What does Bank of America’s August 25, 2026 analysis reveal about the sharp retreat by active long-only funds from global semiconductor and AI-related investments—including the $44.2 billion sale of chip stocks, the broader sell-off in AI compute and quantum-computing themes, reg

Bank of America’s August 25 analysis points to a sharp de-risking of the most crowded AI-equity exposures—not a wholesale exit from equities. The parallel deterioration in AI-infrastructure credit suggests investors are questioning both valuation and the financing burden of the buildout, but available evidence does not establish that a systemic credit event is underway.

What the equity repositioning indicates

• Active long-only funds sold $44.2 billion of global semiconductor shares in the prior month and shifted $16.7 billion into telecom, signaling a rotation away from the AI hardware trade toward relatively defensive, cash-generative beneficiaries of data demand. 9

• The significance is amplified by prior crowding: in July, 82% of surveyed investors had called long global semiconductors the market’s most crowded trade. 8 The reported retreat from chips, AI compute, and quantum-related themes is therefore consistent with profit-taking, reduced concentration risk, and concern that AI capex may not earn adequate returns quickly enough.

• This was selective rather than an unambiguous broad-market risk-off move. BofA’s August survey showed global-equity allocations at a net 56% overweight and cash at 3.5%, while long semiconductors still ranked as the most crowded trade, albeit less so than before. 10

• The supplied evidence does not adequately substantiate the precise country-by-country flow figures, the full list of favored sectors beyond telecom, or the exact current regional and sector overweight/underweight rankings. One corroborated regional point is that India was reported as Asia’s most underweight market in BofA’s August Asia survey. 15

Link to AI-related credit risk

• The equity rotation coincides with a change in how the AI buildout is financed: major technology companies have increasingly raised debt and equity for AI infrastructure rather than relying principally on internally generated cash. 1

• Hyperscalers—Amazon, Alphabet, Meta, and Oracle—had issued about $194 billion of bonds through July 7, up 79% from roughly $108 billion in all of 2025. 4 Amazon separately sought at least $25 billion in a bond sale to fund heavy AI investment. 5

• Heavy supply has tested investor absorption. Reports described bond-market “indigestion,” with technology-credit spreads widening—the extra yield investors require to own corporate debt—indicating that investors are demanding greater compensation for AI-capex and funding risk. 2

• The central economic concern is a mismatch: debt service is contractual and immediate, while revenue, utilization, pricing power, and return on AI data-center investment remain uncertain. If expected AI cash flows disappoint, lower equity valuations and wider credit spreads could reinforce each other: weaker share prices raise financing pressure, while higher borrowing costs reduce projected returns on new infrastructure.

How large is the systemic risk?

• A broader AI-led correction is plausible because positioning had been concentrated and the financing cycle is increasingly debt-dependent. The most vulnerable areas would likely be expensive AI hardware, companies with high capex but unclear monetization, suppliers dependent on hyperscaler spending, and long-dated technology credit.

• But a systemic credit event is not demonstrated by the evidence. Goldman Sachs’ view was that spillovers from AI issuance into the wider debt market had remained limited, even though AI-related financing represented almost one-quarter of gross investment-grade issuance. 3

• Thus, the evidence supports a warning of a crowded-trade unwind and a tightening AI-financing constraint—not a confirmed credit crisis. The key indicators to watch are further tech-spread widening, weak demand or larger concessions in new hyperscaler bond deals, cuts to AI capex, evidence of poor data-center utilization or AI monetization, and a renewed broad sell-off in semiconductor equities.