Front Burner — "If AI is a bubble, what happens when it pops?"
Date: November 17, 2025
Host: Jayme Poisson (CBC)
Guest: Paul Kedrosky (Partner, SK Ventures; Research Fellow, MIT Initiative for the Digital Economy)
Overview
This episode explores mounting fears that the current artificial intelligence (AI) boom is a financial bubble, scrutinizing what’s changed in the world of AI to fuel these anxieties, the real risks within the industry, and what could happen to the global economy if the bubble bursts. Jayme Poisson interviews venture capitalist and academic Paul Kedrosky, who draws parallels between the AI craze and historic bubbles—warning that, if it pops, the fallout may be far-reaching and tangled in not just technology, but also debt, real estate, and government involvement.
Key Discussion Points & Insights
1. Is AI a Bubble? The Two “Bubbles” Explained
[02:54]
- Kedrosky distinguishes between the “bubble” in AI’s technology and the “bubble” in investment and infrastructure.
- The technology is real and transformative (“It's actually a really useful technology. There’s going to be lots of uses for it…”).
- The bubble is in the mania and overspending, particularly on data centers (“…this massive buildout underway…is very expensive…”), driven by investor excitement far outpacing real demand.
2. Shifting Funding Models: From Cashflow to Debt
[04:14]
- The AI industry's infrastructure buildout is now funded by massive, opaque debt rather than operational profits:
- Tech giants used to pay for data centers “out of their profits.”
- Now, with estimates of $4 trillion needed in the next five years, companies finance expansion using private credit and special purpose vehicles—off-balance-sheet structures, which "makes things much riskier."
- Kedrosky likens the system to a “shell game,” moving liabilities out of plain sight.
Quote:
“These are done in this very opaque and sometimes mysterious way… so when it comes time to say, well, how much exposure do you have to all of this? I get to say, hey, not my table, I don't own that, it’s over there. Which is kind of a shell game obviously.”
— Paul Kedrosky [05:21]
3. Circular Money Flows and “Vendor Financing”
[06:17]
- The episode references a Bloomberg diagram with Nvidia at the center, illustrating cash flows looping between chipmakers and AI companies.
- Kedrosky compares this vendor financing to the dot-com era, calling it "circular" and a way to exaggerate growth and risk:
- Companies give credit to their own customers to buy more of their products, which “masks the size of things and creates this illusion of even faster growth.”
Quote:
“It becomes very circular and very hard to tell where the reality is anymore, because all the money just goes… round and round and round.”
— Paul Kedrosky [07:33]
4. Government “Backstopping” and Fears of a Fannie/Freddie Repeat
[08:32]
- OpenAI’s CFO floated the idea of seeking government guarantees or subsidies for AI funding, sparking panic.
- Some see parallels with government-supported housing lenders (Fannie Mae, Freddie Mac) before the 2008 crisis.
- Governments in Canada, the US, Europe, and China are all considering AI as a “sovereign” asset worth strategic intervention.
Quote:
“That’s what Fannie and Freddie were… making it easy for people to make really bad loans. Well, we don’t want to reenact that again, is the thinking.”
— Paul Kedrosky [09:42]
5. Canada’s Push for “Sovereign AI” and the Logic of Subsidies
[11:25]
- Canadian political leaders frame data centers and AI as the new “economic payday,” arguing for major public investment.
- Kedrosky cautions this narrative is misleading—AI capacity could be supplied globally with a handful of data centers, and domestic investment mainly shifts private costs onto public books.
Quote:
“I defy you to come up with a rationale for why it’s strategically important to spend billions of dollars on data centers other than it takes the cost… off the balance sheet of the technology companies themselves.”
— Paul Kedrosky [14:18]
6. Comparisons to Past Bubbles: Bigger, Broader, Riskier
[15:42]
- Unlike earlier bubbles, the AI infrastructure mania combines tech, credit, real estate, and government all at once. That, Kedrosky argues, compounds systemic risk.
- The bubble is not just about tech stocks but is intertwined with national policy, land values, public funds, and ever-more complicated debt structures.
Quote:
“This one has all of the pieces that made the worst bubbles in the last 150 years into one neat package.”
— Paul Kedrosky [16:51]
7. Will Today’s Overbuild Become Tomorrow’s Backbone?
[17:04]
- Poisson notes that overbuilding in the dot-com era (e.g., fiber optics) paid off later. Kedrosky flatly disagrees.
- Railways and canals lasted a century; GPU-based data centers become obsolete in years, often even faster if used for AI “training.”
- Overbuilt GPU farms won’t be useful long-term (“You’re going to have to make all those capital expenditures a second time.”).
Quote:
“The notion that we can build all of this and… figure it all out four or five or six years down the road? No, it’s actually the reverse.”
— Paul Kedrosky [19:16]
8. Potential Consequences: Recession, Wealth Loss, Debt Contagion
[19:43]
- If the bubble bursts, the pain could be deeper than 2008 or 2000 because AI is so central to current markets and growth.
- Over half recent S&P 500 gains are AI/tech-related.
- Unwinding would swamp markets in negative wealth effects, recession, and potentially a debt crisis as private credit is “sliced up and ends up everywhere.”
Quote:
“The idea that somehow we can remove all of this… and have that not have consequences, is naive.”
— Paul Kedrosky [21:42]
9. Are Warning Signs Already Here?
[22:49]
- Poisson presses on timing. Kedrosky notes that credit default swaps on major tech names like Oracle have doubled in three months—historically “unthinkable.”
- This signals the market now sees tech company debt as risky, and “it’ll get riskier before the wheels really come off.”
Quote:
“Technology companies’ risk of defaulting on debt could double in three months. There you go. There’s a sign that… something dramatic has changed.”
— Paul Kedrosky [23:32]
Notable Quotes & Memorable Moments
-
On bubble mechanics:
“It’s sort of the bubble of bubbles… all the things that made the worst bubbles in the last 150 years into one neat package.”
— Paul Kedrosky [16:51] -
On government’s role:
“Now, saying you’re going to backstop loans maybe have been a one step too far. But listen, the US… is really pushing the idea this is the new global international arms race… so it really is just saying the quiet part out loud.”
— Paul Kedrosky [10:40] -
On survivability of data centers:
“60% of the cost of a data center is the chips and their lifespans are very short… So, most of what you spent previously will be useless…”
— Paul Kedrosky [19:29] -
On the unwinding:
“I think it’ll get riskier before the wheels really come off. But is it more than two years away? A year away? I doubt it.”
— Paul Kedrosky [23:35]
Timestamps for Key Segments
- [02:54] – Defining the two types of AI bubbles
- [04:14] – Shift from profit-funded to debt-funded expansion
- [06:17] – The “circular” vendor financing problem (Nvidia example)
- [08:32] – OpenAI CFO’s comments on government backstopping; policy alarm
- [11:25] – Canada’s sovereign AI investment pitch
- [15:42] – Distinguishing the current bubble from 2008 and others
- [17:04] – Will today’s AI infrastructure be tomorrow’s asset?
- [19:43] – Potential impacts if the bubble pops
- [22:49] – Real-time signs: tech credit spreads widening
Tone and Style
- The discussion is analytical, occasionally urgent, with Kedrosky often punctuating tough economic analysis with wry, matter-of-fact observations.
- Jayme Poisson maintains a probing, curious tone, frequently surfacing analogies and counter-arguments, drawing out clarification for a non-specialist audience.
Final Thoughts
The episode starkly illuminates how the AI industry’s explosive growth comes with structural risks eerily reminiscent of past crises—but now concentrated and amplified. Government subsidies, off-balance-sheet debt, and the ultra-fast obsolescence of AI infrastructure mean that, if the bubble pops, repercussions could reach further and cut deeper than most observers expect. As Kedrosky warns, “The idea that…we can remove all of this… and have that not have consequences, is naive.”