Sep 23, 2026 · 1h 13m
AI spending props up growth while recession risks accumulate
The Hidden Recession Beneath The AI Bubble w/ Paul Kedrosky
The episode argues that AI infrastructure investment may be masking weak productivity while increasing exposure to debt, power constraints, and financial shocks.
- 1AI capital spending is boosting GDP, debt issuance, trade, and market activity without proving that underlying productivity has improved.
- 2Data-center economics face pressure from expensive borrowing, scarce power, concentrated customers, and uncertain returns on inference.
- 3A failed debt auction or other unexpected financial break could abruptly end the AI boom and deepen a broader downturn.
Don't miss
Paul Kedrosky explains how an unexpected break in debt markets, rather than a gradual slowdown, could end the AI boom.
The brief
Ed Zitron and economist Paul Kedrosky examine how AI capital spending has become a major force in GDP growth, debt issuance, trade, and financial markets—possibly masking weakness elsewhere.
Kedrosky argues that infrastructure spending can distort productivity statistics, with energy costs and examples such as hospital AI scribes complicating claims that AI is already transforming output.
The central risk is a feedback loop: weak productivity and layoffs meet rising interest costs as AI-related borrowing competes with U.S. Treasuries for investors.
Physical limits—including power shortages and data-center costs—intersect with financial limits such as falling bond values, collateral pressure, and hyperscalers’ dependence on continued borrowing.
The standout warning is that the boom may not unwind gradually: a failed debt auction, sovereign issuance event, or other unexpected financial break could trigger the snap.
Kedrosky closes by connecting concentrated customers, uncertain inference profits, delayed liquidity, and investor FOMO to Hyman Minsky’s theory of bubbles and sudden breaks.
What was said on this episode
38 statements · 5 positive · 31 negative · 1 mixed · 1 neutral
AI capital expenditure contributed 30%–70% of GDP growth over 12–18 months.
“It's been consistently between 30% and 70% of GDP growth.”
Listen at 5:46
AI-related debt issuance began affecting the U.S. Treasury market.
“AI CapEx issuance actually had become so large that it was beginning to bleed into the treasury market.”
Listen at 7:31
AI-related debt issuance caused a Treasury sell-off during the market freakout.
“one of the reasons why we had this treasury market freakout is because of the unprecedented amount of AI-related debt being issued, and that in turn was causing a sell-off in treasuries.”
Listen at 7:38
AI financing became the largest segment of investment-grade and high-yield markets.
“it's now the largest piece of the investment-grade marketplace. It's now the largest piece of the high-yield marketplace. It is literally taking over global debt markets.”
Listen at 8:49
AI capital expenditure financing began competing with the dollar in financial influence.
“AI CapEx as a currency was beginning to compete with the dollar in a really loose sense.”
Listen at 9:28
AI-related trade accounted for almost 55% of global trade growth.
“Almost 55% of global trade growth was... AI-related.”
Listen at 11:42
AI growth underlies both major components of recent U.S. inflation.
“both pieces of inflation have underneath them as predicates this incredible and unprecedented growth in AI”
Listen at 14:13
The United States will not default on its debt because it issues dollars.
“The U.S. prints dollars. They are not going to default.”
Listen at 17:17
Hyperscaler debt issuance raises Treasury yields through competition for yield-sensitive investors.
“This is literally an artifact of competition at the margin from hyperscaler issuance in a newly yield-sensitive market, causing people to say, I'd rather own X than Y. And that has sovereign consequences for the U.S.”
Listen at 17:28
AI is being treated as both the source of debt problems and the productivity solution.
“AI is not just the problem, it's the solution.”
Listen at 19:15
Recent productivity gains largely reflect capital spending rather than labor productivity.
“right now that's almost entirely an artifact of... of capital spending, not labor.”
Listen at 20:52
AI-scribe hospital outcomes may reflect wealth and patient-health selection effects.
“It's exactly what you would expect to have happen if you found a new variable that filtered hospitals for healthy patients and wealthy hospitals.”
Listen at 22:07
Excluding AI-related inflation and energy effects, U.S. prices are falling about 0.25%.
“by my math, the U.S. is actually in a deflationary mode, about a quarter of a percent.”
Listen at 30:32
The U.S. is positioned for a prolonged Japan-like balance-sheet recession.
“We've got all the pieces in place right now for a very long balance sheet recession analogous to what happened in Japan in its lost decade.”
Listen at 31:38
Some AI infrastructure firms may become insolvent when refinancing debt within five years.
“they face a different problem, which is as they try to roll over their debt, it may come on terms over the next five years. force them into some species of insolvency.”
Listen at 32:35
Corporate deleveraging will intensify economy-wide hiring declines.
“the effects overall on hiring will be even more dramatic because companies will be focused entirely on deleveraging.”
Listen at 34:06
The economy will show increasing weakness over the next year.
“over the next year or so, we'll start seeing increasing signs that the economy is much weaker than people expect.”
Listen at 34:48
The economy will probably enter a longer-than-expected recession.
“we'll probably be heading into a relatively longer recession”
Listen at 34:55
High sovereign debt is increasingly limiting fiscal stimulus capacity.
“Fiscal policy is being increasingly rendered impossible because of the indebtedness of major sovereigns around the world.”
Listen at 36:08
AI investment outcomes imply a multi-year recession bordering on depression.
“You're into a multi-year recession, and then because of this balance sheet problem, something that teeters on the edge of being a depression”
Listen at 37:08
Available power cannot support data centers at the scale implied by GPU sales.
“the math doesn't work in terms of our ability to power up data centers on a scale commensurate with the numbers of GPUs being sold in the data centers.”
Listen at 38:19
Data-center financing economics will break within six to twelve months.
“we're within 6 to 12 months on the outside of that aspect of it all breaking”
Listen at 40:32
Falling hyperscaler bond values reduce holders’ ability to conduct other economic activity.
“As this debt becomes less valuable, it reduces my ability to other things in the economy as a functioning economic actor.”
Listen at 41:28
The AI-finance system will likely break unexpectedly, possibly through a failed Treasury auction.
“My guess is it breaks in somewhere, because this is the way these things tend to go, it breaks in somewhere unexpected, like we have a botched 10-year treasury auction.”
Listen at 42:53
A failed Treasury auction would widen spreads and raise ten-year rates.
“The spreads on 10 years would blow out, meaning they would get much wider, they would get much larger, and the rates would go higher”
Listen at 43:18
A failed Treasury auction could make data-center debt financing economically impossible.
“it would explode your ability to raise debt at any economic price in the world of data centers”
Listen at 43:25
The AI-finance system will inevitably crack.
“There's no question in my mind that it'll crack”
Listen at 45:05
The eventual break will involve a technical problem with major sovereign debt issuance.
“it's going to be a sovereign issuance problem, where someone's going to have a... technical issue with a major issuance at around probably sort of a 10-year duration”
Listen at 45:11
Hyperscalers will continue borrowing because unilateral restraint creates competitive disadvantage.
“on goes the race because then everyone sees that and they all defect and we're right back where we were before.”
Listen at 51:50
The median investment-grade data-center debt hurdle rate is about 7.2%.
“The median is now around 7.2%.”
Listen at 52:09
OpenAI and Anthropic will not win the industrial-inference market.
“it's fairly straightforward to see who the winners are going to be, and it ain't going to be OpenAI and Anthropic.”
Listen at 55:22
A late-year IPO announcement has at least a 50% chance of slipping into the following year.
“my general thesis is that even in a good year, if you say you're going to go public late in the year, odds are at least 50% it won't happen that year”
Listen at 56:07
Anthropic’s repeated IPO delays are unusual for a marquee company.
“this is really unusual.”
Listen at 58:10
Large-scale public-market financialization of frontier AI companies will trigger a break.
“that's the moment where I think everything starts to crack and break”
Listen at 1:01:35
A sufficiently small public float can support almost any apparent valuation.
“at a small enough float, you can do anything. I can get any valuation I want in the public markets.”
Listen at 1:03:42
Higher OpenAI private valuations will likely lead to smaller, price-manipulative IPO floats.
“my expectation is the higher the valuation we see today, the smaller the float will be, and the more cynical the eventual IPO will be in terms of trying to manipulate the price.”
Listen at 1:04:03
Misinterpreting AI-driven economic data will make the eventual crisis more severe.
“the crack-up becomes sharper.”
Listen at 1:06:06
Investors will continue entering the AI market before an eventual crash.
“people, despite all of the things I'm saying, are going to walk right into this and then the crack will happen”
Listen at 1:07:14
Statements are attributed to the speaker as said on the episode and reflect their view at the time, not PodLume's. They are not advice.