The biggest junk bond investor in the world also admits he was wrong! After experiencing Claude, Howard Marks candidly says he misjudged the AI bubble

Oaktree Capital co-founder Howard Marks memo 《AI Hurtles Ahead》, to some extent, can be seen as a correction to his own AI views from late last year. This investment veteran, known for risk management, credit cycles, and warnings about bubbles, acknowledges in the memo that, in just the past 11 weeks, AI’s pace of progress has already become so fast that it outstripped what he originally imagined. And if he had to place a bet now, he even believes AI’s potential is more likely to be underestimated by the market than overestimated.

This shift is noteworthy not only because Howard Marks has long been viewed as one of the most influential bond and risk thinkers on Wall Street, but also because Oaktree itself is one of the world’s most representative distressed-debt and high-yield bond investment firms. Before founding Oaktree, Howard Marks was responsible for distressed debt, high-yield bonds, and convertible bond investments; and in recent years, Oaktree has also raised one of the largest distressed-debt funds in history.

As a result, Howard Marks’s adjustment to his stance on AI this time cannot be treated as merely a tech-observation memo. It’s really telling the credit market: even investors who are the best at doubting and who prioritize down-side risk have started to take AI seriously as something more than a theme, more than a valuation story—something that is already a real force reshaping cash flows, capital expenditures, and debt structures.

What’s truly penetrating is what Howard Marks writes at the end of the memo: a friend recently told him, “I’d rather be an optimistic person who’s wrong than a pessimistic person who’s right.” And his response was, “Me too.”

AI changing the world is real, but it doesn’t mean related assets are fairly priced

Howard Marks’s most striking change in the piece is that he is no longer only asking, “Is AI a bubble?”—he has begun to openly acknowledge that he may have underestimated this technology. He wrote that AI has proven itself to be real, capable of doing a lot of knowledge work, and a technology that is growing extremely fast.

If he had to guess, he would likely say that AI’s potential is more likely to be underestimated than overestimated. But at the same time, he still keeps the vigilance of a bond investor, emphasizing that “AI is very real” does not mean “AI assets are cheap,” nor does it mean that all AI investments have already been fairly priced.

Howard Marks publicly says he’s been convinced by AI

Another key point of this memo is that Howard Marks has, to a large extent, been persuaded by AI itself. In the memo, he mentions that he asked Anthropic’s Claude to help organize some AI teaching materials, and that Claude’s performance made him feel “awe.” He describes the output as if a well-acquainted friend or colleague had written private notes for him—not only echoing the frameworks in his past memos such as interest rates and investor psychology, but also anticipating his skepticism, proactively admitting AI’s limitations, and even adding a bit of humor.

For an old-school investor known for words and judgment, this is almost a public concession.

In terms of argumentation, his most obvious “admission” isn’t that he originally was completely wrong, but that he acknowledges his earlier concerns may have been overly conservative. He previously emphasized whether AI can truly think—whether it’s only reorganizing existing knowledge. But this time, he further accepts a more practical benchmark: for companies and investors, the real issue isn’t whether “AI has consciousness,” but whether “AI can actually get the work done.”

I’d rather be an optimistic person who’s wrong than a pessimistic person who’s right

Howard Marks also makes it clear that AI is no longer just a time-saving tool; it is evolving into agents that can complete an entire task independently. The framework he cites divides AI into three layers: in 2023, it was purely conversational AI; in 2024, it entered a stage where it can use tools to carry out tasks; and now it is approaching Level 3—where, as long as you give a goal, it can complete, check, and deliver the results on its own. Marks emphasizes that the difference may sound subtle, but in reality it determines whether AI is merely a productivity tool or whether it is beginning to replace labor.

But what truly gives this memo market penetration is his line at the end, which can almost be seen as a statement of attitude. Howard Marks wrote that a friend recently told him, “I’d rather be an optimistic person who’s wrong than a pessimistic person who’s right.” And his response was, “Me too.”

When the world’s biggest junk-bond investor admits he was wrong

That’s exactly why this memo is especially important for the bond market. Because now AI expansion is not only a stock story—it’s also a debt story. Google’s parent company Alphabet is funding AI infrastructure expansion through global bond issuances totaling more than $30 billion, including the rare 100-year corporate bonds. As hyperscalers massively expand data centers and compute infrastructure, the amount of new debt issuance by the five largest cloud providers in 2026 could exceed $300 billion

In the past, most of these large tech companies relied on strong cash flow to self-fund. But when AI infrastructure investment easily reaches hundreds of billions or even trillions of dollars, the bond market becomes a new ammunition depot. That’s also why Howard Marks’s shift in tone is worth amplifying and examining: when one of the most famous junk-bond investors in the world starts acknowledging that he may have underestimated AI, it is essentially reminding the entire credit market that this wave of capital expenditures is not just hype—it is a force capable of changing financing structures and pricing credit risk.

And along this debt chain, one of the tightest names for the market is Oracle. Fitch this February maintained Oracle’s rating at BBB, and S&P also kept it at BBB but issued a negative outlook; Moody’s rated it Baa2 and likewise maintained a negative outlook. All three rating agencies are still within investment grade, but they have clearly moved closer to the junk-bond threshold—especially Moody’s Baa2, which is only two notches away from junk.

The reason the outside world is particularly focused on Oracle is that in the AI cloud and data center race, it is noticeably more reliant on debt leverage than cash-flow-heavy giants like Google and Microsoft. Market reports show that Oracle’s current debt level is already in the range of more than $95 billion to $100 billion, and free cash flow has also come under pressure due to large-scale AI capital expenditures.

This article World’s biggest junk-bond investor also admits he was wrong! After Howard Marks’s experience with Claude, admitting he got the AI bubble wrong—first appeared in the Link News ABMedia.