After a sharp pullback in global AI computing theme stocks, particularly popular semiconductor names tied to memory chips, top research institutions including Bank of America, Nomura, and the well-known market research platform TrendForce have provided fresh fundamental growth evidence for bargain-hunting: the peak in memory chip industry revenue and earnings may prove more durable than current share prices reflect.

In the week ending November 6, Micron Technology (NASDAQ: MU) fell roughly 22% cumulatively, and its subsequent rebound was hit by hotter-than-expected inflation and rate hike expectations, creating a stark contrast between valuation adjustments in memory stocks and industry prosperity. Bank of America's latest assessment offers a fundamental basis for buying on dips: even using lower price-increase assumptions than TrendForce and factoring in price corrections for some products in 2027, global memory industry revenue is still expected to reach $1.47 trillion in 2027 and $1.58 trillion in 2028, continuing to drive record-level revenue and earnings.

What supports the bullish calls from top Wall Street institutions like Bank of America is essentially the RSI-led AI training paradigm, the expanding use of AI agent applications centered on efficient AI inference that continues to boost demand for HBM, server memory, and enterprise SSDs, while advanced process nodes, capacity conversion, and yield ramp-up limits constrain supply release, and long-term supply agreements further improve sales visibility.

The potential "buy-the-dip investment value" created by this sharp pullback in AI computing and semiconductor stocks depends critically on whether the market's pessimistic discount for "peak earnings" far exceeds the actual deterioration in supply and demand: if memory leaders can continue to deliver cash flow and shareholder returns, the low valuations after the sharp decline could become a new starting point for repositioning in the memory supercycle.

Bank of America expects global memory industry revenue to grow from $970 billion in 2026 to $1.47 trillion in 2027 and $1.58 trillion in 2028, representing growth of about 51.5% and 7.5% in the latter two years respectively. Even using price-increase assumptions significantly below TrendForce's, BofA analysts still judge that the memory industry is poised to extend record-level revenue and earnings through 2028.

Meanwhile, after Micron Technology and SanDisk experienced last week's rout and Monday's rebound, they were hit again on Tuesday by inflation and rate hike expectations. Notably, however, the scale of this pullback was clearly milder — the VIX fear index even fell from 19.92 to 19.46, largely indicating that the market is reassessing the pressure from the Fed's rate path rather than repeating Thursday's panic selling. These signs highlight that the market appears to be re-evaluating whether earnings created by supply shortages can continue to offset valuation pressure from rising rates and adjustments to AI capital expenditure expectations.

BofA Firmly Expects Memory Supercycle to Last Through 2028

In its latest bullish research report on memory chip theme companies, Bank of America's strategist team cited the view of TrendForce senior analyst Avril Wu, saying the memory chip industry's current upcycle may last through 2027 and avoid entering a downcycle in 2028. Speaking at a webinar hosted by BofA on November 10, Wu said memory chip prices may continue to rise, projecting that DRAM and high-bandwidth memory prices will increase 10% to 20% quarter-over-quarter in the fourth quarter of 2026, while NAND flash prices will rise 8% to 13%. She also expects HBM prices to more than double in 2027, while long-term supply agreements may help reduce price volatility after 2028.

BofA broadly agrees with this optimistic industry outlook but adopts more conservative price assumptions. The firm expects fourth-quarter DRAM prices to rise 8% quarter-over-quarter and NAND prices to rise 3%, with price corrections potentially beginning in early 2027. Its strategist team also expects HBM price gains in 2027 to be below 50%, contrasting with TrendForce's judgment of prices more than doubling.

BofA strategists also said their latest forecasts show global memory industry revenue reaching $970 billion in 2026, $1.47 trillion in 2027, and $1.58 trillion in 2028, indicating the industry is poised to sustain record revenue and earnings levels through 2028. With Meta Muse and OpenAI's GPT-6 Astra fully igniting the AI agent wave, "AI FOMO trades" — where investors worry about missing out on gains in AI-related assets and chase or rebuild previously reduced positions — are expected to continue. From the perspective of the underlying architecture of AI computing infrastructure, Muse and Astra are expected to further expand the scope of AI agent usage, and the massive expansion of AI inference workloads they bring will simultaneously increase demand for model inference, tool execution, and state management, while deepening competitive pressure on some software companies.

From a technical mechanism perspective, AI agents expand a single answer into repeated reasoning, retrieval, tool execution, and state preservation. GPUs and TPUs handle model computation, CPUs manage tools and task orchestration, HBM holds high-bandwidth accessed data, server DRAM hosts the runtime environment, and SSDs and HDDs serve high-performance access and large-capacity retention respectively. At the same time, network infrastructure, high-speed optical interconnects in data centers, power supply, and cooling determine the actual utilization of expensive accelerators. Therefore, accelerating cloud revenue can support multiple hardware segments simultaneously; how much profit a specific company captures still depends on performance advantages, supply bottlenecks, and bargaining power.

For the epicenter of the AI trade and tech stock selloff — the hard-hit memory chip sector — optimism about market fundamentals remains strong. In addition to Bank of America's optimistic earnings and memory price trend research, KeyBanc's recent Asia survey data shows that AI data centers are still broadly absorbing memory and other AI computing resource supply, driving price increases through 2027. After surveying Asian markets, KeyBanc forecasts that HBM blended prices could surge 80%–100% in the first quarter of 2027; on the DRAM side, KeyBanc predicts a 15%–20% quarter-over-quarter increase in the fourth quarter of 2026 on top of a strong third-quarter base, followed by another 10%–15% gain in the first quarter of 2027; for NAND pricing, it forecasts roughly 20% quarter-over-quarter growth in the fourth quarter of 2026 on a strong third-quarter base, followed by another 10%–15% increase in the first quarter of 2027.

Memory Rebound Hit Again by Rate Hike Shock: Stocks First Price In Inflation and Sustained Fed Rate Hike Risk, but Earnings and Orders Are Still Accelerating

On November 5 U.S. Eastern Time, the Nasdaq 100 plunged 6.36%, the largest single-day drop since March 2020, with constituent stocks collectively losing about $2.3 trillion in market value — a single-day record; the Philadelphia Semiconductor Index plummeted 12.25%. After Democrats won control of both chambers of Congress, investors worried that stricter AI regulation, data center approval requirements, and electricity cost-sharing mandates could delay computing project launches and chip procurement. These policy concerns, combined with existing high valuations and crowded positioning, amplified the market's repricing of the timeline for AI investment returns.

The memory sector bore a deeper pullback. Based on closing prices from October 30 to November 6, Micron Technology and SanDisk fell 22.01% and 27.44% respectively, while Samsung Electronics and SK Hynix dropped about 16.13% and 20.17% respectively, all weaker than the roughly 4.91% and 12.15% declines in the Nasdaq 100 and Philadelphia Semiconductor Index over the same period. On November 9, global tech stocks broadly rebounded, with the Nasdaq 100 and Philadelphia Semiconductor Index rising 3.40% and 6.85% respectively, while Micron Technology and SanDisk gained 6.25% and 6.16%. Funds bought back into memory leaders, but this rally has not yet offset previous losses, also showing that the "cash flow safe haven" argument from some Wall Street analysts is a judgment on operations and valuation, not directly equivalent to short-term stock resilience.

Tuesday's inflation data raised the bar for a rebound again. U.S. October headline and core CPI rose 0.4% and 0.3% month-over-month respectively, both above market expectations of 0.3% and 0.2%, with core CPI rising 0.3% for a third consecutive month. Waller said before the data release that if core CPI reaches 0.3% or higher, a 50 basis point rate hike in December should be considered. After the data, market pricing for such a larger hike rose to about 11.8%, but 25 basis points remained the mainstream expectation; bets on cumulative hikes through mid-2027 increased from two 25 basis point moves to three, including this December. Deutsche Bank also added a January rate hike forecast between its existing December and next March projections. The logic behind extended Fed rate hike expectations is that employment and economic growth remain resilient, while service prices, energy cost pass-through, and AI construction demand jointly maintain inflation pressure. For memory chip stocks, higher rates both raise the discount rate on future profits and may increase customer financing costs and extend data center construction timelines. On Tuesday, the Nasdaq 100 and Philadelphia Semiconductor Index fell about 0.60% and 0.35% respectively, while Micron Technology and SanDisk dropped about 1.1% and 3.0%: this also shows to some extent that the market continues to absorb rate repricing without repeating Thursday's decline.

Can the Memory Supercycle Survive the Valuation Cold Wave? The Key Is How Long High Profit Numbers Can Be Sustained

The common judgment of BofA, TrendForce, KeyBanc, and other institutions is that the memory upcycle still has support, with disagreements centered on the magnitude of price increases and the timing of corrections. TrendForce expects fourth-quarter DRAM and HBM prices to rise 10%–20% quarter-over-quarter and NAND to rise 8%–13%, while BofA forecasts only 8% and 3% for DRAM and NAND respectively; for 2027 HBM, the former expects prices to more than double, while the latter expects gains below 50%. A more conservative price path can still support revenue and earnings at high levels, meaning BofA strategists' optimistic bullish call also depends on significant shipment growth, product upgrades, and order sustainability. This is also the most investment-relevant part of BofA's "memory supercycle through 2028" thesis.

Technical support comes from AI systems' simultaneous demand across different memory tiers: HBM provides high-bandwidth data supply for GPUs and ASICs, server DDR5 supports main memory demand, and enterprise SSDs handle data retention, retrieval, and caching. At the same time, HBM competes with conventional DRAM for limited advanced process and wafer resources, and new product yield and packaging output ramp-up takes time. Notably, TrendForce's earlier forecast of a 121% increase in 2027 HBM blended average selling prices also includes a higher mix of more expensive HBM4 and the introduction of HBM4e, and should not be interpreted as all same-spec products doubling in price.

Long-term agreements are improving demand visibility. SK Hynix recently disclosed that it has signed long-term agreements with 14 major customers, typically with five-year terms, and secured contract volumes covering more than half of its 2027–2028 capacity. Contract prices adjust within agreed ranges based on the market, so long-term agreements mainly enhance visibility for capacity utilization, sales, and earnings rather than fully locking in future profits. For capital-intensive memory chip companies, this change helps reduce mismatches between capacity expansion and customer procurement pace.

BofA analyst Vivek Arya's team previously explained the current opportunity from a stock price cycle perspective: as of November 5, the Philadelphia Semiconductor Index had fallen about 38% from its June high, exceeding the average drawdown of about 32% in the past four major downcycles; however, the 136-day adjustment period was only about 54% of the historical average of 253 days. BofA analyst Vivek Arya's team sees early 2027 as a potential window for the selloff to end, based on the fact that stock prices typically lead actual sales downturns by 6–9 months. The forward P/E fell to 13.7 times, about a 22% discount to the S&P 500, indicating that a large amount of pessimism is already in the price; however, BofA also estimates that if 2027 AI capital expenditure is 20% below expectations, related chip companies' earnings per share could face 35%–40% downward revision pressure.

Nomura's CW Chung team places more emphasis on cash flow allocation: cloud vendors bear upfront construction investment, while memory suppliers accumulate cash through scarce products and price increases. Based on the stock prices and earnings forecasts used in its report, Samsung and SK Hynix have 2027 expected P/E ratios of about 3.2 times and 2.8 times respectively; supply tightness extending to 2028, increased long-term agreements, and expanding shareholder returns are the basis for its view that the market has over-discounted.

Citi analyst Atif Malik also reiterated a "Buy" rating and $1,300 target price for Micron Technology on October 29, arguing that investors underestimate the degree of sustained DRAM and HBM supply tightness in 2027–2028, while customer long-term agreements enhance revenue visibility. Morgan Stanley analyst Shawn Kim's latest adjustment reveals another layer of disagreement: a November 9 report raised 2027 operating profit forecasts for Samsung and SK Hynix but cut their target prices by 19% and 38% respectively, while maintaining "Overweight." The concern is that if new capacity released by the end of 2027 coincides with delays in data center deliveries, longer-term profit margins will be pressured. This shows that institutional debate has extended from whether money can be made now to how many years high profits can be sustained.

Overall, considering supply and demand, long-term agreements, and valuation changes, this sharp pullback in memory chip stock prices and valuations has created a more fundamentally grounded bargain-hunting window for leaders with strong HBM competitiveness, server-grade high-performance DRAM/NAND product portfolios, and cash accumulation capability. Investment value depends largely on how much free cash flow companies can retain after more conservative price and margin assumptions. If demand materializes, product upgrades continue, and shareholder returns keep validating earnings resilience, the current discount for "peak earnings" could become the source of the next re-rating.

Based on closing prices in their respective markets on November 10, recently verifiable institutional target prices still imply substantial upside for memory leaders. U.S. memory chip giant Micron Technology (MU.US) closed at $724.77, and BofA maintained its "Buy" rating and $1,550 target price on November 9, implying potential upside of about 113.9%; Citi maintained "Buy" and a $1,300 target price on October 29, implying about 79.4% upside. Both institutions emphasized DRAM and HBM supply constraints in 2027–2028 and the support from long-term supply agreements for sales and earnings visibility.

For SanDisk (SNDK.US), Morgan Stanley maintained an "Overweight" rating and $1,750 target price on October 30, implying about 103.5% upside from the November 10 close of $859.91, bullish on data center demand growth improving its business mix while using cross-cycle earnings to measure valuation. The latest ratings for South Korea's two memory giants reflect an institutional stance of "staying bullish on near-term earnings while reassessing long-term margins." In its November 9 report, Morgan Stanley cut its target price for Samsung Electronics common shares from 381,000 won to 309,000 won and SK Hynix's target price from 2.6 million won to 1.6 million won, maintaining "Overweight" on both; based on November 10 closing prices of 198,900 won and 1.212 million won respectively, the implied upside is 55.4% and 32.0%. The firm also raised 2027 operating profit forecasts for both companies, with the target price cuts mainly stemming from a reassessment of new supply around 2028, data center delivery pace, and long-term profit margins.