Retail buying shows a "great narrowing"! JPMorgan fund flows reveal Nvidia and SanDisk attracting inflows against the trend, while US Treasury ETFs return to retail investors' view.

JPMorgan's latest "Retail Radar" research report shows that retail investor fund flows in the U.S. stock market are displaying the latest retail trends and directions: "overall retail capital is cooling, stock buying is increasingly concentrated in a handful of AI computing leaders, and long-duration U.S. Treasuries are beginning to rise in investment appeal after suffering record-selling pressure." JPMorgan's exclusive retail fund flow statistics through September 30 reveal that retail investors have not, as the long-term flow trends this year would suggest, simultaneously expanded their buying of the entire U.S. stock market technology sector and Philadelphia Semiconductor Index constituent stocks: recent advances in frontier AI agents/AI large models represented by Muse, Astra, and Anthropic Claude can be said to provide an important technological foundation for the large-scale commercial expansion of AI applications across industries and the continued surge in AI computing demand. Therefore, Nvidia and SanDisk, the two leaders in the AI computing industry chain, are still favored by retail fund flows, while Intel, SpaceX, and some AI computing infrastructure stocks have unexpectedly seen net selling. JPMorgan's estimated retail fund flows show that excluding the Magnificent Seven, technology is still the only sector to receive net buying, while AI-related hot stocks such as Intel, SpaceX, and Oracle have been reduced, highlighting capital divergence within the same AI computing industry chain investment theme. From September 24 to 30, JPMorgan's latest estimates show that retail investors net bought $4.1 billion, about 40% below the average weekly level of the past 12 months, with individual stock net buying at only $700 million; however, Nvidia and SanDisk respectively received $1.286 billion and $327 million in retail net buying support, totaling $1.613 billion, meaning that about $913 million in net selling of other individual stocks offset part of the buying. Memory chip components for AI data center server clusters, as well as AI GPUs, remain the clearest supply bottlenecks at the AI computing industry chain level. Market research firm TrendForce estimates that in 2026, server DRAM contract prices will cumulatively rise about 270%, and enterprise SSD prices will cumulatively rise about 235%; in 2027, HBM contract prices may still rise 70%140%, continuing to show doubling growth. These data reflect the combined effect of continued expansion in AI computing demand and rising memory chip prices. TrendForce estimates also show that in 2027, NVL72 rack shipments covering the Blackwell and Vera Rubin platforms are expected to grow more than 50% year over year; its market research chart shows that related system output value is expected to rise from about $226 billion in 2026 to $711 billion in 2027, a sharp year-over-year increase of 214%. At the same time, long-duration U.S. Treasuries, whose prices have been under pressure since September, have already seen significant contrarian allocation increases from retail investors, with long-term Treasury ETFs showing a record standardized buying tilt. The ETF focused on long-duration U.S. Treasuriesthe Treasury ETF with ticker TLThas become the most prominent focus of retail funds. "Standardized buying tilt" refers to using historical volatility as a yardstick to measure how strong retail net buying is relative to historical norms. JPMorgan's latest estimate for TLT reached +6.1z, meaning retail investors' dip-buying tendency is unusually strong. "Stock buying sharply narrowing" can be said to be the most apt summary of retail fund flowsincremental retail capital is weakening, while stock selection is concentrating in a small number of AI computing super leaders, memory chip leaders, and large technology names with strong cash flow fundamentals. On the bond market side, another clear clue has emerged: as intensifying Middle East geopolitics drives energy inflation higher and pushes long-term U.S. Treasury prices sharply lower, ultimately keeping yields on long-duration U.S. Treasuries of 10 years and above at high levels (Treasury yields and Treasury prices move inversely), long-term U.S. Treasury ETFs unusually received about $260 million in retail net buying during the week. Retail investors can be said to be simultaneously participating in the AI super bull market and the long-duration yield tradethe former seeking strong earnings realization under the explosive expansion of the AI computing industry, and the latter seeking the highest U.S. bond yields in more than two decades as well as the Treasury price elasticity from future yield declines. Retail buying "narrows the front": in stocks, pick leaders; in bonds, pick beaten-down long-duration U.S. Treasuries JPMorgan's "Retail Radar" research report shows that retail investors are still net buyers, but the incremental capital supporting the market has clearly weakened. The reason retail fund flows deserve increasingly close attention from investors is that they provide marginal buying in the spot market and also influence price elasticity through options trading: JPMorgan's report lists brokerage-channel transaction proxy indicators that, in the latest data available for June 2026, accounted for fully 25% of total U.S. stock and ETF trading volume, while retail options market participation as of the end of September remained near historic highs. The most investment-valuable signal in this report is that amid continued expansion of overall AI computing industry demand, retail stock buying has instead become more concentrated in AI computing leaders, while long-duration U.S. Treasuries have become another clear allocation direction. Micron's latest earnings report shows that fiscal 2026 fourth-quarter revenue reached $54.229 billion, up about 379% year over year, with next-quarter revenue guidance of $61.5 billion, plus or minus $1.5 billion; the Philadelphia Semiconductor Index subsequently rebounded 1.59% on October 1, providing a performance and price-level echo of the strong prosperity in AI memory and computing. Even more significant, Micron executives said bluntly on the earnings call that they "cannot see the end of supply-demand balance"26 long-term agreements have now been signed locking in about $150 billion in long-term orders, and the memory chip supply-demand market in 2027 and 2028 is expected to be even tighter than the record tightness of 2026. Supported by strong cash flow, Micron announced capital expenditure of $25 billion in the first half of fiscal 2027 and promised to return 100% of excess cash to shareholders in the future. JPMorgan estimates show that from September 24 to 30, retail investors net bought $4.1 billion, about 40% below the average weekly level of $6.8 billion over the past 12 months; of that, ETFs saw $3.4 billion in net buying, while individual stocks saw only $700 million. Weekly total retail fund flows in the U.S. stock market were at the 12th percentile historically, with ETF and individual stock flows at the 3rd and 35th percentiles respectively; daily buying roughly stayed in the 15th30th percentile, with limited response to that week's stock price changes, ultimately making September the weakest month for retail activity since December 2024. "Weak" here refers to a decline in net buying intensity; total capital remains a positive factor. Within market allocation, large-cap broad-based stock ETFs absorbed about $1.4 billion in retail funds, while covered-call strategy ETFs, multi-cap broad-based ETFs, and EAFE international stock ETFs absorbed $188 million, $169 million, and $148 million respectively, but overall flows into international stocks, crypto assets, and commodity ETFs weakened compared with before. JPMorgan's research report also lists non-retail category futures traders buying about $33 billion during the week, mainly concentrated in S&P 500 and Nasdaq futures, highlighting that observing rebound momentum requires considering different trading groups simultaneously. In terms of market structure, a decline in retail net buying weakens support for the rally to broaden to more stocks, while the index may still be driven by large-cap heavyweight stocks and other capital channels. Long bonds have become a prominent direction for contrarian retail positioning, behind which is a repricing of interest rates and financing costs. Long-term U.S. Treasury ETFs received $260 million in net buying during the week, with the long-end buying tilt recorded in the report reaching +5.4z and TLT reaching +6.1z. However, it should be noted that these figures measure abnormal strength relative to history and cannot be interpreted as records for yields, allocation ratios, or absolute dollar net inflows. In the environment described in JPMorgan's research report where U.S. Treasury yields are testing 22-year highs, buying TLT means increasing exposure to long-duration U.S. Treasury yields and positioning for a rebound in the prices of U.S. Treasuries with maturities of 10 years or more. It can be understood as simultaneously seeking higher yields and price elasticity from future rate declines; the actual motivation still cannot be determined from fund flows alone. The high-rate/high-yield curve also draws a financing-cost dividing line within stocks: retail fund flows into small companies that rely more on short-term, floating-rate bank financing are clearly under pressure, and the trend of retail net buying Russell 1000 and net selling Russell 2000 has accelerated since March; median short interest ratios for Russell 2000 and S&P 500 constituents are at the 99.8th and 96.2nd historical percentiles respectively, and idiosyncratic indicators for both types of stocks are also near the 92nd percentile, highlighting individual stock divergence. Another cooling appears in the stock market's energy sector: after the Saudi east-west pipeline resumed, buying of energy stocks and ETFs turned to slight net selling; although oil prices remain around $100, Middle East crude oil exports have recovered to 98% of pre-war levels, while refined product exports have recovered to only 58%. This indicates that retail investors are separately trading AI growth, expectations for a major rebound in U.S. Treasury prices against the backdrop of surging yields, and energy supply recovery, with clear structural differences in capital direction. Computing demand is spreading, but retail stock buying is converginghas the AI trade entered the "name-calling era"? The AI computing theme that has supported the U.S. stock market's super bull market trajectory since 2023 remains an important stock mainline for retail investors, but having an "AI technology label" is no longer enough to explain the buy list. The report clearly points out that retail investors prefer semiconductors and hardware, while software lags relatively; excluding the Magnificent Seven U.S. stock market technology giants, the technology sector still saw $194 million in net buying during the week, while all other sectors saw large net selling. As of that week, the top five retail net buys were Nvidia at $1.286 billion, Tesla at $514 million, SanDisk at $327 million, Alphabet at $153 million, and Amazon at $146 million. On September 30, Micron received $19.8 million in single-day retail net buying, and the report found no abnormal pre-earnings position increases. Especially noteworthy is that Nvidia alone saw net buying exceeding the $700 million total net buying of all individual stocks in the market, indicating that selling in other stocks offset a considerable portion of the buying strength in AI computing leaders. On the other hand, Intel, SpaceX, Oracle, Nebius, and Bloom Energy were respectively net sold by retail investors last week by $199 million, $171 million, $88 million, $60 million, and $54 million; Apple, Microsoft, and Meta also saw net selling. Within AI investment theme baskets, AI computing core infrastructure supply, data center construction/computing leasing, data center electrification, growth stocks, AI software monetization, and U.S. companies with high China revenue exposure remain areas of long-term retail attention. In the derivatives market, although the one-month rolling average of retail options trading share has left its high, it remains at the 93.5th historical percentile, with an actual share of about 23%, still at a historically high level. Tesla, Meta, Micron, Nvidia, AMD, and SanDisk remain options trading focuses. Nvidia tops the spot net buying list but also ranks high in options Delta selling; SpaceX saw spot net selling but remains active on the options Delta buying list, showing divergence among trading instruments. JPMorgan said the latest social media discussions among retail investors focused on names such as CIFR, WULF, HUT, AKAM, and CoreWeave, but actual retail selling in high-short-interest stocks surged, reducing overall short-squeeze risk somewhat; ALOY, EU, and EVTL are event-specific cases requiring separate observation. Clear divergence also exists outside AI: Carnival shares rose 13% on positive earnings, yet retail investors instead net sold $14.4 million; Veeva made progress with a large pharmaceutical client and its shares rose, yet it still saw net selling; KOD attracted $5 million in buying after successful clinical trials, with abnormal strength reaching 11.4z; MGM considering acquiring PPLI, which holds about 27% of its shares, activated an event trade with potential buyback characteristics. "Computing demand is spreading, stock buying is converging" is the summary wording JPMorgan's research report uses for retail fund flow direction. Nvidia, SanDisk, and other beneficiaries of AI computing core infrastructure and memory chips amid the explosive expansion of AI computing demand have become representatives of concentrated fund buyingNvidia and SanDisk correspond respectively to AI computing and NAND memory segments; at the same time, long-duration U.S. Treasuries, whose prices have remained under pressure, have already seen significant contrarian retail allocation increases. OpenAI is in talks to raise funds at a pre-money valuation of about $1.4 trillion; potential IPO investors in Anthropic have assigned a valuation judgment of $1.8 trillion$2 trillion, with expectations that it could match or exceed SpaceX's issuance scale. These still belong to financing negotiations and Silicon Valley venture capital's IPO listing expectations. Compared with AI application valuation narratives, the AI computing supply chain already has more concrete evidence of strong demand: media disclosed that Anthropic has about $518 billion in infrastructure arrangements over roughly the next decade, of which about 80% is non-cancelable or payable as agreed. Anthropic's 2025 revenue grew to about 12 times the prior year, approaching $4.6 billion, with operating losses exceeding $8 billion. Computing and infrastructure spending reached $7.33 billion, about three times that of 2024, accounting for about 58% of total operating expenses of $12.65 billion; Nvidia's latest quarterly data center revenue was $89 billion, up 117% year over year; South Korea's September semiconductor exports were $60.3 billion, up 262.8% year over year, with officials also noting increases in memory export volumes and contract prices. From the perspective of massive AI inference workloads, Muse's continuous background execution and Astra's improved capabilities on complex computer tasks have expanded the scope of work AI can undertake; multi-step tasks, tool calls, and parallel agents may also increase model calls and context processing per user. Anthropic has observed in its research systems that token usage for multi-agent tasks is about 15 times that of ordinary chat. From this, it can be inferred that under the full penetration of frontier AI agents like Muse, AI computing demand will spread along GPU computing, HBM and DRAM capacity and bandwidth, KV cache and SSD storage, CPU tool execution, and network transmission; total resource demand ultimately depends on the combined effect of task volume growth and per-task efficiency improvement. This also explains why AI hardware still has fundamental appeal, yet cannot guarantee that all related stocks will simultaneously receive incremental capital. JPMorgan's data supports that "retail investors continue to selectively buy AI computing and memory leaders while increasing long-term U.S. Treasury exposure"; for a full return to technology, high-beta small caps, or all AI infrastructure stocks, the retail fund evidence does not yet hold.