Hedge Funds Shift to Underweight AI Holdings as Deleveraging Hits Largest Scale in 15 Years
nashnova research
Hedge funds have turned AI-related holdings to underweight for the first time since 2023. Morgan Stanley data shows gross exposure fell 8% from its June peak through October — the largest sustained drop in 15 years — signaling a systematic retreat from the AI trade.
How severe is this deleveraging?
Hedge fund gross exposure — the sum of all long and short positions — fell 8% from its June peak, the largest sustained decline on record.
This means → this is not a routine position trim but a four-month systematic unwind stretching across the entire summer.
Morgan Stanley strategist Nick Savone compared it to "the last drop on a log flume." In plain terms = the deleveraging is accelerating downhill and hasn't hit the splash pool yet.
What are funds buying and selling?
This week's buying was driven by short covering — closing out bearish bets — pushing net leverage up 5 percentage points to 51%.
Gross exposure, however, only dipped 1 percentage point to 207%, still at the 92nd percentile since 2010. In plain terms = the net position is rising, but total "chips on the table" remain near historic highs — only the direction is shifting.
Funds added to memory, agentic AI, internet, and cybersecurity while trimming hyperscalers, energy, materials, and real estate. This reflects a rotation within AI — from infrastructure buildout to niche verticals — not a full exit.
Are fund managers making money this year?
35% to 40% of long-short funds are flat or negative year-to-date.
October's scorecard was worse: Americas long-short funds fell 1.7% while the S&P 500 rose 1.0%; European long-short funds fell 1.7% against a 2.4% gain for the Stoxx 600; Asia long-short funds dropped 4.0% versus a 5.0% decline in MSCI Asia-Pacific.
This means → hedge funds underperformed their benchmarks across all three regions. In a high-dispersion market, stock-picking skill became a drag rather than an edge.
Why is single-stock volatility breaking records?
Over the past two weeks, S&P 500 constituent dispersion — how far individual stocks diverge from each other — sat at the 88th percentile over five years, with single-stock volatility at the 83rd percentile.
The ratio of single-stock volatility to index volatility has risen to its highest level since records began in 2005, surpassing peaks during the financial crisis and the pandemic.
At the same time, six-month pairwise correlation among S&P 500 stocks has fallen below 10%, a historic low. In plain terms = the market has split into two separate trading worlds — "AI stocks" and "everything else" — and index-level moves mask extreme divergence underneath.
When can positions be rebuilt?
Savone says funds are "unwilling to rebuild exposure" for a simple reason: every time they try to add, the positions they buy immediately pull back hard.
This means → deleveraging has become self-reinforcing — the more they sell, the less they dare to buy, and the harder reversal becomes.
When this cycle breaks and the window to rebuild opens remains, in Savone's view, the market's biggest unresolved question.
市场有风险，内容仅供研究参考，不构成投资建议。