Despite recent price corrections among leading memory chipmakers and the broader AI semiconductor sector, major Wall Street institutions remain bullish on the long-term trajectory for stocks tied to the unprecedented AI infrastructure boom and the ensuing "memory super cycle."
Nomura, a prominent Wall Street investment firm, has published a report refuting the "semiconductor peak" theory. Meanwhile, a recent report from Bank of America indicates that global capital expenditure for cloud computing and AI-related infrastructure is projected to reach $1.5 trillion by 2027. The report suggests that the current summer pullback in AI semiconductor stocks, including memory chips, represents a healthy reset rather than any structural change in demand for AI computing power.
In the AI compute stack, GPUs are responsible for generating intelligence, HBM/DRAM provides high-speed data feeding, enterprise-grade NAND/eSSD handles hot data and caching, while HDDs manage the long-term storage of massive cold/warm data. Consequently, Wall Street giants like Goldman Sachs argue that the AI computing arms race led by cloud giants is transforming memory chips from cyclical commodities into scarce strategic assets. Price increases for DRAM/NAND in 2026 are not seen as the end but potentially the initial phase of a super cycle.
Whether it's Google's massive TPU clusters or Nvidia's immense AI GPU clusters, all require fully integrated HBM memory systems equipped with AI chips. This is compounded by tech giants' accelerated construction and expansion of AI data centers, necessitating large-scale purchases of server-grade DDR5 memory and enterprise-grade high-performance SSDs/HDDs. Samsung Electronics, SK Hynix, and Micron Technology are uniquely positioned across these three core memory domains: HBM, high-performance server DRAM (including DDR5/LPDDR5X), and high-end data center SSDs. They stand to be the most direct beneficiaries within the "AI memory and storage stack," reaping the "super dividends" of the AI infrastructure wave.
The stellar preliminary Q2 results just disclosed by South Korea-based memory chip leader Samsung Electronics serve as a clear profit snapshot of this memory super cycle. Operating profit for April to June is estimated to have soared roughly 19-fold year-over-year to approximately 89.4 trillion won (about $58.4 billion), setting a new quarterly record and marking a 56% sequential increase from a strong prior quarter. This figure exceeded the analyst consensus estimate of around 84.2 trillion won. Revenue for the period is expected to reach 171 trillion won, surpassing market estimates of 169.2 trillion won and representing about 129% year-over-year growth. The company plans to release its full financial report on July 30, which will include net profit and segment breakdowns. Samsung Electronics' quarterly operating profit has now surpassed Nvidia's last quarter operating profit of $53.536 billion (about 82 trillion won), making it the company with the highest quarterly operating profit globally.
On Wall Street, analysts are collectively optimistic about further record-breaking surges for these three memory chip manufacturers—SK Hynix, Samsung Electronics, and Micron. The core logic is that against the backdrop of a fervent global AI data center construction boom, demand for HBM, high-capacity DRAM, and enterprise NAND memory chips continues to explode due to unrelenting demand for AI compute infrastructure. AI servers, high-performance computing, and cloud infrastructure build-outs are persistently driving up memory demand, while new capacity additions struggle to keep pace with demand growth.
In Goldman Sachs' view, the AI bull market is far from over; it is transitioning from the "AI chip purchasing frenzy" into the second phase of "massive AI factory construction." This means the next wave of excess alpha returns will not be confined to the strongest leaders in AI GPU/AI ASIC but will systematically spread across the full-stack AI compute infrastructure layer of the "AI factory," including data center high-performance CPUs, DRAM/NAND/HBM memory, AI PCBs, liquid cooling systems, data center optical interconnects, ABF substrates/glass substrates, MLCCs, electronic fabric, and broad foundry services.
Goldman Sachs believes the global bull market surrounding the AI compute chain is far from finished. The market's main narrative has evolved from the long-standing "programming/code-driven software and light-asset valuation expansion" since 2008 to a "re-pricing of a series of physical assets constituting AI compute infrastructure."
Goldman Sachs' latest estimates suggest hyperscale cloud providers' total AI infrastructure-related investments could exceed $6 trillion by 2030. The global AI capital expenditure base model is projected to grow from $765 billion annually in 2026 to $1.65 trillion annually by 2031, with cumulative capex from 2026-2031 estimated at around $7.6 trillion. U.S. data center power demand is expected to rise from 31 GW in 2025 to 66 GW by 2027.
Bank of America Champions AI Semiconductors
An analyst team at Bank of America, led by veteran strategist Vivek Arya, stated: "The AI semiconductor-driven market rally is not over. After a record 88% surge in Q2, the Philadelphia Semiconductor Index (SOX) pulled back 11% in Q3, aligning with its historically weakest seasonal period. We view the current pullback as a healthy reset, not a structural change in AI demand. This reset is expected to be a summer phenomenon, with a potential rebound in the fall. Short-term leadership may favor lower-beta names like Nvidia (NVDA), Texas Instruments (TXN), Analog Devices (ADI), and the two leading chip design EDA firms, Cadence Design Systems (CDNS) and Synopsys (SNPS). However, history suggests that after consolidation periods, new momentum often emerges as investors regain strong confidence in the next earnings and capex growth cycle."
The analysts also noted that ongoing consolidation announcements, such as Texas Instruments' (TXN) acquisition of Silicon Labs (SLAB) and ON Semiconductor's (ON) active pursuit of Synaptics (SYNA), could become another enduring theme in the fragmented analog chip sector.
Arya's team emphasized in their report that the $1.5 trillion scale of cloud and AI compute infrastructure capital expenditure keeps the AI compute super-demand cycle intact. They stated: "We expect global cloud and AI compute infrastructure capex to approach $1.5 trillion by 2027, implying potential for another 40% to 50% year-over-year growth. This is strongly supported by continued token scale growth, a surge in enterprise AI agent adoption, and constrained infrastructure supply. Importantly, the focus for hyperscalers remains on maximizing utilization and AI-driven performance growth trajectories, not on optimizing depreciation."
The analysts added that as visibility into 2027 cloud capex improves in the second half of 2026, they anticipate renewed leadership and momentum in the following areas: Micron Technology (MU) in memory chips; Advanced Micro Devices (AMD) and x86 server CPU leader Intel (INTC) in the CPU compute space around analyst events in late 2026; semiconductor equipment firms Applied Materials (AMAT), Lam Research (LRCX), KLA (KLAC), and Teradyne (TER); MACOM Technology Solutions (MTSI) in data center high-speed optical interconnects; and Credo Technology (CRDO) and Marvell Technology (MRVL) in AI data center high-performance networking infrastructure.
Arya's team noted that AI data center memory chips/components currently account for about 35% to 40% of cloud AI capex, which is two to three times historical levels, yet memory chip-related stocks are trading at valuation levels below their warranted growth profile. The analysts stated: "Investors remain skeptical about pricing durability, expectations for increased supply scale, and customer concentration. We believe the market underestimates the strong trend toward longer-term product supply agreements (LTAs) and more predictable pricing. As memory chips evolve from cyclical commodities to strategic AI-era enabling assets, valuation multiples should expand significantly."
Bank of America's analysts reiterated their most bullish "Buy" rating on U.S. memory leader Micron (MU), naming it a top pick with a price target of $1,550. They stated that Chinese AI models represent a trend of U.S.-China AI model competition, not a capex risk. Arya's team pointed out that China's open-source weight models, such as Z.AI's (formerly Zhipu AI) GLM, Moonshot AI's Kimi, DeepSeek, and Alibaba's (BABA) open-source model Qwen, have rapidly narrowed the gap with leading frontier AI lab models from the U.S. while offering significantly lower inference costs.
The analysts added that the robust rise of capable, low-cost models raises legitimate questions about the overall economics of future global industry expansion and AI software margins. However, they stated: "We view this as a significant positive for adoption expansion expectations. Lower-cost AI agent token expenditure trajectories expand actual usage, broaden AI application deployment, and ultimately increase robust demand for CPU/GPU compute, HBM/DRAM/NAND memory chips, networking infrastructure, data center optical interconnect systems, and power infrastructure. In our view, the greater risk lies in the model economics framework, not AI semiconductor demand."
The analysts added that investors remain overly focused on large model benchmark leadership and daily large language model (LLM) headlines. The ultimate goal of AI is not the best model but the best labor productivity growth outcome. As industries globally shift from models to AI agents, agentic workflows, and enterprise automation, value creation should increasingly shift, at scale, toward AI applications and the compute infrastructure supporting them.
Arya's team stated: "Similar to how the internet era commoditized information yet created enormous software value, open-source AI large language models can accelerate the adoption and penetration of cutting-edge AI applications like AI agents, while frontier labs continue to push the boundaries of AI technology."
From Cyclical Commodity to AI Strategic Asset
Nomura is currently among the most aggressively bullish large financial institutions on the memory chip sector. In a recent research report, Nomura raised its price target for Samsung Electronics from 340,000 won to 590,000 won and for SK Hynix from 2.34 million won to 4 million won, implying potential upside of approximately 118% and 120%, respectively.
Nomura's core bullish logic is that AI has transformed memory from a traditional PC/mobile cyclical commodity into a long-term growth asset for data centers: agentic AI inference requires massive key-value (KV) caching, and HBM supply lags significantly behind demand. It projects global data center capital expenditure to grow from $1.16 trillion last year to $6.13 trillion by 2030, with memory's share of data center investment potentially rising from the current 9% to 23%. Therefore, the approximately 6x forward 12-month P/E ratios for Samsung and SK Hynix are seen as significantly undervalued, with room for re-rating toward TSMC's roughly 20x valuation framework.
The most optimistic Wall Street price target for Micron currently comes from Gil Luria, a senior analyst at asset management firm DA Davidson, who maintains a "Buy" rating and has raised his price target substantially from $1,500 to $2,000 per share. Based on Micron's current price around $975.56, this target implies roughly 105% potential upside. If the stock reaches $2,000, the potential market capitalization would be approximately $2.29 trillion.
Luria's core bullish thesis can be summarized as follows: AI compute infrastructure expansion is lengthening this memory upcycle. Micron is no longer just a traditional, strongly cyclical DRAM company but is becoming an AI compute bottleneck asset with stronger earnings visibility and higher valuation multiples, reshaped by its position in HBM, server DRAM/NAND, and long-term supply agreements. He has previously emphasized that this memory cycle differs from the old paradigm of "capacity expansion -> oversupply -> price collapse," as AI compute build-out could extend the demand upswing indefinitely. Simultaneously, Micron's multi-year HBM/memory sales agreements enhance future revenue and EPS visibility.
The current key debate in the memory chip sector is not whether AI demand will disappear, but whether the pace of price increases is nearing a peak and whether the market is willing to assign higher valuation multiples to memory companies.
Bank of America's narrative is clear: after the Philadelphia Semiconductor Index surged 88% in Q2, its ~11% pullback in Q3 resembles a seasonal "healthy reset," not a structural deterioration in AI demand. The firm expects global cloud and AI infrastructure capex to approach $1.5 trillion by 2027, growing another 40%-50% year-over-year, supported by token growth, agent adoption, and constrained infrastructure supply. More crucially, memory chips/components now account for about 35%-40% of cloud AI capex, reaching 2-3 times historical levels, yet memory chip stock valuations overall remain below warranted levels. Therefore, Bank of America maintains its "Buy" rating and $1,550 target for Micron, with the core judgment being that memory is transitioning from a cyclical commodity to AI strategic infrastructure.
Morgan Stanley's caution does not equate to bearishness on the memory cycle but serves as a reminder to the market that the "rate of price change" may be peaking, not the "absolute price level." This distinction is important. Recent forecasts from TrendForce predict DRAM contract prices will rise 13%-18% in Q3 2026 on a strong Q2 base, with NAND Flash expected to rise 10%-15%, representing a clear deceleration from previous increases of around 60%. In other words, slowing marginal price increases may suppress short-term valuation expansion for high-beta memory stocks, but AI inference, hyperscale cloud data centers, and enterprise storage continue to maintain tight supply-demand conditions.
The notion that "massive Korean factory investments equal a memory cycle peak" also requires scrutiny. Large-scale capacity expansion is certainly a mid-to-long-term risk, especially given the memory industry's history of downturns following high景气 periods due to excessive capital expenditure. However, current new Korean clusters are more about capacity planning for the 2030s rather than an immediate supply shock for 2026-2027. SK Hynix's investment plans in Korea include the Yongin semiconductor cluster, Cheongju NAND and HBM packaging expansion, and a still-in-planning southwestern semiconductor cluster. The first fab in Yongin is expected to commence production in 2027, with the full four fabs completed by 2033, while the southwestern cluster is a longer-term plan. For the current investment cycle, what truly determines stock prices are hyperscale cloud providers' capital expenditure guidance for the next four to six quarters, the execution of HBM/server DRAM long-term agreements, enterprise SSD demand, and NAND price sustainability—not nominal capacity blueprints a decade out.
Nomura released a report this week refuting the "semiconductor peak" theory. The key to Nomura's rebuttal is not simply stating that AI chips will continue to rise, but pointing out that AI cloud infrastructure demand is spreading from a single-point GPU shortage to a systemic component mismatch. According to Nomura's research framework, AI server revenue is projected to grow 78% in 2026 and 76% in 2027. The number of global data center projects is expected to increase from 240 to 280, with about 50 being gigawatt-scale projects. New compute deployment in 2027 is projected to reach 32 GW, with 23 GW already visible for 2028. However, the real bottlenecks are now spilling over from Nvidia AI GPU and Google TPU产能, and TSMC's CoWoS advanced packaging, to memory chips, wafer-level substrates, AI PCBs, copper-clad laminate (CCL), electronic fabric, MLCCs, glass substrates/ABF substrates, IC载板, high-end capacitors, power management chips, and data center optical high-speed interconnect components.
Comments