To The Moon

Executive Summary

The Philadelphia Semiconductor Index ( $费城半导体指数(SOX)$ ) has risen more than 50% in the past 25 trading days. This is the longest winning streak in the index's 32-year history, and the fastest rate of ascent since March 9, 2000.

Memory and storage stocks are leading the charge. $Micron Technology(MU)$ gained 38% in a single week, 84% over the past month, and now commands a market capitalization exceeding $700 billion. $SanDisk Corp.(SNDK)$ has risen more than 3,000% over the past year. $Western Digital(WDC)$ has risen more than 1,000% over the same period. Optical networking stocks have delivered similar violence— $Applied Optoelectronics(AAOI)$ is up 328% year-to-date, $Lumentum(LITE)$137%.

The prevailing narrative is the "AI Memory Supercycle." NAND contract prices rose 65–70% quarter-over-quarter in Q1. Q2 is projected to deliver another 70–75% increase. Gartner forecasts full-year 2026 NAND average selling prices will rise 234%, with no meaningful pricing relief expected until late 2027.

Demand is real. The supply gap is physical. But has the rate of price appreciation decoupled from the pace of fundamental delivery?

This report does not answer "can it keep going up." It answers a colder question: What assumptions are currently embedded in the price? What is the verification path for those assumptions? And if any single assumption fails, what is the downside?

The Physical Reality of Demand: Establishing the Foundation

Before auditing price, we must first acknowledge fact.

The 2026 storage demand explosion is not narrative-driven. It rests on three layers of physical support.

The first layer is the absolute scale of AI infrastructure spending. The five major North American hyperscalers have revised their full-year 2026 CAPEX expectations upward to $690 billion—nearly double the $405 billion deployed in 2025. Google Cloud's quarterly backlog has roughly doubled to over $460 billion. These orders ultimately require storage-layer fulfillment.

The second layer is architectural migration. The proliferation of RAG (Retrieval-Augmented Generation) architectures is shifting AI systems from "compute-intensive" toward "data-intensive." Enterprise-grade PCIe Gen5 NVMe SSDs with capacities exceeding 122.88TB are now shipping at commercial scale in Q1 2026. System architects are deploying dense eSSD arrays to replace portions of expensive HBM caches—an energy arbitrage strategy that trades storage capacity for compute cycles.

The third layer is the physical supply-demand mismatch. 2026 NAND demand growth is estimated at 20–22%, while supply growth is limited to 15–17%. New wafer fab construction cycles run 2–3 years. The supply gap cannot be closed before 2027.

This is not a 2000-style bubble built on revenue-free speculation. Micron's most recent quarter (Fiscal Q2) delivered revenue of $23.9 billion, up 196% year-over-year. Gross margin reached 75%—a company record. Q3 guidance is even stronger: revenue of $33.5 billion, gross margin approximately 81%, EPS approximately $19.15.

The fundamentals are real.

The Micron Valuation Audit: A Counterintuitive Discovery

This is the most critical section of this report.

Intuition suggests that a stock which has risen 84% in one month must be expensive. The ledger tells a different story.

As of early May, Micron's trailing P/E ratio stands at approximately 25x. But based on Q3 guidance (EPS $19.15) annualized, the forward P/E compresses to roughly 5.9x. This is among the lowest forward P/E ratios in the entire S&P 500.

What does this number mean?

It means the market is still applying a "cyclical stock" valuation framework to Micron. The defining characteristic of cyclical stocks is that P/E ratios are lowest at peak earnings—because the market expects earnings to decline. If the market believed that 75–81% gross margins were sustainable, the forward P/E should be 15–20x. If the market believes this is merely a cycle peak, a forward P/E of 5–6x is rational.

The real question, therefore, is not "is Micron expensive?" It is:

Are current earnings levels a cycle peak, or a new normal?

Arguments supporting "new normal":

• AI-driven storage demand is structural, not one-time

• Supply expansion is constrained by physical limitations (2–3 year fab construction cycles)

• Industry concentration is extreme (three manufacturers control 95%+ of global share), enabling coordinated pricing

• Gartner projects no meaningful price relief until late 2027

Arguments supporting "cycle peak":

• The memory industry has never sustained 75%+ gross margins for more than 3–4 consecutive quarters in its history

• In the 2017–2018 supercycle, Micron's gross margin collapsed from 60% back to 30% within four quarters

• When price increases reach 234% (Gartner's full-year forecast), downstream customer tolerance reaches physical limits

• Samsung, SK Hynix expansion plans, once executed, will release supply rapidly

My assessment: the truth likely lies between these extremes. AI has genuinely altered the structural demand profile for storage. But 75–81% gross margins remain a cyclical extreme. A reasonable "new normal" gross margin likely sits in the 45–55% range—well above the historical average of 30–35%, but well below current levels.

If gross margins revert from 81% to 50%, even with revenue sustained at elevated levels, EPS would decline from $19 to approximately $10–12. At 15x P/E, that implies a stock price of $150–180. At 20x, approximately $200–240.

The current stock price is $746.

This does not mean Micron will necessarily fall to $200. But it means: the current price implicitly assumes that 75%+ gross margins will persist for at least 2–3 years, AND that the market will award growth-stock multiples (15–20x) rather than cyclical-stock multiples (5–8x).

If both assumptions hold simultaneously, the current price is reasonable—perhaps even cheap. If either assumption fails, the downside is substantial.

The Optical Networking Signal: May 7th

The optical networking sector has delivered extraordinary returns in 2026. Applied Optoelectronics is up 328% year-to-date, Viavi Solutions 148%, Lumentum Holdings(LITE) 137%, and $COHERENT(COHR)$ earned inclusion in the $S&P 500(.SPX)$ 500.

The demand logic is clear: 800G to 1.6T transceiver upgrades, NVIDIA's $500 million optical partnership with Corning, and Bank of America's designation of an "Optics Supercycle."

But on May 7th, something notable occurred.

Lumentum, Coherent, and Fabrinet simultaneously reported earnings that exceeded expectations. Revenue hit records. Profits beat consensus. Yet after-hours prices collapsed—Applied Optoelectronics fell 14%, Coherent dropped 10%, Lumentum declined 7%.

This is a textbook "sell the news" signal.

When positive news fails to drive prices higher, it indicates that the positive news was already fully embedded in the price. Marginal buyers entered before the earnings release. After the release, no new buyers remained to provide continuation.

This does not mean optical networking demand has disappeared. The 800G to 1.6T upgrade cycle is a genuine physical requirement. But it means: current stock prices have already discounted 2–3 years of future growth in a single move. Entering at these levels means the return depends not on "fundamentals improving," but on "sentiment continuing to escalate." That is a different kind of bet.

The Speed Problem: Direction Is Not the Issue—Velocity Is

The SOX has risen 50% in 25 days. 32% over six weeks. Approximately 60% year-to-date.

Morningstar has noted: the last time SOX delivered this magnitude of 25-day performance was March 9, 2000.

I do not believe 2026 is a replay of 2000. The distinctions are clear.

In 2000, semiconductor companies had no real AI revenue. In 2026, Micron has 196% revenue growth. In 2000, CAPEX was aspirational. In 2026, $690 billion is being physically deployed. In 2000, there were no supply constraints. In 2026, wafer fab construction takes 2–3 years and transformer lead times stretch to 48 months.

But velocity itself is risk.

When an index rises 50% in 25 days, it does not consume fundamentals. It consumes future expectation space. Each day of gains compresses the potential return available to new entrants.

A simple calculation: if the "reasonable" annualized return for SOX is 15–20% (based on constituent earnings growth), then a 50% gain in 25 days means the index has already borrowed 2–3 years of future returns.

This does not guarantee a decline. Momentum can persist far longer than rationality suggests. But it means: the risk-reward ratio for initiating new positions at current levels has deteriorated materially.

Physical Signals of Cycle Reversal: What to Monitor

I do not predict tops. But I can identify the conditions under which the cycle may be approaching an inflection.

First, NAND contract price increases begin to decelerate. From 70–75% quarter-over-quarter down to 30–40%, then to 10–15%. Prices do not need to decline. They only need to decelerate. Stock prices will lead the reaction.

Second, memory manufacturers announce major capacity expansion. Any announcement from Samsung, SK Hynix, or Micron regarding new fab construction or significant CAPEX increases signals that supply relief is approaching.

Third, channel inventory begins to accumulate. When downstream customers shift from "panic buying" to "strategic stocking" to "inventory digestion," the cycle has entered its latter phase.

Fourth, downstream customers begin deferring orders. Similar to the HBM push-out dynamics observed in early 2026—when data center power constraints prevent equipment from being energized, storage orders face corresponding delays.

Historical reference: during the 2017–2018 NAND supercycle, Micron's stock rose from $20 to $60 (+200%), then fell back to $30 within six months of manufacturers collectively announcing expansion plans. Stock prices peaked 3–6 months ahead of fundamentals.

Conditional Framework: Positioning by Current Exposure

For investors already holding memory, optical, or semiconductor positions:

Demand is real. There is no reason for panic selling. But a "physical verification timeline" must be established. If Q3 NAND contract price increases decelerate from 70% to below 30%, while stock prices continue making new highs, that is the signal that price has decoupled from its physical anchor. At that point, gradual position reduction is warranted—before prices have already declined.

For investors considering entry:

Current prices embed extremely optimistic assumptions. Using Micron as the reference case, a $746 stock price requires 75%+ gross margins to persist for 2–3 years to be justified. Entering at this level means the return depends on "the market continuing to award growth-stock multiples"—not on "the business improving." This is a bet on sentiment, not on enterprise value.

For investors seeking structural opportunity:

The genuine expectation gap may not reside in companies that have already risen 30x. It may exist in upstream bottlenecks within the supply chain that have not yet been fully priced—testing equipment for storage devices, specialty gases, photoresist materials, and other input layers. Order growth in these segments is equally certain, but valuations remain far from the extremes observed downstream.

Final Observation

SOX rose 50% in 25 days. The last time this velocity occurred was March 2000.

That time, fundamentals were hollow. This time, fundamentals are real.

But price never reflects fundamentals alone. It also reflects the unanimity of expectations about the future.

When consensus reaches its extreme, marginal buyers disappear.

Demand does not disappear. But pricing power transfers.

From those who chase. To those who wait.