The Memory Chip Cycle: Why the Same Stocks Boom and Bust on Repeat

Definition

Memory chips store data. DRAM holds it while a device is running; NAND holds it when the power is off. One manufacturer’s DDR5 module does the same job as another’s, which means buyers select on price and no producer sets its own.

That single characteristic — no pricing power — combined with factories costing tens of billions of dollars produces earnings swings that have no equivalent in consumer staples, software or banking. Micron reported a net loss of $5.83 billion in fiscal 2023 and net income of $28.24 billion in the single quarter ended 28 May 2026. Same company, same product, three years apart.

How the Memory Cycle Works

The loop has four phases and it has repeated for five decades.

Phase 1 — Shortage. Demand outruns installed capacity. Contract prices rise. Because the cost of producing a wafer barely changes, almost every incremental dollar of price falls to the bottom line. Margins go vertical.

Phase 2 — Capacity response. Record profits fund record capital expenditure. Every producer expands simultaneously, because each is responding to the same price signal. A new fabrication plant takes roughly two to three years and $15–25 billion to bring online.

Phase 3 — Glut. The new capacity arrives together, and it arrives late — typically after the demand surge that justified it has cooled. Supply overshoots. Contract prices collapse, often below cash cost.

Phase 4 — Discipline. Producers cut utilisation, delay equipment, write down inventory and post losses. Capex stops. Supply growth flattens while demand keeps compounding. Eventually the overhang clears and Phase 1 begins again.

The agricultural analogy is exact: a good wheat year prompts every farmer to plant extra, the following harvest floods the market, and prices collapse. The difference is that a wheat farmer’s response takes one season and a memory producer’s takes three years, which is why the memory overshoot is larger and the correction is deeper.

Micron’s own capital expenditure tells the story in real time: $19.60 billion in the first nine months of fiscal 2026 against $10.20 billion in the same period of fiscal 2025. The capacity response to the current boom is already funded and under construction.

Why the Swings Are So Violent: Operating Leverage

A memory fab’s cost is overwhelmingly depreciation on the building and equipment, plus the engineers who run it. Those costs do not shrink when the selling price halves, and they do not grow when it triples. The clearest possible demonstration sits in Micron’s own income statement:

Revenue multiplied by 4.5. The cost of making the product rose by a tenth. Everything in between became profit. That is not a management achievement — it is what happens to a fixed cost base when the market clearing price rises.

Now run it backwards. In fiscal 2023 the same fixed costs met collapsing prices:

Fiscal 2023 revenue fell 49.5% and the company recorded $1.83 billion of inventory write-downs — chips manufactured at a cost above what they could be sold for. Cluenex AI ingests exactly these inputs across semiconductor names, so a memory stock’s sentiment score reflects where the cycle sits rather than the last reported quarter.

The Trap: Cyclicals Look Cheapest at the Top

Because a P/E ratio divides price by trailing earnings, a cyclical company posting peak earnings displays a low P/E — right before earnings fall. The same company posting a loss has no meaningful P/E at all, or an enormous one, right before earnings recover.

How to Analyse a Memory Stock in Practice

1. Average earnings across a full cycle, never a single year. Micron’s nine-month fiscal 2026 net income of $47.27 billion and its fiscal 2023 loss of $5.83 billion are the same business. Take a 5–7 year average of earnings or free cash flow and value against that.

2. Track capex as the leading indicator of the next bust. Industry-wide capital expenditure peaks 18–36 months before supply does. Micron’s nine-month fiscal 2026 capex of $19.60 billion, nearly double the prior year, is the signal that new capacity is being committed now.

3. Watch contract pricing, not spot headlines. The bulk of memory volume moves on quarterly and annual contracts. Contract price direction, and any shift from annual to quarterly repricing, tells you more about the cycle than a single spot quote.

4. Separate HBM from commodity DRAM. High-bandwidth memory sold into AI accelerators is qualified per customer, supply-constrained and priced under longer agreements. Standard DDR5 and NAND are not. A company’s exposure mix determines how commodity-like its earnings actually are.

5. Read the balance sheet at the peak, because it has to survive the trough. Micron ended the May 2026 quarter with $30.2 billion of cash and marketable investments and $5.14 billion of long-term debt, down from $14.02 billion the prior August. Cyclicals that deleverage during booms survive the busts.

Common Mistakes and Misconceptions

Example: Samsung’s June 2026 Quarter

The cleanest illustration of what a commodity cycle does to a diversified company came from Samsung Electronics in the quarter ended June 2026.

How Cluenex Uses Cycle Data

Cluenex AI ingests semiconductor cycle inputs — pricing direction, inventory levels, capital expenditure trends and lead times — as part of the model that calculates predicted short-term and long-term price movement across the top 1,000+ US-listed stocks. Those inputs are not displayed as raw indicators on the platform; they are digested into the sentiment scores and forecasts shown for each name.

For memory and semiconductor equities specifically, this matters because a single reported quarter is close to meaningless. Cluenex’s owner earnings and discounted cash flow tools value a business from the cash its operations generate over time, and the moat analysis flags whether a company’s advantage is structural or is simply the current phase of a supply cycle.

Frequently Asked Questions

• What makes memory chips a commodity rather than a differentiated product? DRAM and NAND are manufactured to common JEDEC standards, so modules from different producers are functionally interchangeable in most applications. Buyers therefore select on price and availability, and no single producer can set its own price above the global clearing level for long. High-bandwidth memory is the partial exception, because it is qualified per customer platform and supply is concentrated.

• Why do memory companies lose money instead of just producing less? Because most of the cost is fixed. A fabrication plant’s depreciation, maintenance and engineering payroll continue whether output is high or low, so idling capacity raises the cost per chip rather than lowering total cost. Producers often run fabs at high utilisation into a glut, which is why Micron recorded $1.83 billion of inventory write-downs in fiscal 2023 — chips manufactured above the price they could be sold for.

• How long does a memory cycle last? Historically three to five years peak to peak, driven by the roughly two-to-three-year lag between committing capital to a new fab and that capacity reaching volume production. The current upcycle began in fiscal 2024 and Micron guided to $50 billion of revenue for the quarter ending August 2026, so the capacity response funded during it will not fully arrive until 2027 and beyond.

• Why does a low P/E ratio not work for cyclical stocks? A P/E divides price by trailing earnings. For a cyclical, trailing earnings are highest exactly when the next several years of earnings are most likely to fall, so the ratio looks cheapest at the point of greatest risk. Micron’s fiscal 2023 loss makes its P/E for that year undefined and its fiscal 2026 record earnings make its trailing P/E look very low. Neither number describes the business.

• Has AI demand actually broken the memory cycle? There is genuine evidence on both sides and no settled answer. In favour: HBM is technically difficult, supply is concentrated among three producers, and Micron has moved a portion of volume onto multi-year Strategic Customer Agreements that management says improve the durability and predictability of results. Against: every prior demand shock produced the same claim, and Micron’s own capital expenditure rose from $10.20 billion to $19.60 billion across the first nine months of fiscal 2025 and 2026 respectively — the standard precondition for the next glut.

• What should I look at instead of quarterly earnings? Industry capital expenditure, contract pricing direction, customer inventory levels and the balance sheet. Capex is the leading indicator of future supply; contract pricing is the near-term revenue driver; customer inventory signals whether demand is real consumption or restocking. For valuation, use mid-cycle earnings averaged over a full boom and bust, or price-to-book, which is far more stable than earnings.

• How does memory pricing affect companies outside the chip sector? It has become a material input cost for AI infrastructure buyers. Microsoft attributed roughly $25 billion of its approximately $190 billion 2026 capital expenditure to component price inflation, and Meta raised its 2026 capex guidance citing higher component pricing and data centre costs. Rising memory prices transfer profit from hyperscalers to memory producers, which is why the two groups can move in opposite directions on the same news.

Related Concepts

• How to Analyze Semiconductor Stocks: Wafer Starts, Lead Times, and Inventory Cycles — the broader semiconductor cycle framework
• Forward P/E vs Trailing P/E: Which One Actually Matters — why the denominator choice decides the answer for cyclicals
• Gross Margin vs Operating Margin vs Net Margin — reading the operating leverage in an income statement
• Value Trap vs Bargain: How to Tell If a Cheap Stock Is Broken — the low-multiple problem generalised
• How the AI Boom Turns Into a Stock Boom: Following the Capex Chain — where the demand driving this cycle originates
• Concentrated Stock Positions: The Risk Hiding in Your Best Winner — sizing a position whose earnings swing this far