Half the S&P 500 now sits in 20 stocks. Fund managers just called it a bubble.

Forty-five percent of the world's fund managers just named an AI bubble as the biggest tail risk in markets, according to Bank of America's July 2026 Global Fund Manager Survey. That is a 17-point jump in a single month, from 28% the prior survey, and it displaced second-wave inflation from the top spot. The same respondents ranked long positions in global semiconductors as the world's most crowded trade and hyperscaler AI infrastructure spending as the most likely source of a credit event.

A "tail risk" is the worst-case scenario investors watch for, not the one they expect. A "crowded trade" is a position so many funds hold at once that unwinding it would move prices. Together, those two rankings say the same thing: the same people running the AI infrastructure trade are also publicly flagging it as the trade most likely to break.

The stakes sit inside the index. J.P. Morgan Global Research estimates the top 20 S&P 500 stocks now account for roughly 50.8% of total market capitalization, the heaviest concentration in roughly 50 years. The five largest hyperscalers, Microsoft, Google, Amazon, Meta, and Oracle, are expected to deploy more than $1 trillion across 2025 and 2026 combined on AI infrastructure. These are the handful of cloud and platform companies now building AI data centers at scale. Goldman Sachs pegs 2026 AI investment at more than $1 trillion in a recent Insights forecast, and the same research notes annualized AI-related spending could exceed $800 billion by year-end.

That is why the survey number lands harder than it reads. Owning the S&P 500 today is no longer a bet on 500 companies. It is a bet on roughly 20, and most of them are AI infrastructure plays. The mechanical link between hyperscaler capex, earnings, and index weight is the lever the survey is pointing at: the capex is what builds the data centers, the earnings are what justify the valuations, and the index weight is what forces every pension and index fund to hold the same names.

The institutional warnings extend beyond the buy side. The Bank for International Settlements, the Basel-based forum for the world's central banks, warned earlier in 2026 that Big Tech's spending spree risks becoming a prolonged investment bust, a frame Morgan Stanley's 2026 AI Market Trends report uses to flag the same risk from the sell side. The label matters: the BIS is not a Wall Street desk, and the warning is not one analyst's view.

On the earnings side, a pseudonymous analyst writing as "Mac10" noted on August 8, 2026, that forward earnings growth is accelerating at a record pace only because of unprecedented AI capex, much of which lands on profit statements as one-time boosts rather than sustainable operating performance. If that reading holds, the same AI capex driving concentration and crowding is also what the market is paying the highest multiple for, and the same spending the BofA survey ranks as the most likely source of a credit event.

The household layer is the second-order point. Equity allocation as a share of household wealth is elevated relative to past cycles, so any sharp drawdown would land harder on a population that holds more stock, directly or through retirement accounts, than at any point in the post-war era. The dot-com parallel the BofA survey itself invites is the right one: in 2000, concentration was the warning, and the warning was correct, even as the buildout kept going for another quarter, and then another.

The watch item is not a date. It is whether four institutional layers stay aligned: the BIS warning, the BofA survey, the JPM concentration estimate, and the Goldman spending forecast. The BofA survey is published monthly. The next reading is the cleanest tell.