"AI Bottleneck Trade" Ignites Upstream Sector—Who’s Raking in the Profits?

PANews

joined discussion ·

From HBM to Cold Storage: The Six Storage Giants Reach a Tipping Point—What Will Drive the Next Market Cycle?

Written by DaiDai, MSX MaiTong

Editor: Frank, MSX MaiTong

The current memory cycle is not an ordinary price cycle; rather, AI infrastructure demand is expanding from HBM to server DRAM, enterprise SSDs, NAND, and HDDs,and the market’s focus has shifted from 'faster storage' to 'more, cheaper storage that requires long-term retention';

Six companies represent distinct investment theses:SK Hynix (SKHY.M) trades on its leading HBM position and order visibility; Samsung reflects expectations of technological catch-up and market share recovery; Micron serves as the most direct U.S.-listed proxy for HBM and DRAM exposure; SanDisk is most sensitive to NAND and enterprise SSD price increases; Seagate and Western Digital benefit from demand for high-capacity HDDs, data lakes, and cold storage;

Since the start of 2025, all six companies have undergone significant re-rating, but as of mid-July 2026, they have generally retraced approximately 20%–30% from their recent peaks,The sector is now transitioning from a phase of 'demand surge and earnings recovery' into one characterized by high expectations, elevated valuations, and heightened volatility.

Historical performance shows that catalysts capable of driving sustained rallies are typically not product launches or sample shipments, but ratherearnings beats, certification by key customers, volume deliveries, secured orders for the following year, and upward revisions to guidance on revenue, pricing, and profit margins;

the core of the next rally will hinge onwhich companies can convert demand into long-term contracts, achieve higher product pricing, execute controlled capacity expansion, and deliver consistently growing profits and free cash flow;

In June 2026, Micron Technology (MU.O) released earnings results with virtually no discernible weaknesses.

The company reported quarterly revenue of $41.456 billion, non-GAAP gross margin rose to 84.9%, adjusted free cash flow exceeded $18.3 billion, and guidance for next quarter’s revenue was further raised to approximately $50 billion,in other words, whether it’s product pricing, profit margins, or order visibility, all indicators point to the same conclusion: AI-driven memory demand remains robust.

Yet the market’s reaction was not as straightforward as in the past.

Micron initially surged after its earnings release but quickly gave back those gains. SanDisk, SK Hynix, Samsung Electronics, Western Digital, and Seagate also retreated from their June highs. On a consistent total-return basis, as of July 22, all six companies had generally pulled back about 20%–30% from their recent peaks, with some individual stocks down more than 30%.

Paradoxically, fundamentals did not suddenly weaken, nor did AI data centers halt their expansion. Investors simply stopped questioning whether AI would drive greater memory demand and instead began asking how much longer product prices could keep rising—and whether new capacity additions in 2027–2028 might once again shift the supply-demand balance.

The memory storage sector has thus reached a new inflection point.

The memory storage rally since 2025 was initially ignited by HBM.

Large-scale model training requires GPUs, and large GPU clusters in turn demand high-bandwidth, low-latency memory systems. HBM, by vertically stacking multiple layers of DRAM, significantly boosts data throughput and has rapidly become one of the most critical, highest-margin, and tightest-supply components in AI accelerators.

SK Hynix (SKHY.M), leveraging its lead in HBM3E products, strong customer relationships, and mass production advantages, was the first to achieve a re-rating of its earnings baseline.

In the third quarter of 2025, the company stated that discussions for next year's HBM supply were largely finalized, HBM4 would begin shipping in the fourth quarter, and demand coverage for next year’s DRAM and NAND output was already substantially secured. By the first quarter of 2026, the company’s revenue, operating profit, and net income continued to hit record highs.

Micron (MU.M) has become the most liquid U.S.-listed proxy for trading the supply-demand dynamics of HBM and server DRAM—particularly as HBM4 enters volume shipments, server memory prices rise, and multiple long-term customer agreements collectively drive higher revenue, gross margins, and cash flow.

Therefore,For U.S. equity investors, Micron’s earnings reports are no longer just about the company itself; thus, when Micron raises its guidance on revenue, pricing, and margins, the market typically reassesses the entire memory storage industry’s profit potential.

Samsung Electronics, meanwhile, has taken a different path.

In 2025, the market primarily priced in Samsung’s lag behind SK Hynix in terms of HBM product progress, customer certifications, and yield rates. Entering 2026, as HBM4 began commercial shipments, HBM4E entered the sampling phase, and profitability in the memory business improved significantly, Samsung’s valuation narrative shifted from 'lagging' to 'catching up.'

The company’s preliminary results for the second quarter of 2026 showed quarterly sales of approximately KRW 171 trillion and operating profit of about KRW 89.4 trillion. Of course, this figure includes businesses such as smartphones, foundry, displays, and consumer electronics, and thus cannot be directly interpreted as standalone performance from the memory division.

In fact, what truly drove the sustained broadening of this rally was not just HBM itself.

As manufacturers allocated more wafers, capital expenditures, and advanced packaging resources toward high-value AI products while maintaining relatively disciplined supply discipline, the supply-demand dynamics for server DDR5, standard DRAM, and NAND also improved.

Meanwhile, as AI moved beyond training into large-scale inference, memory demand began spreading from GPU-adjacent components to the entire data infrastructure.

Model weights, vector databases, key-value caches, inference contexts, and frequently accessed data are driving increased capacity and performance demands for enterprise SSDs; meanwhile, training data, video, multimodal content, inference logs, historical model versions, and compliance archives require lower-cost, scalable high-capacity storage solutions.

SanDisk (SNDK.M) emerged during this phase as the highest-beta stock in the sector.

Following its spin-off from Western Digital (WDC.M) in February 2025, SanDisk transformed from being part of a hybrid HDD-and-flash company into a pure-play NAND and SSD play on U.S. equity markets. In its latest fiscal quarter, the company reported a 233% sequential increase in data center revenue, signed multiple new commercial agreements with financial safeguards, and announced a share repurchase program of up to USD 6 billion after debt repayment.

As AI data continues to accumulate, demand is further expanding from high-speed memory into high-capacity storage.

Since model training data, video, multimodal data, inference logs, historical versions, and compliance archives do not all need to reside permanently in high-cost SSDs, a significant volume of less-frequently accessed—but still long-term retained—data ultimately flows into nearline HDDs and tiered storage systems.

Western Digital reported a non-GAAP gross margin of 50.5% in its latest fiscal quarter, generated $978 million in free cash flow, and expects mid-to-high teens or higher year-over-year growth in exabyte shipments of nearline HDDs over the next three to five years.

Seagate (STX.M), meanwhile, has gained a clearer technological premium through the commercialization of HAMR. In its latest quarter, the company posted a non-GAAP gross margin of 47% and $953 million in free cash flow. Its Mozaic 4+ products, with capacities up to 44TB, have already begun volume shipments to two leading hyperscale customers.

Therefore, the current storage market rally cannot be simplistically characterized as merely a price increase; rather, it resembles a layered expansion of storage demand radii driven by the deepening of AI infrastructure build-out, wherein HBM addresses bandwidth constraints around GPUs, server DRAM handles computational workloads, enterprise SSDs support high-frequency data access, and HDDs accommodate ever-expanding data lakes and long-term archival needs.

To some extent, the market initially priced in the idea that 'AI requires faster storage,' and later began pricing in that 'AI also requires more storage.' This industry logic—extending from HBM to server DRAM, NAND, enterprise SSDs, and HDDs—is fully reflected in the share price performance of these six companies.

Based on a consistent total-return-adjusted basis, from the beginning of 2025 or the first trading day following their respective spin-offs until July 22, 2026, all six companies underwent significant revaluation, though the magnitude of gains and timing of rallies varied:

SK Hynix and Micron were the earliest beneficiaries of the upcycle in HBM and server DRAM;

Samsung subsequently began pricing in its HBM technology catch-up and market share recovery; SanDisk, following its spin-off, emerged as the most responsive stock in this cycle due to its pure-play exposure to NAND and enterprise SSDs;

Western Digital and Seagate then took the lead in the first half of 2026, as the market started incorporating nearline HDDs, high-capacity products, and cold data storage demand into the valuation framework for AI infrastructure;

If we segment the past 18 months of market action, it can roughly be divided into four phases:

In the first half of 2025, the market initially traded HBM.SK Hynix and Micron became the most direct beneficiaries, as investors began to recognize that HBM was not just a short-term inventory restocking cycle but a long-term, high-margin product driven by the expansion of AI accelerators;

In the second half of 2025, the rally spread to conventional DRAM and NAND.HBM’s heavy demand for advanced wafer, packaging, and testing capacity tightened supply for server DRAM and other memory products, prompting the market to revise upward earnings expectations for companies like Samsung and SanDisk.

In the first half of 2026, enterprise SSDs and HDDs took over the momentum.As AI shifted from training to inference, storage demand expanded beyond high-speed memory near GPUs, with requirements for training data, inference logs, and long-term archiving driving a repricing of HDDs and cold storage.

By June–July 2026, the market entered a phase of heightened expectations and elevated volatility.The peak share prices of the six companies were almost all concentrated in June; thereafter, even though companies like Micron continued to deliver strong earnings reports, the sector failed to sustain its prior one-way rally;

This pullback indicates that the market’s pricing logic is shifting, as the earlier stock gains were primarily underpinned by three reinforcing consensus views:AI-driven memory demand will continue to expand, industry supply will remain tight, and product pricing and margins still have room to rise.

However, after a significant repricing, investors have started to consider several additional issues simultaneously—such as whether current valuations have already priced in future growth, whether positioning has become overly crowded, whether new capacity scheduled for 2027–2028 will gradually come online, and how much of the strong demand will ultimately translate into long-term orders and free cash flow.

Thus, this collective pullback appears more like a marker of the memory sector’s transition from phase one to phase two, with future performance increasingly dependent on which companies can offer greater order visibility, stronger pricing power, and more sustainable profit growth.

If judged solely by share prices, all six companies appear to be part of the same AI-driven memory rally.

Yet, in terms of industry position, earnings elasticity, and next-phase catalysts, they actually represent six distinct pricing logics.

SK Hynix remains the company with the strongest industry position and highest order certainty among the six.

Its core advantage lies not only in its leading market share in HBM but also in its ability to lock in customers, pricing, and next year’s capacity earlier than peers. Compared with typical product launches, this advance order coverage enhances visibility into future revenue and margins and makes sustained rallies more likely.

However, following SK Hynix’s Nasdaq listing in July 2026 under the ticker SKHY—priced at USD 149 and closing around USD 168 on its first trading day—the stock introduced a new layer of trading dynamics. Due to limited initial float in the U.S. market, the ADR briefly traded at a noticeable premium to its Korean ordinary shares. Naturally, as conversion and arbitrage mechanisms gradually open up, this premium is likely to narrow.

It is therefore necessary to analyze the company’s fundamentals separately from ADR supply-demand dynamics.

The Korean ordinary shares primarily reflect HBM orders, pricing, and margins, whereas SKHY is also influenced by U.S. investor access, float size, ETF allocations, and arbitrage mechanisms. This means that even if operational performance remains robust, the ADR could experience price movements independent of fundamentals due to increased supply.

Micron is the most direct U.S. equity proxy and also the easiest target for front-running trades.

Its strength lies in its exposure across HBM, server DRAM, NAND, and enterprise SSDs, along with support from U.S. domestic manufacturing and supply chain policies. When Micron raises product prices, gross margins, and earnings guidance, the market often revises upward profitability expectations for the entire memory sector. Its earnings report serves not only as a catalyst for the company itself but also as a key pricing signal for the global memory industry.

However, its drawback is equally evident: market expectations for Micron adjust extremely quickly.

Put simply, when valuations are low and industry sentiment is weak, a better-than-expected earnings report can drive sustained re-rating. But once the stock enters a high-expectation phase, merely delivering strong results is no longer sufficient—the share price demands pricing above even the most optimistic forecasts, longer order visibility, and significantly higher margin revisions.

This also explains why Micron continues to exhibit significant single-day price swings after earnings releases, yet the sustainability of its post-earnings rally has started to weaken.

Samsung Electronics has lagged in terms of price appreciation but offers the clearest upside potential from market share recovery.

It is not the purest play on fundamentals in this cycle—its foundry, smartphone, consumer electronics, and display businesses dilute the impact of memory profit improvements on overall group earnings. Additionally, its slower progress in HBM customer certifications earlier in the cycle caused its stock to underperform SK Hynix and Micron.

Conversely, Samsung also holds the most clearly defined 'expectation gap' among the six companies. The market already recognizes SK Hynix as the HBM leader but remains uncertain about how much high-end HBM business Samsung will ultimately secure.

Therefore, Samsung’s next meaningful catalyst hinges on whether core customer certifications are completed, whether its products enter volume procurement, whether HBM revenue contribution rises, and whether both market share and memory segment profitability improve in tandem. Once these indicators begin to materialize, Samsung’s investment thesis will evolve from 'technology catch-up' to 'market share recovery.'

SanDisk, meanwhile, is gradually shifting from trading on NAND price beta to trading on contract visibility.

It is the company with the highest stock price elasticity in this market cycle, and also the one facing the greatest risk when expectations retreat. On one hand, its NAND and enterprise SSD businesses are highly pure-play, making them extremely sensitive to product pricing and industry inventory fluctuations; on the other hand, its newly independent valuation framework post-spin-off, growing data center revenue, and long-term business agreements further amplify its earnings upside potential.

However, SanDisk is undergoing a notable shift. Historically, the market viewed it primarily as a high-beta proxy for NAND price increases. Recently, with the rise in agreements featuring minimum purchase commitments, financial guarantees, or long-term collaboration mechanisms, investors have begun assigning it a higher valuation based on improved visibility into its revenue and cash flow.

The sampling of BiCS10 marks an important product milestone, but it may not directly translate into sustained market momentum. What truly matters is whether the product can pass enterprise customer certifications, enter volume procurement, and ultimately be reflected in average selling prices, data center revenue, and gross margins.

After all, for high-valuation cyclical stocks, technological advancement is merely the starting point—order fulfillment is the finish line.

Western Digital and Seagate both benefit from HDDs, but their re-rating paths differ, driven by distinct underlying rationales.

Western Digital’s re-rating leans more toward its financial structure. Following the spin-off of its flash memory business, WDC has become a more focused pure-play HDD company. A higher mix of high-capacity products, improved per-terabyte pricing, disciplined industry supply, and operating leverage collectively drive gross margins and free cash flow upward.

As a result, WDC’s share price sometimes doesn’t fully react on the first day after earnings. The market needs confirmation that margin improvements stem not from inventory shifts, foreign exchange effects, or one-off factors, but from sustainable changes driven by long-term orders, product mix, and industry supply discipline.

Seagate’s re-rating, by contrast, is more tied to technology and capacity value. HAMR enables higher per-drive capacity without significantly increasing the physical number of drives, lowering cost per terabyte and enhancing the sales value per drive. As 40TB+ products enter volume shipments, the market is no longer just trading HDD shipment growth, but rather next-generation product penetration, scarce capacity, and long-term profitability.

This is also why Seagate’s earnings catalysts often exhibit stronger persistence—investors revise not only current EPS upward, but also the capacity upgrade trajectory and earnings baseline for the coming years.

Overall, over the past year and a half, the valuation reset of memory companies has primarily accomplished two things:The market has confirmed that AI-driven memory demand is not a one-time inventory restocking event and has started to acknowledge that memory companies’ earnings power may be structurally higher than in traditional cycles.

However, with all six companies having experienced significant share price gains, the bar for the next phase has clearly risen.

First, demand must translate into binding long-term orders.

The market will increasingly focus on contract value, minimum purchase commitments, execution timelines, pricing adjustment mechanisms, and customer prepayments. Merely announcing strategic collaborations, joint R&D initiatives, or long-term partnerships will no longer suffice to justify upward revisions to future earnings forecasts. Only contracts that lock in volume, pricing, and cash flows can genuinely dampen cyclicality and warrant higher valuation multiples.

On this front, SK Hynix benefits most from its long-term HBM supply arrangements; Micron can enhance revenue visibility through strategic customer agreements; and SanDisk needs to demonstrate that its new business collaboration model can mitigate the impact of NAND spot-market cyclicality on profitability.

Second, product pricing must continue to flow through to margins and free cash flow.

A common pitfall in the memory industry is focusing solely on quoted prices without examining profitability. Whether rising prices for HBM, DRAM, NAND, and HDD ultimately translate into higher gross margins depends on product mix, yield rates, depreciation, capital expenditures, and customer contract terms.

For SK Hynix, Micron, and Samsung, the key metrics are HBM4 average selling price, yield rate, and revenue contribution; for SanDisk, they are NAND contract pricing, enterprise SSD mix, and inventory levels; and for Western Digital and Seagate, they are per-terabyte pricing, penetration of high-capacity products, and free cash flow generation.

Ultimately, the market doesn’t buy stories about price increases—it buys upward revisions to earnings forecasts.

Third, capacity expansion must not outpace the realization of demand.

HBM, advanced DRAM, NAND, and HDD are all currently benefiting from supply discipline, but high profitability will likewise stimulate capital expenditure.

If new wafer fabs, packaging capacity, and high-capacity HDD production lines are all ramped up simultaneously between 2027 and 2028, the market will reassess how long supply tightness can last. Consequently, capital expenditure plans, equipment installation progress, capacity ramp-up speed, and the duration of customer order coverage will gradually become more important indicators than quarterly revenue alone.

For Samsung, the market is focused on whether new capacity additions come with market share gains; for SK Hynix and Micron, the key question is whether their capacity expansions still lag behind HBM demand; for Seagate and Western Digital, the focus is on whether high-capacity HDD capacity continues to be pre-booked by customers.

Notably,Late July to mid-August will also serve as a new concentrated verification window for the memory sector:Seagate will report earnings on July 28, and SK Hynix will release its quarterly results on July 29; SanDisk and Western Digital both plan to announce earnings on August 5, with SanDisk also holding an investor day on August 13.

At that time, the market will inevitably need answers to critical questions: How far have 2027 orders progressed? By how much can product prices still rise? When will new capacity come online? Can high profitability consistently translate into free cash flow? And how do management teams plan to allocate that cash—through buybacks, dividends, or further capacity expansion?

From HBM to server DRAM, and from enterprise SSDs to HDDs, this cycle has demonstrated that AI’s impact on the memory industry extends far beyond just a high-bandwidth memory module adjacent to the GPU.

The larger the models, the more frequent the inference, and the more data generated, the greater the new demand across the entire storage hierarchy.

However, clear industry demand does not automatically translate into assured stock returns.

While the market is still skeptical about AI storage demand, investors are buying exposure; once that demand becomes consensus, the market buys realization—Only those who can secure orders in advance, raise product prices, control new capacity additions, and truly convert revenue into profit and free cash flow will continue to command a premium.

141K Views