Aswath Damodaran on the AI Spending Spree: Bubble, Boom, or Both? | #619

Episode Summary

Executive Summary: Azwath Damodaran argues that today’s mega-cap AI leaders are still investable but increasingly expensive, overconfident, and potentially overbuilding. He warns that the bigger risk is not valuation alone but financing excess CapEx with debt and private credit. He also highlights trust erosion, gray-area private markets, sports teams as trophy assets, and prediction markets as both useful and dangerous.

Main Topics: Mega-cap AI valuations and the Mag 7 (Priority: 5/5): Damodaran discusses his continued holdings in five of the seven Magnificent 7 names, while exiting Tesla and NVIDIA. He argues Tesla became a political stock, and NVIDIA’s future growth no longer justifies its stretched valuation. AI disruption and the innovator’s dilemma (Priority: 5/5): He frames AI as a general-purpose technology that will mechanically replace repetitive work, especially in software. Incumbents face the classic dilemma: cannibalize themselves or be disrupted by newcomers. CapEx boom, overconfidence, and financing risk (Priority: 5/5): He thinks large tech firms are overinvesting in AI infrastructure because executives believe they will be among the few winners. The key danger is not just overbuilding, but borrowing to fund it, especially through private credit. Private markets, business models, and the gray zone (Priority: 4/5): He argues that private companies are scaling to massive valuations without clear business models, and that the boundary between public and private markets is blurring as private firms access public-style capital without public-company governance. Trust erosion, gold, and changing asset preferences (Priority: 4/5): Damodaran links rising interest in gold, silver, crypto, and other non-traditional assets to a broader loss of trust in institutions, central banks, and developed-market stability. Sports teams as trophy assets (Priority: 3/5): He says professional sports franchises are priced like luxury trophies for billionaires rather than cash-flow investments, which explains why valuations keep rising despite weak fundamental justification. Prediction markets and gamification (Priority: 3/5): He sees crowdsourced prediction markets as often better than experts, but warns they can be manipulated, thinly traded, and self-reinforcing when used for politics or public sentiment.

Key Arguments: Tesla was sold because it became a political investment, not because valuation alone was the issue; political backlash can distort business demand. NVIDIA was partially sold over several years because it already captured much of the AI architecture upside, leaving limited incremental growth to justify the price. At trillion-dollar scale, investors should reverse-engineer required revenue growth and compare it to the total addressable market; if the math fails, the stock can still trade higher but cannot be fundamentally justified. OpenAI is being treated too much like a public company valuation story without a clear business model; subscription revenue alone cannot support the implied numbers. AI is mostly a continuation of older trends—more compute plus more data—and will hit businesses built on mechanical, repetitive tasks first. Software incumbents have high margins because of stickiness and switching costs, but AI threatens those advantages by making many tasks easier and cheaper to replicate. The AI spending boom is driven by executive overconfidence and winner-take-all beliefs; overinvestment is normal, but debt-financed overinvestment can create systemic fallout. The real risk is not capex itself but how it is financed; private credit exposure could transmit losses from tech overreach into the broader economy. Current market returns are rich but not outrageous; the market may be pricing a transition to a new global order more smoothly than reality will allow. The boundary between public and private markets is increasingly blurred, with giant private firms raising public-like capital without public accountability. Historical market data and factors are useful but not conclusive; long samples can still be artifacts of a specific era rather than timeless laws. Holding cash is not about maximizing return; it is about stability, optionality, and preserving sleep. Sports franchises should be valued as trophies, not investments; billionaire scarcity, not cash flow, sets prices. Prediction markets can outperform experts, but thin markets can be gamed and may create dangerous feedback loops.

Data Points: Podcast episode milestone: 600 episodes - The guest’s prior appearance set the record for most downloaded episode ever for the show. Mag 7 holdings: 5 of 7 - Damodaran says he still owns five of the Magnificent 7 after selling Tesla and NVIDIA. Tesla exit timing: Fairly early - He sold Tesla because it became a politically charged consumer choice. NVIDIA entry price: $1.80 adjusted, split-adjusted basis - He bought NVIDIA in 2018 and calls it one of his best lifetime investments. Trillion-dollar companies: 8 - He says there were eight trillion-dollar companies in the world at the time, including Broadcom. Market-growth hurdle for trillion-dollar valuation: 15% to 20% growth - He reverse engineered implied growth for most trillion-dollar companies and found the market was often assuming plausible but demanding growth rates. OpenAI projected revenue: Almost $150 billion in 2029 - Used as an example of a very aggressive growth forecast. OpenAI revenue base: About $4 billion in 2024 - Damodaran cites this as the starting point for the forecast discussion. Historical sample size: 18,900+ firm-period observations - Referenced from the paper discussed regarding historical growth patterns in public companies. Average compound growth rate: 7% - From the cited U.S. public-company history sample. Standard deviation of growth: 10% - Used in the paper’s normal approximation of growth outcomes. OpenAI forecast rarity: Roughly 9 standard deviations - The cited paper suggests the forecast is extraordinarily unlikely under public-company history. AI capex estimate: $600 billion - He cites expected collective AI capex as a large scale spending boom. Private credit market size: $250 billion to $300 billion - He warns that AI spending financed through private credit could create ripple effects. Google bond maturity: 100-year bond - He calls this inconsistent with corporate-finance asset-liability matching. Start of 2026 implied equity return: About 8.41% - His model-agnostic implied equity return estimate from market prices. Implied equity risk premium: About 4.1% - He says this is roughly in line with the 75-year average. Historical stock risk premium over T-bonds: About 5.4% - He cites this as the long-run average historical premium. Standard error of historical risk premium: 2.1% - He uses this to emphasize uncertainty around long-run averages. Gold return in 2025: Almost 70% - He uses gold’s rise as evidence of lost trust, not just inflation hedging. Silver return in 2025: 150% - He highlights silver’s outsized move as part of the trust narrative. Bitcoin valuation reference: $120,000 peak - He references a prior Bitcoin price peak when discussing treasury allocation debates. Sports franchise valuation multiple: About 8x revenues - He says collective sports-team pricing is hard to justify on fundamentals.

Pivotal Quotes: "This is not a narrative told by a business person, this is a narrative told by a trader." — Azwath Damodaran: His criticism of Sam Altman/OpenAI’s response about business model and stock-price skepticism. "I think this is the big market delusion." — Azwath Damodaran: His description of AI overinvestment driven by overconfidence and winner-take-all expectations. "You hold cash to stabilize the process, to be as a precaution, something you pull on if you need the cash." — Azwath Damodaran: His explanation for why corporate treasuries should not be placed into Bitcoin.

Implications: Investors should separate story from fundamentals, watch financing risk as closely as valuation, and be skeptical of private-market hype. AI may be transformative, but excess leverage and trust erosion could create slow-motion corrections.