On October 22, U.S. stocks closed with AI-related trades facing another wave of concentrated selling, driven by surging AI financing cost pressures and deteriorating Middle East geopolitical conditions. However, a divergence is emerging on the other side of the market: while U.S. semiconductor stocks suffered heavy losses, Chinese and South Korean tech stocks staged a dramatic "deep-V" reversal on October 23, plunging intraday before rallying to close higher.
Expectations of memory chip price increases, sustained expansion of AI computing demand, and commercialization progress in certain AI applications continue to provide support for tech stock bulls. Notably, whether this rebound can be sustained will soon face a new test. Next week, Microsoft, Google parent Alphabet, Amazon, and Intel will release earnings between October 28 and 29. With AI infrastructure spending continuing to climb and the semiconductor sector experiencing violent swings, these companies' results and guidance will provide fresh observation windows for the market: whether cloud giants can convert AI investments into revenue growth, whether chip demand remains strong, and whether corporate capital expenditures can generate commensurate returns.
Chinese and South Korean Tech Stocks Stage "Deep-V" Counterattack, Semiconductor Bull Narrative Not Yet Extinguished
On October 22, U.S. tech stocks faced concentrated selling. The Nasdaq 100 index fell 2.90%, while the Philadelphia Semiconductor Index plunged 6.72%. Memory chip and related hardware stocks came under broad pressure, with Micron Technology falling 6.26%, Seagate Technology dropping 7.86%, Western Digital declining 6.43%, and SanDisk sliding 8.51%.
However, this selloff did not mean the entire U.S. market experienced indiscriminate declines. Approximately two-thirds of S&P 500 constituents still closed higher, with value-oriented stocks like Coca-Cola and Berkshire Hathaway showing relative resilience. This indicates that market pressure was more concentrated in tech stocks and AI-related trades, as investors reassess the financing costs, capital expenditure, and return risks facing high-valuation tech assets.
On October 23, Asian tech stocks showed a clear reversal. South Korea's KOSPI index fell more than 4% intraday before recovering losses to close up 0.30%; SK Hynix gained 4.42%, and Samsung Electronics rose 0.47%. News that Solidigm halted its U.S. listing also eased some concerns about the potential impact of dual listing.
China's ChiNext index similarly experienced violent swings, falling 3.21% intraday before staging a strong rebound to close up 4.16%. Daily market turnover reached approximately 1.96 trillion yuan, with more than 4,500 individual stocks advancing. Meanwhile, order conditions at optical communication supply chain companies like Zhongji Innolight and Tianfu Communication drew attention, with related enterprises in the "Yizhongtian" supply chain continuing to be market trading focal points; buyback plans from optical module companies further attracted investor attention.
From a trading perspective, the counterattack in Chinese and South Korean tech stocks demonstrates that the previous day's U.S. semiconductor plunge was insufficient to directly disprove the growth logic of the AI supply chain. However, violent intraday fluctuations also signal that market sentiment remains fragile: capital continues betting on computing infrastructure demand on one hand, while remaining vigilant about high investment, rising financing costs, and excessive valuations on the other.
Memory Chip Price Increase Expectations Extend to 2027, AI Infrastructure Demand Remains Strong
Memory chips are among the most closely watched segments in the current AI infrastructure investment chain. As data center construction continues, AI servers' demand for high-bandwidth memory (HBM), DRAM, and enterprise-grade solid-state drives keeps increasing, strengthening memory suppliers' pricing power.
Recent Asian research data from KeyBanc shows that AI data centers continue to absorb memory and other AI computing resource supplies on a large scale, driving price increases through 2027. The firm predicts HBM blended prices could surge 80%-100% in Q1 2027. For DRAM memory prices, KeyBanc forecasts a 15%-20% quarter-over-quarter increase in Q4 2026 on top of Q3's strong pricing base, followed by another 10%-15% increase in Q1 2027. For NAND future pricing, it predicts approximately 20% quarter-over-quarter increases in Q4 2026 on Q3's strong base, with another 10%-15% gain in Q1 2027.
TrendForce's forecast for conventional memory is slightly more moderate but similarly believes AI inference demand supports upward pricing. The market research firm expects Q4 conventional DRAM contract prices to rise 10%-15% quarter-over-quarter, with NAND contract prices up 15%-20%. On the DRAM side, agent workloads drive companion general-purpose server and memory procurement, while advanced process, packaging testing, and capacity allocation face constraints. On the NAND side, real-time retrieval, caching, and vector databases expand enterprise SSD demand. TrendForce predicts enterprise SSD bit demand will grow more than 80% year-over-year in 2026, with most new supply already booked, meaning Q4 price increases for this category could even exceed Q3's.
HBM price increases also compound with product upgrades and supply bottlenecks. TrendForce recently significantly raised its 2027 HBM blended average selling price forecast to 121% year-over-year growth, citing competition between HBM and conventional DRAM for limited advanced process and wafer capacity, time needed for new generation product yield ramps, and the rising share of higher-priced HBM4 with HBM4e gradually ramping. Therefore, this figure includes both price increases and product mix upgrades.
Both authoritative institutions' judgments jointly reinforce semiconductor bulls' core argument—AI demand is extending suppliers' pricing advantages, and the duration of elevated selling prices and their ability to convert into profits is more investment-relevant than simply whether the percentage of price increases is decelerating. However, memory price increases do not mean all semiconductor companies will benefit equally. Supply capacity, product mix, customer bargaining power, and capacity expansion pace all affect the ultimate profit increment companies can capture. For investors, what truly deserves attention is whether price increases can sustainably convert into revenue growth and earnings improvement.
Philadelphia Semiconductor Plunges While China and South Korea Rally Strongly! Wall Street Still Views Nvidia as the Bullish Pricing Anchor for AI Super Counterattack
As broad-based ETF trading volumes expanded, CATL rose 7.80% and Zhongji Innolight gained 3.82%, with new energy and computing hardware weights jointly driving index recovery, while the humanoid robot sector strengthened in tandem. From the AI computing supply chain investment logic perspective, this reversal reflects the resonance of order confirmation, buyback support, and risk appetite recovery. It also echoes the continuous surge in AI application-related revenue from Meta and SpaceX, further expanding AI computing resource investment, as well as the industry backdrop of KeyBanc and TrendForce's bullish memory pricing views—global semiconductor and broad tech stock valuation pressures do not necessarily mean AI demand is simultaneously declining. Those strong fundamental AI computing supply chain leaders capable of converting overseas capital expenditure into actual deliveries, profits, and cash flows—such as AI chip superpower Nvidia—are gaining opportunities for the market to reassess their value.
The dramatic "deep-V" reversal in Asian tech sectors on Friday, dominated by the massive turnaround in Chinese and South Korean markets, was essentially not driven by news flow. Macro-level positive catalysts and favorable news throughout the Asian trading session were extremely sparse, with virtually no important macro-level developments emerging. This rebound appeared more driven by the South Korean market, with the latest market commentary from Wall Street analysts suggesting—on one hand, after concentrated selling over the past two weeks in South Korean tech stocks, some investors may have already taken profits by closing short positions, while dip-buying capital began entering. When new selling pressure weakened, short covering and bottom-fishing capital emerged simultaneously, potentially causing Friday's sharp rebound in the South Korean market. On the other hand, more capital recognized the severely undervalued nature of South Korea's KOSPI index, with its forward 12-month price-to-earnings ratio dropping to just 4.2 times, far below the Nasdaq and Philadelphia Semiconductor Index.
Friday's dramatic "deep-V" reversal in Chinese and South Korean tech stocks is understood by some analysts through changes in trading forces after consecutive selloffs. After concentrated declines in South Korean tech stocks, short-covering trajectories began emerging, and dip-buying capital started absorbing shares, particularly as most bearish forces may have cashed in previously strong short-selling gains by buying back stocks. When marginal selling pressure weakened, active buying and potential short covering combined, even with limited new positive catalysts, could amplify intraday gains. Meanwhile, the recovery in South Korean chip stocks and Chinese A-share tech weights echoed each other, with company-level order, buyback, and listing arrangement news further improving sentiment, pushing the market from concentrated position reduction toward oversold recovery.
The core of this adjustment is that the market maintains extremely optimistic expectations for AI computing resources and AI agent-driven AI application demand, yet has raised requirements for investment scale, financing arrangements, and return realization speed to a higher level. On Thursday, Brent crude oil rose 4% to $113.17, reinforcing concerns about inflation and persistently high AI-related funding costs; however, safe-haven buying that day pushed 10-year U.S. Treasury yields down about 19 basis points to 4.95%. Therefore, the sharp decline in U.S. tech stocks cannot simply be attributed to Fed rate hike expectations and elevated Treasury yields amid the day's oil and natural gas price surges—AI financing costs near historic highs constitute a persistent constraint, and the stock selloff itself pushed capital toward Treasuries, indicating investors are reducing risk exposure.
Nvidia, with a market capitalization of approximately $5.11 trillion, remains Wall Street's important bullish pricing anchor amid this intense bull-bear divergence. Citigroup maintained its "Buy" rating and $315 target price in its October 22 Berlin GTC commentary, believing the main constraint on AI expansion remains manufacturing and deployment speed. Morgan Stanley restored its "Top Semiconductor Pick" status on October 2, maintaining "Overweight" and a $300 target price. Bank of America's August report also listed it as a top pick with a $350 target price, while Jefferies reiterated "Buy" and $300 at the end of August. Supporting these judgments are Rubin and Vera CPU ramps, expanding customer scope, and Nvidia's ability to increase computing output and system value per unit of power delivered. Based on Nvidia's 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; assuming approximately 24.1 billion total shares outstanding remain unchanged, this corresponds to market capitalizations of approximately $7.23 trillion, $7.59 trillion, and $8.44 trillion. This pricing expresses judgments about future earnings expansion, with the realization path still centered on next-generation AI computing cluster super platform deliveries, customer computing activation, and continued procurement.
The highest target price in the above statistics comes from Raymond James analyst Simon Leopold's $515 target, representing 137.91% upside from $216.47. He reiterated this judgment in a recent Bloomberg interview. His model expects revenue could approach $1 trillion in the fiscal year ending January 2029, valued at approximately 22 times 2028 calendar year earnings, corresponding to earnings per share of approximately $23.41. Supporting this judgment are supply improvements releasing delivery capacity, Rubin and Vera expanding system revenue, and customer growth across enterprise, sovereign AI, and professional cloud segments. Therefore, this most optimistic target price is primarily built on significantly higher earnings forecasts. The capacity expansion agreements, modular deployment, and industry application expansion conveyed at Berlin GTC correspond to key links determining whether these forecasts can be realized. However, target prices still depend on whether revenue growth, profit margins, and valuation assumptions can be achieved, and do not mean stock prices will necessarily reach corresponding levels.
AI Computing Demand Remains Very Strong, But Market Increasingly Asks Who Can Quickly Recoup Investments
The market interpreted Tesla's reduction in annual capital expenditure directly as weakening AI industry computing demand, completely ignoring its investment structure and project pacing. JPMorgan's strategist team's AI capital expenditure forecasts and outlook show hyperscale cloud providers' capital expenditure at approximately $950 billion in 2026, approximately $1.4 trillion in 2027, and approximately $3 trillion in 2030. Tesla's Q3 revenue was $28.046 billion, essentially flat year-over-year, with adjusted earnings per share of $0.48; management adjusted full-year capital expenditure expectations from over $25 billion to approximately $20 billion, but still expects Robotaxi, Optimus, solar manufacturing, and AI computing investment to expand in coming years. Its capital expenditure covers automotive, energy, and robot manufacturing, and is not a direct indicator of cloud computing demand. Terafab's large equipment investment is mainly in subsequent years, Tesla and SpaceX cost-sharing remains undetermined, and a potential merger is still in special committee exploration stages. Therefore, observing Tesla's spending in isolation makes it difficult to infer the ultimate total investment in related projects.
Tesla's earnings also show that profit improvement requires decomposing sources: automotive gross margin excluding credit revenue rose from 16.3% in Q2 to 18.1%, partly supported by improved delivery region mix; energy gross margin rose from 20.4% to 33.1%, clearly benefiting from tariff refunds and warranty reserve reductions. Q3 free cash flow turned positive at $1.531 billion, driven by increased operating cash flow and quarter-over-quarter capital expenditure declines. These changes are neither sufficient to prove that AI application and computing resource demand has fallen into recession as some pessimistic strategists describe, nor can they directly prove that autonomous driving and robotics have formed mature profit engines.
In contrast, Facebook and Instagram parent Meta provides more direct AI commercialization evidence. Q3 revenue was $64.601 billion, up 26% year-over-year, including advertising revenue of $63.061 billion; ad impressions grew 10%, with average ad prices up 14%. Q4 revenue guidance is $73-76 billion, with a midpoint of $74.5 billion, above the approximately $73.65 billion market expectation. Although legal expenses caused GAAP operating profit to decline 37% year-over-year, excluding this impact it still grew 11%. More explanatory is the company's estimate that advertisers using AI business assistant recommendations achieve nearly 5% additional conversions at the same cost per conversion, with ad spending nearly 10% higher than comparable non-adopters—Muse-led cutting-edge AI agent technology is supporting Meta's existing business growth by improving advertisers' return on investment.
Muse has thoroughly opened new commercialization space for Meta—launched approximately six weeks ago, over 10 million people in the U.S. and Canada have set up personal agents, with approximately 25% of set-up users using them on any given day; shopping features cover more than a quarter of weekly active users. Subscriptions, transaction services, enterprise APIs, Muse Code, and business agents constitute subsequent revenue paths. Management expects new AI businesses to begin contributing incremental revenue in 2027 and plans to improve computing cost efficiency by more than tenfold in the coming year. Especially noteworthy is that existing advertising business already reflects AI returns, while Muse's focus is converting usage growth into sustainable unit economics.
After earnings disclosure, Bank of America and Citigroup raised Meta target prices to $845 and $905 respectively, both maintaining "Buy" ratings; relative to Thursday's closing price of $648.85, this corresponds to potential upside of approximately 30.2% and 39.5%. However, revenue growth has not yet offset cash consumption from expansion. Meta's Q3 capital expenditure was $34.91 billion, up approximately 80% year-over-year; the full-year guidance lower bound was raised from $130 billion to $135 billion, with the upper bound maintained at $145 billion. After deducting the first nine months' investment, Q4 capital expenditure is estimated at approximately $49.2-59.2 billion, an increase of approximately 41%-70% from Q3. Meanwhile, Meta's Q3 free cash flow was negative $2.12 billion, with debt book value of approximately $83.67 billion, slightly higher than $82.44 billion in cash and marketable securities. The addition of a new non-voting share class provides a tool for future financing; proportional stock distributions themselves do not generate cash. Growth in third-party cloud and AI token procurement costs indicates Meta continues purchasing external services beyond self-building. This explains why its stock price faces pressure—AI computing supply chain computing resource suppliers receive strong AI computing demand support, while Meta shareholders need to wait for new revenue to cover expanding depreciation, operating expenses, and capital investment.
SpaceX provides another set of revenue realization evidence—the company's Q3 revenue was $14.575 billion, up 161% year-over-year; superintelligence business revenue including cloud services, software, and advertising was $7.560 billion, up 830% year-over-year, with operating profit turning positive at $431 million. Management said short-term computing contract renewal prices continue to rise and expects at least 13 gigawatts of computing online by the end of 2027. Bank of America raised its target price from $235 to $275 after earnings, maintaining "Buy," emphasizing Q1 2027 revenue guidance of $33-37 billion significantly exceeding the approximately $22 billion market expectation before earnings; relative to Thursday's closing price of $144.75, the target price corresponds to approximately 90% potential upside. SpaceX's strong earnings and outlook also provide demand-side support for Nvidia's Berlin GTC computing expansion narrative. Musk further stated that year-end computing capacity is expected to reach 2.5-3 gigawatts, with confidence in adding another 10 gigawatts in 2027, with next year's deployment primarily using Nvidia's Vera Rubin platform. This expansion plan echoes GTC's demonstration of DSX POD accelerating on-site deployment and key supply chain capacity expansion: customers need faster access to operational computing, while Nvidia focuses on shortening the distance between hardware delivery and computing service revenue generation. For the semiconductor market, AI business revenue realization and next-generation computing procurement plans mutually confirm each other, providing more specific support for order growth expectations across the entire AI computing supply chain including Nvidia AI GPUs and HBM, data center high-speed optical interconnect, power supply, and liquid cooling chains, and supplementing fundamental catalysts for the "AI bull market narrative."
Financing Still Supported, Memory Still Has Pricing Power: Where Does the Semiconductor Super Counterattack or Even New Bull Market Confidence Come From?
JPMorgan's latest financing calculations draw a line between "individual companies facing cash pressure" and "entire industry investment stalling": globally listed tech companies are expected to generate approximately $2.2 trillion in operating cash flow this year, and by the bank's accounting standards, still have approximately $400 billion in financing surplus after covering related investments; the gap between new cloud providers' approximately $30 billion operating cash flow and $80 billion investment is approximately $50 billion. Its judgment is that financing pressure may change construction entities and financing methods, with companies having more robust balance sheets still capable of undertaking projects, and can also release funds by adjusting buyback arrangements. Meanwhile, semiconductors' approximately 13% market capitalization weight in global indices contributes approximately 21% of risk, with concentrated positioning combined with rising volatility easily triggering passive selling; positions declining from September highs means technical pressure that previously amplified declines is being digested.
Morgan Stanley's chief equity market strategist Michael Wilson's recent positioning approach focuses on earnings quality and financing capacity. Morgan Stanley calculations show the quality factor has risen approximately 6% since mid-August, while the equal-weighted S&P 500 declined nearly 4% over the same period; earnings revisions for high-quality versus low-quality companies have further diverged, leading him to lean toward preparing to increase risk exposure. Wilson's strategist team expects 2027 hyperscale cloud providers and new cloud enterprises' AI capital expenditure at approximately $1.58 trillion, with AI-related company earnings expected to grow approximately 25% further, versus approximately 4% for non-AI companies. This means adjusted selection criteria should place more emphasis on order realization, cash generation, and earnings revision capability; Wilson's team's latest "top picks" screening list including Nvidia, Broadcom, and Micron reflects this direction.
The market's bull-bear debate over the AI computing theme's storm center—the memory chip sector—centers on whether decelerating price increases mean profits are about to collapse. Another Wall Street financial giant Bernstein's rebuttal has three layers: Q4 NAND and conventional DRAM contract prices are still expected to rise 15%-20% and 10%-15% quarter-over-quarter respectively, at historic highs; on a high base, NAND's absolute price increment per GB remains considerable, which can significantly boost gross profit when unit costs are relatively stable; long-term agreement price caps also mechanically suppress blended average price increases. For example, assuming 40% of volume has reached long-term agreement price caps and other volume market prices rise 30%, overall average price increase would only be 18%, and this cannot be used to conclude demand has suddenly declined. SanDisk's data center bit shipment share rose from 12% to 38%, and long-term agreement price floors also strengthen downside protection for some volume, constituting the basis for the firm's opposition to pricing at traditional cycle-top low multiples. Bernstein analysts also noted that nearline hard drive average price per EB rose only 13.6% year-over-year, still having more ample price recovery space compared to DRAM and NAND's previous significant price increases; therefore maintaining Seagate as top pick, with Western Digital and SanDisk both receiving "Outperform," with all three companies' 2027 fiscal year earnings forecasts approximately 12% above consensus.
Bank of America's bullish logic further covers technology upgrades: advanced GPU and ASIC updates drive HBM, SOCAMM, and enterprise SSD demand, and even if 2028 earnings growth decelerates, absolute earnings are still expected to approach 2027's highs. From the underlying architecture perspective, multi-step agents simultaneously increase model inference, tool execution, and state management loads: GPUs and specialized accelerators handle model computation, CPUs execute browser, database query, and task orchestration, HBM and server DRAM carry models, context, and concurrent environments, enterprise SSDs store frequently accessed indices and task states, high-speed networking and optical interconnect support distributed switching, and power and liquid cooling determine the scale at which clusters can continuously operate. Long-term retained data that still grows after filtering, compression, and deduplication supports HDD capacity demand. Therefore, demand expansion can propagate through computing, memory, storage, interconnect, and infrastructure segments, and configuration changes in a single GPU are insufficient to represent procurement changes across the entire data center.
This transmission has already been echoed on the manufacturing and deployment side—"chip foundry king" TSMC's Q3 dollar revenue grew 41.2% year-over-year, and lithography giant ASML stated its existing 2027-2028 supply plans are still insufficient to fully meet customer demand. The DSX POD released at Berlin GTC, through factory pre-integration and validation, shortens on-site deployment to as fast as 14 days once site power, cooling, and network are ready; Nvidia and partners like Amkor expanding packaging, testing, and optical engine capacity improves future delivery capability. For Nvidia, Broadcom, AMD, Micron, SK Hynix, TSMC, and Amkor, investment value is determined by each company's product share, delivery pace, and profit margins, but all benefit jointly from computing demand converting into deliverable systems.
Next Week's Market Focus: Cloud Giants and Intel Earnings Take Center Stage, AI Investment Quality Faces Test
After this week's U.S. semiconductor plunge and Chinese and South Korean tech stock counterattack, next week brings a new round of key earnings. Microsoft, Alphabet, Amazon, and Intel will report results between October 28 and 29. From cloud services to chip manufacturing, these companies' results and management guidance will provide new clues for investors assessing AI demand, capital expenditure, and semiconductor supply chain conditions.
For Microsoft, Alphabet, and Amazon, market focus will be on Azure, Google Cloud, and AWS business growth rates, and whether AI-related revenue can keep pace with infrastructure investment growth. As data center construction continues to push capital expenditure higher, investors will also watch the three companies' subsequent investment plans, computing supply progress, and cash flow performance. If cloud business growth remains strong and AI commercialization progress becomes more evident, it will help support market judgments about AI infrastructure demand; conversely, if capital expenditure continues climbing while return realization remains unclear, related stocks and their upstream supply chains may continue facing valuation pressure.
Intel's earnings will provide another observation perspective for the semiconductor industry. Investors need to watch its data center and client business performance, product competitiveness, manufacturing and foundry business progress, and management's judgment on future demand and operating outlook. Unlike cloud giants that primarily answer "whether AI investment can convert into revenue," Intel's results are more directly related to the traditional CPU market, data center processor demand, and its own business transformation progress. Whether its earnings can release positive signals may also affect overall market sentiment toward the semiconductor sector.
Therefore, the key to next week's earnings season is not just whether companies' revenue and earnings per share exceed expectations, but whether different segments can provide mutually confirming signals: whether cloud giants continue expanding AI investment, whether AI services can bring clearer revenue growth, and whether chip companies can improve operating performance from demand expansion. If these signals are broadly positive, the recent semiconductor selloff may be viewed as a sentiment and valuation adjustment; but if capital expenditure returns, business growth, or earnings guidance fall short, market divergence over the AI supply chain may further intensify.
Overall, memory chip price increases, advanced packaging capacity expansion, and continued AI data center construction still provide fundamental support for the semiconductor supply chain; the strong counterattack in Chinese and South Korean tech stocks also shows some capital has not abandoned the AI growth narrative. However, strong demand does not equal stock prices only rising. As AI infrastructure investment scale continues expanding, the market is shifting from purely chasing computing demand toward placing greater emphasis on earnings quality, financing conditions, and investment returns. Next week's earnings from Microsoft, Alphabet, Amazon, and Intel will provide a new round of verification for this divergence: whether AI computing expansion can continue supporting the semiconductor bull market narrative, or whether the market will further demand companies provide clearer revenue and cash flow evidence.
Comments