Semiconductor stocks have been among the most popular institutional trades since the AI boom reshaped Wall Street portfolios, and the momentum has continued into 2026.
Data from Goldman Sachs‘ prime brokerage division reveal a shift in hedge fund positioning that complicates the surface-level picture of broad institutional AI buying.
A divergence has opened up between hedge fund flows and the rest of the technology trade as stocks continue to reach new highs.
Goldman’s own analysts, however, offer a very different reading of what the selling means for the artificial intelligence investment story.
Chip stocks rank as the most net-sold U.S. subsector in Goldman’s data
Semiconductor and equipment stocks topped every other U.S. subsector in net selling among hedge fund clients over the prior month, with reductions concentrated in long holdings rather than new shorts.
Selling consisted primarily of reductions in long holdings rather than new short positions, as detailed in a May 22 prime brokerage client note.
The sector has shifted into net-sold territory on a year-to-date basis, reversing the first quarter’s trend of steady institutional inflows. Funds were selling into a sustained rally to manage growing semiconductor positions, Co-head of Prime Insights and Analytics Vincent Lin explained.
The mechanical growth in portfolio weights matters significantly for managers bound by internal sector exposure limits, Lin noted in the report.
When a holding rallies sharply while the rest of a portfolio stays flat, the position’s share of the total book expands without additional purchases.
Goldman’s AI semiconductor basket has outperformed the S&P 500 by more than 50% since the start of 2025, the bank’s tracking data showed.
Gains of that magnitude in a concentrated sector generate rebalancing pressure across most institutional portfolios with sector exposure limits, the firm’s analysis indicated.
AI conviction stays near record highs among hedge funds
Despite the semiconductor selling, hedge fund exposure to U.S. artificial intelligence stocks within Goldman’s technology, media, and telecommunications tracking basket remains close to all-time peaks.
The selling does not reflect a fundamental shift away from AI as an investment theme for hedge funds, Lin noted in the report.
The gap between trimming a winning position and losing confidence in a multi-year thesis matters for interpreting hedge fund flow data, the report indicated.
More AI:
- Micron sits at the center of a red-hot chip rally
- IBM CEO sends blunt message on AI and quantum computing
- Anthropic CEO makes shocking admission about AI
Chip stocks remain the primary vehicle through which hedge funds have expressed their conviction in AI infrastructure as a sector.
Cumulative hedge fund purchases of semiconductor names since early 2025 rank among the highest of any U.S. subsector, the Goldman report found.
Hedge funds also appear to be rotating capital geographically rather than retreating from equities as a whole, the firm’s broader positioning data showed. Asia ranked as the most net-bought region recently, according to the firm’s flow data.
Hedge fund macro hedging climbs to a 10-year high amid rising bond yields
Alongside the semiconductor selling, hedge fund short positions in equity indexes, ETFs, and macro instruments have climbed to their highest levels in a decade.
These hedging tools allow portfolio managers to reduce their overall market exposure without having to liquidate individual stock positions they want to hold. Lin characterized the hedging buildup as calibrated risk management rather than directional selling.
In the middle of this substantial price rally in the group, hedge funds have not been chasing…They’ve been selling down their exposure in the sector. It is a reflection of hedge funds taking profits, taking some chips off the table.
Gross leverage across the firm’s hedge fund client base has climbed to a fresh five-year maximum this month, the bank reported.
Net leverage, which captures the directional bias of hedge fund positions, has held comparatively stable at the 58th percentile on a three-year lookback.
Higher inflation readings and climbing bond yields have both contributed to the increased hedging activity among Goldman’s institutional client base, the report found.
The broader market environment has pushed managers to add downside protection, even as they maintain significant long positions in AI-linked semiconductor companies.
Semiconductor earnings growth and AI demand still support the long-term case
The constructive stance on artificial intelligence has held firm on the research side of the bank, even as prime brokerage data reveal hedge fund selling activity.
Agentic AI could drive a 24-fold increase in token consumption, to roughly 120 quadrillion tokens processed per month, between 2026 and 2030, senior equity analyst Jim Schneider projected in a May 20 Goldman Sachs Research note.
The AI infrastructure buildout continues to drive a multi-year capital expenditure cycle supporting the chip sector, King Lip, chief strategist at BakerAvenue Wealth Management in San Francisco, told Reuters in a May 13 analysis on the semiconductor rally.
Worldwide chip revenue is projected to climb 64% to an estimated $1.3 trillion during 2026, research firm Gartner reported in an April 8 press release.
S&P 500 semiconductor and equipment companies are expected to boost earnings by roughly 95% this year (up from a 62% increase expected as of Jan. 1), according to Tajinder Dhillon, head of earnings and equity research at LSEG Data & Analytics in the May 13 analysis.
Earnings projections of that magnitude help explain why hedge funds have tightened risk controls without abandoning their semiconductor exposure, even as the prime brokerage data show heavy net selling on the surface.
Goldman’s positioning data indicate a market in which professional money managers are tightening risk controls while keeping their fundamental AI allocations largely intact.
Related: Goldman Sachs doubles down on S&P 500 message for 2026
