UBS: Last Week's US AI Sector Saw Deleveraging, Not De-Risking
nashnova research
UBS trader Torsten Sippel says last week's US equity move was a targeted position reset, not a broad liquidation — longs were cut but new shorts stayed limited, style rotation from momentum-growth to quality-value is well underway, and structural fragility remains unresolved.
What's the difference between de-grossing and de-risking?
De-grossing means shrinking position size. De-risking means systematically abandoning exposure. This means → funds are lightening up, but they haven't walked away from the table.
Long-only funds ran a buy/sell ratio of 36/64 — far more selling than buying. Hedge funds came in at 42/50/8 (buy/sell/short), with relatively few new single-stock shorts.
In plain terms = everyone is putting fewer chips on the table, but nobody has flipped it over.
Who got hit and who caught a bid?
AI and momentum names were the epicenter: memory stocks fell 16% in a single week, speculative growth dropped 12%, and momentum long-short fell 8% — with almost no dip-buying.
Meanwhile, consumer staples rose 5% and value long-short gained 6%. This means → capital rotated out of the most crowded high-momentum positions into cheaper, steadier names. The style switch is already pronounced.
ASML and TSMC both posted strong fundamental data, yet neither eased the selling pressure. This reflects a market that now prices AI on a standalone financing-sustainability logic — even good orders don't matter if the funding question is unresolved.
Why is gross exposure still dangerous?
Hedge-fund net exposure has dropped to the 30th percentile of its five-year range — seemingly low. But gross exposure remains above the 90th percentile.
In plain terms = net positioning shrank, but total leverage is still very large. If losses continue, markets could face another forced unwind.
The VIX spiked on Friday; the 1-month / 3-month VIX term structure inverted for the first time in six months. This means → hedging demand has spread from AI-specific risk to broader market risk. Fear is widening.
What was Friday afternoon's indiscriminate sell-off?
Sippel's read: Friday's blanket tech selling fits the pattern of ETF redemptions, options expiry, and hedge rolls — not fundamental, stock-by-stock selling.
Macro acted as an amplifier, not a trigger — inflation and rate shocks accelerated the systematic sell-off, while Middle East escalation increasingly transmitted into volatility markets.
In plain terms = no one studied a company and decided to sell. Positioning hit a structural tipping point, and macro headlines sped up the clock.
Why has CoreWeave become the bellwether?
As CoreWeave credit concerns widen, the market debate has shifted from "Is AI demand strong?" to a more fundamental question: Can AI infrastructure economics and financing structures support the next round of capex?
Hyperscaler earnings due shortly face a higher verification bar — investors need to see cloud revenue, order backlogs, and cash generation that credibly match the scale of AI capex commitments.
80 companies report over the next five trading days, covering 13.8% of the S&P 500 by market cap — including Netflix, Meta Platforms, Tesla, and GE. Sippel flags that Meta reports a week earlier than usual, asking pointedly: "Getting ahead of a closing equity-issuance window?"
When does this de-grossing actually stop?
Earnings season so far is decent: 10.4% of S&P 500 market cap has reported, Q3 EPS growth stands at 26.5%, and 75% of companies have beaten expectations.
Yet strong bank results and positive AI data points have failed to absorb the selling pressure. This means → earnings are currently subordinate to positioning logic — good news cannot outrun the momentum of de-grossing.
Sippel's bottom line: whether AI names can find a genuine buy signal in hyperscaler earnings is the key test for whether this de-grossing wave stops here.
市场有风险，内容仅供研究参考，不构成投资建议。