PIMCO podcast · · 29 min
AI Boom or Credit Risk?
With Lotfi Karoui, Nidhi Nakra
ChorusBring the market's conversation to your AI. Learn more
Summary
PIMCO's Lotfi Karoui argues that the AI infrastructure boom is being financed increasingly with debt, with four of five hyperscalers roughly at negative free cash flow, but that credit has so far absorbed the supply shock. He sees the AI issuance wave over the next 12 to 18 months as at least as large as before, with more predictable capex in 2027 as spending growth slows, and he expects slower build-out to favour high-quality hyperscaler credit over hyperscaler equity. On rates, he says the roughly 5% base yield and real-yield-driven moves have restored bonds' income and diversification role, while CCC borrowers face pressure from higher funding costs and credit dispersion is driving active selection. He also contends AI bond supply does not textbook-crowd out Treasuries, and that Europe may benefit from rerouted investment only slowly, constrained by energy and grid capacity, with Germany's fiscal room and Nakra's view of concentrated European data-centre planning (Frankfurt, London, Amsterdam, Paris, Dublin) as the main regional angles.
Key takeaways
- Lotfi Karoui on AI infrastructure & data centers: Democratic midterm majority adds permitting friction, slowing the AI buildout pace but not its ultimate size.
- Lotfi Karoui on AI infrastructure & data centers: Most hyperscalers have negative free cash flow, so AI capex depends heavily on external debt and equity funding.
- Lotfi Karoui on US Treasuries & yields: Starting yields are the most attractive in about 25 years; real yields above 2.5% restore bond income and protection.
- Lotfi Karoui on Credit markets: AI debt supply has been absorbed through wider spreads on AI issuers, not across the broader corporate credit market.
- Nidhi Nakra on AI infrastructure & data centers: Europe's data-centre pipeline is concentrated in five cities, so catch-up and opportunity exist beyond them.
What was said, by company and theme
18 views from 2 speakers. Every quote is checked against the episode's transcript (how).
AI infrastructure & data centers
Where everyone stands on AI infrastructure & data centers
Lotfi Karoui · PIMCO
Democratic midterm majority adds permitting friction, slowing the AI buildout pace but not its ultimate size.
The likely Democratic congressional majority from the US midterms makes permitting for AI infrastructure build-out more complicated than initially expected, which is the most immediate implication. Karoui's view is that this is more likely a change in the pace of the build-out than in its ultimate size: the destination is the same, but the journey takes longer. He says size is unlikely to change while the pace will likely slow somewhat. He adds that this has implications for relative value and for where value in the build-out accrues.
“My view would be that it's probably more likely that this is just a change in the pace of the build out without necessarily impacting the ultimate size. And so in other words, the destination is the same, the journey might take a little longer.”at 2:06
Lotfi Karoui · PIMCO
Most hyperscalers have negative free cash flow, so AI capex depends heavily on external debt and equity funding.
Karoui says four of the five hyperscalers are essentially negative free cash flow today. That makes reliance on external funding critical to sustaining the current pace of capex. He notes that a decent portion of every incremental dollar of capex is funded with debt. This ties the AI ecosystem into a negative feedback loop: capex drives debt supply, and credit has been caught in it.
“Today, four of the five hyperscalers are essentially negative free cash flows. And so the reliance on external funding to continue to sustain the current pace of CapEx is really critical.”at 2:52
Lotfi Karoui · PIMCO
Multi-year, multi-trillion AI capex forecasts are unreliable; confidence is greater for the 2027 outlook.
Karoui is cautious about long-term capex projections. He says they rest on assumptions that may or may not materialise, especially monetisation, so he prefers to take them one year at a time. He has more confidence in the 2027 capex outlook than in a multi-year, multi-trillion cycle. He points to consensus 2026 capex expectations having more than doubled since the start of the year, which made the supply pipeline hard to predict. He expects 2027 to feel better and 2028 to be more predictable.
“I'd be a little careful with long term projections of CapEx. And so I would take things like one year at a time a little bit. I have greater confidence in the 2027 outlook for CapEx, as opposed to a multi year, multi trillion type of cycle”at 13:59
Lotfi Karoui · PIMCO
Next AI debt issuance wave likely at least as large as the last twelve months, clearing at higher risk premium.
Despite permitting friction, Karoui says the next wave of AI-related debt over the next twelve to eighteen months is very reasonably expected to be at least as large, maybe larger, and faster. It will likely come in more flavours across structures. He expects it to clear with some additional build-up of risk premium. The next wave should be more predictable than the last one, because capex growth is already projected to slow.
“for the next twelve months or eighteen months, despite the friction that is introduced in permitting, it's very reasonable to expect the next wave to possibly be at least as large, maybe larger, faster.”at 15:37
Lotfi Karoui · PIMCO
Europe offers diversification against the AI buildout, though energy constraints limit how much capex it can attract.
Europe has not been at the epicentre of the AI build-out. If permitting friction in the US becomes more than a temporary slowdown, Europe could benefit from rerouted investment, and it offers diversification for investors wanting to hedge the build-out. Karoui cautions that rerouting is constrained by energy and grid capacity, as Europe is a net energy importer. Rerouting also takes multiple quarters or years. Euro-market hyperscaler issuance is far below dollar-market levels. The same energy argument applies to emerging markets.
“And so Europe in some ways does provide diversification for investors that are looking for ways to sort of hedge the AI infrastructure build out, that's a good thing.”at 22:37
Nidhi Nakra · PIMCO
Europe's data-centre pipeline is concentrated in five cities, so catch-up and opportunity exist beyond them.
Nakra says about two-thirds or more of European data-centre planning is concentrated in Frankfurt, London, Amsterdam, Paris and Dublin. Europe is much more than five cities for innovation, so there is a lot of catch-up to do and therefore opportunity. Madrid and Milan are working on more data-centre planning. Some Nordic countries and France are energy-sufficient. Spain has relaxed renewables requirements to allow non-renewable power out to 2029, which supports data-centre build-out there. She sees fertile ground for opportunities in the near future.
“in Spain, they relaxed some of the requirements around using renewables. So they've allowed nonrenewables out to 2029. So that's been very supportive for the data center build out over there.”at 23:44
Credit markets
Where everyone stands on Credit markets
Lotfi Karoui · PIMCO
Higher funding costs are testing low-quality borrowers, with CCC credits under pressure since late April.
Higher yields are bad news for low-quality borrowers, and this is visible in the CCC part of the high-yield market, which has been under pressure since late April. Karoui attributes part of this to the higher cost of funding. Capital structures built when policy rates were near zero, assuming a base yield of roughly 2.5 to 3%, no longer work at today's rates. Other idiosyncratic factors also play a role. He thinks the ability of low-quality borrowers to sustain current leverage will be tested and probably diminished.
“sort of bad news for low quality borrowers across the board. I mean, as you know, we're seeing it in the high end market, the CCC part of the market.”at 10:34
Lotfi Karoui · PIMCO
European sovereign credit is under pressure as higher yields hit weak growth and poor fiscal policy.
European sovereign credit has seen pressure, not entirely because of higher yields but partly because of them. Higher yields put pressure on sovereign balance sheets already suffering from low growth and poor fiscal policy. Karoui notes this is another area where the bar for normalisation is probably quite high. The pressure is about specific sovereigns rather than systemic risk, which is why selectivity matters.
“sovereign European credit is another area where you've seen pressure, not entirely due to the reset higher in yields, but somewhat due to it. When you reset yields higher, you do put pressure on sovereign balance sheets that are sort of suffering from the intersection of [low] growth and poor fiscal policy”at 12:00
Lotfi Karoui · PIMCO
Higher base yields reduce the incentive to take credit risk, creating a quality decompression in credit.
When the cost of funding resets higher, through the base yield or the spread, incentives to move down the quality spectrum fall, all else equal. High-quality fixed income can now generate a seven to 7.5% yield, so there is no need to take excessive credit risk. Karoui says this feels like a decompression environment, in credit jargon, where spreads between quality tiers widen. Higher costs of capital are the main reason. Active credit selection therefore matters more, and dispersion has been the biggest source of alpha.
“you essentially reduce incentives to go down in the quality spectrum, all else equal... there's no need to take an excessive amount of credit risk. And so this, in credit jargon, this feels like a decompression environment all else equal.”at 13:07
Lotfi Karoui · PIMCO
AI debt supply has been absorbed through wider spreads on AI issuers, not across the broader corporate credit market.
Over the last twelve months, the AI debt supply shock has been absorbed at its source, through wider spreads on the AI ecosystem itself. Karoui splits the investment-grade index into AI-related and non-AI debt. The non-AI portion has been remarkably stable, with spreads still tight and oscillating in a narrow range. He concludes there is little pricing evidence that AI issuance is crowding out other parts of credit.
“this AI supply shock into debt markets has actually been absorbed at its source so far. It's been absorbed via wider spreads for that AI ecosystem itself, as opposed to the broader corporate credit market.”at 15:22
Lotfi Karoui · PIMCO
Scarcity from slower AI buildout favours hyperscaler credit over equity, since supply falls and credit has a technical tailwind.
Slowing infrastructure build-out creates scarcity, and the question is where the premium for that scarcity goes. Karoui thinks it is a more positive story for hyperscaler credit than hyperscaler equity. Credit has a technical tailwind, supply, which equity lacks to the same degree. Slower capex means slower debt supply. He sees high-quality balance sheets as the clear relative winner, since they can afford to wait, while not everyone benefits from the scarcity.
“In my view, I think it's a more positive story for hyperscaler credits relative to hyperscaler equities, because in credit, there's a technical tailwind that doesn't exist on the equity side, or at least not to the same degree, and that's supply.”at 24:39
US Treasuries & yields
Where everyone stands on US Treasuries & yields
Lotfi Karoui · PIMCO
US yield rise is mostly a real-yield move reflecting reset growth expectations, not an inflation repricing.
Decomposing US nominal ten-year yields, the vast majority of the move higher came from real yields rather than breakeven inflation. Karoui says this means the main driver of the backup was a reset higher in growth expectations, at least before October. In euro and sterling markets the split is roughly 60/40, with 60% real yields and 40% inflation expectations, which he calls a less friendly backdrop. He argues that the drivers matter more than the backup itself, which is why risk assets digested the rise until early October. Duration then rallied on a reset lower in growth after the tech sell-off, again through real yields.
“That's really important because that tells you that the primary driver of that backup has been actually a reset in growth expectations at least before October.”at 7:13
Lotfi Karoui · PIMCO
Starting yields are the most attractive in about 25 years; real yields above 2.5% restore bond income and protection.
Using the dollar market as a benchmark, which he says generalises across G10, the base yield is around 5%. Fixed income can again deliver income, and the risk-reward of high-quality fixed income has improved dramatically. Real yields above 2.5% mean duration can again protect in a bad state of the world. October's equity sell-off showed this, as bonds rallied while stocks fell, restoring the negative stock-bond correlation. Karoui says the starting level is perhaps the most attractive in 25 years, which usually bodes well for forward total returns.
“But the base yield is around 5%, give or take 10 basis points. And so the value proposition of high quality fixed income from a risk reward standpoint has improved quite dramatically.”at 9:03
Lotfi Karoui · PIMCO
AI bond issuance is unlikely to crowd out the $32 trillion Treasury market; event study found no abnormal returns.
Skeptics argue that long-dated AI bond supply is large relative to long-dated government bonds, and Karoui does not dispute the flow numbers. He says it does not fit the textbook definition of crowding out, which usually runs from an oversupply of safe assets onto the private sector. Treasuries have superior liquidity and are held for protection, not income. His team's event study of Treasury returns around AI deal days found no abnormal returns. Real yields have risen partly because of AI, but through an investment-equals-savings channel that would be the same whether capex is debt-funded or funded from retained earnings, not portfolio rebalancing.
“I struggle to see an AI infrastructure build out, however big it is, put pressure on a $32,000,000,000,000 market, which is the treasury market.”at 19:36
Power demand & the grid
Where everyone stands on Power demand & the grid
Lotfi Karoui · PIMCO
Power and grid capacity is a major constraint on AI buildout, limiting how much can be rerouted to Europe.
Beyond permitting, Karoui says the other major constraint on the build-out is the grid, meaning power and energy. He notes energy is not abundant in the rest of the world, using Europe as an example, since it is a net importer of energy. That caps how much capacity can be relocated outside the US, so rerouting is a slow and limited response. He views power as the key differentiator for attracting marginal data-centre investment, including in emerging markets.
“permitting is one friction, but there are other constraints. I think the other major constraint to the build out is the grid system, power, energy, etcetera.”at 4:20
AI: overall outlook
Where everyone stands on AI: overall outlook
Lotfi Karoui · PIMCO
Q3 earnings showed clear AI monetisation on revenue, though spending is still growing faster.
Karoui says the revenue numbers from the latest earnings season were good, and there is clear evidence that AI is already being monetised. The issue is that the spending side continues to move faster than revenue, which is why the market reacted the way it did. He says this matters because continued heavy capex depends on the ability to monetise, and he sees early evidence that it is happening.
“the numbers were good. I mean, the issue is that the spending side of that equation continues to move faster, which is why the market reacted the way it did. But on the revenue side, there's clear evidence of monetization already.”at 17:49
Europe's economy
Where everyone stands on Europe's economy
Lotfi Karoui · PIMCO
Germany still has decent fiscal space to stimulate, a positive for Europe.
Karoui concedes that European balance sheets are generally strained. Disaggregating the picture, though, Germany still has decent fiscal space, and it is the largest economy in Europe. That gives Europe the ability to stimulate if it chooses. He presents this as a silver lining alongside Europe's relative absence from the AI build-out.
“I would say we do have fiscal capacity in Germany. ... there is decent fiscal space in Germany, which is still the largest economy in Europe, so that is good news.”at 21:55
AI chips & semiconductors
Where everyone stands on AI chips & semiconductors
Lotfi Karoui · PIMCO
Slower AI capex is a negative for picks-and-shovels, semis and memory, as chip demand gets destroyed.
If capex slows, companies buy less of what goes into the data center shell. Karoui sees this as demand destruction for picks-and-shovels suppliers, semiconductors and memory, which will have to happen now. Scarcity keeps compute demand strong for longer, but that does not solve the funding issues or the longer-term question marks. Neoclouds are a case-by-case story: strong demand may linger, but funding and secular questions remain.
“I think it's probably a negative for picks and shovels, semis, [memory], etcetera, because that is demand destruction that will have to happen now. If you slow down the CapEx, that means that you're buying less of whatever goes into the shell”at 24:57
Topics
- AI bond supply crowding out Treasuries
- AI debt issuance pipeline
- AI debt supply and spreads
- AI monetisation vs spending
- Europe data center buildout
- Europe's role in AI buildout
- European sovereign credit
- German fiscal capacity
- bond yields and income
- capex slowdown and chip demand
- credit quality decompression
- data center permitting and capex pace
- drivers of yield moves
- high-yield CCC pressure from funding costs
- hyperscaler credit vs equity
- hyperscaler funding needs
- multi-year capex forecasts
- power constraints for AI buildout
More from PIMCO podcast
- Gordon Brown: Power and Politics in a Fragmented World · November 3, 2026
- The Top Client Questions About the U.S. · October 30, 2026
Ask your AI about this episode
Chorus gives Claude, ChatGPT and other assistants every view from this episode and every other show Chorus follows. Ask who agrees, who changed their mind, and what else was said about the same companies.
Chorus reports what speakers said. It isn't investment advice. How Chorus works · About Chorus