Howard Marks: AI, Debt vs Equity & The Next 40 Years Of Investing | Nikhil Kamath

ELI5/TLDR

Howard Marks, co-founder of Oaktree and patron saint of distressed debt, sits with Nikhil Kamath for what is mostly a career retrospective wrapped around three live ideas: where we are in the cycle (middle-aged, no obvious excesses except maybe AI), why bonds were the better fit for his temperament than equities (predictable contracts, conservative parents, lucky to be standing in the right line at the right time), and what AI actually changes about investing (pattern matching is fine, but pattern creation and second-level thinking are still human work). The Marks memos and books are the core text; this conversation is the abridged audio version.

The Full Story

Mujo, and the case against prediction

Marks opens with a piece of his Wharton minor — Japanese literature — and the concept of mujo, the inevitability of change. Change is inevitable, unpredictable, and uncontrollable; you accommodate it, you don’t outsmart it. This becomes the foundation for his most-cited memo:

“About 20 odd years ago I wrote a memo called ‘You can’t predict, you can prepare.’ We can’t predict the future. We can’t control the future. But we can prepare for the future.”

The contradiction is obvious — how do you prepare for what you can’t predict? The answer is portfolio construction that suboptimises across several plausible futures rather than maxing out for one. You don’t simultaneously prepare for the positive tail and the negative tail; you crowd the middle of the distribution and accept that you won’t be the hero in any single scenario.

Kamath presses him on the tension with the coin-flip example — if past flips don’t predict the next, why does past behaviour predict markets? Marks’s answer is the cleanest one in the conversation. Coin flips are independent. Human behaviour is not. History is the history of what people did, and what people did has implications for what they’ll do next. Cycles aren’t sine waves. They’re excesses and corrections around a trend line, driven by optimism and pessimism rather than by physics.

Where are we in the cycle

Kamath references a 2024 interview where Marks said “somewhere near the middle.” Two years later, the bull market ran into early 2026, the recovery is somewhere between six and seventeen years old (depending on whether you count the COVID quarter as a real recession — Marks doesn’t, really; it was government-induced, demand was suppressed not destroyed, and the third quarter of 2020 was the best quarter in GDP history right after the worst).

His diagnostic is almost folksy:

“When I go to a city, I look for construction cranes in the sky to tell me if there’s a building boom. Most booms are followed at some point by busts. I don’t see it in most of our cities. I don’t see inventories growing.”

The one place he does see the boom: AI infrastructure. Compute, energy, datacentres — the modern equivalent of cranes. He won’t call it a bust-in-waiting because he doesn’t know enough, but he names it as the area of obvious capital excess.

Why debt, not equity

The longest answer in the conversation is to Kamath’s question — why spend a career in an asset class with capped upside and binary downside? Marks’s answer breaks into three pieces.

First, he didn’t choose it. In 1978 the new CIO at Citibank told him to start a convertible bond fund. He said yes. Three months later a phone call asked whether he’d figure out what “high yield bonds” — Milken’s junk bonds — were. He said yes again. He calls this Gladwellian luck: standing in the front of the line because someone pointed at it, not because he’d worked out the line was the right one.

Second, the temperament fit. His parents were adults during the Depression, which meant his upbringing was an unbroken chant of don’t-put-all-eggs-in-one-basket. Bonds reward conservatism: predictable outcomes, contractual returns, almost-every-time payoffs. He pushes back hard on Kamath’s framing that downside in distressed debt is binary:

“I’ve been in high yield bonds for 48 years. In our experience, 99% of the bonds have paid interest and principal as promised.”

The math matters. If 10% of your bonds default and recover nothing while you earn 10% on the rest, your return is zero. To compound at meaningful rates in distressed debt you have to be right on something like 97-99% of names, which Oaktree has been because they were early, they have institutional memory, and analysts can stay analysts for life rather than clawing toward the PM seat.

Third, the field was unloved. The word “junk” did the marketing work for him — bias drove competitors out and let him buy mispriced risk. (Today’s euphemism: “opportunistic credit.” The companies they rescue-loan don’t want to be called distressed.)

The Oaktree edge in performing credit is one number: when the market thinks default probability is 5% and they think it’s 2%, the spread on offer prices the 5% world. If they’re right, the difference is excess return. The skill is bottom-up — predicting default better than other people, name by name. Their long-run default rate has been roughly a third of the high yield universe’s 3.6-3.7% annual default rate.

The AI memo, and the question Claude can’t answer

Marks wrote two AI memos in late 2025/early 2026, after an “extensive conversation with Claude” that he says shocked him. He has not fired anyone because of AI, and won’t, for a specific reason:

“AI’s excellence is in discovering past patterns, extrapolating them, and applying them with discipline. But I think there’s more to investment excellence than that. It’s the innovation of new patterns. It’s seeing the potential of things that have never… how will it deal with something where there’s no pattern?”

He revives two questions from his 2015-ish “Investing Without People” memo:

“Can AI sit down with five business plans and figure out which one is Amazon? Can AI sit down with five CEOs and figure out which one is Steve Jobs?”

He’d bet not. Then immediately concedes: neither can most humans. Which means AI doesn’t kill active management — it kills average active management. Mediocre stock pickers were already losing to indexation; AI just accelerates their extinction. The remaining slot is for people who can do the things AI can’t, which is the same slot that always existed for second-level thinkers.

The clarifying line on indexation:

“Indexation has taken over as it has not because it’s so good, but because active management was so bad.”

Kamath floats the obvious counter: maybe everyone uses AI in the future, in which case nobody has an edge. Marks works through it live. If all AI is equally smart and trained on the same history, they all write the same investment process and produce the same result — which means everybody makes 10% and nobody makes 12. But why pay AI to deliver the average? Investing rewards superiority, and superiority requires either differential insight or a model genuinely better than every competitor’s. He doesn’t claim to know which world we’re heading into.

Second-level thinking, and the limits of asking how

The most-quoted Marks idea makes its appearance. If you think the same as everyone else, you act the same, and you get the same returns. To outperform you have to deviate from the herd — and be right about the deviation. Different and wrong is worse than consensus and average.

“You can’t coach height. All the coaching in the world will not make your team taller. I can tell you that you must become a second-level thinker to be successful as an investor. I just can’t tell you how.”

This is also why he refuses Kamath’s “how do you stay sharp at 80” question. Hows are the hard ones. He can describe the destination — superior insight, the willingness to change your mind, intolerance for your own intellectual atrophy — but the path is yours. He gives credit to his son Andrew for pushing him, and to a lesson Andrew taught him during the COVID lockdown when three generations lived under one roof:

“Readily available quantitative information about the present cannot hold the key to success because everybody has it.”

This is in the Something of Value memo (January 2021). It’s also why Buffett’s edge in the early decades was sitting in Omaha reading Moody’s manuals on onion-skin paper while everyone else waited on annual reports to arrive in the mail.

India, briefly

Kamath asks the polite home-team question: India growing 7%, GDP per capita rising, can he just stay 80% in equities for 30 years, and is there a sector Marks likes? Marks declines, completely:

“I will not hold myself out as knowing anything about the Indian stock market.”

That’s the whole India answer. Oaktree has deployed roughly $4B into India recently, but Marks hasn’t written a memo on it and won’t pretend.

The 1929 echo, and a private credit memo coming

Marks mentions that he just finished Andrew Ross Sorkin’s 1929 and has a memo on private credit dropping the same week, drawing parallels. He doesn’t elaborate on the specifics in the conversation — the memo is the artefact, not the chat.

The dentistry test

The closing frame is the cleanest summary of who should do this work. Taleb (whom Marks ran into more than Kamath expects, given that Taleb buys tail protection and Marks crowds the middle of the distribution) makes the distinction in Fooled by Randomness: dentists fill cavities the same way every time and succeed every time. Civil engineers calculate steel and concrete and the bridge stands. Investing is not like that.

“If there’s something in your makeup that says you’re going to be unhappy if you’re not right all the time, don’t become an investor.”

You’re solving a puzzle to a better degree than other people, with no underlying laws to lean on. If that frustrates you, become an engineer.

Key Takeaways

You can’t predict, you can prepare. Build portfolios that do fine across several plausible futures rather than perfect in one. You can’t simultaneously hedge both tails — pick the middle and accept the tradeoff.
Cycles = excesses and corrections, not ups and downs. Average S&P return is 10%, but actual annual returns are almost never between 8 and 12. The norm is not the average. Greed and fear keep the system off-trend.
Where we are now (early 2026): middle-aged recovery; US economy healthy; no obvious excesses (no crane-bloom, no inventory build) — except AI capex, which is the area of unmistakable boom.
The Oaktree default rate framework: lend when market prices in 5% default risk and you assess 2%. The spread is excess return. Long-run high yield default rate ~3.7%/yr; Oaktree ~1/3 of that.
99% repayment rule: in 48 years of high yield, 99% of bonds paid interest and principal. To compound in this asset class you need 97-99%+ accuracy — getting “9 of 10 right” with full loss on the 10th gives you zero return.
Indexation won by default. Active management was bad and expensive, not just expensive. Even at zero fees, inferior active loses to passive.
AI does pattern recognition. It does not (yet) do pattern creation. It can’t pick Amazon out of five business plans or Steve Jobs out of five CEOs. Neither can most humans, which is exactly the gap superior investors live in.
Second-level thinking: different from consensus and better. Different and wrong is the worst place to be. Marks can tell you the destination but can’t teach the route.
Andrew Marks’s insight: present-tense quantitative data has no edge because everyone has it. The edge is in interpretation, in seeing what the data implies.
Mark Twain (allegedly): history doesn’t repeat, it rhymes — and what rhymes is human nature. Fight-or-flight from the watering hole still drives behaviour at the trading desk.
The dentistry test: if you need to be right every time, do something with laws — dentistry, civil engineering. Investing has no laws.
On India: Marks won’t say anything. He’s deployed capital, not opinions.

Claude’s Take

Kamath is a soft interviewer — friendly, deferential, prone to setting up Marks rather than testing him — and the first 20 minutes are throat-clearing about Wharton, Queens, and how to stay sharp at 80. Worth filtering. The signal density picks up sharply once Kamath asks the bond-vs-equity question, and the AI section is where Marks is doing genuine new thinking rather than reciting the back catalogue.

For someone who’s read Marks before, the framework material — second-level thinking, cycles as excesses, you-can’t-predict-you-can-prepare — is restatement, not revelation. The fresh content is two-fold. First, the live diagnostic on where we are in early 2026 (middle-aged recovery, AI as the only obvious bubble candidate, COVID quarter probably shouldn’t count as a recession). Second, the concession — repeated several times — that AI has changed his thinking enough that he wrote two memos about it, that the experience of talking to Claude “shocked” him, and that the surviving moat for human investors is specifically the things AI can’t do: pattern creation over pattern matching, and managing other humans’ psychology in stress.

What he doesn’t engage with: the specifics of the private credit memo he says is dropping that week (which is the actually interesting current call, given that private credit has been the fastest-growing corner of the credit world and Oaktree has competitors everywhere). He also waves off the India question entirely, which is the right move epistemically but disappointing for the audience.

The Taleb tangent at the end is the best moment of the conversation — Marks engaging with someone whose worldview is genuinely opposed to his (Taleb crowds the tails, Marks crowds the middle) and articulating the disagreement crisply. More of that, less of the origin-story coverage, would have made the interview great rather than good.

Score: 7/10. Reliable Marks; nothing here that supersedes a careful reading of the memos.

Howard Marks: AI, Debt vs Equity & The Next 40 Years Of Investing | Nikhil Kamath | People by WTF