Episode summary
How to Navigate the AI Debt Bubble
Money For The Rest Of Us
· · 29 min
Summary
The host argues that the AI buildout has entered a new phase: after being funded largely by free cash flow, hyperscalers and related companies are now relying heavily on debt to finance data centers, power infrastructure, and compute. Estimates cited from JP Morgan, McKinsey, and Goldman Sachs put total AI-related capex at $6 trillion to $7 trillion, with JP Morgan suggesting 75% could be debt-funded, implying roughly $4 trillion of new borrowing. The host says that scale is large enough to matter for corporate credit markets, even though most of the debt is still investment grade.
The episode compares the AI boom to the capital cycle in railroads, canals, and the internet, where excess investment eventually produced oversupply, widening spreads, and losses even though the infrastructure remained. The host says AI demand is real, but returns may not justify the level of spending if chip cycles shorten, competition intensifies, or open-weight models catch up to frontier models.
Most of the AI debt is said to sit in investment-grade bonds, bank lending, private credit, insurers, and off-balance-sheet SPVs, while leveraged loans and CLOs have relatively little direct exposure. The host warns that investors may have hidden exposure through ETFs, active credit funds, and REITs, and says current credit spreads are still too tight to compensate for the risks.
Key takeaways
- The host estimates AI infrastructure could require $6 trillion to $7 trillion of capex, with about $4 trillion potentially funded by debt.
- Investment-grade bonds are the main public market channel for AI debt, while bank loans, private credit, insurers, and SPVs also carry exposure.
- The host says leveraged loans and CLOs have little direct AI-data-center exposure because the debt mix does not fit variable-rate financing well.
- A key risk is that short-lived chips and uncertain demand are being matched with long-dated debt, creating refinancing and obsolescence risk.
- The host says current corporate credit spreads are still narrow, so investors are not being paid much to take AI-related credit risk.
Notable quotes
“there's some mismatches in terms of the longevity of the chips that are used to to run the compute in the data centers versus the the length of the maturity of the debt”
“they estimate, and this is JP Morgan specifically, 75% of this CAPEX will be funded with debt. That's about $4 trillion of new debt outstanding linked to the AI infrastructure buildout”
“The largest by far and is investment-grade bonds. And they are found in in ETFs”
“One of the criticisms has been circular finance”
“right now is not a great time to be taking a lot of credit risk because you're not being compensated for it”
How the episode opens
I recently got an email from a member of money for the rest of us plus and he was worried to what extent is all this debt that is being used to finance AI infrastructure, new data centers, all the backbone for this AI bubble we're in. As hyperscalers build out all this infrastructure, all this compute for frontier models, for non-frontier models, as we are in this grand experiment to implement AI. As individuals and businesses use AI, figure out how this new general technology can help us in our daily lives, can help businesses boost productivity, to innovate more.
Now, I've talked a lot about AI and the AI bubble and in the the very high valuations of AI-related companies. And as I record this, few episodes ago, we did a episode on SpaceX, its IPO and the exorbitant valuation of that IPO. Now, it's selling for 20% less than the IPO price and other AI-connected companies had have sold off.
We're we're getting near almost a 10% type correction. Now, this is a really volatile area. But this member was concerned, well, where's all this debt going that is being used to finance these data centers?
Because something has happened in the past 18 months. Previously, the these companies like Meta and Amazon and others generated enough free cash flow to essentially build out infrastructure, data center infrastructures. But now that the level of capital expenditures is so high that they're having to borrow.
And a lot a lot of these companies are are investment grade and they're out borrowing trillions of dollars essentially. Or will be trillions of dollars to to fund this grand AI experiment. And and there's some mismatches in terms of the longevity of the chips that are used to to run the compute in the data centers versus the the length of the maturity of the debt.
From the episode’s captions: roughly the first two minutes, lightly cleaned. Keep watching on YouTube →
Topics
- ai capex boom
- ai-related debt size
- fixed income vehicles
- ai debt risk transfer
- ai infrastructure buildout risks
Watch the episode
Watch “How to Navigate the AI Debt Bubble” on YouTube
Summary, takeaways and quote selection by Octagon. Quotes are short excerpts; the full conversation is in the episode.
More from Money For The Rest Of Us
-
10X Your Productivity While Doing Your Best Work -
Stock Pickers Keep Losing, Who's Winning? -
Stocks Are Cheaper Than You Think - It All Depends On What You Measure -
What If You Only Invested at Stock Market Peaks? Or Invested at Stock Market Bottoms? -
Can You Really Earn 68% per Year Selling Puts? How Option Income Strategies Work -
The Financial System Will Be Hacked and What To Do About It
Money For The Rest Of Us, this episode and its artwork belong to J. David Stein. Octagon is not affiliated with the show; this page is an independent summary.
