Howard Marks: AI's IPO Euphoria is a Warning Sign
Howard Marks warns that current AI company IPOs exhibit bubble characteristics similar to previous tech booms, with valuations disconnected from predictable earnings. He argues investors cannot confidently forecast AI company profitability 10 years out, and recommends a spectrum approach: hyperscalers with established moats as lower-risk, pure-play AI companies as medium-risk, and startups as lottery-like high-risk plays.
Summary
Howard Marks, founder of Oaktree Capital and prominent value investor, discusses the current IPO frenzy surrounding companies like SpaceX, Anthropic, and OpenAI on the Prof G podcast. Marks argues that predicting these companies' net earnings 10 years forward is nearly impossible, making IPO investments closer to speculation than analytical investing. He demonstrates this through discounted cash flow analysis, showing SpaceX's current valuation at roughly six times his optimistic intrinsic value estimate of $440 billion.
Marks contrasts current AI investing with traditional value investing methodology, which requires examining current earnings and building realistic 5-10 year projections. He emphasizes that successful forecasting requires two components: a forecast AND confidence in that forecast's probability of accuracy. With AI, both are highly uncertain. He compares the current environment to historical tech bubbles including railroads (1860s), radio (1920s), automobiles, computers (1950s-60s), and the internet (2000), noting that while those innovations had specifiable uses, AI's "unimaginable, unlimitable upside" creates unprecedented uncertainty.
Marks highlights the infrastructure spending surge, with Goldman Sachs projecting $5.3 trillion in capex from Meta, Microsoft, Amazon, and Alphabet through 2030, and these companies planning $725 billion in 2026 capital expenditures alone (77% year-over-year increase). He argues that in every previous bubble, too much capital flowed in, too much infrastructure was built, prices were paid too high, and capital providers ultimately lost money.
For investors wanting exposure to AI without maximum risk, Marks describes a spectrum approach: hyperscalers (Google, Microsoft, Amazon) offer established businesses with moats and cash flow; established AI companies (Anthropic, OpenAI, Nvidia) have proven success probability but single-product risk; and AI startups function as lottery tickets with massive upside but likely total loss for most participants. He also suggests looking at less-disrupted sectors (energy, food, timber, home building, metals, mining, paper, chemicals) that offer more predictable growth with less AI disruption risk.
Key Insights
- Marks argues that if someone predicts Anthropic's net earnings in 2036 within 50% accuracy, they would be exceeding probability; therefore, IPO investments in such companies are closer to speculation than analytical investing
- Successful forecasting requires two distinct components: creating a forecast AND forming a judgment about the probability that forecast is correct; most investors make the mistake of high confidence in AI predictions without acknowledging uncertainty
- Marks contends that AI has greater uncertainty and less specifiability than previous transformative technologies (railroads, radio, automobiles, computers, internet), despite potentially being the most powerful innovation
- Every major technological bubble in the last 150 years resulted in too much capital flowing in and too much infrastructure being built, with capital providers ultimately losing money; Marks predicts this AI cycle will follow the same pattern if exuberance continues
- Marks describes a three-tier spectrum for AI exposure: hyperscalers with established moats and cash flow (lower risk), pure-play AI companies already generating revenue (medium risk with high success probability), and AI startups functioning as lottery tickets where most lose money but winners become incredibly wealthy
Topics
Transcript
[0:00] If this exuberance doesn't produce a money-losing bubble, it'll be the first. As Buffett says, it's only when the tide goes out that we find out who's been swimming naked. >> And there is Howard Marks, an incredibly successful Buffett-style investor, founder of Oaktree Capital, one of the world's most successful asset management firms. And recently, he went on the Prof G podcast to explain exactly what he thinks of the IPO frenzy we're currently seeing. So, I'm talking SpaceX, Anthropic, and OpenAI. Honestly, it was a really fantastic interview. I really [0:30] encourage you guys to check out the full thing. I'll leave it linked in the description, but in it, we get a really good sense of…
Full transcript available for MurmurCast members
Sign Up to AccessMore from New Money
Michael Burry's Warning for the IPO Bubble.
Michael Burry warns that massive AI IPOs (SpaceX, Anthropic, OpenAI) raising unprecedented capital at extreme valuations mirror the 2000 dot-com bubble, suggesting markets are near euphoric peaks. Historical IPO data shows capital floods coincide with market tops, not bottoms, indicating investors should scrutinize valuations rather than get swept up in hype.
Buffett's 7-Step Investing Strategy that Made Mohnish Pabrai Rich.
Mohnish Pabrai outlines his 7-step value investing strategy focused on finding undervalued companies with high uncertainty but low risk, emphasizing infrequent big bets, rigorous checklists, and the importance of holding long-term outlier positions. He stresses that even Warren Buffett's hit rate is only 3-4%, making portfolio asymmetry and avoiding premature selling critical to investment success.
'Big Short' Investor Explains How the AI Bubble Will Burst
Steve Eisman, known for predicting the 2008 financial crisis, argues that the AI industry faces severe concentration risks, with 70% of hyperscaler AI revenue dependent on just OpenAI and Anthropic, while these companies bleed money despite massive capital raises. He believes the AI moat narrative is overblown, competition from Chinese models is intensifying, and companies are manufacturing AI existential risk hysteria to lobby for favorable regulation.
Is the U.S. Running Out of Lenders?
The US Treasury has doubled its buyback program for long-term debt to $4 billion per operation, officially citing improved market liquidity but raising questions about whether it's attempting to manipulate interest rates downward as yields approach multi-decade highs. While the Treasury claims the buybacks are routine maintenance, skeptics argue this is the Treasury fighting bond vigilantes demanding higher yields due to America's $40 trillion debt, massive deficits, and inflation concerns.
Steve Eisman: This is How the AI Narrative Collapses
Steve Eisman warns that the AI narrative is vulnerable to a major correction due to three interconnected dominoes: LLM providers lacking competitive moats face pricing pressure from cheaper Chinese alternatives, hyperscalers have massive revenue backlogs dependent on OpenAI and Anthropic's continued spending, and the entire market is overexposed to AI as a single trade. He recommends monitoring the financial health of these LLM providers and favors hyperscalers over pure-play LLM companies.