JPMorgan: Retail Net Buying Drops to Historic Lows, While NVIDIA and Western Digital Buck the Trend
nashnova research
J.P. Morgan's latest Retail Radar shows U.S. retail net buying at $4.1 billion for the week ending September 30 — roughly 40% below the 12-month weekly average and at the 12th historical percentile. As fresh capital dries up, buying concentrates in Nvidia and SanDisk while long-duration Treasuries draw record inflows.
How weak is retail activity right now?
Weekly retail net buying hit $4.1 billion, versus a 12-month weekly average of $6.8 billion — a roughly 40% drop.
The weekly flow sits at the 12th historical percentile. September is the weakest month for retail activity since December 2024.
This means → Retail hasn't left the market (net flows are still positive), but fresh ammunition is drying up — the direction hasn't flipped, the momentum has.
Where is the money going?
Net single-stock buying was just $700 million for the week. Nvidia alone absorbed $1.286 billion; SanDisk took $327 million. Combined: $1.613 billion.
In plain terms = Strip out those two names and the rest of the stock universe was a net $913 million sell. Most stocks are bleeding, not gaining.
Top five net buys: Nvidia $1.286B, Tesla $514M, SanDisk $327M, Alphabet $153M, Amazon $146M.
This reflects a narrowing, not a fading, of AI conviction — retail has gone from "buy the sector" to "buy only the two or three surest names."
Who is being dumped?
Intel led net selling at $199 million, followed by SpaceX $171M, Oracle $88M, Nebius $60M, and Bloom Energy $54M.
Apple, Microsoft, and Meta also saw net selling — even the Magnificent Seven are not uniformly favored.
J.P. Morgan notes that tech remains the only net-bought sector even after stripping out the Mag Seven, but capital divergence within the same AI-compute supply chain is stark — "carrying an AI label" no longer explains the buy list.
Why are retail investors piling into long-duration Treasuries?
Long-duration Treasury ETFs drew roughly $260 million in net buying. The standardized buy-skew — a measure of net buying intensity against historical volatility — hit +5.4z for long-end Treasuries overall, and +6.1z for TLT alone. Both are record highs.
This means → Retail is buying aggressively into falling bond prices and elevated yields, betting on locking in high coupons + capturing price upside when rates eventually drop.
The backdrop: Middle East geopolitical tensions are fueling energy-driven inflation; yields on 10-year-plus Treasuries remain elevated.
Put simply = The retail thesis is "bonds have fallen far enough, rates must come down eventually, so now is the best entry." J.P. Morgan cautions, however, that actual motivation cannot be determined from flow data alone.
What is happening to small caps and energy?
Since March, retail has consistently net bought the Russell 1000 (large cap) and net sold the Russell 2000 (small cap). High rates are drawing a clear funding-cost dividing line.
The Russell 2000's median short-interest ratio sits at the 99.8th historical percentile; single-stock idiosyncratic dispersion is near the 92nd percentile — small caps are not just being sold, they are being shorted to extreme levels.
Energy: after Saudi east-west pipeline restoration, energy equity and ETF flows flipped to slight net selling. Crude exports have recovered to 98% of pre-conflict levels, but refined-product exports are only at 58%. Retail enthusiasm for the energy-supply-recovery trade is clearly cooling.
What different story is the options market telling?
Retail options volume as a one-month rolling share still sits at the 93.5th historical percentile, roughly 23% of total volume — off its peak but still very elevated.
Nvidia tops the spot net-buy list, yet also ranks among the top options-delta sellers — long in cash, hedged in options. Retail is not blindly one-directional.
SpaceX saw net spot selling, but remains active on the options-delta buy list — opposite directions across instruments.
This means → Retail is not one voice: the same cohort is making opposing directional bets across different instruments. Flow data increasingly needs tool-by-tool decomposition to be readable.
市场有风险，内容仅供研究参考，不构成投资建议。