JPMorgan: Retail Buying Contracts, Capital Concentrates into AI Computing Core Names Like NVIDIA
nashnova research
JPMorgan's *Retail Radar* report shows weekly retail net inflows fell to just $2.5 billion — down 63% from the 12-month average — yet Nvidia alone absorbed nearly $1.2 billion, three-quarters of all Mag-7 buying. Total flow is shrinking; what remains is concentrating on the hardest links of the compute chain.
How much did overall retail buying shrink?
In the week of September 10–16, retail net inflows totaled only $2.5 billion, versus a 12-month weekly average of $6.8 billion — a roughly 63% drop.
The report ranks the week's overall inflow intensity at the 2nd percentile historically. This means → across the past year, almost no week was weaker.
ETF inflows hit a one-year low; single-stock flows sat at the 12th percentile. Both channels narrowed simultaneously.
Where did the smaller pool of money go?
The Magnificent Seven drew $1.601 billion in net buying, accounting for 64% of total retail inflows. Nvidia alone captured $1.196 billion — roughly 74.7% of the Mag-7 total.
In plain terms = when retail wallets tightened, almost every spare dollar went to a single name: Nvidia.
The other six all logged net inflows: Tesla $201 million, Amazon $87 million, Apple $63 million, Alphabet $25 million, Microsoft $19 million, Meta $10 million — each step down dramatically from the last.
Beyond the Mag-7, who was bought and who was sold?
Three non-Mag-7 names joined the top-five net-buy list: SanDisk $202 million, ASML $119 million, Oracle $102 million.
This means → the names retail chose map to storage, semiconductor equipment, and cloud infrastructure — key chokepoints on the AI compute chain.
Top-five net sells: SpaceX-related $249 million, Marvell $95 million, Intel $76 million, NuScale Power $60 million, AMD $52 million — all tech-adjacent, but outside the "core compute chain" as retail currently defines it.
Why did the report single out Oracle?
Oracle posted 30% year-over-year revenue growth, triple-digit cloud-infrastructure revenue growth, and a $26 billion sequential increase in remaining performance obligations — contracts signed but not yet recognized as revenue.
On September 11 alone, retail net-bought $87 million of Oracle shares. This reflects that verifiable earnings growth still draws dip-buying, even when the broader flow is contracting.
In plain terms = Oracle is the report's case study for one point — retail isn't unwilling to spend; it's only willing to spend on companies that can show the numbers.
How did ETFs and sectors break down?
Sector ETFs posted their third-largest weekly net outflow on record, led by tech products. Broad large-cap ETFs still drew $1.3 billion in net buying; precious-metals ETFs took in $188 million.
Excluding the Mag-7, consumer staples was the only sector with net inflows — just $11 million. Industrials, communications, and tech saw net outflows of $555 million, $493 million, and $405 million respectively.
The one-month rolling average of retail options-trading share sat at the 96.8th percentile. This means → engagement is still elevated; it is spot buying — not participation — that is shrinking, and directional bets are diverging.
Can this "shrinking-but-concentrating" pattern last?
JPMorgan argues that as long as earnings stay strong and geopolitical risk is contained, a shallow rate-hike cycle and persistently high long-term Treasury yields can still be absorbed by equities.
The report cites SemiAnalysis: pre-training — training a large model from scratch on massive data — now accounts for less than 15% of compute allocation; post-training and continuous inference demand are rising. This reflects a shift in AI compute needs from centralized pre-training toward post-training, inference, and agent execution.
Nvidia, SanDisk, ASML, and Oracle map to successive links of one demand chain: accelerated compute → memory & storage → semiconductor equipment → cloud infrastructure. Whether retail flows can sustain selective support for core compute names amid continued aggregate contraction is the key variable for testing the durability of an AI-theme rebound.
市场有风险，内容仅供研究参考，不构成投资建议。