The Philadelphia Semiconductor Index closed up 5.53% yesterday at 12,876.91, setting a new all-time high. The S&P 500 rose 0.61% to 7,519.12, while the Nasdaq gained 1.19% to 26,656.18 — both indices reaching fresh record closing highs. In contrast, the Dow Jones Industrial Average fell 0.23% to 50,461.68, highlighting the stark divergence between weakness in traditional industries and enthusiasm surrounding technology stocks.

At the individual stock level, Sandisk rose 7.50% to a record high, with trading volume reaching US$20.177 billion. Qualcomm gained 4.48% to close at US$248.82. Meanwhile, Western Digital climbed more than 8%, AMD surged 7.78% to a new all-time high, Marvell Technology gained over 6%, Seagate Technology rose more than 4%, and Intel advanced over 3%.

The biggest winner was Micron Technology, which surged 19.29%, marking its largest single-day gain since November 2011. Its market capitalization officially surpassed US$1 trillion for the first time, making Micron the latest member of the semiconductor industry’s “trillion-dollar club,” alongside Nvidia, Microsoft, TSMC, and Broadcom. Since the beginning of this year, Micron shares have gained 210%, while the stock has risen more than tenfold from its lows over the past year and a half.

The Macro Background Behind the Surge

US Secretary of State Marco Rubio stated that the US and Iran are still engaged in “extensive back-and-forth discussions” over the wording of an initial agreement draft, and that reaching a final deal may still require several more days. President Trump described the negotiations as “progressing well,” while also warning that additional military action remains possible if talks fail. According to semi-official Tasnim News Agency reports, one major point of discussion involves approximately US$24 billion in frozen Iranian assets, with Tehran reportedly seeking the release of half those funds after signing a deal.

Oil prices subsequently declined, while gold fell below US$4,500 per ounce. As traditional safe-haven assets weakened, risk appetite capital accelerated its rotation back into AI-related trades. With geopolitical risks beginning to ease at the margin, AI semiconductor stocks quickly entered another explosive rally phase.

Why Did Micron Surge 19%?

First, the AI narrative itself is shifting from “training” toward “inference and deployment.” Training large AI models primarily requires computing power. However, AI inference and real-world AI agent deployment require full-stack data center infrastructure — servers require memory, data centers require storage, networks require high-speed interconnects, and power systems require efficient management. The AI investment narrative is now evolving from “stockpiling GPUs to train models” toward “running AI agents and inference workloads using CPU + GPU infrastructure.” As a result, Nvidia is no longer the only beneficiary. The broad sector-wide rally on May 26 reflected this shift in AI market positioning.

Micron’s 19.29% single-day surge was driven not only by the broader sector rally, but also by a research report from UBS analyst Timothy Arcuri. UBS raised its Micron price target from US$535 all the way to US$1,625 while maintaining its “Buy” rating. A US$1,625 share price would imply a market capitalization of approximately US$1.8 trillion. If achieved, Micron’s valuation would surpass Tesla, Meta, and Berkshire Hathaway, placing it behind only Apple, Microsoft, Amazon, and Broadcom.

Tripling a price target overnight is extremely rare on Wall Street. More importantly, the logic behind UBS’s call attempts to overturn not merely a few quarters of earnings expectations, but the entire valuation framework that has governed the memory chip industry for the past three decades.

Layer One: LTAs — From “Three-Year Boom-and-Bust Cycles” to Locked-In Long-Term Pricing

Historically, the memory chip industry has repeatedly followed the same cycle: boom → capacity expansion → oversupply → price collapse → losses → production cuts → recovery. DRAM prices have historically been capable of falling by half from peak to trough, while NAND prices have often suffered even steeper collapses. As a result, memory manufacturers typically endured a brutal “profit-to-loss-to-recovery” cycle roughly every three years, creating highly volatile cash flows and structurally depressed valuations.

UBS’s core argument now is that this pricing model is beginning to fundamentally change. Nearly 30% of industry DDR shipments are expected to move under long-term supply agreements (LTAs) containing partially fixed pricing structures. These contracts typically span three to five years under either “2+3 structures” (2 years fixed + 3 years adjustable) or “3+2 structures.” Large cloud providers have reportedly already locked in roughly 60%–70% of DDR5 server memory supply through such long-term contracts.

This means Micron may sacrifice some short-term upside pricing flexibility in exchange for significantly higher demand certainty and smoother long-term profitability. For a company whose historical earnings volatility could destabilize almost any valuation model, certainty itself becomes the most valuable premium.

Layer Two: The HBM Shortage — Structural Supply Deficits Could Extend Into 2028

The LTA story addresses pricing stability, but the deeper fundamental issue remains supply and demand. Since Goldman Sachs’s February report earlier this year, the supply-demand structure of the memory industry has undergone a major transformation. Global DRAM supply deficits for 2026 are now estimated at 4.9%, while NAND Flash deficits are projected at 4.2%. HBM shortages are even more severe at 5.1% — the highest levels since 2011.

The HBM shortage is being driven primarily by explosive demand from Nvidia and the three major US cloud providers. Micron has already publicly stated that its entire 2026 HBM production capacity is fully sold out. HBM4 products have already entered mass production, with ramp-up speeds reportedly twice as fast as HBM3. Next-generation HBM4E mass production is expected to begin in 2027. According to Mizuho Securities, HBM4 and HBM4e pricing could rise another 70%–100% year-on-year by 2027. Mizuho further estimates that supply shortages in non-HBM memory markets for major customers currently range between 30%–50%, with the imbalance expected to persist “well beyond 2026.”

UBS’s quantitative forecasts estimate Micron’s HBM average selling price could rise 50% year-on-year to US$18.60 per GB by 2027. Gross margins could eventually climb toward approximately 75%, while the structural supply shortage window may extend into the second quarter of 2028.

Layer Three: Valuation Re-Rating — Benchmarking Against Nvidia

This represents the most disruptive element of the UBS report. UBS proposed a remarkably straightforward framework: apply a 15x forward earnings multiple over the next 12 months — and explicitly state that “there is no reason Micron’s valuation multiple should differ dramatically from Nvidia’s.”

Institutional ownership data from Q1 also confirms that major capital has already been aggressively accumulating Micron shares. Approximately 2,440 institutional investors disclosed new Micron positions during the quarter, including Rockefeller Capital Management and Schroders.

When combining the 2,440 institutional purchases in Q1 with UBS’s new US$1,625 target price, a clear trend emerges: institutional capital had already heavily accumulated Micron long before the UBS report arrived. UBS’s research simply provided the theoretical justification for positions institutions had already built.

The deeper logic is that AI is transforming memory chips from cyclical commodities into core infrastructure components of the AI economy. Once GPUs become tightly integrated with HBM memory, memory manufacturers no longer function merely as “component suppliers.” Instead, they become co-creators of AI computing power itself. From that perspective, memory companies eventually receiving valuation multiples comparable to GPU manufacturers may no longer seem unreasonable.