Original report: BofA Global Research "Global Memory Tech", July 2, 2026

Compiled & organized: DaiDai, MSX Maitong

Editor: Frank, MSX

Key points at a glance:

• Bank of America believes that the recent concentrated correction in storage stocks mainly comes from risk narratives such as Meta orders, Changxin Storage entering Apple’s supply chain and South Korea’s production expansion, and does not mean that the fundamentals of the industry have reversed;
• Meta's provision of external data centers or cloud services is more likely to monetize computing power and diversify its business than to significantly reduce storage requirements. Its demand for HBM, LPDDR5 and enterprise-class SSDs is still growing;
• It is difficult for Changxin Memory to become an important DRAM supplier for Apple in the short term, and Apple is more likely to use it as a bargaining chip in price negotiations with Samsung, SK Hynix and Micron;
• Japan’s industry chain survey shows that DRAM and NAND prices are still expected to rise quarter-on-quarter in the third and fourth quarters, and the industry may continue to be in shortage in 2027, with manufacturers’ capital expenditures and wafer output remaining relatively restrained;
• The storage industry is still in a strong cycle, but as product prices and related stocks rise sharply, subsequent market trends will rely more on profit realization, and sector fluctuations and individual stock differentiation may increase significantly;

Over the past week, global storage stocks have experienced a significant correction.

The market quickly found three seemingly reasonable explanations for this decline: Meta is preparing to sell some computing power to the outside world, which may mean that there is a surplus in previous data center construction; Apple is evaluating DRAM from Changxin Memory, which may break the supply pattern of Samsung Electronics, SK Hynix and Micron; South Korea has announced a large-scale semiconductor industry cluster plan, further exacerbating concerns about future oversupply.

The three narratives ultimately point to the same conclusion: Demand may have peaked, supply is about to expand, and the storage super cycle may have come to an end.

But the judgment given by Bank of America in the latest "Global Memory Tech" report is exactly the opposite.

In his view, the above risks are not completely non-existent, but the market has clearly overestimated their impact on the short-term supply and demand pattern. Whether it is the capital expenditures of cloud manufacturers, South Korean semiconductor exports, or the spot and contract prices of DRAM and NAND, they have not yet shown that the storage cycle has reversed direction.

What has really changed is not that the fundamentals have turned from strong to weak, but that after experiencing substantial price increases and stock price revaluations, the industry has entered a new stage in which fundamentals are still strong but trading difficulty has increased significantly.

1. Take stock one by one. Are market concerns valid?

1.Meta’s takeout computing power does not mean reducing storage orders

The market's concerns about Meta come from a seemingly reasonable inference: If Meta starts to open its data center to external customers or sell cloud services, does it mean that the company has purchased too many servers before and that its internal business can no longer absorb the existing computing power?

If the answer is yes, demand for AI hardware such as GPUs, HBM, server DRAM, and enterprise SSDs is likely to decline.

However, the Bank of America report stated that judging from feedback from the industry chain, storage chip manufacturers believe that Meta will continue to more actively adopt high-performance storage products such as HBM, LPDDR5 and enterprise-grade SSD in AI data centers. Therefore, the market speculation that "Meta will rent out AI servers or cloud infrastructure that has been over-invested in the early stage" lacks sufficient basis.

Evensome NAND controller chip and packaging substrate material manufacturers said that Meta’s chip and component orders are still increasing. Therefore, Meta’s opening of its own data center to external customers is more likely to be an attempt to realize assets and business diversification, rather than being forced to deal with a serious excess of computing power.

2. Changxin Storage’s entry into Apple’s supply chain is more like a bargaining chip in price negotiation

The Bank of America report predicts that the probability of Apple’s large-scale adoption of Changxin Memory DRAM in the short term is still low.

To break it down, there are three main constraints:

• The first is policy and supply chain restrictions: Apple needs to consider the relevant US restrictions on China’s semiconductor industry and the resulting compliance and supply chain risks;
• The second is the technical specifications: Apple has high requirements for the transmission speed, power consumption and reliability of mobile DRAM, including a transmission speed of more than 10Gbps, a low-power design of about 1.1V, and ECC error correction capabilities. Whether Changxin Storage can meet these requirements stably and on a large scale in the long term still needs further verification;
• Finally, there is the risk of intellectual property rights: DRAM core patents have long been concentrated in the hands of leading manufacturers such as Samsung, SK Hynix and Micron. If Apple adopts products with insufficient patent coverage on a large scale, it may face potential litigation and supply interruption risks;

Changxin Storage can theoretically win orders for low-end iPhone 18e, but considering the size of the relevant models in the Chinese market, the actual purchase volume is expected to be limited.

Compared with truly restructuring the supply chain, Apple is more likely to use this to enhance its bargaining power in contract price negotiations in the second half of 2026 or 2027. Therefore, this incident is more likely to affect the pricing expectations of Samsung, Hynix and Micron in the short term, rather than immediately changing the global DRAM supply and demand pattern.

3. South Korea’s large-scale production expansion does not mean that short-term supply is out of control

Another recent concern comes from South Korea’s new round of semiconductor industry cluster planning.

Some investors believe that the South Korean government plans to invest approximately 800 trillion won in building a new storage fab cluster in the southwestern region, which may mean that the storage cycle is nearing the top. However, the Bank of America report takes a negative attitude towards this and predicts thatit will be difficult for this project to form a large-scale effective supply before the early 2030s, and at this stage, priority still needs to be given to the expansion of the Yongin and Pyeongtaek industrial clusters in 2026-2035.

Therefore, an industrial plan spanning more than ten years cannot be directly equivalent to the fact that supply will be out of control in the next two or three years. Long-term production expansion is worthy of continuous tracking, but it is not enough to be a direct basis for judging the peak of this storage cycle.

4. Japan’s industrial chain research is still optimistic

Bank of America’s recent industrial chain survey in Japan has also continued to strengthen its optimistic judgment on the storage industry.

Japanese investors generally recognize the current industry boom, but as product prices and related stocks rise rapidly, the market has begun to pay more attention to potential downward cycles. Compared with investors’ caution, the judgments given by industry chain management are still positive:

• Storage ASP performance was strong in the second quarter, especially NAND;
• ASP in the third and fourth quarters is expected to still be higher than that in the second quarter;
• DRAM and NAND may continue to be in shortage in 2027;
• The number of long-term supply agreements has increased, but they are still mainly based on purchase volume agreements;
• Capital expenditure and wafer production continue to remain restrained, especially by Japanese NAND manufacturers;

This means that although the market has begun to discuss the next round of supply cycle in advance, judging from the actual expansion of manufacturers and customer purchasing behavior, the industry has not yet entered an obvious stage of out-of-control supply.

5. Samsung’s storage business may still exceed expectations

In a report released on July 2, Bank of America predicted that Samsung Electronics' overall operating profit in the second quarter may be slightly lower than the market's more optimistic expectations due to special bonus payments and pressure on smartphone business profit margins; however, due to strong average selling prices of DRAM and NAND, the operating profit calculated separately for the storage business is still expected to be higher than market expectations.

Five days after the report was released, Samsung announced preliminary second-quarter results on July 7: consolidated sales of approximately 171 trillion won and operating profit of approximately 89.4 trillion won, a year-on-year increase of 129.3% and 1810.3% respectively. Among them, operating profit was higher than the market’s previous expectation of about 86 trillion won, which means that Bank of America’s judgment that the group’s overall profit may be slightly lower than optimistic expectations did not ultimately materialize.

However, what Samsung disclosed this time is still preliminary performance at the group level, and detailed profit data for storage, wafer foundry and mobile businesses have not yet been released. Therefore, whether the storage department alone has exceeded expectations still needs to wait for confirmation of the complete financial report. Combined with the rise in DRAM and NAND prices in the second quarter and the substantial growth in South Korea's semiconductor exports, the storage business is likely to remain the core force driving Samsung's current profit jump.

2. What signals are released by exports, ASP and product prices?

1. South Korea’s semiconductor exports increased significantly

In June 2026, South Korea's semiconductor exports reached US$44.8 billion, a month-on-month increase of 21% and a year-on-year increase of 199%, and has achieved triple-digit year-on-year growth for six consecutive months.

This figure is approximately three times the monthly average of US$14 billion in 2025, reflecting that current storage price increases have begun to be significantly transmitted to export revenues and corporate profits.

Of course, the increase in exports does not completely represent a simultaneous increase in shipments. A large part of the increase comes from the rapid increase in the average selling price of products. However, this also shows that the core contradiction in the current industrial chain is still rising prices and tight supply, rather than inventory backlogs or obvious shrinking demand.

South Korea's semiconductor export volume and year-on-year growth rate: Exports jumped significantly in June 2026 (page 2 of the original report)

2.DRAM is still the strongest storage category at present

TrendForce has raised its DRAM ASP forecast for the third quarter of 2026 from a quarter-on-quarter rise of 3% to 8% to 13%-18%; Bank of America predicts that DRAM ASP will rise by 53%, 17%, and 7% quarter-on-quarter respectively in the second to fourth quarters of 2026.

The specific caliber of the two sets of forecasts are different, but they all point to the same trend, that is, DRAM prices will continue to increase in the second half of the year, but as the price base increases, the month-on-month increase may slow down quarter by quarter.

As of early July 2026, the spot price of 16Gb DDR5 is about US$47 and 16Gb DDR4 is about US$75, both significantly higher than the price highs of the previous storage cycle. The core reason is not just end customers replenishing inventory, but storage manufacturers continue to shift wafer production capacity to HBM and server DRAM with higher profit margins.

After advanced production capacity is absorbed by AI-related products, the supply available for traditional DDR4 and conventional DDR5 decreases simultaneously.

Especially DDR4. As leading manufacturers gradually withdraw from mature products, there is an obvious structural shortage of DDR4. The contract prices of 16Gb DDR4 and DDR5 have both risen to the US$35-40 range, and the long-standing technology premium of DDR5 over DDR4 has basically disappeared.

This does not mean that the market prefers the older generation of DDR4, but that manufacturers are exiting faster than customers can complete product switching, causing mature products to become more scarce.

3. NAND price increases have slowed down, but absolute prices are still high

The marginal change in NAND prices is more pronounced than that of DRAM.

After reaching a stage high in March 2026, the spot price of 512Gb NAND wafers gradually stabilized or fell slightly from April to June, but it still rose by more than 50% during the year, which is about eight times the low in February 2025.

The NAND contract price is about $25, about ten times the low of $2.50 in February 2025. After experiencing sharp increases in the fourth quarter of 2025 and the first quarter of 2026, the single-month increase in NAND contract prices from April to June has fallen back to about 1%-5%.

This does not mean that NAND prices have reversed, but it means that customers' ability to bear high prices is gradually approaching the limit, and the price increase rate is returning to normal.

The changes in client SSD prices are particularly intuitive. As of June 2026, the price of 512GB client SSD has risen from US$73.1 at the end of 2025 to US$137.5, nearly doubling, reflecting that upstream NAND price increases are continuing to be transmitted to end products.

Thus, a more accurate current state of NAND is that absolute prices remain high, but sequential increases are slowing.

4. Server memory continues to hit highs

Server memory also continues to be strong.

64GB server DRAM module prices are at an all-time high, with DDR5 around $1,400 and DDR4 around $1,100. In June 2026, DDR5 server DRAM contract prices rose again, while DDR4 prices remained basically flat.

This shows that even if the price increase of some consumer-grade storage products begins to slow down, demand for high-end storage related to AI servers and data centers remains strong.

3. Cloud capital expenditure is still the demand anchor, but the investment logic is changing

1. Hyperscale cloud vendors are still expanding

Hyper-scale cloud vendors such as Amazon, Microsoft, Alphabet and Meta are becoming the most important sources of new storage demand. The report predicts that the total capital expenditures of the four companies in 2026 will be approximately US$700 billion, a year-on-year increase of approximately 80%; from 2027 to 2028, the annual capital expenditures may further approach US$1 trillion.

And Bank of Americahas not seen significant signs of major cloud vendors cutting capital expenditures in 2027, which means that these investments will ultimately translate into more AI accelerators and HBM, more server DRAM, more enterprise-class SSDs, more data centers and AI inference infrastructure.

Capital expenditure, revenue and gross profit margin trends of major U.S. hyperscale cloud vendors (page 3 of the original report)

The report predicts that the overall revenue of the four major technology companies is expected to grow by 15% to 20% from 2026 to 2028, and the year-on-year growth rate of cloud business revenue may reach 35% to 40%.

Among them, AWS operating profit margin is expected to remain above 35%, Azure may exceed 40%, and Google Cloud is expected to reach 30%-35%.

As long as the cloud business can maintain high revenue growth and profit margins, technology giants will still have the business motivation to continue to expand investment in AI infrastructure.

Cloud revenue and cloud business operating profit margin trends (page 3 of the original report)

2. This cycle is no longer just about consumer electronics restocking

The biggest difference between this storage cycle and the past is that demand no longer relies mainly on smartphones and personal computers to replenish inventory.

In the past, storage cycles were often driven mainly by changes in the inventory of personal computers and smartphones, forming typical cycle characteristics, such as rising terminal demand, customers replenishing inventory, and rising storage prices; then manufacturers expanded production capacity, inventory gradually accumulated, and prices entered a downward cycle again.

However, the structure of this round of storage cycle is more complex. The current demand has already expanded from single consumer electronics replenishment to:

• HBM;
• Server DRAM;
• Enterprise-grade SSD;
• AI inference infrastructure;
• Capital expenditures of hyperscale cloud vendors;
• Structural shortage caused by the withdrawal of DDR4 production capacity.

This means that just observing PC and smartphone sales is no longer enough to judge the entire storage cycle - even if some consumer electronics demand is under pressure due to high prices, AI servers and data centers may continue to absorb high-end production capacity, keeping overall supply tight.

But this also means that the differentiation within the plate will become increasingly obvious. In short,HBM, server DRAM, enterprise SSD and advanced packaging related companies may continue to benefit from stronger orders and profit margins; manufacturers that are overly dependent on clients, mobile phones and consumer NAND may feel the decline in demand elasticity earlier.

3. From general industry gains to profit realization

NAND, HDD, and DRAM-related stocks have generally experienced significant gains since 2026.

Charts from the original report show that SanDisk and Kioxia once rose more than 800% during the year, and DRAM manufacturers and some semiconductor companies also saw significant gains.

In this case, even if there is no directional change in industry fundamentals, any news about customer orders, capital expenditures, new supply, or price negotiations may trigger violent fluctuations.

Therefore, Bank of America’s judgment on the storage industry can be summarized as that the fundamentals are still bullish, but the stock price is no longer at a stage where risks can be ignored. Overall, the future performance of the sector will be more dependent on three factors:

• First, whether storage prices can be maintained at a level sufficient to drive earnings upward;
• Second, whether the company’s actual profit growth rate can absorb the previous valuation expansion;
• Third, whether new capital expenditures will continue to be restrained to prevent long-term supply expectations from deteriorating too quickly;

This is whyThe correction in the past week does not necessarily mean that the storage super cycle is over, but it may mean that the sector is shifting from "general industry gains" to "profit verification" and "individual stock screening."

Write at the end

Objectively speaking, Meta's external provision of cloud services, Changxin Storage's entry into Apple's supply chain, and South Korea's launch of a large-scale production expansion plan are not risks that can be completely ignored.

But at least judging from the data presented in this Bank of America report, they have not changed the three most important facts about the storage industry:

• Capital expenditures of hyperscale cloud vendors are still growing;
• DRAM and NAND prices remain at historically high levels;
• It will still take a long time for new advanced production capacity to form effective supply;

So rather than saying the pullback over the past week represents a "top of the cycle," the market is trying to reprice for the next phase.

This round of storage cycle has not yet seen a clear fundamental turning point, but after high prices, high expectations and high growth rates appear at the same time, the questions that investors need to answer next are not limited to whether the storage industry will continue to rise, but expanded toWhich products are still in short supply, which companies can realize profits, and which stocks have overdrawn the future in advance?

The fundamentals of the industry are still strong, but the stage of indiscriminate revaluation of the entire sector may be gradually passing.