Howard Marks: Don't go all-in, don't stay all-out (2026)

In December 2025, after client meetings in Asia and the Middle East, Howard Marks wrote Is It a Bubble? to answer a question he kept hearing about artificial intelligence. He opens with a separation that does most of the work: one issue is how AI companies are behaving; another is how investors are behaving around AI. He leaves the industry's aggressive buildout largely alone and stays with investor psychology and the prices that psychology produces. 1

The memo is long. The passage worth carrying is the close, where Marks refuses both the all-in and the all-out answer.

The excerpt

"Struggling with whether to apply the 'bubble' label can bog you down and interfere with proper judgment; we can accomplish a great deal by merely assessing what's going on around us and drawing inferences with regard to proper behavior."
"On the other hand, 'inflection bubbles' based on revolutionary developments accelerate technological progress and create the foundation for a more prosperous future, and they destroy wealth. The key is to not be one of the investors whose wealth is destroyed in the process of bringing on progress."
"Since no one can say definitively whether this is a bubble, I'd advise that no one should go all-in without acknowledging that they face the risk of ruin if things go badly. But by the same token, no one should stay all-out and risk missing out on one of the great technological steps forward. A moderate position, applied with selectivity and prudence, seems like the best approach." 1

Marks is offering a way to act while the label stays unsettled.

The label is the wrong fight

Marks notes that bubbles usually form around something new — a financial structure or a technology — because newness removes history as a restraint. In the absence of history, the future can look limitless, and a limitless future can justify prices far beyond ordinary earning power. 1

That is why the binary question — bubble or not? — feels urgent and still stalls the analysis. Settling the label commits the speaker to a verdict the market will only settle later. Marks puts the useful work earlier and smaller: watch the behavior, watch the price paid for that behavior, and decide how much capital should sit under that uncertainty.

He cites his own earlier observations around the internet boom and the pre-crisis credit boom. In both cases, he says, the value came less from expertise in the underlying instruments than from describing the folly in the surrounding behavior. 1

Progress can incinerate capital

The memo's most uncomfortable distinction comes from the literature on "inflection bubbles" versus "mean-reversion bubbles." Mean-reversion episodes rise and fall on a financial fad with little claim to lasting progress. Inflection episodes form around real technological change: railroads, electricity, the internet. They can speed the buildout of infrastructure that later generations use. 1

Marks accepts that point and then turns it against the investor who confuses social progress with personal return. An inflection bubble can leave the world better equipped and leave many of the people who financed the equipment poorer. Technology promoters may welcome the speed that frenzy buys. An investor still has to decide whether to be the capital that gets burned to buy that speed.

That is the link between the bubble discussion and Marks's broader defensive style. Belief that AI may transform economies still leaves price, concentration, and leverage as open questions.

What remains unknown on purpose

Marks lists the unknowns that make certainty expensive: who the lasting winners will be; whether AI vendors or AI users capture the profits; how long chips and data centers stay productive enough to repay the debt used to buy them; whether circular deals exaggerate the boom; how much debt the buildout will ultimately carry. 1

He also quotes Sam Altman summarizing the tension: investors may be overexcited about AI, and AI may still be the most important development in a long time. Marks's response is that both statements can be true at once, which is why the market will hand out a clean verdict only in retrospect. 1

Under that combination, full participation that ignores a ruin case is reckless. Full abstention that ignores the cost of missing the upside is also a choice. The middle path Marks names keeps the book from being dictated by a single narrative — either "bubble" or "this time is different."

Three questions before the next AI bet

Marks's close gives a long-term investor a short checklist that still works while the bubble verdict stays open.

What price is the story already demanding? Separate belief in the technology from the return required by today's price.
What happens to the portfolio if the enthusiasm was excessive? Size the position so a painful correction is survivable rather than existential.
Where is the selectivity? A moderate stance still requires choosing among names, capital structures, and entry points; "some AI exposure" still underwrites a story if every holding needs the same boom to work.

These questions leave the stock choice to the reader. They ask whether the decision rests on a sober price judgment or on the need to settle an argument about labels.

Marks ends the investment section with a reminder that even the fashionable objects of the boom — data centers among them — still require ordinary underwriting. Shortage or oversupply can appear. Rents can surprise either way. Intelligent investment in AI, he writes, still requires sober judgment and skillful implementation. 1

The transferable habit is older than this cycle. When the technology is large and the unknowns are larger, stop waiting for permission from the word "bubble." Judge the behavior, require a price that still works under disappointment, and leave room on both sides of the bet.

Fuentes de referencia

1
Is It a Bubble? — Howard Marks
oaktreecapital.com