On the day the US semiconductor sector suffered a sharp selloff on Thursday, capital in Asian markets was actually buying up popular AI computing power and semiconductor stocks at lower levels.
On October 22, the Philadelphia Semiconductor Index fell 6.72%, the Nasdaq Composite dropped 2.60%, and NVIDIA (NVDA) and Micron Technology (MU) declined 4.40% and 6.26%, respectively. The next day, South Korea's KOSPI index reversed from an intraday drop of more than 4% to close up 0.30%, with SK Hynix rising 4.42% and Samsung Electronics gaining 0.47%; China's ChiNext index rebounded from an intraday fall of more than 3% to close up 4.16%, with total turnover across the Shanghai, Shenzhen, and Beijing markets reaching approximately 1.97 trillion yuan. The recovery of memory chip and optical communication leaders became an important driving force behind this round of tech stock reversal in China and South Korea. SK Hynix's suspension of its evaluation of a Solidigm listing eased investor concerns about duplicate listings and valuation discounts; Xinyisheng disclosed a buyback plan of 800 million to 1 billion yuan, and several optical communication companies released operational updates, reinforcing market confidence in orders and product upgrades. Bargain hunting at low levels after oversold conditions and positive company-level news reinforced each other, pushing some tech leaders to recover first. A senior analyst said that after concentrated selling of tech stocks in Chinese and Korean markets over the past two weeks, some investors may have already taken profits on short positions, while funds looking to buy at low points began entering the market. When new selling pressure weakened, short covering and bargain-hunting capital appeared simultaneously, potentially causing the sharp rebound in semiconductor and broader tech stocks seen in Chinese and Korean markets on Friday.
LSEG Lipper fund flow data showed that equity fund subscription data provided a broader capital backdrop. In the week ending October 21, global equity funds saw net inflows of $28.16 billion, higher than the previous week's $19.88 billion, marking the fifth consecutive week of inflows; Asian equity funds in particular saw massive net inflows of $11.53 billion, the highest level since at least December 2022, while European equity funds attracted $9.14 billion and US funds attracted $5.07 billion. However, although US equity funds received net subscriptions for the second consecutive week, the scale decreased by about 62% from the previous week's $13.37 billion. This data reflects relatively optimistic global allocation behavior before Thursday's sharp US stock decline — that is, global investors are still increasing equity exposure, especially aggressively targeting Asian stock markets that had significantly underperformed the US and fallen into corrections, while reducing their chase of the US stock market that recently hit new highs. Undoubtedly, the two core variables supporting the radiation of stock market buying power are the downward trajectory of 10-year US Treasury yields and the profit realization of semiconductor and cloud computing giants associated with the AI computing power theme. The US 10-year Treasury yield has fallen nearly 38 basis points from its 24.5-year high of 5.3645% in early October and dropped below 5%; of the approximately 24% of S&P 500 companies that have reported earnings, about 82% beat expectations, and of the approximately 18% of MSCI All Country World Index constituents that have disclosed results, about 73% exceeded expectations. Although the MSCI World Index still fell 1.31% on October 22, the strong profit expansion of the semiconductor sector closely tied to the AI computing power theme and the easing of discount rate pressure have already provided a basis for capital to reassess tech stocks.
For the epicenter of the AI trade and tech stock selloff — namely the recently battered memory chip sector — LSEG Lipper fund flows showed global technology sector funds saw net inflows of $5.05 billion that week, reversing the previous week's large outflow. Optimism on the fundamental side remains strong, with KeyBanc's recent Asia survey data showing that AI data centers are still absorbing memory and other AI computing power resource supplies on a large scale, pushing price increases through to 2027. The firm predicts that HBM blended prices could rise significantly by 80% to 100% in the first quarter of 2027. On the DRAM memory price level, KeyBanc forecasts a 15% to 20% quarter-over-quarter increase in the fourth quarter of 2026 on top of a strong third-quarter price increase base, with a further 10% to 15% rise in the first quarter of 2027; for NAND future pricing, it predicts about a 20% quarter-over-quarter increase in the fourth quarter of 2026 on a strong third-quarter base, with another 10% to 15% rise in the first quarter of 2027.
Tech capital returns, credit divergence: funds begin selecting who can survive financing tests
LSEG Lipper fund flow data shows that global capital is increasing tech exposure while raising requirements for corporate scale, liquidity, and financing capacity. Global technology sector funds saw net inflows of $5.05 billion that week, reversing the previous week's net outflow of $4.77 billion; healthcare funds attracted $1.83 billion, while consumer discretionary funds saw outflows of $1.29 billion. Within the US market, large-cap and mid-cap funds attracted $8.63 billion and $1.92 billion respectively, while small-cap funds saw outflows of $4.24 billion. US sector funds ended two consecutive weeks of redemptions, with total inflows of $1.8 billion, of which metals and mining, technology, and industrials received net inflows of $1.18 billion, $429 million, and $269 million respectively. The most noteworthy allocation tendency here is that global capital is more willing to take on investment risk in stock markets with strong earnings support, while maintaining distance from companies with weaker financing capacity, which also explains why Asian stock markets — home to the vast majority of fundamentally strong leaders in the global AI computing power supply chain — are receiving increasingly strong capital inflows. The bond market also shows divergence. Global bond funds attracted inflows for the fifth consecutive week, with net inflows of $9.44 billion, of which Swiss franc-denominated bond funds saw inflows of $6.8 billion, the highest since at least November 2024, and short-term bond funds attracted $2.78 billion; high-yield and government bond funds saw outflows of $1.68 billion and $550 million respectively. US bond funds saw net outflows of $969 million, ending 26 consecutive weeks of net subscriptions: long-term government and Treasury funds saw outflows of $3.3 billion, while short-to-medium-term同类 funds and municipal bond funds saw inflows of $1.76 billion and $1.56 billion respectively. The benchmark rate expectations amid falling Fed rate hike expectations, along with the downward trajectory of 10-year and 30-year US Treasury yields, have significantly improved the valuation environment, but investors are still actively controlling credit risk and duration risk.
Cash and safe-haven assets have also not lost their investment appeal. Global money market funds shifted from an outflow of $47.51 billion the previous week to an inflow of $15.86 billion, with the US shifting from an outflow of $45.93 billion to an inflow of $14.82 billion; gold and other precious metals funds attracted $1.29 billion, the fifth net inflow in nearly six weeks, and energy funds received $41.5 million. Data covering 29,174 emerging market funds showed equity funds attracted $6.28 billion, the highest since August 5, while bond funds saw a small outflow of $48.72 million. The overall allocation presents a portfolio characteristic of "selective equity additions, cash retaining flexibility, and strict credit screening."
The global capital flow dynamics and major trends revealed by LSEG Lipper fund flow data also partly explain why JPMorgan and UBS still see room for bargain hunting and position building in the AI computing power theme and broader semiconductor and tech stocks. According to JPMorgan's tech industry sample and model estimates, this year's cash flow is approximately $2.2 trillion, with corresponding investment of about $1.8 trillion, forming a financing surplus of about $400 billion; new cloud vendors facing financing pressure may change who builds projects but may not end the entire AI capital expenditure cycle. Meanwhile, the bank estimates that semiconductors contribute about 21% of global equity portfolio risk while accounting for only about 13% of market cap weight, and concentrated positions and risk limit adjustments can amplify short-term declines. UBS pointed out that the relative strength indicators for the Hang Seng Tech Index and CSI 300 fell to about 27 and 23.9 respectively, with US stock short positions elevated, increasing the possibility of oversold recovery and short covering. Both firms' positive judgments jointly point to: selling pressure is being released, and those fundamentally strong AI computing power supply chain leaders with cash flow and order support are better positioned to be repriced first.
Against the backdrop of AI monetization and computing power order relay, NVIDIA remains the absolute optimistic pricing anchor for Wall Street bulls
The earnings report from Facebook and Instagram parent Meta Platforms, Inc. (META) presented the core contradiction of the current AI trade — namely that commercialization around the AI wave is accelerating while infrastructure spending is consuming cash even faster. The company's third-quarter revenue was $64.601 billion, up 26% year-over-year, with advertising revenue reaching $63.061 billion; excluding $9.89 billion in legal expenses, operating profit rose 11% year-over-year, with fourth-quarter revenue guidance of $73 billion to $76 billion. But capital expenditure for the quarter reached $34.91 billion, free cash flow was negative $2.12 billion, and the full-year capital expenditure guidance floor was raised from $130 billion to $135 billion, with the ceiling maintained at $145 billion. Future spending commitments of approximately $825 billion further extend the time frame for the market to assess investment returns; however, this represents跨年度 leasing and procurement arrangements, not single-year capital expenditure. AI revenue support can already be observed from advertising effectiveness: after Meta expanded its GEM advertising model application, conversions for off-site conversion ads and app ads increased by 1.7% and 9.6% respectively. Muse, after about six weeks of launch in the US and Canada, attracted over 10 million sign-ups, with about 25% of sign-ups using it daily. Meta is advancing transaction service fees, personal and business subscriptions, and expects to improve Muse's computing cost efficiency by more than tenfold over the next year. The advertising business provides current cash generation capability, while Muse provides new user and transaction growth space; Muse has not yet separately disclosed revenue, and user shopping amounts cannot be directly regarded as Meta revenue. The newly created non-voting shares were initially implemented through allocation to existing shareholders, equivalent to a stock split and not equivalent to current equity financing, but their future financing use deepens market attention to financing costs and capital needs.
From an infrastructure mechanism perspective, expanding the use scope of agents like Muse will simultaneously increase model inference, tool execution, and state management requirements: GPUs and custom accelerators handle model computation, CPUs carry browsers, database queries, and business orchestration, HBM and server DRAM accommodate models, context, and concurrent working environments, enterprise SSDs store indexes and persistent state, HDDs handle long-term retained data after filtering, compression, and deduplication, and networking, optical interconnect, power, and liquid cooling jointly support continuous operation. This corresponds to opportunities for NVIDIA (NVDA) and AMD in accelerated computing, Broadcom in custom chips and interconnect, Micron Technology (MU) and SK Hynix in memory, Seagate and Western Digital in high-capacity storage, and TSMC and Amkor in manufacturing and packaging/testing. Capital expenditure expansion strengthens the demand foundation of the entire supply chain, but specific orders and profitability still depend on each company's product share and delivery capability. Wall Street financial giant Bank of America's positive judgment on the memory industry also revolves around HBM4 introduction and SOCAMM and enterprise SSD demand brought by advanced GPU and ASIC updates. Different levels of memory and storage handle different tasks, and configuration changes for a single GPU cannot be directly equated with a decline in overall data center demand. The DSX POD released at Berlin GTC integrates computing, networking, power, and liquid cooling forward into the manufacturing stage, shortening on-site deployment time to as fast as 14 days at sites where power, cooling, and networking are ready, with delivery planned for early 2027. Its investment implication is improving the conversion speed of "equipment delivery — computing power online — service revenue" rather than eliminating constraints from power supply and construction conditions. SpaceX provided another set of evidence for revenue realization — the company's third-quarter revenue was $14.575 billion, up 161% year-over-year; superintelligence business revenue including cloud services, software, and advertising was $7.56 billion, up 830% year-over-year, with operating profit turning positive to $431 million. Management said short-term computing power contract renewal prices continue to rise and expects at least 13 gigawatts of computing power online by the end of 2027. Bank of America raised its target price from $235 to $275 after the earnings report, maintaining a "buy" rating, citing first-quarter 2027 revenue guidance of $33 billion to $37 billion significantly exceeding the pre-earnings market expectation of about $22 billion; relative to Thursday's closing price of $144.75, the target price implies about 90% potential upside. SpaceX's strong earnings and outlook also provide demand-side support for NVIDIA's Berlin GTC computing power expansion narrative. Musk further stated that year-end online computing capacity is expected to reach 2.5 to 3 gigawatts, and he is confident of adding another 10 gigawatts in 2027, with next year's deployment mainly using NVIDIA's Vera Rubin platform. This expansion plan echoes GTC's DSX POD accelerating on-site deployment and key supply chain capacity expansion: customers need faster access to operational computing power, while NVIDIA focuses on shortening the distance between hardware delivery and computing power service revenue generation. For the semiconductor market, AI business revenue realization and next-generation computing power procurement plans mutually confirm each other, providing more concrete support for order growth expectations across the entire AI computing power supply chain, including NVIDIA AI GPUs and HBM, data center high-speed optical interconnect, power, and liquid cooling chains, and adding fundamental catalysts to the "AI bull market narrative." NVIDIA (NVDA), with a market cap of about $5.1 trillion, remains an important bullish pricing anchor on Wall Street amid this round of long-short divergence. Citigroup maintained a "buy" rating and $315 target price on October 22; Morgan Stanley maintained a $300 target price and restored its "top semiconductor pick" status in October; Bank of America's August target price was $350. Based on the October 22 closing price of $212.21, $300, $315, and $350 correspond to potential upside of approximately 41.4%, 48.4%, and 64.9%, respectively. The core bet behind these target prices is that next-generation platform delivery, customer computing power onboarding, and continuous procurement can drive profit growth. For the semiconductor sector and tech stock trajectory, capital return and oversold recovery provide conditions for a counterattack, while order conversion and cash returns determine how far the rally can go.
Comments