PIMCO Pod Sep 21, 2026
With Lotfi Karoui, Managing Director and Multi-Asset Credit Strategist at PIMCO
Roughly $1 trillion of hyperscaler capital spending is expected this year, and Lotfi Karoui's claim is that the way it is paid for changes nothing about where real interest rates end up.
The popular version of the story says otherwise. AI companies are issuing unprecedented amounts of debt into a Treasury market that is already supplying heavily, both draw on the same pool of investor capital, and so yields have to rise to clear it. Karoui's answer is that this is not how crowding out works in the textbook, and that when he went looking for the effect in the data he could not find it.
"Can private-sector financing, however large, crowd out a $32 trillion Treasury market?"
Karoui is a managing director and multi-asset credit strategist at PIMCO, and was chief credit strategist at Goldman Sachs before he joined. The piece is PIMCO's own Credit Market Lens, and the test at the center of it is an event study he built around a year of large AI-related bond deals.
The full episode is covered here so you can skip it.
Here are the 8 arguments that matter.
Key Takeaways
The textbook version of crowding out runs the other way — government borrowing absorbs savings and squeezes out private investment, not the reverse
The AI capex boom can still push real yields up, but through desired saving against desired investment, not through anyone selling Treasuries to buy AI paper
A dollar of debt-financed capex and a dollar funded from retained earnings consume the same labor, power and construction
Cash already sitting on a balance sheet is not "pre-funded" in macroeconomic terms — spending it converts saving into investment
Six deals over the past year were large or badly timed enough to surprise the market, identified by their own bonds underperforming by more than one standard deviation
None of the three tests found a footprint in Treasuries — not in 10-year yields, not in the term premium, not in swap spreads
The one apparent exception, Amazon's 10 March deal, lines up with a 22bp move in 2-year yields, which is a policy story rather than a supply story
Term premia have moved up since Kevin Warsh's first FOMC meeting, which is the rise in yields that does have an identifiable source
1. Crowding Out Runs Backward
Karoui opens with the argument he is about to take apart, and he is fair to it: it can appear compelling. AI issuers are expected to keep borrowing unprecedented sums while Treasury supply stays elevated, and both are competing for the same investors.
The popular story is that AI borrowing is pushing Treasury yields up
A popular narrative for the rise in bond yields over the past few months is that the debt-funded AI capital expenditure cycle is crowding out the Treasury market.
Lotfi Karoui
The objection is a definitional one. Crowding out, as economists use the term, describes what the government does to the private sector, not what one private borrower does to the government.
The textbook mechanism points the other way
The textbook mechanism runs in the opposite direction: Government borrowing absorbs a finite pool of savings, pushes up interest rates, and crowds out interest-rate sensitive private investment.
Lotfi Karoui
That leaves the question of scale, which he poses directly.
The Treasury market is $32 trillion deep
Can private-sector financing, however large, crowd out a $32 trillion Treasury market?
Lotfi Karoui
2. Real Yields Must Adjust
Karoui concedes the part of the story he thinks is right. An AI capital spending cycle, like any large capital spending cycle, can put upward pressure on inflation-adjusted yields — just not for the reason most investors give.
Equilibrium real yields are the price that balances how much the world wants to save against how much it wants to invest. Hyperscalers bidding for labor, power and construction capacity move the investment side of that up.
The capex boom shifts desired investment upward
As hyperscalers ramp up spending and compete for labor, power, and construction capacity, the AI capex boom shifts desired investment upward.
Lotfi Karoui
Something then has to give, and he lists what.
Either saving rises, or investment gets squeezed, or both
Unless desired saving rises commensurately, equilibrium real rates must increase until desired saving rises, some investment is crowded out, or both.
Lotfi Karoui
He points at the chart of 10-year and 5y5y forward real rates to show that this is what has been happening while AI investment has gathered pace.
3. The Open-Economy Margin
The savings-investment identity does not have to close inside the United States. It is the world's largest open economy, so higher domestic investment can be met by importing capital instead of by saving more at home. A second adjustment runs through the currency.
A weaker dollar reduces how much foreign capital is needed
A weaker U.S. dollar can improve the trade balance over time, reducing the amount of foreign capital required to finance higher investment.
Lotfi Karoui
Karoui immediately limits the argument. The country already runs a large current account deficit and already absorbs a lot of the world's savings, so there is not much room left on that margin. And the prospect of unusually high returns on AI investment could pull in foreign money by itself, which would take the pressure off the exchange rate and put it back on yields. His expectation is that the adjustment splits across all three channels — real yields, capital inflows and the dollar — with real yields doing an important share of the work.
4. Financing Mix Is Neutral
This is the part of the argument Karoui says the AI capex debate keeps missing. The pressure on real rates does not depend on how the spending is funded.
The same real resources are consumed whatever the liability
The roughly $1 trillion in hyperscaler capex expected this year consumes the same real resources whether it is funded with debt, retained earnings, or equity issuance.
Lotfi Karoui
The case that trips people up is retained earnings, because a company spending cash it already holds looks as though it has paid for the investment in advance. Karoui says the accounting and the macroeconomics part company there.
Spending cash on hand converts saving into investment
Deploying retained earnings into AI infrastructure simply converts saving into investment and consumes the same labor, power, and construction resources as any other form of financing.
Lotfi Karoui
What the financing choice does change is who owns the claim and who carries the risk. That matters for asset prices and for market technicals — which is the second channel, and the one he tests next — but not for the arithmetic linking desired saving, desired investment and equilibrium real rates.
5. Finding Surprise Deals
The portfolio rebalancing channel is the one where financing genuinely could matter: a wave of AI bond issuance forces investors to sell something to make room, and Treasuries are the obvious something. Measuring that directly is hard, so Karoui narrows the question to whether unexpectedly large deals leave a mark he can see.
The first problem is separating issuance the market had already priced from issuance that caught it out. His proxy is how an issuer's existing bonds trade when a new deal is announced.
An abnormal move in an issuer's own bonds marks a deal as a surprise
When those bonds generate abnormal (larger than a 1 standard deviation move) negative returns relative to the broader index, we treat that as evidence that the size or timing of the offering was not fully priced in.
Lotfi Karoui
Six deals in a year cleared that bar
Using this approach, we identify six surprise deals over the past year.
Lotfi Karoui
He grants that the proxy is imperfect, since bonds can underperform for reasons that have nothing to do with supply — bad news about another issuer in the same sector, for instance. He checked, and says that was not what happened in this sample.
6. Test 1: Nominal Yields
The first test measures the two-day change in 10-year Treasury yields around each of the six events, from the close before the announcement to the close the day after. The window is deliberately short, but long enough for a large deal to be priced, allocated and absorbed.
Surprise AI issuance does not lift Treasury yields by a statistically significant amount
The data show surprise AI debt issuance does not produce a statistically significant increase in Treasury yields
Lotfi Karoui
One event does move, and Karoui names it rather than leaving it in the scatter.
Amazon's March deal is the outlier
The main outlier is Amazon's 10 March deal, which coincided with a meaningful backup in nominal yields.
Lotfi Karoui
He then gives the reason to discount it. The short end moved too, and the short end is not where a supply shock in corporate bonds would show up.
The 2-year moved 22bp in the same window, which points at policy
Two-year Treasury yields, a proxy for policy rate expectations, rose roughly 22 basis points over the same window.
Lotfi Karoui
7. Test 2: The Term Premium
The second test swaps nominal yields for the estimated 10-year term premium, on the Christensen-Rudebusch model. The result does not change: surprise AI issuance does not produce a statistically significant increase in term premia either.
Karoui says that fits the wider pattern since the third quarter of last year, in which the rise in yields has mostly been about expected policy rates rather than about compensation for holding duration. He does name one thing that moved the term premium.
Term premia rose after Warsh's first meeting in the chair
Most of the rise in yields has reflected higher expected policy rates, although term premia have moved up more noticeably since Fed Chair Kevin Warsh's first FOMC meeting.
Lotfi Karoui
8. Test 3: Swap Spreads
The third test is the cleanest of the three, because swap spreads are observable rather than estimated. If investors really were dumping Treasuries to absorb unexpected fixed-rate AI bond supply, Treasuries should get cheaper against swaps even when the signal in outright yields is too noisy to read.
Swap spreads are the robustness check that needs no model
If investors were selling Treasuries to absorb unexpected fixed-rate AI bond supply, Treasuries should cheapen relative to swaps even if the signal in outright yields is noisy.
Lotfi Karoui
There is no systematic response here either
Once again, we find little systematic response around surprise issuance
Lotfi Karoui
Bonus Insights
The two channels are not the same claim, and Karoui keeps them apart
The piece makes one concession and one refusal, and the distinction between them is the whole argument. Real yields can rise because the economy is being asked to fund more investment out of the same saving. That is not the same as investors selling Treasuries to buy AI bonds, and only the second one is what market participants usually mean by crowding out.
The figures are dated to the same day
Every chart in the piece — the real rates series, the two-day yield changes, the term premium and the swap spreads — is marked as of 15 September 2026, six days before publication, and sourced to Haver Analytics and Bloomberg alongside PIMCO's own work.
Two colleagues are credited
Michael Puempel and Gabriel Cazaubieilh contributed to this report.
Lotfi Karoui
Karoui's bottom line is that the AI capital spending boom is a real-resources story rather than a bond-supply story: it can lift equilibrium real yields through the saving-investment channel, but across nominal yields, term premia and swap spreads, the narrower claim that AI bond supply is directly pushing Treasury yields higher is, as he puts it, hard to find in the data.
Products, Companies & Tools Mentioned
PIMCO (Karoui's firm, and the publisher of the Credit Market Lens series this piece belongs to)
Amazon (Its 10 March bond deal is the one surprise issuance in the sample that coincided with a meaningful rise in nominal yields — a move Karoui attributes to policy expectations rather than supply)
The Federal Reserve (Term premia have moved up more noticeably since Kevin Warsh's first FOMC meeting as chair)
Books & Resources Mentioned
The Christensen-Rudebusch term premium model (The estimate Karoui uses for the 10-year Treasury term premium in his second test)
Listen to the full episode
If this was worth your time, send it to someone who has to have a view on this.
Get the latest market chatter as it happens: