TL;DR
After the semiconductor sell-off on June 5, the market's attention quickly shifted from "why did it drop" to another question: after the drop, who will rise first.
The answer was not uniform. According to Reuters, the market value of U.S.-listed chip stocks once evaporated by over $1 trillion, with the Philadelphia Semiconductor Index plunging nearly 8.5% during trading. At the individual stock level, Micron fell by about 13.25%, NVIDIA by about 6.2%, AMD by about 10.86%, and Broadcom by about 7.92%. However, by June 8, Micron quickly rebounded by nearly 10%; on June 9, in the Korean market, SK Hynix and Samsung Electronics also strengthened simultaneously.
Funds did not leave the AI semiconductor sector but rather underwent internal reshuffling within the sector. As valuations came under scrutiny, market attention also shifted from "who has the AI narrative" to "who can most quickly translate AI demand into profit." Compared to some AI hardware segments still trading on future product cycles, customer adoption, and capex expansion expectations, the growth in storage demand has more directly manifested in orders, prices, and earnings reports.
This is also why storage was the first to attract fund inflows. What the market bought back was not just storage itself but the more easily validated EPS growth narrative behind it.
The Sell-Off Indicates a Repricing of High Expectation Trades
One of the triggers for this risk-off move was a guidance miss after Broadcom's earnings report.
In absolute numbers, Broadcom's fundamentals are not weak. According to the company's announcement, FY2026 Q2 revenue was $22.2 billion, a 48% year-over-year increase. The company expects total revenue of approximately $29.4 billion for FY2026 Q3 and anticipates AI semiconductor revenue of $16 billion, representing over a 200% year-over-year growth.
But the market chose to sell. The reason was not a sudden disappearance of AI demand, but the fact that AI semiconductor assets had already accumulated high expectations over the past year. Even a fundamentally strong company experienced selling pressure because its AI revenue guidance was below some expectations, indicating that the market's pricing threshold had shifted. Simply belonging to the AI supply chain was no longer sufficient; the growth rate, profit realization, and next-quarter guidance all needed to align with the valuation.
This is the significance of the sharp drop on June 5th. It was not a demand collapse stress test but a pressure test for high-expectation trading.
In the past, the main theme of AI semiconductors was more like "who is closer to AI CAPEX." Whether it was GPUs, ASICs, high-speed optical modules, copper interconnects, or equipment materials that could be incorporated into the AI cluster expansion chain, valuations would receive a premium. However, when the market began to worry about crowded trades, overvaluation, and the pace of guidance realization, the issue shifted from "who has an AI story" to "who can most quickly turn AI demand into financial reports."
For the stock market, what ultimately determines valuation is not the orders themselves but whether the orders can be converted into earnings per share (EPS). Because, in the long run, stock prices are fundamentally a pricing of a company's profit-making ability. When the market starts to focus on next-quarter profits rather than a story three years down the line, changes in EPS often become more critical than the narrative itself.
Thus, Broadcom's role also carries a signal. It is one of the core assets in the AI ASIC and networking chipchains. Precisely because it is strong, the stock price reaction after the financial report indicates that the AI semiconductor chain is undergoing a higher validation standard.
Why Storage: Price and Profit Already in the Model
The advantage of storage is that the EPS propagation chain is shorter.
What AI server demand initially changes is the supply-demand relationship of high-value-added products such as HBM, server DRAM, and enterprise SSDs. Cloud providers and AI system manufacturers need more computational power, which in turn requires more GPU-accompanying memory, higher-capacity server memory, and larger-scale data center storage.
After storage manufacturers shift their capacity to HBM and high-end server products, the supply of traditional DRAM and NAND will also be further compressed, leading to higher contract prices. This chain is not entirely dependent on long-term imagination but will quickly enter revenue, gross margin, and EPS.
Micron's financial report has already reflected this change. According to the company's announcement, in FY2026 Q2, it achieved multiple records in revenue, gross margin, EPS, free cash flow, etc. Data center-related revenue saw a significant year-on-year growth, with guidance for FY2026 Q3 continuing to reach new highs. For Micron, AI storage is no longer a distant vision but a revenue source that is reflected in the quarterly financial statements.
SK Hynix's report is more straightforward. According to the company's announcement, the 1Q26 revenue was 52.5763 trillion South Korean won, operating profit was 37.6103 trillion South Korean won, and the operating profit margin reached 72%. The company attributed the growth to high-value-added products such as HBM, high-capacity server DRAM modules, and eSSDs. For investors, this profit margin reflects the product mix, supply-demand dynamics, and pricing power presented in the report.
Industry price data also supports the same logic. TrendForce expects 2Q26 conventional DRAM contract prices to increase by 58% to 63% QoQ, and NAND Flash contract prices to rise by 70% to 75% QoQ. Their report also indicates an 81% QoQ growth in 1Q26 DRAM industry revenue.
Price does not equal profit, but in a phase of supply constraints, product mix upgrades, and strong demand, price increases will enhance market expectations for EPS over the next several quarters. South Korean export data also provides early industry validation. According to Reuters and South Korean media reports, South Korea's May 2026 exports hit a record high, with semiconductor exports growing by 169.4% YoY to around $37.16 billion, with chips accounting for over 40% of total exports for the first time.
This does not directly equate to SK Hynix's or Samsung Electronics' earnings per share, but it indicates that the storage sector's prosperity has been reflected in the country's export-driven revenue acceleration.
Storage is not a stronger narrative, but a quicker validation
In this round of reassessment, the difference between storage and other AI semiconductor directions is not about whether there is growth, but how growth is validated.
NVIDIA remains the gatekeeper of AI demand. The GPU platform iteration determines AI server architecture, HBM capacity requirements, and supply chain qualifications. However, the market is already highly familiar with NVIDIA's growth and profits, with valuation long concentrated on its strongest AI assets. In the short term, it is more susceptible to export controls, supply chain constraints, platform transition pace, and expectation divergences.
The ASIC direction also has valid logic. Cloud providers' in-house chip development, custom accelerators, and rising AI inference demand are all driving long-term potential for assets like Broadcom and Marvell. However, ASICs are more like project-based business, with customer concentration, single-project import pace, production windows, and next-gen platform transitions affecting market visibility into revenue.
Optical modules and copper interconnects also have an EPS realization path. Companies like Coherent and Credo benefit from AI cluster internal bandwidth upgrades, and the shift to 1.6T, 3.2T optical modules, and cluster interconnect architecture changes will drive demand. But the pricing in these directions is more dependent on future architectural roadmaps, customer certifications, shipment pace, and capital expenditure cycles. When the market is willing to pay a premium, they exhibit high elasticity. However, when the market starts seeking validation, they are also more likely to be questioned about when orders will translate into revenue.
By contrast, storage now has a more direct pricing basis. HBM demand is driving high-end products, capacity shifts are squeezing traditional DRAM/NAND supply, contract price increases are improving revenue, a shift in product mix is raising gross margins, ultimately leading to EPS.
This chain does not mean there is no risk, but it is easier to validate in the next quarter's financial report compared to the idea of "a future generation architecture will bring large-scale orders." This is what is meant by storage being easier to model. It is not saying that storage is more important than GPU, ASIC, or optoelectronic modules, but rather that after this AI semiconductor derisking, the market prefers assets that can be collectively validated through pricing, orders, profit margins, and export data.
EPS Logic is Strengthening, But Has Not Yet Become Consensus
A one-day or two-day rebound does not prove that AI semiconductor trading has completely switched from P/E expansion to EPS validation.
Micron, on June 5, saw a nearly 13% decline; in the rebound on June 8, nearing 10%, there may have been technical repairs, short covering, and a return of risk appetite. SK Hynix's rise was also catalyzed by news related to data center cooperation with NVIDIA. In short-term trends, news, positions, and fundamentals are often overlapping, and not all price increases can be attributed to EPS certainty.
Storage itself remains a cyclical industry. Rapid increases in DRAM and NAND prices will improve supplier profitability, but they may also stimulate supply expansion or suppress the purchasing intentions of some end customers. HBM's annual contracts, yield ramp-up, customer qualifications, and market share allocation are still evolving, and it cannot be simply assumed that all price increases will seamlessly flow into the profit and loss statement.
SK Hynix and Micron are already highly watched AI storage targets in the market; stock price elasticity and fundamental resilience do not always move in sync. If the future slope of DRAM/NAND price increases slows, HBM market share falls short of expectations, or repeated orders from customers are disproven, the logic of EPS upside revision will also face challenges.
Similarly, this does not mean we should disregard ASICs, optoelectronic modules, copper interconnects, and equipment materials. If these areas provide stronger orders, clearer customer introductions, or guidance that exceeds expectations, the market may still reassign a valuation premium. AI semiconductors are not just about storage; at the current stage, storage is easier to buy back because it can be more easily explained through financial reports.
A more prudent assessment of this round of trends is that the plunge on June 5 has raised the market's validation threshold for AI assets. The recovery from June 8 to 9 shows that funds within the AI chain prefer sections with a shorter path to EPS realization. Storage happens to be in a position where orders, prices, capacity, and profit margins are all simultaneously visible.
Welcome to join the official BlockBeats community:
Telegram Subscription Group: https://t.me/theblockbeats
Telegram Discussion Group: https://t.me/BlockBeats_App
Official Twitter Account: https://twitter.com/BlockBeatsAsia
