AI is the most crowded trade on Wall Street. Half of investors still say it isn't a bubble.

In Bank of America's July 2026 Global Fund Manager Survey, 82% of respondents call long global semiconductors the most crowded trade on record, and roughly half of the same group still say we are not in an AI bubble. That contradiction is not a pollster's error. It is the trade.

The survey, fielded July 2 to 9 and published July 14 by Bank of America, polled roughly 200 to 400 institutional investors and asked, in effect, two questions at once: how worried are you about an AI bubble, and how much of the AI trade do you already own. The answers moved in opposite directions. AI bubble risk rose to the top tail risk at 45%, up from 28% in June, while the share of respondents calling the trade "crowded" climbed for a third straight month. BofA notes that "no one is short."

When 82% of professional money managers already hold the same position, the question stops being whether the trade is right and becomes who blinks first. Bank of America's own Bull & Bear Indicator, a composite sentiment gauge, sits at 9.4, above the 8.0 threshold that historically flags a contrarian sell signal. Business Insider calls it the most extreme reading in roughly five years. Cash allocations fell to 3.6% from 4.1% in June, a level BofA labels "uber-low," and that drop also triggered the bank's contrarian cash rule.

Both sides have reasons to feel certain. Bulls argue the spending will be repaid by future growth, productivity gains, and profit: 61% of investors do not expect AI hyperscalers, the largest cloud and AI infrastructure operators, to cut capital spending this year, against 28% who expect a cut. Bears argue the cost of compute is too high, that open-source and Chinese models will siphon demand, and that returns will disappoint. Neither camp is bluffing. The record 54% of respondents expecting a "no landing" for the global economy, against only 2% expecting a hard landing, tells you how confident both groups feel about the macro backdrop.

The clearest way to see the underlying mechanism is BofA's "generational transfer" phrase, which the bank's strategists use to describe a directional flow of free cash from U.S. tech megacaps to chipmakers. The Magnificent Seven, the seven largest U.S. tech megacaps, have spent $234 billion on capital expenditure year to date in 2026, according to BofA strategists as reported by The Globe and Mail. The stocks have barely moved. Investors are watching hyperscaler free cash flow turn negative, and that is the point. The trade is no longer a call on AI demand. It is a call on who captures the spending, and on whether the spending pays back before the cycle turns.

Q2 U.S. earnings season, which starts in roughly a week, is the proximate test. Hyperscaler guidance on capital spending, free cash flow trajectories, and any signs of demand erosion from open-source or Chinese alternatives will land directly on the question both camps are asking. Investors have also raised U.S. equity allocations to the highest overweight since December 2024, cut their end-2026 oil forecast to $71 from $86, and told BofA that 83% of them do not expect the Federal Reserve to hike rates before the November midterms. CryptoBriefing and InvestingLive both confirm the bullish tilt alongside the record crowding read.

The paradox in the survey is the position. The fund managers who call AI the most crowded trade and the fund managers who deny an AI bubble are, increasingly, the same people. Whether the earnings calendar breaks the trade or entrenches it, the conviction on both sides is the trade.