After Two-Day Bear Market, Capital Returns to Storage First

• After U.S. chip stocks significantly de-risked on June 5, Micron rebounded nearly 10% on June 8; on June 9, the Korean market followed suit with notable gains in SK Hynix and Samsung Electronics.
• Combined with earnings reports, rising DRAM/NAND prices, and South Korea’s semiconductor export data, storage is currently more likely to be revalued by the market based on upwardly revised EPS.
• Underlying Assets: Micron, SK Hynix, Samsung Electronics, Western Digital, SanDisk, NVIDIA, Broadcom, Marvell, Coherent, Credo, SOXX Semiconductor ETF, SMH Semiconductor ETF.

After the semiconductor crash on June 5, market attention quickly shifted from "why it fell" to another question: who will rebound first after the decline.

The responses were uneven. According to Reuters, the market value of U.S.-listed semiconductor stocks once evaporated by over $1 trillion, with the Philadelphia Semiconductor Index falling nearly 8.5% intraday. Individual stocks declined as well: Micron dropped about 13.25%, NVIDIA fell about 6.2%, AMD declined about 10.86%, and Broadcom dropped about 7.92%. However, by June 8, Micron rebounded nearly 10%; on June 9, SK Hynix and Samsung Electronics in the Korean market also strengthened simultaneously.

Capital has not left the AI semiconductor sector but is being reallocated within it. As valuations begin to be tested, market focus has shifted from “who owns the AI story” to “who can most quickly turn AI demand into profits.” Compared to certain AI hardware segments still trading on expectations of future product cycles, customer onboarding, and capital expenditure expansion, demand growth in memory is already more directly reflected in orders, pricing, and earnings reports.

This is also why Storage was the first to receive capital inflows. The market isn't just buying Storage itself, but the more verifiable EPS growth logic behind it.

One of the triggers for this risk-off move was the earnings surprise following Broadcom's report.

In absolute terms, Broadcom's fundamentals are not weak. According to the company’s announcement, revenue for FY2026 Q2 was $22.2 billion, representing a 48% year-over-year growth. The company expects total revenue for FY2026 Q3 to reach approximately $29.4 billion, with AI semiconductor revenue projected at $16 billion, an increase of over 200% year-over-year.

But the market chose to sell. The reason is not that demand for AI has suddenly disappeared, but that AI semiconductor assets have accumulated high expectations over the past year and a half. When a company with strong fundamentals still faces selling pressure due to AI revenue guidance falling short of some expectations, it indicates that the market’s pricing threshold has shifted. Simply being part of the AI supply chain is no longer enough—growth trajectory, profit realization, and next-quarter guidance must all align with valuation.

This is what the crash on June 5 meant. It was not a test of demand collapse, but a stress test of high-expectation trading.

Previously, the main theme in AI semiconductors was more about “who is closest to AI CAPEX (capital expenditures).” GPU, ASIC (custom chips), high-speed optical modules, copper interconnects, and equipment materials—all of which could be integrated into the AI cluster expansion chain—received valuation premiums. But as the market begins to worry about crowded trades, overvaluation, and the pace of guidance realization, the question has shifted from “who has an AI story” to “who can most quickly turn AI demand into earnings.”

For the stock market, what ultimately determines valuation is not the order itself, but whether the order can be converted into earnings per share (EPS). Over the long term, stock prices essentially reflect the pricing of a company’s profitability. When the market begins to focus on next quarter’s earnings rather than stories three years out, changes in EPS often matter more than the narrative itself.

Broadcom's role also carries symbolic significance, as it is one of the core assets in the AI ASIC and networking chip supply chain. Precisely because of its strength, the stock's reaction after the earnings report indicates that the AI semiconductor chain is now being held to higher validation standards.

The advantage of storage is that the EPS transmission chain is shorter.

The demand for AI servers first alters the supply and demand dynamics of high-value-added products such as HBM (High Bandwidth Memory), server DRAM, and eSSD (enterprise-grade solid-state drives). Cloud providers and AI system manufacturers require more computing power, which in turn drives demand for greater GPU memory, higher-capacity server memory, and larger-scale data center storage.

After storage manufacturers shift production capacity toward HBM and high-end server products, supply of traditional DRAM and NAND will be further constrained, leading to higher contract prices. This chain of effects does not rely solely on distant future expectations but will quickly impact revenue, gross margin, and EPS.

Micron’s earnings report has already reflected this shift. According to the company’s announcement, Q2 of FY2026 set records in revenue, gross margin, EPS, and free cash flow, with data center-related revenue growing significantly year-over-year. The company also forecasts continued significant growth in Q3 of FY2026. For Micron, AI storage is no longer a distant vision—it has become a current revenue source reflected in this quarter’s financial results.

SK Hynix's financial report is more straightforward. According to the company’s announcement, first-quarter 2026 revenue reached KRW 52.5763 trillion, with an operating profit of KRW 37.6103 trillion, resulting in an operating profit margin of 72%. The company attributes this growth to high-value-added products such as HBM, high-capacity server DRAM modules, and eSSDs. For investors, this margin reflects the combined impact of product mix, supply-demand imbalances, and pricing power reflected in the financials.

Industry price data also supports this same logic. TrendForce expects conventional DRAM contract prices to rise 58% to 63% quarter-over-quarter in 2Q26, and NAND Flash contract prices to increase 70% to 75% quarter-over-quarter. Its report also shows that DRAM industry revenue grew 81% quarter-over-quarter in 1Q26.

Price does not equal profit, but during phases of supply constraints, product mix upgrades, and strong demand, price increases improve market modeling of EPS over the next several quarters. South Korean export data also provides industry-level leading validation. According to Reuters and South Korean media, South Korea’s exports in May 2026 reached a record high, with semiconductor exports surging 169.4% year-over-year to approximately $37.16 billion, marking the first time chips accounted for over 40% of total exports.

This cannot be directly equated to the earnings per share of SK Hynix or Samsung Electronics, but it indicates that the memory boom is already reflected in accelerated export revenues at the national level.

In this round of revaluation, the difference between storage and other AI semiconductor sectors isn't whether there is growth, but how that growth is validated.

NVIDIA remains the central valve for AI demand. GPU platform iterations determine AI server architecture, HBM capacity requirements, and supply chain qualifications. However, the market is already highly familiar with NVIDIA’s growth and profitability, and its valuation has long been concentrated on its strongest AI assets. In the short term, it is more susceptible to export controls, supply chain constraints, the pace of platform transitions, and expectations gaps.

The ASIC sector also has solid underlying logic: cloud providers developing their own chips, custom accelerators, and rising demand for AI inference are all expanding the long-term potential for companies like Broadcom and Marvell. However, ASICs resemble project-based businesses—customer concentration, the timing of project adoption, mass production windows, and transitions to next-generation platforms all affect market perceptions of revenue visibility.

Optical modules and copper connections also have EPS realization pathways. Companies such as Coherent and Credo benefit from increased intra-AI-cluster bandwidth demands, with rising demand driven by 1.6T and 3.2T optical modules and evolving cluster interconnection architectures. However, pricing in these areas depends heavily on future architecture roadmaps, customer certifications, shipment timing, and capital expenditure cycles. When the market is willing to pay a premium, their earnings elasticity is strong; but when the market begins demanding validation, they are more susceptible to questions about when orders will hit revenue.

In contrast, the current pricing basis for storage is more straightforward. Demand for HBM drives higher-end products, capacity shifts reduce supply of traditional DRAM/NAND, rising contract prices improve revenue, a higher product mix boosts gross margin, and ultimately lifts EPS.

This chain does not imply zero risk, but it is easier to verify in the next quarter’s earnings report than the notion that “some future-generation architecture will generate massive orders.” This is what is meant by storage being easier to model—it does not mean storage is more important than GPUs, ASICs, or optical modules, but rather that, after this AI semiconductor de-risking, the market now favors assets that can be jointly validated by pricing, orders, profit margins, and export data.

A one- or two-day rebound does not prove that AI semiconductor trading has fully shifted from PE expansion to EPS validation.

Micron's stock dropped nearly 13% on June 5 and rebounded close to 10% on June 8, possibly driven by technical corrections, short covering, and improved risk appetite. SK Hynix's rally was also catalyzed by news related to its data center collaboration with NVIDIA. In short-term market movements, news, positioning, and fundamentals often overlap; not all of the price gains can be attributed solely to EPS certainty.

Storage remains a cyclical industry. Rapid price increases in DRAM and NAND may improve supplier profits but could also stimulate supply expansion or dampen purchasing intent among some end customers. HBM annual contracts, yield ramp-up, customer qualification, and allocation shares are still evolving, so it cannot be assumed that all price increases will fully flow through to the income statement without impact.

SK Hynix and Micron are already highly watched AI storage stocks, but their stock price volatility and fundamental flexibility do not always move in sync. If the pace of DRAM/NAND price increases slows, HBM market share falls short of expectations, or customer repeat orders are proven false, the rationale for upward EPS revisions will also come under pressure.

Similarly, one cannot inversely negate ASICs, optical modules, copper interconnects, and device materials. If these areas deliver stronger orders, clearer customer adoption, or upside guidance, the market may still reassign a valuation premium. AI semiconductors are not limited to memory alone; however, at this stage, memory is easier to justify through financial statements as to why it should be bought back.

A more prudent assessment of this market cycle is that the sharp decline on June 5 raised the validation threshold for AI assets. The recovery from June 8 to 9 indicates that capital is favoring segments within the AI value chain with shorter paths to EPS realization. Storage is uniquely positioned where orders, pricing, capacity, and profit margins are all clearly visible.