Investors Are Pulling Back From the Crowded AI Trade

Bank of America’s latest fund-flow analysis shows a meaningful change in how active managers are expressing the AI investment thesis. The move is less a verdict that artificial intelligence or semiconductor demand will collapse than a repricing of crowded positions, financing risk and the time investors may have to wait for infrastructure spending to generate returns. 2

The headline: $44.2 billion leaves global semiconductors

Active long-only funds sold $44.2 billion of global semiconductor equities in the latest month, according to Bank of America data reported by Investing.com. The selling marks a sharp reversal after semiconductors became one of the market’s most concentrated AI-linked trades. 2

That distinction matters. The data describe portfolio positioning by active managers—not a forecast that AI spending is ending. Investors can reduce exposure to the broad semiconductor sector while retaining large positions in companies they view as better positioned or more attractively valued.

AI compute and quantum computing face the strongest pressure

The heaviest selling over the past year was concentrated in AI compute and quantum computing themes. At the same time, managers directed more money toward energy storage, precision medicine, space and cybersecurity. 2

The pattern suggests that investors are moving away from themes whose eventual economics remain difficult to time and toward areas with more immediate commercial or infrastructure narratives. It also reflects a preference for selectivity: ASML and TSMC can remain important benchmark-relative holdings even as funds reduce their overall semiconductor exposure. 2

Bank of America’s July fund-manager survey had already identified long global semiconductors as the market’s most crowded trade, with 82% of respondents naming it the most crowded position. 1415

The rotation is broader than technology

The latest regional flows reinforce the idea that this is a wider portfolio adjustment rather than a single-sector trade:

• US equities: $64.8 billion in outflows
• European equities: $25.9 billion in outflows
• Japanese equities: $13.9 billion in inflows

On a year-to-date basis, Japan had attracted $34.1 billion, while global energy received $28.2 billion and materials received $25.7 billion. Global telecommunications recorded a $16.7 billion inflow in the latest month—the largest sector inflow in the data reported. 2

The preference is consistent with a search for exposure that is less dependent on a small group of high-growth technology companies continuing to spend at an accelerating pace.

Where managers are overweight

By theme, portfolios are most overweight in:

• Precision medicine
• Luxury lifestyle
• Grid modernization

The largest dollar overweights relative to regional benchmarks include:

• United States: Philip Morris and Meta Platforms
• Europe: ASML and British American Tobacco
• Japan: Keyence and Hitachi
• Emerging markets: TSMC and Samsung Electronics 2

These positions show why “selling semiconductors” should not be read as a uniform rejection of every company connected to AI. Managers appear to be distinguishing between individual companies, regional exposure and broad thematic concentration.

Where portfolios are underweight

The deepest theme underweights are quantum computing and AI platforms. 2

That positioning points to a central investor concern: not whether AI will matter, but whether current spending levels and valuations can be justified soon enough. A company may have a credible long-term AI strategy while still being a less attractive portfolio holding if the required investment is rising faster than visible cash returns.

Why the bond market is adding to the concern

The reassessment is also appearing in credit markets. Alphabet, Amazon, Meta, Microsoft and Oracle issued approximately $220 billion of bonds through August 10, 2026, compared with $12.5 billion during the comparable period a year earlier, according to data cited by Reuters. 21

That borrowing surge matters because AI infrastructure is increasingly being financed through debt as well as operating cash flow. Reuters reported that technology corporate bond spreads had widened to about 89 basis points, roughly 9 basis points above the broader investment-grade market. Wider spreads mean investors are demanding more compensation to absorb the additional supply and perceived uncertainty. 18

Amazon’s $25 billion long-dated bond sale priced at roughly 120 basis points over comparable US Treasuries, another sign that bond buyers are becoming more selective as AI-related issuance grows. 18

The credit-market evidence does not prove that AI investment will fail. It does show that the financing burden is becoming part of the investment debate. Equity investors are asking when capital spending will translate into earnings, while bond investors are asking for more yield before funding the expansion.

What the flow data mean for the AI trade

Taken together, the evidence points to a change in portfolio construction:

1. Less indiscriminate exposure: Managers are reducing broad, crowded semiconductor and AI positions.
2. More attention to timing: The question is shifting from whether AI has potential to when returns can justify the infrastructure bill.
3. Greater preference for tangible beneficiaries: Grid modernization, energy storage, telecom networks, materials and selected industrial and health-care themes offer exposure to the consequences of AI investment without relying entirely on the most crowded software or compute narratives.
4. More regional diversification: Japan and other non-US allocations are gaining attention as funds reduce exposure to US and European equities. 2

The most defensible conclusion is therefore not “investors are abandoning AI.” It is that the market is moving from owning the entire AI capital-expenditure chain at almost any price toward demanding clearer valuation support, nearer-term cash flows and more selective exposure. The next test will be whether AI-related revenue and productivity gains arrive quickly enough to validate the scale of spending now being financed.