struggle to keep up with fixed debt/interest costs, leading to potential defaults and write-offs.
- “The economics of data centers...were bad in the first place and getting worse because of this collapsing token economics problem.” (12:15, Paul)
4. Where Are We in the Cycle?
- Current Stage: Late-Bubble Exuberance: Like 1999 for dot-coms—rapid new IPOs, speculative risk-taking, retail investor FOMO.
- Turning Point in Sentiment: October 2025 marked a shift away from endlessly rewarding capital expenditure; market is demanding actual ROI.
- “There was a regime change…people stopped being rewarded for exec of capex.” (15:40, Paul)
5. The Broader Economic Impact
- AI Infrastructure Is Driving GDP Growth: Over half of US GDP growth in 2025 was due to data center capex, distorting economic fundamentals.
- “More than half of it was coming from this anomalous spending on…the single thing that we call data centers.” (20:34, Paul)
- Hidden Risks: Misunderstandings about what’s driving macroeconomic numbers can lead policymakers astray.
6. The Party Ends: What Causes the Crash?
- Collapse Triggers: Spiking interest rates, drying up of debt, cascade of failed IPOs, and a realization that core AI revenues can't keep pace with declining costs.
- “Things that can't go on must stop…It must end because that's what happens with all of these capex frenzies.” (38:36, Paul)
- Widespread Fallout: Not just tech giants, but insurers, sovereign wealth funds, regional economies (lured by data center promises), and the construction industry will all be affected.
7. Commoditization and Diminishing Returns
- LLMs Are Becoming Interchangeable: Little improvement between new models; major tech CEOs tacitly admit models are commodity products.
- “You even heard…Satya Nadella…the Microsoft CEO…you're like a huge investor at OpenAI. You just said one of your largest investments is kind of meh to you. I mean, what the hell?” (32:39, Paul)
- Overbuilding for Imaginary Demand: Most capacity is used not for serving consumers, but for endless, often unproductive model training and internal tool use.
8. Long-Term Optimism: What Survives Will Transform Society
- Cognition Becomes (Nearly) Free: Post-crash, the infrastructure unleashes cheap intelligence for all kinds of applications—medicine, science, robotics.
- “Cognition is going to become free…It's a cheat code for unlocking cognition for a huge population of people…” (46:24, Paul)
- Analogy to Past Infrastructures: Like cheap power, cheap cognition opens new domains previously closed by cost or expertise.
- Social Disruption and Workforce Realignment: The decline of “training-intensive” occupations will be painful; vocational and trades may flourish.
- “We've got this huge misallocation of people towards occupations that are shrinking…that’s where everything has to change.” (49:13, Paul)
- Offsetting Demographic Headwinds: Aging populations will push adoption of robotics and AI augmentation.
Notable Quotes & Memorable Moments
-
On the Cycle’s Inevitability:
"You don't know if you've built enough until you've built too much...People coming into this thinking that there's some kind of rational, clean way...are naive."
(27:48, Paul) -
On Token Economics:
“Imagine being General Motors and the cost of cars is falling 60% year over year. I better find some new planets to sell cars on because I'm in big trouble selling on this planet.”
(10:35, Paul) -
On Retail Investors’ Fate:
“There's a very cynical answer...if I can unload at the IPO at a significant premium…then I don't really care what their underlying unit economics look like. That's a problem for retail investors in future.”
(24:48, Paul) -
On Overbuilding:
“You could satisfy all of global inference, consumer chat related global inference from a single data center in Virginia.”
(30:19, Paul) -
On the Data Center Bubble’s Reach:
“Regional economies [will be] hurt by this that end up with these white elephants, data centers that are either underused or unused.”
(44:19, Paul) -
On Optimism Post-Crash:
“It's this fabric that really radically changes the nature and cost of everything we do...the atomic units of cognition...are these tokens. Token prices are collapsing. They will be ubiquitous everywhere and essentially costless.”
(46:24, Paul)
Timestamps for Important Segments
- AI boom & bubble comparisons: 02:12 – 05:04
- Winners/losers in the AI buildout: 05:30 – 07:39
- Token economics explained: 09:16 – 11:08
- Data center risk & debt cycle: 11:08 – 12:38
- Stage of the bubble & market shift: 13:02 – 16:24
- AI’s impact on GDP & risk misperception: 20:15 – 22:29
- Crash mechanics and historical echoes: 38:05 – 41:12
- Broader economic/social fallout: 41:12 – 45:44
- Long-run positive outlook: 46:13 – 51:34
Tone & Language
The episode maintains a sardonic, skeptical tone—typical of Paul Kedrosky’s writing and conversation—but concludes with genuine optimism about the potential of cheap cognition. The conversation is rich in analogies (toll booths, car markets, Oscar Wilde), accessible explanations, and a refreshingly frank view of incentives and rationality in tech and finance.
Final Takeaway
Near-term: Expect “AI crash” ramifications to ripple far beyond OpenAI and its peers—affecting economic growth, regional development, and retail investors.
Long-term: The massive (and wasteful) infrastructure boom will yield a world where intelligence is accessible, affordable, and transformative for nearly every domain of human progress.