Oaktree's Howard Marks Says AI Progress Does Not Guarantee Investment Returns

A Bubble Question in Two Parts

Howard Marks, co-founder of Oaktree Capital Management, has set out why rapid progress in artificial intelligence does not by itself guarantee returns for the investors funding the build-out. In a December 9, 2025 memo titled “Is It a Bubble?”, the veteran credit investor argued that two questions must be kept apart: whether AI will genuinely transform industries, and whether the companies building the technology are worth what markets currently pay for them. He stopped short of declaring the AI market a textbook bubble, and framed the distance between technological change and investment payoff as the core of the debate. Marks wrote that he lacks the expertise to judge whether the aggressive capital spending by AI companies will prove justified, so he directed his analysis at investor psychology surrounding the theme rather than at corporate investment behavior or trading volume data. Sentiment, in his reading, is the more workable gauge of whether a bubble is forming, and he placed the burden of proof on prices rather than on the technology itself. The heart of the memo is a framework: he sorts bubbles into two categories. The first, mean-reversion bubbles, covers assets whose prices rise on a new financial fashion and fall back toward earlier levels once enthusiasm fades. The second, tipping-point bubbles, leaves something behind: even after prices break, the technology and infrastructure built during the run-up remain in place and can reshape the structure of the economy. That distinction explains why investment performance cannot be read off technological success alone. Even if AI keeps advancing and remakes industry after industry, the gains may not accrue to the companies currently absorbing the capital. Whether those firms' outlays convert into earnings, whether productivity improvements reach corporate bottom lines, and whether today's share prices already discount that potential were all left in the memo as open questions.

The December memo was part of a sequence rather than a standalone comment. Marks returned to the subject in a follow-up memo, “AI Hurtles Ahead”, published on February 26, 2026, which took up the pace of AI development and the change in what the systems can do. In it he suggested that AI may move beyond its current role of supporting human productivity toward performing work on its own. The capability jump, in his framing, sharpens the original question instead of settling it. The warning from December carries through unchanged: a technology's growing potential does not by itself establish that the prices attached to it are right. Read together, the two memos give investors a way to hold the threads apart. In a mean-reversion bubble, the risk sits in timing, because prices reflect a financial fashion and give back most of their gains when the fashion passes. In a tipping-point bubble, an investor can be wrong on price and still end up in a world transformed by the technology, since the infrastructure survives the drawdown. That is the ambiguity Marks said makes this episode hard to call, and it is why he resisted issuing a single verdict on whether AI stocks are overvalued. He kept his own conclusions deliberately narrow. The memos do not endorse current valuations, do not predict which companies will capture the gains and do not put a figure on what the sector is worth. Marks confined his judgment to what he said he could assess, the behavior of investors chasing the theme, and left the accounting questions, whether spending becomes earnings and whether share prices already reflect the upside, unresolved. The series, as of the February memo, ends on that gap between capability and price.

For investors allocating across technology themes, the memos leave a test rather than a forecast. Marks's point is that adoption can be real while returns still disappoint, so the two have to be measured separately: capability announcements answer the first question, cash flows and market capitalizations answer the second. The February 26, 2026 memo is the latest word in the series, and it ends where the December one began, with the valuation question unresolved and with AI, in his description, moving from assisting work to performing it. Whether today's prices survive that transition is the open position every investor in the theme carries.