Oaktree Capital co-founder Howard Marks' memo "AI Hurtles Ahead" can be seen to some extent as a revision of his views on AI from late last year. The veteran investor, known for his risk management, credit cycles, and bubble warnings, admitted in the memo that over the past 11 weeks, the pace of AI advancement has exceeded his original expectations. If forced to place a bet now, he even believes that AI's potential is more likely underestimated by the market rather than overestimated.

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

Therefore, Marks' adjustment in his attitude toward AI cannot be viewed merely as a tech observation memo. It is actually telling the credit market: even the most skeptical investors who prioritize downside risk are beginning to seriously acknowledge that AI is not just a theme or a valuation story, but a real force that is already reshaping cash flows, capital expenditures, and debt structures.

AI's Impact on the World is Real, But It Doesn't Mean Related Assets Are Priced Reasonably

The most distinct change in Marks' memo is that he is no longer just asking "Is AI a bubble?" but is starting to positively admit that he might have underestimated the technology. He wrote that AI has proven itself to be a real, fast-growing technology capable of doing a massive amount of knowledge work.

If he had to guess, he would now say that AI's potential is more likely underestimated than overestimated. But at the same time, he retains the vigilance of a bond investor, emphasizing that "AI is real" does not equal "AI assets are cheap," nor does it mean that all AI investments are reasonably priced.

Howard Marks Publicly States He Was Convinced by AI

Another key point of this memo is that Marks was largely convinced by AI itself. He mentioned in the article that he asked Anthropic's Claude to help organize some AI educational content, and Claude's performance left him in "awe." He described the output as a personal note written by a familiar friend or colleague, which not only echoed the frameworks of interest rates and investor psychology from his past memos but also anticipated his doubts, proactively admitted AI's limitations, and even included a touch of humor.

For an old-school investor known for his writing and judgment, this is almost a public concession. In his reasoning, his most obvious admission of error is not saying he was completely wrong initially, but admitting that his previous concerns might have been too conservative. He originally emphasized whether AI could truly think or if it was just recombining existing knowledge, but this time he accepted a more realistic criterion: for companies and investors, the truly important question is not "Does AI have consciousness?" but "Can AI actually get the job done?"

Rather Be a Wrong Optimist Than a Right Pessimist

Marks also explicitly pointed out that AI is no longer just a time-saving tool, but is evolving into agents that can autonomously complete entire tasks. The framework he cited divides AI into three tiers: 2023 was purely conversational AI, 2024 entered the phase of using tools to execute tasks, and now it is approaching Level 3, where "given a goal, it can complete, check, and deliver results on its own." Marks emphasized that this difference sounds subtle, but it actually determines whether AI is just a productivity tool or is starting to become a labor substitute.

But what truly gives this memo market penetration is his near-confessional statement at the end. Marks wrote that a friend recently told him: "I'd rather be a wrong optimist than a right pessimist." And his response was: "Me too."

When the World's Largest Junk Bond Investor Publicly Admits a Mistake

Now, AI expansion is not just a stock story, but a debt story. Google's parent company Alphabet has issued over $30 billion in global bonds to fund AI infrastructure expansion, including a rare 100-year corporate bond. As hyperscalers massively expand data centers and computing facilities, the new debt issuance by the top five hyperscalers in 2026 could exceed $300 billion.

In the past, these large tech companies mostly relied on abundant cash flows to self-fund. But when AI infrastructure investments easily reach tens or hundreds of billions of dollars, the bond market becomes the new arsenal. This is why Marks' change of tone is worth zooming in on: when one of the world's most famous junk bond investors begins to admit he might have underestimated AI, it is equivalent to reminding the entire credit market that this wave of capital expenditure is not just hype, but a force capable of changing financing structures and credit risk pricing.

And in this debt chain, one of the names the market is most nervous about is Oracle. Fitch maintained Oracle's rating at BBB in February this year, and S&P also maintained it at BBB but gave a negative outlook; Moody's rated it Baa2 and also maintained a negative outlook. All three ratings are still within investment grade, but they are clearly approaching the high-yield bond threshold. The reason the outside world is particularly focused on Oracle is that in the AI cloud and data center race, it is clearly more dependent on debt leverage than giants with thicker cash flows like Google and Microsoft. Market reports show that Oracle's current debt scale has exceeded the $95 billion to $100 billion range, and its free cash flow is also under pressure due to massive AI capital expenditures.

Disclaimer: None of the information contained here constitutes an offer (or solicitation of an offer) to buy or sell any currency, product or financial instrument, to make any investment, or to participate in any particular trading strategy.