GMO blames share deluge for tomorrow's bust

Ben Inker, co-head of asset allocation at GMO, which oversees about $80B, made an unusual claim in his third-quarter letter: AI enthusiasm looks like a classic bubble to him, but the trigger will not be weak earnings — it will be a flood of new shares hitting the market within 12 to 18 months. I take the claim seriously because he argues about the fuse, not the existence, of the bubble, and the figures in the letter back him up.

Every transforming technology arrives with a wealth promise: canals, railways, autos, electricity, the internet. Everyone sees the world changing and wants a seat. The trouble is that nobody knows yet where the profit will land. Early days bring hot demand, fat returns, and headline wins, which pull in floods of fresh investment. Yet high returns are born of scarcity; where funds pour in, returns dry up. To me the current AI race replays that old film frame by frame.

The first London-Manchester rail line earned a fortune, yet once six lines served the route nobody earned anything. That is the harsh math of bubbles: the weak new bet fails, and the earlier sound bets burn with it. The lesson for any investor is blunt — where everyone rushes in, few walk out winners. I like Inker's railway example because today's accelerator build-out mirrors it almost exactly.

GMO's figures hit hard: when new share supply equals 1% of market value, stocks trail their average by about 4% over the next year. Scale it simply and a 6% supply wave means a 24% headwind. Why? Cash for new shares usually comes from selling old shares, and when former owners steer proceeds toward taxes, debt, or fresh ventures, no new money reaches the market. Many shares plus few buyers leaves one outcome — lower prices. I picture a street market: add ten stalls to the same lane and shoppers stay flat while prices slide.

SpaceX is the first and loudest source of that wave. In June the firm priced its record $75B IPO at $135, floating only about 5% of its shares. Reuters reports that the first lockup release freed more than 911.5M shares, with roughly 40% of the equity tradeable by December. Musk stays locked until June 2027, yet the final batch unlocking then runs near $2T on its own. My note is short: supply will set the price, not the other way round.

Two more giants wait in line. OpenAI and Anthropic each submitted confidential IPO filings in June at talks near $1T apiece, together worth about 5% of America's investable equity value. Reuters says the two rivals race for investor appetite, with the first mover catching the wind. Bloomberg adds that each listing would rank beside JPMorgan or Exxon in size. Should all three giants land back to back, the cash pulled from the market would dwarf anything seen before.

A deeper break hides beneath: for years US buybacks beat new issuance, shrinking share supply by about 1% a year. That pattern has flipped, and outside crisis years the market faces net dilution — issuance beating buybacks — for the first time in 20 years. The culprit is hyperscaler spending: when cash runs short, firms sell shares, and sold shares swell supply. In my eyes this is the bubble's second self-feeding engine.

Why cannot the market absorb that flow as smoothly as before? The buyer base changed. An S&P 500 fund never sells just because prices rose; it needs a rule. Leveraged single-share funds do not sell into strength either — they buy. Career-risk active managers hug their benchmarks to avoid lagging the herd. What remains are quick traders such as Citadel and Jane Street, who smooth seconds-long wobbles but carry no buckets. Patient buyers have thinned out, so new supply moves prices more.

The internet bubble was the trailer for this film. Fiber glut built quietly while stocks still rose, and the true trigger was a rain of internet shares. The years 2000 and 2021 marked the largest supply peaks of the last fifty years, each landing at peak euphoria, and what followed is history: the Nasdaq fell more than 80%. Inker's warning chills for that reason — prices can turn before results sour. While everyone watches demand, I watch the supply schedule.

The credit side of the story runs quieter than the equity side yet burns just as hot. Borrowing setups built for 0% rates do not work at 5% rates. Debt flowing toward AI themes has grown so large that single-theme weight inside corporate credit looks unmatched since the railway era. Lenders now ask for extra yield, and as that yield climbs, firms rolling debt forward carry heavier loads. Once that loop starts turning, the weakest balance sheets pay first.

Now read that abundance story in reverse. Suppose half of all accelerator power today serves training of new systems, and that training share drops toward 10%. A flood of idle capacity hits the market at once. Five-year-old accelerators turn to scrap, since new hubs deliver far more answers per watt of electricity. Old hubs face faster write-downs while their owners still owe lenders.

Missed loan payments follow, equity values sink, and hyperscaler investment returns collapse with them. Inker asks whether that picture costs the market 18%. His answer is no — the fall lands between 35% and 50%. I read that math as a fire drill, not a horror film.

Hyperscaler balance sheets fund this film's budget. CNBC reports 2026 spend near $700B across four giants: Amazon near $200B, Alphabet near $185B, Meta near $135B. Alphabet's free cash has melted by nearly 90%, and Meta looks little different. Bloomberg puts the total at $725B. Money runs from silicon to buildings, cooling, and grids. Cloud growth stays strong but trails the pace of spend. As an investor I ask one thing: if returns lag, who pays the bill? Usually the share price.

The first pain lands outside AI. Housing crawls, while autos and durables live on credit. Rates rose from the floor, and heavy AI borrowing pushes them higher still. Every activity beyond AI goes thirsty. A BusinessInsider report notes the same point from the GMO team: in today's passive, benchmark-hugging market, fewer buyers stand ready to absorb fresh shares. My image is a garden: water one tree with the whole hose and the rest dries out.

Memory producers confuse investors most right now. Micron closed a record year with gross margins above 80% while customers signed $22B of purchase deals. The firm's investor page glows. Yet Inker warns that memory is a commodity trade where the market sets the price. At record margins, shares look cheapest against forward profit — but if margins normalize, the firm may ship more units for less revenue. My rule stands: cheapness priced off peak profit usually proves dear.

A profit scan shows the same split personality. Large-cap earnings sit at records while small caps languish near the 20th percentile of history. At a normal cycle peak, small fry would party; today they barely stir. Cause: a handful of giants holds stronger bargaining power as rivalry fades. Few set the terms, few set prices. That shape fattens top profits yet widens fragility — when few hands hold such power, their tremor chills everyone.

Asset mix sits on the table too. With the classic 60-40 mix of shares and bonds wobbling, investors hunt for fresh balance. Across history, energy stands out as a rare diversifier that often zigs when both stocks and bonds zag. GMO's recipe for the flexible investor leans that way: non-US shares, deep-value shares, and liquid alternatives can chase fair returns without hugging the bubble. Their Benchmark-Free sample mix targets 6.5% real gains. I like the recipe: sheltered if right, decent if wrong.

To sum up: bubbles burst on flooding supply, not fading demand. A rain of shares looms, lockup dates sit in public view, and the credit tap flows toward one theme, so caution reads as common sense. I would trim US weight, widen bonds and non-US shares, and keep cash ready. Should markets prove me wrong, my loss stays small; should they prove me right, these lines will have earned their keep.