According to ZhiTong Financial APP, just as the semiconductor sector closely tied to global AI computing power continues to decline in global stock markets, especially with storage chip stocks that have long been comprehensive beneficiaries of an unprecedented AI infrastructure boom seeing sharp price drops recently, Wall Street financial giant Bank of America has essentially delivered a much-needed "bullish tonic" to storage chip stocks and the broader global AI computing theme that AI and tech stock bulls have been desperately craving.

After storage chips experienced a deep correction, the market has already been pricing in advance a significant slowdown in future price increases, supply expansion, and AI financing pressures. However, Bank of America believes that spot price resilience, HBM generational upgrades, and tiered storage demand driven by the AI agent inference frenzy can still support strong profit expansion for a considerable period. Just as some cautiously positioned institutional investors begin putting real money behind pricing "AI bubble burst" and "storage chip supercycle disillusionment," the most important contrarian bullish judgment in Bank of America's in-depth research report titled "Global Storage Technology — P/E and P/B Ratios Fall to Cyclical Lows, Investor Views Diverge, Focus on Q4 Guidance and Spot Price Stability" is that some institutional investors have already been pricing "boom-to-bust" for global storage chip stocks such as SK Hynix and Samsung Electronics, as well as broader AI computing investment theme targets, yet fundamental indicators including storage chip prices and momentum, orders, and profit trends have not shown any corresponding degree of deterioration.

Bank of America's research report shows that after approximately three and a half months of adjustment, valuations of some storage chip super leaders have approached historical cyclical troughs, while 2027-2028 earnings are still expected to remain at elevated levels. However, the market has recently been concentrated in trading certain extremely pessimistic logics — last week the Nasdaq 100 fell about 5.5%, and South Korea's KOSPI index, known as the "storage chip and AI computing barometer," plunged about 14.5% for the week; both have declined about 4% and 17% respectively so far in October; as of the 16th, year-to-date gains remain at approximately 16% and 35% respectively, which sufficiently illustrates that the massive profits accumulated previously are experiencing rapid giveback under the negative pessimistic pricing trajectory of "rising financing costs associated with AI infrastructure" and "AI bubble burst."

The Philadelphia Semiconductor Index, which encompasses the world's top chip giants including Nvidia, AMD, Intel, and Micron, saw a rare sharp drop of about 11.4% last week. Rising U.S. Treasury yields, which continue to sit at their highest levels in more than two decades, have significantly escalated AI financing pressures, further amplifying market concerns about pricing and supply-demand rapidly balancing after substantial storage chip capacity expansion, as well as forward profit expectations.

Overall, the storage chip sector's persistently low valuations, spot price resilience, and contrarian buying are constituting positive signals, but what can truly drive a sustained counterattack is whether the earnings season and chip giants' forward outlooks can positively demonstrate that high profits can be converted into cash, and whether credit pressures will stop eroding customer purchasing power.

It is worth noting that just as the global semiconductor sector plummeted, institutions significantly reduced directional allocation exposure, but retail investors' long-preferred semiconductor leveraged ETF (the triple-leveraged bullish ETF tracking the Philadelphia Semiconductor Index — U.S. ticker "SOXL") saw massive inflows. Potential short-covering demand and leveraged capital have together amplified the market's sensitivity to strong fundamental catalysts from the earnings season. Positive earnings and management's potential upside guidance and strong shareholder return trends could become catalysts for underweighted institutions to re-add positions — if cloud computing giants' AI computing resource-related orders and cash flows, and storage manufacturers' selling prices and profit margins continue to exceed expectations, positions that institutions had previously reduced due to financing pressures and cyclical concerns could translate into potential buying.

Meanwhile, retail investors' overall capital continues to pour into semiconductor leveraged ETFs, indicating that some investors have already begun betting on a rebound in advance; when earnings expectations improve and institutional re-allocation resonate together, the AI computing sector including storage chips could shift from valuation compression to a strong recovery supported by earnings.

Stock prices were the first to fall into a "cyclical winter," yet storage profits have not simultaneously entered winter. In this research report, Bank of America defines the current adjustment as valuation-first compression rather than a confirmed storage recession. The report calculates that SK Hynix has retreated about 57% from its June high, exceeding the average 51% decline of major historical adjustment cycles. Based on Bloomberg consensus estimates, Samsung and Hynix's 2027 P/E ratios are approximately 3.0x and 2.7x respectively, further declining to 2.8x and 2.3x in 2028; based on 2028 forecast net asset values, both have P/B ratios of only about 0.9x. Bank of America's own quantitative model appears relatively more conservative, projecting the two companies' 2027-2028 P/E ratios at approximately 3.5-3.7x. It is worth noting that these extremely cheap P/B ratios are based on future profit and net asset realization and cannot be directly equated to being "below book value" based on current book value.

Bank of America acknowledges that 2027 quarterly growth will moderate and 2028 EPS may decline slightly year-over-year, but expects that the absolute scale of revenue, operating profit, and EPS will continue to approach historical highs. Its bullish core is the duration of elevated profits, not ever-accelerating price increases. Price data indeed does not yet support a comprehensive recession judgment: the report lists 16Gb DDR5 spot prices at approximately $60.6, up 3% weekly and 107% year-to-date; October DDR5 contract prices are expected to rise about 17% month-over-month, with DDR4 also expected to gain more than 15%.

Meanwhile, some DDR4 specifications have stabilized after falling from highs, NAND main specifications were essentially flat week-over-week with some products seeing monthly average price declines; server memory module prices remain elevated after previous increases. Consumer SSD price recovery and broadly stable panel prices also indicate that end demand is not experiencing asynchronous prosperity. Another Wall Street giant KeyBanc's recent Asia survey data shows the institution forecasts HBM blended prices could rise significantly by 80%-100% in Q1 2027; on the DRAM memory price front, KeyBanc forecasts a 15%-20% quarter-over-quarter increase in Q4 2026 on top of Q3's strong price increase base, with a further 10%-15% increase in Q1 2027; for NAND future pricing, it forecasts approximately 20% quarter-over-quarter increase in Q4 2026 on Q3's strong base, with a further 10%-15% increase in Q1 2027.

The current more accurate description is "differentiation and decelerating growth on a high price plateau," rather than storage prices having comprehensively collapsed. What Bank of America's top semiconductor stock analyst team is betting on is precisely that — even if HBM/DRAM/NAND storage chip price increases narrow, higher selling prices and improved product mix can still sustain considerable profit margins and return on capital near record levels. What is truly suppressing stock valuations is the market beginning to work backward along the financing chain to infer future demand.

CoreWeave's 2031-maturity bond yield briefly exceeded 16% last week, and the market-implied cumulative default probability corresponding to five-year CDS quotes is close to 63%; however, it should be noted that the latter includes risk premiums and recovery rate assumptions and cannot be treated as actual default probability. For new cloud companies, data center, power, and GPU procurement first form cash expenditures, with customer acceptance, revenue recognition, and collection materializing subsequently; rising financing costs compress the tolerance for delays and low utilization rates in projects, while GPU iteration adds uncertainty to collateral asset values. However, the market's increasingly intense concerns and "AI bubble burst" pessimism lie in the fact that — even if the extraordinarily strong order demand associated with AI computing infrastructure truly exists, financing gaps may force customers to slow expansion.

Furthermore, in the view of investors holding cautious positions on the AI computing theme, it is also worth noting that long-end U.S. Treasury yields have seen a risk-aversion-driven pullback after previous increases, with the 10-year yield at approximately 5.05% on October 19; credit spreads may not necessarily narrow in tandem, so falling risk-free rates do not automatically relieve AI project financing pressures.

From a single GPU to an entire AI computing supercluster, AI storage demand is being redistributed. Bank of America's investor exchanges in Singapore show that hedge funds are overall pessimistic, while most long-term asset management-type funds and retail investor representatives remain biased toward long-term bullishness, with disagreements centered on new cloud financing, China's 4F² DRAM technology, declining per-GPU storage configuration, and slowing AI capital expenditure and storage price growth.

From a technical mechanism perspective, multi-turn agent inference, tool calling, and long context simultaneously increase demand for model weight reading, KV cache, and state preservation; these workloads are distributed across HBM, host DRAM, LPDDR5/SOCAMM, and enterprise SSDs. Cache tiering and sharing can reduce redundant storage but also shift demand to higher-capacity, lower-cost tiers. Therefore, a decline in per-GPU HBM configuration cannot directly lead to a conclusion of declining storage demand for the entire cluster; the key still lies in cluster scale, concurrency, and context length. Similarly, 4F² means improved storage cell area efficiency, but its supply impact still needs to materialize through yield, mass production, and customer certification; therefore, Bank of America's analyst team judges near-term impact to be limited and emphasizes storage replacement demand brought by GPU and ASIC upgrades.

Meta's opening of Muse Spark Open can be said to provide new technological impetus for application diffusion and inference demand growth. As Meta Muse and OpenAI's GPT-6 Astra comprehensively ignite the AI agent mega-wave, combined with the latest developments showing the release of Muse Spark Open — "the most capable open-weight AI large model in the U.S. to date" — which also focuses on cutting-edge AI agent workloads, the "AI FOMO trade" (referring to investors fearing missing out on gains in AI-related assets and therefore chasing or re-adding previously reduced positions) is likely to make a comeback in the near term.

From the perspective of AI computing infrastructure's underlying architecture, Muse and Astra, along with Meta's latest open-source Muse Spark Open, are expected to jointly further expand the scope of AI agent usage. The massive expansion of AI inference workloads they bring will simultaneously increase demand for model inference, tool execution, and state management, while deepening competitive pressures on some software companies. GPUs, TPUs, and other accelerators handle model computation; high-performance server CPUs run browsers, virtual machines, product retrieval, database queries, and transaction orchestration; HBM and server DRAM carry model data, context, and concurrent work environments; enterprise SSDs store product indices, task records, and persistent state, while high-speed networks and optical interconnects support distributed data exchange.

Another Wall Street financial giant JPMorgan's recently published research report shows that at the stock price level, trading in advance a cooling of the storage chip cycle and the global AI computing industry chain's earnings supercycle — including storage chip components — maintaining strong momentum for an extended period can occur simultaneously. JPMorgan's Asia-Pacific analyst team led by Jay Kwon is accordingly bullish on Samsung Electronics, SK Hynix, Kioxia, as well as Samsung Electro-Mechanics, ISU, and ISC over the next six months, betting on the storage upcycle lasting longer and the continuation of related substrate, PCB, and testing demand.

JPMorgan explains from the industry-side strong earnings trajectory why the storage pullback may create opportunities: market concerns about AI capital expenditure sustainability may not have been equivalently translated into deterioration in storage orders and profits. AI inference requires continuous reading of model weights, saving context and KV cache, and handling data demand at different tiers through server memory and SSDs; HBM upgrades also consume wafer, packaging, and yield ramp resources, making it difficult for effective supply to quickly keep up with demand. JPMorgan's semiconductor research report on the Korean stock market shows that macro-level concerns about AI investment sustainability are clearly disconnected from the storage industry's bottom-up supply-demand data, so it views Q3 as an "expectation reset quarter," with core overweight positions over the next six months in Samsung Electronics, SK Hynix, Kioxia, extending to Samsung Electro-Mechanics, as well as ISU and ISC — leaders in AI PCB and test sockets respectively.

The "duration trade" here is betting on the upcycle lasting longer and high profits being sustained for longer: even if DRAM/NAND price increase slopes slow, as long as HBM, server DRAM, and enterprise SSD demand continues to substantially absorb new capacity and achieve significant growth trajectories, the rapid profit decline previously priced into the market may need to be revised. This also explains why storage stocks can decline before earnings peak, yet may rebound early once earnings duration is validated; what determines trading opportunities is the continuously evolving gap between actual earnings paths and stock price-implied expectations.

Bank of America's research report shows that global trade and equipment data provide cross-validation for the resilience of storage chip and broader global semiconductor demand driven by the AI computing infrastructure boom, but also reveal growth structure: South Korea's semiconductor exports in the first ten days of October were approximately $12.2 billion, up 171% year-over-year; China's September integrated circuit import value grew 75% year-over-year with volume growing only 2%, meaning price and product mix contributions far exceeded volume expansion. Meanwhile, TSMC's Q3 revenue significantly beat expectations with approximately 51% year-over-year growth, with full-year capital expenditure planned at $60-64 billion; ASML (the lithography giant) reported Q3 revenue of approximately €11.7 billion and expects Q4 to reach €14.2-15.2 billion.

This also explains why equipment orders may continue to be strong during storage stock adjustments: stocks are pricing in advance future supply increases and price deceleration, while equipment companies are realizing customers' previously initiated capacity expansion and process upgrades. However, the two are not permanently decoupled; if customers ultimately cut capital expenditure, equipment orders would also be affected. The recent validation window is concentrated on Q4 guidance. Nanya Technology has already released a positive outlook, and Bank of America expects Hynix to disclose stronger selling prices and profit margins on October 28, 12-layer HBM4 mass production ramp-up, and increased 8-layer product output for new orders, and may announce a special cash dividend in addition to the recent 40 trillion won buyback. The report believes that 2027-2028 dividends and buybacks have the potential to equal more than 10% of current market capitalization, but this represents potential shareholder returns, not a committed cash dividend yield. Longer term, Bank of America suggests the global storage market may expand from approximately $1 trillion in 2026 to over $2 trillion by 2030, and discusses scenarios where industry operating profit reaches $1 trillion and market capitalization reaches $10 trillion. These positive outlooks are long-term bullish scenarios provided by Bank of America's analyst team; whether recent stock prices can recover, in the view of Bank of America's analyst team, still depends first on positive realization of DRAM/NAND price increases, HBM delivery data, and cash distribution.

Institutional selling versus leveraged bottom-fishing: conditions for an AI computing super-counterattack are accumulating, with earnings and guidance determining the final outcome. Wall Street legendary investor — Oaktree Capital Co-Chairman Howard Marks' latest judgment provides a reverse observation angle for current pessimism: truly dangerous bubbles are usually accompanied by loss of vigilance toward risk, and continuous discussion of bubbles means investors have not completely abandoned constraints. Current forward P/E ratios of leading tech companies are significantly lower than during the internet bubble era, which is also related to companies having already formed scaled profits. This intersects with Bank of America's storage judgment: whether low P/E ratios can constitute opportunities depends on whether earnings can be maintained; if the market underestimates the duration of high profits, valuations may recover; if earnings forecasts are too high, low multiples may just be a cyclical trap.

On the capital flow side, there appears to be a strong two-way force. According to the latest Goldman Sachs statistics, approximately $9.5 billion flowed into popular leveraged ETFs in the U.S. stock market during October 12-16, of which approximately $7 billion entered the triple-leveraged semiconductor bull ETF SOXL; meanwhile, institutional net selling of tech stocks hit a record, with U.S. fundamental long-short funds' net exposure falling to the lowest historical percentile, while gross exposure remained at 213.9%, at a relatively high level. Particularly noteworthy is that these triple-leveraged high-leverage ETF investment products pursue daily leverage targets, with rebalancing during sustained declines reducing risk exposure and potentially increasing exposure during rallies, thereby amplifying market moves.

Furthermore, for certain institutional investors with relatively optimized positions, better-than-expected earnings may increase their willingness to significantly re-add positions, strengthening pressure to catch up with benchmark returns. Positive earnings are expected to become a catalyst for underweighted institutions to re-add positions: if cloud company orders and cash flows, and storage manufacturers' selling prices and profit margins continue to materialize, positions that institutions had previously reduced due to financing pressures and cyclical concerns could translate into potential buying. Meanwhile, capital continuing to flow into semiconductor leveraged ETFs indicates some investors have already begun betting on a rebound in advance; when earnings expectations improve and institutional re-allocation resonate together, the AI computing sector could shift from valuation compression to recovery supported by earnings.

In the view of Bank of America analysts, the contrarian bullish logic for storage chips and the broader AI computing investment theme is taking shape, with the most powerful fundamental basis cited being significant valuation compression while prices, deliveries, and earnings have not yet simultaneously broken down. Goldman Sachs emphasizes that earnings confirmation needs to be observed along the entire industry chain: whether large cloud computing giants and new cloud companies can convert orders into utilization rates, collections, and sustainable cash flows; whether storage chip super giants can maintain profit margins, control capacity expansion, and deliver shareholder returns after price growth slows; and whether credit markets can stop further raising financing thresholds. If all three pieces of evidence improve simultaneously, underweighted positioning recovery and earnings revaluation could together drive a counterattack; if only leveraged buying increases while cash flows and financing conditions continue to deteriorate, the rebound could still be interrupted by another round of deleveraging.