Building on Derek Thompson’s dictum that “AI could be the railroad of the 21st century,” Marks leaned on the Mark Twain-esque idea that “history rhymes,” arguing that the script of exuberance, overbuilding, and painful correction has repeated itself from railroads to radio, aviation, and the dot-com era. For him, the most important question is not whether AI is important – “I haven’t met anyone who doesn’t believe that” – but whether today’s enthusiasm has crossed the invisible line from optimism to a bubble.”
Flexion bubbles that change the world and yet destroy wealth
Marks highlighted a distinction he finds “fascinating” between “mean-reversion bubbles” and “inflection bubbles,” as outlined by Byrne Hobart and Tobias Huber. The first – subprime mortgage bonds, portfolio insurance, the South Sea Company – promise returns without real social progress and simply rise and fall, destroying wealth. The latter, linked to the railways or the Internet, “accelerate technological progress and lay the foundation for a more prosperous future, and they destroy wealth,” with the caveat that he adds: “The key is not to be one of the investors whose wealth is destroyed in the process of creating progress.” Quoting Ben Thompson and Carlota Perez, Marks emphasized that bubbles “can be important catalysts for technoscientific progress” because they “create opportunities to deploy the capital needed to finance and accelerate such developments.” large-scale experiments.” Bubbles, he said, are “bombarding a new area of opportunity” with money, compressing decades of development into a few years, even as much of the capital is “burned” along the way.
AI’s grip on the markets
As for the current market, Marks pointed out that AI is now responsible for a “very large portion” of total corporate capital investment, accounts for a large portion of U.S. GDP growth, and has driven the “vast majority” of the S&P 500’s profits. Citing a Fortune headline — “75% of Earnings, 80% of Earnings, 90% of Investments — AI’s Grip on the S&P is Total and Morgan Stanley’s Top Analyst is ‘Very Concerned’” — he said the enthusiasm around AI has likely lifted non-AI stocks as well.
Nvidia epitomizes the boom: from a market cap of $626 million at its 1999 IPO to briefly becoming the first company to be valued at $5 trillion, a valuation of “about 8,000x,” or roughly 40% per year for more than 26 years. At the same time, Marks highlighted what he calls “lottery thinking” in early-stage AI betting, citing Etched CEO Gavin Uberti’s claim that “if transformers disappear, we will die,” but “if they survive, we will be the greatest company of all time,” and the underlying logic that a one-tenth of a percent chance of a $1 trillion outcome can “force” investors to play.
Circular money, SPVs and the debt machine
One of Marks’ sharpest warnings was reserved for the financing of structures around the development of AI. He flagged “circular deals” reminiscent of the telecom boom of the late 1990s, where OpenAI “stands to receive billions from tech companies, but also send billions back to the same companies” for computing, and where Nvidia’s $100 billion investment in OpenAI is coming back in the form of chip purchases. These arrangements could account for 15% of Nvidia’s revenue next year, according to Goldman Sachs estimates.
Marks also pointed to OpenAI’s $1.4 trillion investment commitments to industry counterparties, to be paid from the same parties’ revenues. Marks wondered whether the industry has developed “a perpetual motion machine” and reminded readers how difficult it is to understand what “a trillion dollars” really means. He then turned to the rise of off-balance sheet Special Purpose Vehicles (SPVs) in data center financing, echoing Azeem Azhar and Paul Kedrosky’s concerns that vendor financing, thin coverage ratios, and the proliferation of SPVs are classic signs of a “Minsky moment” in which credit expansion moves from good to bad projects.
The winner gets the most technology
From Oaktree’s perspective, Marks said the “right way” to play most technological revolutions is “through equity, not debt,” citing Bob O’Leary’s view that in the “winner-takes-all” or “winner-takes-most” markets, the single big stock winner more than offsets losses on the losers. In contrast, for a “diversified pool of debt exposures,” lenders only earn the coupon on the winner, which is “grossly insufficient” to offset impairments on bankrupt borrowers.He emphasized three hard truths about debt in speculative booms: it magnifies losses, increases the chance of failure in times of recession, and ultimately puts lenders’ capital at risk when tough times become severe. A glut of overbuilt data centers could drive some owners into bankruptcy, he notes, allowing a new cohort to buy assets “for pennies on the dollar” from exclusionary lenders, a classic cycle of “creative destruction” that stabilizes the industry but leaves early, leveraged players taking big losses.
Why this bubble could be different
Marks has carefully explained why AI is not a copy of the dot-com era. Unlike 1999, he noted, AI products “already exist at scale,” demand is “exploding” and revenues are rising rapidly, with coding-focused companies like Anthropic and Cursor having grown revenues “tenfold” two years in a row and are on track to become $1 billion businesses from virtually zero. He also points out that today’s leading AI names – Nvidia, Microsoft, Alphabet, Amazon, Meta – have built businesses, profits and cash flows, and are trading at price-to-earnings ratios well below the extremes we saw at Microsoft, Cisco and Oracle during the dot-com bubble or Nifty-Fifty era.
Yet he also listed the similarities that skeptics cite: a change-the-world technology, exuberant speculative behavior, FOMO, suspect circular deals, SPVs and even $1 billion seed rounds like Mira Murati’s Thinking Machines, which raised $2 billion at a $10 billion valuation – and is now reportedly looking to fund $50 billion without a public product. A second example he cites is Ilya Sutskever’s Safe Superintelligence raising $2 billion at a $32 billion valuation “despite having no publicly released product or service,” reinforcing his sense that capital chases stories as much as cash flows.
Marks’ result is deliberately unsatisfactory for anyone looking for a binary call. Borrowing Alan Greenspan’s quote, he accepted that there is clear “exuberance” around AI, but emphasized that “virtually no one can say for certain” whether it is irrational, because bubbles are “best identified in retrospect” and the technology’s “enormous potential” lies alongside “enormous unknowns”.
Instead, he offered a behavioral prescription: “no one should go all in without recognizing that they risk being ruined if things go badly” – but “no one should stay all out and risk missing one of the great technological leaps forward.” The sensible course, he argued, is a “moderate position, applied with selectivity and caution,” supported by “sober, insightful judgment and skillful implementation,” and a clear eye on how much of your capital you can afford to see destroyed if AI follows the typical “flex bubble.”
(Disclaimer: Recommendations, suggestions, views and opinions expressed by the experts are their own. These do not represent the views of the Economic Times)
#bubble #euphoria #Howard #Marks #warns #bet #house #future