Retail Investors Stage a "Big Shrinking Circle"! JPMorgan Fund Flows Reveal Nvidia and SanDisk Attracting Funds Against the Trend, US Treasury ETF Returns to Retail Investors' Radar

According to the latest "Retail Radar" research report by Wall Street financial giant JPMorgan, retail capital flows in the US stock market reveal the latest trends: "a cooling of overall retail funds, stock buying becoming increasingly concentrated on a few leading AI computing power stocks, and the investment attractiveness of long-term US Treasuries rising after record sell-offs."

JPMorgan’s exclusive retail capital flow statistics as of September 30 show that retail investors have not accelerated their buying across the entire US tech sector and Philadelphia Semiconductor Index constituents as sustained flows this year might suggest. Recently, advances in cutting-edge AI agents/models represented by Muse, Astra, and Anthropic Claude have provided a crucial technology foundation for the massive commercialization of AI applications across industries and the continued surge in AI computing demand. As a result, Nvidia and SanDisk, two leading players in the AI computing power supply chain, remain favored by retail investors. However, Intel, SpaceX, and some AI infrastructure stocks have unexpectedly seen net selling.

JPMorgan’s data shows that excluding the Magnificent Seven, technology remains the only industry with net buying by retail, yet hot AI stocks such as Intel, SpaceX, and Oracle have been reduced, highlighting divergences within the AI computing power investment theme itself. From September 24 to 30, the latest JPMorgan data indicates that retail investors had a net buy of $4.1 billion, about 40% lower than the $6.8 billion weekly average over the past 12 months, with single-stock net buying only $700 million. However, Nvidia and SanDisk received $1.286 billion and $327 million in net retail buying, respectively—a total of $1.613 billion—meaning net selling of other stocks offset part of those inflows by about $913 million.

Storage chip components for AI data center server clusters and AI GPUs remain the clearest supply bottlenecks in the AI computing power supply chain. Market research firm TrendForce projects aggregate contract prices for server DRAM to rise about 270% by 2026, and enterprise SSD prices to increase approximately 235%. HBM contract prices could still rise 70%–140% in 2027, marking continued exponential growth—data that reflects the combined effect of rising AI computing demand and storage chip price hikes. TrendForce also estimates that in 2027, NVL72 rack shipments (covering Blackwell and Vera Rubin platforms) will increase by over 50% year-on-year. Their charts show system output value rising from about $226 billion in 2026 to $711 billion in 2027, a dramatic 214% annual increase.

Meanwhile, long-term US Treasuries, which have faced continuous price pressure since September, have notably seen retail investors increase allocations against the trend, with long-duration Treasury ETFs experiencing record-standardized buying tilts. The TLT ETF, focused on long-term US Treasuries, became the standout focus of retail capital. "Standardized buying tilt" measures how strong retail net buys are relative to historical norms using historical volatility as a gauge. JPMorgan’s latest reading of TLT at +6.1z means retail investors exhibited exceptionally strong dip-buying behavior.

"Significant contraction of stock buying" is a fitting summary for retail capital flows—incremental retail money is weakening, while stock selection is focusing on a handful of AI super-leaders, leading storage chip names, and large technology companies with robust cash flow fundamentals. In the bond market, another clear trend is unfolding: as Middle East geopolitical risk intensifies and energy-driven inflation pushes long-term US Treasury prices under pressure—keeping 10-year and longer Treasury yields high (yields move inversely with prices)—long-term US Treasury ETFs saw a rare $260 million in net retail buying in the same week. Essentially, retail investors are simultaneously participating in both the AI super bull market and the long-duration yield trade—seeking strong AI sector profits while also chasing the highest US bond yields in two decades and the price flexibility that may accompany future yield declines.

Retail Stock Buying "Narrowed Front": Picking Leaders in Stocks, Choosing Beaten-down Long-term US Treasuries in Bonds

The JPMorgan "Retail Radar" research report shows retail investors are still net buyers, but incremental capital inflows propping up the market have clearly weakened.

Why retail capital flow is increasingly worth investors’ close attention is because it not only provides marginal spot market buying but also influences price elasticity through options trading: brokerage channel trading activity listed in JPMorgan’s report accounted for a full 25% of all US stock and ETF market turnover in June 2026 (latest data), and as of the end of September, retail options market participation remains near historical highs. The most valuable investment signal from this report is that as overall AI computing power demand continues to expand, retail stock buying has become even more concentrated among AI computing power leaders, with long-term US Treasuries forming another clear allocation direction.

Micron’s latest financials show Q4 FY2026 revenue of $54.229 billion, up about 379% year-on-year, with next quarter guidance at $61.5 billion, +/- $1.5 billion. The Philadelphia Semiconductor Index subsequently rebounded 1.59% on October 1, reflecting the robust boom in the AI storage and computing power industry. Even more noteworthy, Micron executives stated on the earnings call, "We see no end in sight for the supply-demand imbalance"—having already signed 26 long-term deals locking in roughly $150 billion of long-term orders, and expecting the storage chip supply-demand market to tighten even further in 2027 and 2028 compared to the record-high strain in 2026. With strong cash flow backing, Micron announced capex of $25 billion for the first half of FY2027 and pledged to return 100% of excess cash to shareholders in the future.

JPMorgan’s calculations show that from September 24 to 30, retail net buying totaled $4.1 billion, about 40% lower than the $6.8 billion weekly average of the past 12 months; ETF net buying made up $3.4 billion while single-stock net buying was only $700 million. Retail’s weekly total capital inflow was at the 12th historical percentile, with ETFs and single stocks at the 3rd and 35th percentiles, respectively. Daily buying hovered between the 15th and 30th percentiles, with only limited response to weekly stock price movements, making September the weakest month for retail activity since December 2024. "Weak" here refers to declining net buying intensity—the overall volume remains positive.

In terms of market allocation, broad market large-cap ETFs absorbed about $1.4 billion in retail funds, while covered call ETFs, multi-cap broad ETFs, and EAFE international equity ETFs respectively absorbed $188 million, $169 million, and $148 million. However, overall flows into international equity, crypto asset, and commodity ETFs weakened compared to before. The JPMorgan report also highlighted that non-retail futures traders bought about $33 billion that week, mainly in S&P 500 and Nasdaq futures, emphasizing that understanding rebound momentum requires considering all trading groups. Structurally, declining net retail buying weakens support for broader stock market participation, though indexes can still be lifted by large weights and other capital channels.

Long bonds have become a standout area for contrarian retail positioning, driven by a repricing of interest rates and financing costs. That week, long-term US Treasury ETFs saw $260 million in net inflows; recorded long-end buying tilt reached +5.4z, with TLT at +6.1z. It's important to note these figures measure relative extremes versus history and shouldn't be interpreted as yield, allocation ratio, or absolute dollar flow records.

In the context of JP Morgan's analysis of Treasury yields testing 22-year highs, buying TLT means increasing exposure to long-term Treasury yields and price rebounds for bonds with maturities over 10 years. This can be seen as seeking both higher yields and price elasticity from potential future rate cuts—though actual motivations cannot be determined based on flow data alone.

The high-interest/high-yield curve is making a clear cost-of-funding distinction within equities: small companies that rely more on short-term, floating-rate bank loans are under clear retail flow pressure, with net buying in Russell 1000 and net selling in Russell 2000 speeding up since March. Short interest medians on Russell 2000 and S&P 500 constituents are at the 99.8th and 96.2nd percentiles historically, and both segments’ specific indicators are near the 92nd percentile, highlighting stock-level differentiation.

Within equities, the energy sector is cooling: after Saudi Arabia’s east-west pipeline resumed, energy stocks and ETFs shifted to slight net selling. Though oil is still near $100 per barrel, Middle Eastern crude exports are back to 98% of prewar levels, while refined product exports recovered just 58%. This suggests retail is trading on expectations of a US Treasury price rebound driven by AI growth and yield surges, as well as energy supply recovery, with flows showing a clear structural divergence.

AI Computing Demand Spreads, but Retail Stock Buying Concentrates—Has "Roll Call Era" Begun in AI Trading?

The AI computing power theme, supporting the US super bull market since 2023, remains a top focus for retail investors, but simply "having the AI technology label" is no longer enough to make the buy list.

The report explicitly states that retail prefers semiconductors and hardware, with software lagging. Excluding the Magnificent Seven, the tech sector still saw $194 million in net buying that week, while all other sectors saw sharp net outflows. The top five net retail buys for the week were Nvidia ($1.286 billion), Tesla ($514 million), SanDisk ($327 million), Alphabet ($153 million), and Amazon ($146 million). On September 30, Micron had a single-day net buy of $19.8 million by retail, with no abnormal pre-earnings buildup. Particularly notable is that Nvidia’s net buy alone exceeded the combined $700 million net buy in all other individual stocks, illustrating how selling in other names offset much of the AI leader buying force.

Meanwhile, Intel, SpaceX, Oracle, Nebius, and Bloom Energy were net sold by retail for $199 million, $171 million, $88 million, $60 million, and $54 million, respectively. Apple, Microsoft, and Meta also had net outflows. Within the AI investment basket, core AI infrastructure supply, data center construction/leasing, data center electrification, growth stocks, AI software monetization, and US companies with high China exposure remain long-term focuses for retail investors.

In derivatives, the one-month rolling average share of retail options trades, though off its highs, remains at the 93.5th historical percentile—actual proportion is about 23%, still high historically. Tesla, Meta, Micron, Nvidia, AMD, and SanDisk remain option trading hotspots. Nvidia tops net buying in spot but also ranks high among option delta sellers; SpaceX saw net spot selling but remained active on the option delta buy leaderboard, indicating some decoupling between trading tools.

JPMorgan notes that recent retail social media discussion focuses on names like CIFR, WULF, HUT, AKAM, and CoreWeave, but increased retail selling in high-short-interest stocks has reduced the overall short-squeeze risk; ALOY, EU, and EVTL are cases for individual monitoring. Non-AI segments show sharp divergences as well: Carnival’s strong results drove the stock up 13%, but retail were net sellers to the tune of $14.4 million; Veeva advanced on large pharma wins and stock gains but was still sold; KOD’s clinical trial success attracted a strong $5 million buy (11.4z intensity); MGM’s potential acquisition of PPLI (with about 27% interest) triggered event-driven trades with buyback potential.

"Computing power demand is spreading, but stock buying is concentrating"—this is JPMorgan’s conclusion on retail capital flows. Nvidia and SanDisk, beneficiaries of explosive AI infrastructure and demand growth, are the main focal buys—Nvidia is aligned with AI computing, SanDisk with NAND storage. At the same time, long-term US Treasuries under persistent price pressure are now seeing significant contrarian retail allocation.

OpenAI is reportedly in talks for new fundraising at a pre-money valuation of about $1.4 trillion. Anthropic’s potential IPO investors are valuing at $1.8–2 trillion, with expectations to match or exceed SpaceX’s issuance scale—these remain in the funding negotiation and Silicon Valley venture IPO forecast stage.

Compared with AI application valuation narratives, the AI computing supply chain has more specific strong demand evidence: media reports Anthropic's roughly $518 billion infrastructure planning over the next decade, about 80% of which is non-cancellable or contractually payable; 2025 revenue is expected to jump nearly 12-fold year-on-year to $4.6 billion, with an operating loss over $8 billion; computing and infrastructure expenditure will total $7.33 billion, about triple that of 2024, accounting for about 58% of total operating expenses ($12.65 billion). Nvidia’s latest quarterly data center revenue is $89 billion, up 117% year-on-year; South Korea’s September semiconductor exports hit $60.3 billion, up 262.8% year-on-year—officials also noted both higher export volume and contract chip prices.

From the perspective of massive AI inference workloads, continuous backend execution by Muse and Astra’s improved complex task handling expand the universe of AI tasks. Multi-step tasks, tool invocation, and parallel agents can increase per-user model calls and context handling—Anthropic observed token usage per multi-agent task at ~15x standard chat. This suggests that with the pervasive adoption of leading AI agents like Muse, AI computing demand will expand across GPU computation, HBM and DRAM bandwidth, KV cache and SSD storage, CPU tool execution, and network transfer. Total resource needs will ultimately be a function of workload growth and per-task efficiency improvement. This explains why AI hardware still attracts on fundamentals but not all related stocks see inflows at the same time. JPMorgan's data supports "retail continues to selectively buy AI computing and storage leaders while increasing exposure to long-term US Treasuries"; there is not yet evidence for a comprehensive return to tech, high-beta small caps, or all AI infrastructure names for retail funds.