Micron's net profit in a single quarter has already surpassed its full-year revenue from the previous year.

In its results released on September 30, Micron reported revenue of $54.23B for the quarter ended September 3, up 379% year over year; net profit was $37.7B. By comparison, its total revenue for the entire 2025 fiscal year was only $37.38B.
Adjusted gross margin in the fourth quarter soared to 87%. A capital-intensive manufacturer that has to buy lithography machines, build factories, and endure yield losses achieved profit margins that even software companies would struggle to match.
This earnings report provides a sufficiently concrete answer to where the money from AI infrastructure is ultimately flowing. But for Micron CEO Sanjay Mehrotra, more important than how much the company earns this quarter is how long it can continue earning that money.
He has reason to remain wary of the boom before him. In fiscal 2023, Micron's revenue nearly halved, and it posted a full-year net loss of $5.83B. The memory industry has its own cycle. Shortages, price hikes, and capacity expansion follow; once the new capacity built with tens of billions of dollars comes online, customers begin destocking and prices consequently collapse.
The profits from one cycle are used to pay for the depreciation of the next.
Now, Micron is trying to tear up this script.
This quarter, DRAM bit shipments rose only by the mid-single digits sequentially, while average selling prices jumped by more than ten percent; the gross margin of the division serving mobile phones and PCs was even pushed to 90%. Price contributed far more than shipment volume.
AI's demand for memory has long since spilled over into ordinary memory and flash memory. At the same process node and capacity, producing a single HBM3E chip consumes roughly three times the wafer resources required for DDR5. Limited capacity is being diverted to support large models, squeezing supplies for the consumer market completely dry.
This means that even a phone manufacturer that has never produced an AI server is being forced to foot the bill for this computing power frenzy.
Then the relationship between buyers and sellers changed.
Micron has already signed 26 strategic customer agreements, which are expected to correspond to more than 35% of the company's total revenue through 2030. Without exception, these agreements introduce multi-year take-or-pay arrangements, with customers' combined financial commitments totaling $32 billion, most of which comes directly in the form of cash deposits.
At the absolute peak of its bargaining power, Micron is asking customers to pay for future purchases. Multi-year procurement obligations shift the cost of misjudgment onto customers, giving manufacturers the confidence to schedule tens of billions of dollars in capacity several years in advance.
But this does not mean the cycle has disappeared.
Around three-quarters of these agreements include price floors and ceilings. The price floors provide protection, while the price ceilings likewise cut off excess returns. Moreover, the agreements change who bears the risk but cannot prevent the time lag in capacity coming online.
Micron's next move remains aggressive capacity expansion. Net capex is projected at $27.37 billion in fiscal 2026, followed by another $25 billion in the first half of fiscal 2027, while the new plants will not truly begin producing wafers until 2027 or even 2030.
Computing's appetite for memory is happening now, but the realization of massive capacity lies several years ahead. Several years from now, models' efficiency in utilizing KV cache and the data center budgets of the major players will together determine the reality those new factories face.