The AI Bubble Will Pop from Equity Supply Says This $80 Billion Asset Manager | Ben Inker | GMO

Ben Inker expects equity supply, rather than an abrupt loss of enthusiasm for AI, to be a possible catalyst for a market downturn over the next 12 to 18 months. He estimates that SpaceX, Anthropic, and OpenAI together could add IPO supply equal to roughly 6% to 7.5% of U.S. market capitalization in a year, an unusually large increase. Inker cited a $75 billion SpaceX IPO and said roughly another $2 trillion of SpaceX shares could become available for sale over the following 12 months, with a decent portion likely to be sold. He expects similar opportunities for owners of Anthropic and OpenAI shares.

Staggered share unlocks would be less disruptive than selling all at once, and SpaceX may spread sales out to protect its own share price, Inker said. But he expects nearly all owners to have an opportunity to sell within about a year. Historically, each 1% of market capitalization issued through IPOs in a year has been associated with market performance about 4% worse than normal. Six percent of supply would therefore imply performance about 24% worse than normal, not a guaranteed rout. IPOs accounted for roughly 2.5% to 4% of market capitalization in 1999–2000; the projected current inflow would be the first year above 4%. The largest IPO-supply years, including 1999, 2000, and 2021, were followed by difficult markets.

Inker says markets may now be less responsive to price changes because passive investing has grown from under 10% of the market in the 1980s to perhaps half today, while active management has declined and market concentration has increased. With fewer investors likely to sell as prices rise, a large new supply of shares may require larger price moves to find buyers. High-frequency traders can help manage temporary imbalances, but their risk limits restrict how much they can absorb. The result, in his view, is not a certain crash but a market that could need more pronounced repricing to clear unusually large issuance.

The underlying AI investment boom also carries risks beyond share supply. Inker sees compute and AI models as competitive businesses rather than winner-take-all markets: customers can switch providers, and widespread adoption requires low prices that may constrain returns. If the buildout proves excessive, data centers could be written off and hyperscalers could earn poor returns. Abundant compute could also undermine credit-funded firms and cause some equity-funded companies to fail. In a severe overcapacity scenario, Inker said a 35% to 50% market decline was possible, not certain. He suggested Microsoft could fall but remain viable, while CoreWeave and possibly SuperMicro could fail; markets could turn down before the scale of overinvestment becomes evident.

How much an AI downturn hurts a company depends on its valuation and how large AI is relative to its existing business. Inker said an established business such as Meta’s may retain value, while an expensive stock can fall sharply even if its business remains sound. He sees Oracle’s AI ambitions as large relative to its existing business and considers smaller or AI-focused companies more exposed. GMO favors reasonably priced companies with decent returns on capital and an AI “kicker,” rather than concentrating portfolios in one theme. Inker also says GMO favors non-U.S. stocks and value over growth because the valuation spread is wide; a wider spread can benefit value investors when companies move into the growth group, while leaving growth investors exposed to the premium they paid.

Inker cautions that an AI downturn need not become a broader financial disaster: losses from the internet-era fiber overbuild did not destabilize the system because they were not concentrated in heavily leveraged entities or banks, and he expects the Federal Reserve to have room to cut rates in a recession. Credit spreads were unusually tight at the time discussed, with high-yield and investment-grade spreads around the 13th percentile; triple-C spreads were around the 74th percentile at the end of September, while double-B and single-B spreads were especially tight. He sees limited upside in credit when conditions are good against potentially large losses if they deteriorate, and said some borrowers’ capital structures no longer work at 5% interest rates. Public credit showed little distress, but private-credit default estimates ranged from 1% to 6% to 19%, and he did not know which was right.