Written by: Little Cake

In one quarter, $41.4 billion in revenue, a year-on-year surge of 346%. Gross margin 84.9%, net profit of $28 billion.

These numbers are shocking for any tech company, but for Micron, it seems almost absurd. In the same quarter two years ago, the company’s revenue was $9.3 billion, gross margin 39%, and net profit was less than $1.9 billion.

However, the stock price jumped 15% after hours, indicating that market excitement is no longer focused on Q3 itself. What truly ignited the sentiment was the Q4 guidance: $50 billion in revenue (median), gross margin around 86%, and earnings per share of $31.

The storage supercycle is ongoing.

Behind the numbers: Dissecting a money printer

Breaking down Micron's financial report, every business segment is speaking through exponential growth.

DRAM contributed $31.3 billion in revenue, accounting for 76% of total income, with average prices skyrocketing over 60% quarter-on-quarter. NAND recorded $9.9 billion, also far exceeding expectations. But the most explosive growth lies within the business units: the core data center business had quarterly revenue exceeding $25 billion, annualized over $100 billion, up more than 7 times from $1.53 billion in the same quarter last year. Data center SSD revenue exceeded $5 billion, doubling quarter-on-quarter. The automotive and embedded business also recorded $4.63 billion, growing over 3 times year-on-year.

Operating cash flow was $25.39 billion, adjusted free cash flow was $18.3 billion. Cash and investments on hand totaled $30.2 billion, with net cash of $24.4 billion. Debt was reduced by $4.4 billion in this quarter, and all three major rating agencies have upgraded to BBB+.

This is a financial report that makes the saying "storage is a cyclical industry" feel outdated.

16 SCAs: Micron is rewriting its business DNA

Numbers can be explained by supply-demand imbalance, but the mutation of the business model cannot.

Micron announced on the conference call that it has signed 16 strategic customer agreements (SCAs). These agreements are binding contracts on a "take or pay" basis, covering about 20% of DRAM shipments and one-third of NAND shipments, with terms extending from 2026 to the end of 2030. They include 4 ultra-large clients and 3 mid-sized clients, with the rest coming from the automotive industry.

CFO Mark Murphy disclosed a new metric for the first time: Remaining Performance Obligations (RPO). As of the end of Q3, the RPO was $5 billion; with the new agreements signed after the end of the quarter, this figure jumped to about $100 billion. Management clearly stated that actual revenue will "far exceed" the contract floor price corresponding to the RPO.

More critically, the goal: Micron plans to increase the share of revenue covered by SCAs to over 50%.

What does this mean?

The business model of the storage industry has been based on spot pricing and short-term contracts for the past 40 years. Price volatility is the norm, and valuations have long been suppressed by "cyclical discounts." The essence of the SCAs is to transform storage chips from commodities into pre-sold infrastructure resources, akin to how cloud vendors sign long-term contracts to buy electricity and fiber.

If Micron can achieve a 50% SCA coverage by 2027, its revenue predictability will approach that of an enterprise software company, while its current forward price-to-earnings ratio is just above 10 times. Among trillion-dollar market cap companies, there is nothing cheaper than this.

Supply cannot keep up with demand

CEO Sanjay Mehrotra's remark on the conference call is worth savoring: Micron currently "does not see a time point when supply can catch up with the continually growing demand." He expects shortages of DRAM and NAND to continue beyond 2027.

This is not empty talk. Micron's HBM production capacity for the entire year of 2026 has already been fully locked in based on price and quantity. The ramp-up speed for HBM4's 12-layer stacking production is twice that of the previous generation HBM3E and has already contributed over $1 billion in revenue. Management predicts that the total HBM market will grow at an approximate compound annual growth rate of 40%, climbing from $35 billion in 2025 to $100 billion in 2028, two years ahead of previous expectations.

The constraints on the supply side are physical. Building an advanced storage wafer fab requires 3-4 years and billions of dollars in investment. Micron has already raised its capital expenditure for fiscal year 2026 to about $27 billion (net of government subsidies), with new plants in Idaho and Japan ramping up, each quarter adding $100-200 million in startup costs. However, the release of production capacity is slow, while the demand from AI data centers for storage is exponential.

Global hyperscale cloud providers' capital expenditure for AI data centers in 2026 will exceed $725 billion, and this money will ultimately flow through storage chips.

The timeline of the Anthropic deal is interesting

Two days before the financial report was released, Micron announced a strategic agreement with Anthropic, covering joint design of storage architecture, multi-year supply contracts, deploying Claude within Micron, and strategic investment in Anthropic's Series H financing. Thus, all three global HBM suppliers (Samsung, SK Hynix, Micron) have become strategic investors in Anthropic's Series H.

No AI lab has locked in all three manufacturers of the global HBM supply chain simultaneously. The significance of this deal goes far beyond the supply contracts themselves; it marks the beginning of AI companies viewing the storage supply chain as a strategic asset. As the efficiency of model training and inference increasingly depends on the performance of storage subsystems, the bargaining power of storage manufacturers will only continue to rise.

Concerns remain, but the weighting is changing

The historical lesson of the storage industry is clear: Every supercycle is followed by a collapse caused by overcapacity. Will the synchronized expansion of Samsung, SK Hynix, and Micron ultimately replay the story of 2018 or 2022?

The difference lies in the structural changes on the demand side. Previous storage cycles were driven by consumer electronics; once shipments of smartphones and PCs peaked, demand would plummet. However, the storage demand from AI data centers is continuously additive, with each generation of models being larger, each inference request consuming more tokens, and each Agent requiring a longer context window. Additionally, as AI PCs will standardize memory from 16GB to 32GB, and flagship smartphone DRAM demand rises simultaneously, the demand bottom for storage has been structurally elevated.

The SCA system is Micron's institutional hedge against the risks of this cycle. Even if demand growth slows, take-or-pay contracts can ensure a revenue floor. This cannot eliminate the cycle, but it can significantly narrow the amplitude of fluctuations.

Micron's earnings season has ended, but the questions it raised have just begun: When storage transforms from a commodity into a strategic resource that requires reservation, does the valuation language of this industry need to be rewritten?

Based on a Q4 guidance of $31 EPS annualized, Micron's current forward price-to-earnings ratio stands at around 10 times. Similarly benefiting from AI, with demand exceeding supply, Nvidia's forward price-to-earnings ratio exceeds 30 times. Within this gap lies market pricing on "the cycle will eventually return," alongside an old narrative that is gradually being disproven by the SCA system.