JP Morgan's Michael Cembalest on the Impending Treasury Bust

ELI5/TLDR

JP Morgan’s chief market strategist thinks the US bond market is heading for a real crisis in about three to four years, because around 2030-2031 every dollar of federal tax revenue will be eaten by interest payments plus entitlements, leaving nothing for anything else. He’s not panicking yet, because that’s outside any sensible investment horizon, and preparing for it now would have gotten you fired years ago. The rest of the conversation is a tour through what he’s actually watching: an AI buildout where we have great data on spending and almost none on returns, a private credit market quietly loosening its standards, and the eternal lesson that markets price in disasters long before the headlines stop.

The Full Story

The man and the method

Michael Cembalest joined JP Morgan in October 1987 — two weeks before the crash, as a French and Russian literature major who knew nothing about finance, parked in a back-office data job. He worked his way to the front office and eventually became chief investment officer of the asset management business. His “Eye on the Market” reports, free and famously dense, started in 2005 as a way to explain portfolio decisions to private-bank clients without having to fly around the world doing it.

His one editorial obsession is plain language. He digests other people’s research and routinely calls them out:

I’ve been in this business for over 35 years. And you’ve got a guy there who writes something and I don’t understand more than 30% of it every single time. That’s a problem for you. That’s not a problem for me.

The US overweight, and why it’s ending

JP Morgan went overweight US equities in 2009 and rode it for 15 years while “brave strategists” called the European comeback every year and were consistently wrong. By the end of 2024 the rest of the world traded at nearly a 40% PE discount to the US — a gap he’d never seen. That’s when he started evening the position out.

The irony is that the America First president coming in was the end of America First outperformance.

Last year was a catch-up year for everyone but India and the US. The discount has narrowed to about 30%. His view: it settles around 20%, “as good as it gets” — some US outperformance left, but no return to parity. For the diagnosis of what’s wrong elsewhere, he points to the Draghi report on European productivity, labour and capital mobility, and energy security.

The circularity nobody likes to look at

Nearly half of US consumer spending comes from the top 1%, and the top 1% is essentially a leveraged bet on the stock market. A big correction would mean a “massive unwinding.” Cembalest agrees it’s a real risk but a short-term one — value-based corrections tend to repair themselves quickly. The fragile part is that the high valuations all sit in tech, and tech now means AI. He notes software has already been clobbered 20-30% this year, “a good example of what could happen to the broader market if those profit fundamentals come under siege.”

AI: great data on the spending, almost none on the return

This is the most honest stretch of the interview. The hyperscaler capex is, in his words, “the bet of the century.” The average S&P company spends ~20% of revenue on capex and R&D. Google, Amazon and Microsoft are at 30-35%. Meta is spending 70-80% of revenue on capital spending — “a number that’s unheard of in the entire history of corporate finance.”

The hard data we have is how much being spent. We don’t have a lot of hard data on how much is being earned.

He calls the evidence “very tea leafy” — accounting-firm adoption surveys, rough measures of who’s paying actual money, a Microsoft estimate of agentic AI payoff. Want optimism? Read Wharton. Want despair? Read MIT. Free cash flow margins at the big hyperscalers held stable through Q3 last year (meaning they were funding the spend without degrading margins), but the last two quarters they’ve “started to roll over pretty fast.”

On Meta’s accounting specifically — the off-balance-sheet Blue Owl JV, the extended depreciation lives — he’s less worried than you’d expect, for a counterintuitive reason. In the old days hiding obligations worked because the tools were poor. Now every analyst loads the obligation right back in.

I find it kind of strange that people are willing to pay real cash money to change the perception of something when the information abundance now is clear to everybody… It’s not hidden from anybody. No. Except for them in their own minds.

He uses AI sparingly himself — an AI overview for “what’s the best Bloomberg ticker for sulfur and helium prices” works because he gets instant feedback on whether it’s right. For anything where it would have to stitch together factoids he can’t immediately verify, he does the work himself.

OpenAI, the weak link

Asked why he called OpenAI the weak link in the AI edifice, he names two bridges it has to cross. First, the money: it’s mostly subscription revenue, with no real advertising or robust corporate model — that has to change. Second, energy: over a trillion dollars of build-out commitments requiring something like 30 gigawatts, “the equivalent of 30 nuclear power plants.” Sam Altman’s internal memo forecasting 250 GW by 2033 he treats as Silicon Valley hyperbole — “meant to be more allegorical” — but shocking enough to include anyway.

On the $110bn raised from Nvidia and SoftBank: there are entities with too much money that can’t write small cheques, and a lot of capital moves “based on the vibe.” His comparison for Jensen Huang calling a software release the most important of all time:

That particular statement sounded to me like the chairman of Saudi Aramco saying that the Hummer is the most important vehicle that’s ever been sold to the public.

He distinguishes statements oriented toward “a debate about the future” from ones “mercenarily linked to their own fortunes and shareholders,” and warns against treating everything Altman, Huang and Cook say as the former. On why a pre-IPO round can clear at a wild valuation: institutions working backwards toward a “market weight position,” too scared to hold zero in a company that might become a meaningful chunk of the index.

The diversification America lost

The US used to be the beautifully diversified market — put your money in the S&P and you owned utilities, healthcare, staples, energy, materials, tech. Top-10 concentration is now drifting up “the same way the financial sector numbers drifted up before the financial crisis.” It’s starting to look like the concentrated smaller markets it used to be the alternative to — a reason JP Morgan is diversifying away from being too US-concentrated. The saving grace: today’s tech sector is actually profitable and less cyclical, unlike the dot-com version with no earnings.

The Treasury bust

The headline call. The rewards of being the reserve currency are enormous — no other country could run this fiscal policy and debt burden with so little bond-market reaction. JP Morgan tracks six or seven measures of dollar dominance (FX trading share, reserve assets, SWIFT payments, new debt/equity issuance) and doesn’t see it shifting much on the ground, despite the popular “de-dollarization” thesis. But:

If that reserve currency status ever did change, the US is in a world of trouble. Because they would have to start to monetize the debt that other entities didn’t want to hold.

And he thinks this administration “is doing as much as any to tear down the fundamental supports for that reserve currency system.” His mechanism for the crisis, 3-4 years out:

Around 2030 or 2031, 100% of government revenues… federal tax revenues will be required to pay interest on the debt and entitlements. There won’t be a penny left for anything else.

A year or so ahead of that, rating agencies warn of a downgrade. Then the big sovereign wealth funds — Abu Dhabi, GIC, Temasek — stop buying new dollars at auction (not necessarily selling). That precipitates the crisis: the 10-year jumps 150-200bp in a short window, equities down 15%. From there it forks. The hopeful path is a 2009-TARP-style moment where politicians use the crisis as cover for painful tax-and-spending decisions — “take this tough medicine, you’re correcting some of the imbalances and you move on.” The bad path: the Fed just monetizes the debt.

His portfolio advice is the discipline, not the alarm:

One of the most important things to not do is prepare portfolios for things that are happening outside the time horizon of your investments… A lot of the Cassandras and the Roubinis of the world would have recommended preparing for it then and would have been fired long since.

He frames much of investing as “war chest strategies” — make as much as you can, because at some point you’ll give some back and may not see it coming. The Indian rupee is his example: higher rates, you compound, then a devaluation hands some back, then it climbs again. Investors who accept the cycle beat those constantly trying to game it.

Private credit, quietly rotting

Underwriting over the last two years has been “pretty bad.” A flood of money chasing private credit collapsed its standards. The tells, which “a credit guy would love”: EBITDA add-backs (letting borrowers book hypothetical future synergies as if already realized) migrating from the broadly-syndicated market into private credit, where lenders used to forbid them; and asset-sale proceeds no longer fully sweeping to pay down loans. All signs of borrowers (PE firms) gaining negotiating power over lenders. He notes some private-credit people told him in November their software holdings were the safest, bedrock part of the book — right before “Claude co-work” and similar releases delivered an external shock that blindsided the space, “myself included.” The balance of power won’t shift back without a recession or a real contraction in liquidity, and “there’s just too much capital around.”

The favourite chart

His least favourite chart is the high-yield-spread average line drawn at 600bp — spreads are never there, they’re always much higher or much lower depending on the cycle. His favourite is how markets anticipate the future. In the early-90s S&L crisis, bank stocks bottomed when only 20% of the eventual failures had happened. In March 2009 housing investments bottomed with only ~10% of mortgage defaults realized.

By the time the bad news on bank failures stops, the bank stocks will be 150% up… If you say “I’m going to wait until the clouds clear,” you can do it if you want, but you’re going to miss all of the upside.

The humbling coda: in summer 2008 JP Morgan, sitting in the centre of the financial system, put $2bn of its own capital into Fannie and Freddie preferreds. Three months later the GSEs went into conservatorship and the preferreds went to zero.

If we missed how bad things were, that’s a signal that sometimes the rot taking place in something is not even apparent to people sitting in the middle of it.

Key Takeaways

The Treasury crisis mechanism: ~2030-31, 100% of federal tax revenue goes to interest + entitlements → rating agency downgrade warning → sovereign wealth funds stop buying new dollars at auction → 10-year jumps 150-200bp, equities -15%. The fork is political reform (good) vs Fed debt monetization (bad).
Meta is spending 70-80% of revenue on capex — unprecedented in corporate finance history. Average S&P firm: ~20%. Other hyperscalers (Google/Amazon/Microsoft): 30-35%.
On AI, the asymmetry that matters: hard data exists on spending, almost none on returns. Hyperscaler free cash flow margins held through Q3 last year but rolled over fast the last two quarters.
Off-balance-sheet financial engineering no longer fools anyone — analysts universally load Meta’s Blue Owl JV obligation back into its debt. Companies pay ~100bp extra to hide what everyone already counts.
The “America First president coming in was the end of America First outperformance.” Rest-of-world PE discount peaked near 40% end-2024, now ~30%; Cembalest expects it to settle at ~20%, not parity.
Nearly half of US consumer spending comes from the top 1%, whose wealth is essentially a stock-market bet — a circular fragility, but a short-term risk since value corrections self-repair.
OpenAI’s two unsolved bridges: a real revenue model (currently mostly subscriptions, no ads/corporate base) and ~30 GW of power for its >$1tn build-out commitments.
Pre-IPO mega-rounds clear at high valuations partly because institutions are “working backwards” to a market-weight index position, too scared to hold zero.
Private credit’s degradation tells: EBITDA add-backs migrating in from syndicated loans, and asset-sale proceeds no longer fully sweeping to pay down debt — both signs PE borrowers now outnegotiate lenders. Won’t reverse without a recession.
War chest mental model: make as much as you can because you’ll inevitably give some back unpredictably. Indian rupee carry is the archetype — compound, devalue, repeat; accepters beat gamers.
Don’t hedge risks outside your investment horizon — the perma-bears who prepared for this exact crisis years ago would have been fired by now.
Markets bottom long before the bad news stops. Bank stocks bottomed at 20% of eventual S&L failures; housing at ~10% of mortgage defaults. Waiting for lagging indicators (unemployment) to turn means missing the upside.
Even insiders miss the rot: JP Morgan put $2bn into Fannie/Freddie preferreds in summer 2008; they went to zero within three months.
Career advice for young analysts: finance is jargon-intensive; read the FT and Barron’s daily, circle every term you don’t know, and get it explained. Best investment available.

Claude’s Take

This is a high-signal interview, mostly because Cembalest keeps separating what he knows from what he’s guessing, which is rarer than it should be in macro punditry. The Treasury call is the clickbait, and to his credit he immediately defuses his own headline: the people who’d have made the same call seven years ago are unemployed, the timing is the whole game, and you don’t position for things outside your horizon. That’s the opposite of a doom-monger selling fear — it’s a man telling you why the doom doesn’t change what you should do today.

The genuinely useful parts are quieter. The AI section is the clearest articulation I’ve heard of the actual problem: we have excellent visibility into the spend and almost none into the return, and everyone is reading the same handful of soft surveys with whatever lighting they prefer. The Meta point is sharp and slightly funny — the spectacle of a company paying real money to obscure something that every analyst un-obscures for free. And the private-credit add-back mechanics are the kind of specific, checkable detail that separates someone who’s read the loan docs from someone narrating vibes.

What to discount: the reserve-currency confidence is doing a lot of load-bearing work, and he half-acknowledges it. His whole “no near-term dollar problem” rests on six metrics that are, by construction, slow-moving and backward-looking — exactly the kind of lagging indicators he warns against trusting elsewhere. The crisis he forecasts is a loss of confidence in the dollar’s debt; saying the metrics don’t show it yet is consistent with his own point that the rot isn’t visible until it isn’t. There’s a mild tension there he doesn’t fully resolve.

Score 8: dense, specific, intellectually honest, light on self-promotion (the only soft spots are the host’s house ads). It loses points only because the marquee Treasury thesis is more a known directional concern than a novel insight, and the most original material is buried in the second half.