Micron makes DRAM and NAND memory, and its profits swing with memory prices more than almost any other large chip company. High-bandwidth memory for AI servers is the newer driver. See the stock courses for method.
The Cycle Did Not Die
Micron sells memory. Memory prices go up until they overbuild, then they go down until they underbuild. HBM and AI servers can change the mix and the duration of an upcycle. They do not repeal the cycle. If your MU thesis requires the cycle to be dead, you do not have a thesis.
Size MU as a high-beta semiconductor cyclical, not as a software compounder that happens to make chips.
1. Boise memory maker and the boom-bust cycle
Micron (1978, Boise) survived by consolidation: rivals died or were absorbed, DRAM became an oligopoly, NAND stayed brutal. The 2000s and 2010s taught the same lesson twice: bit growth is not profit growth when ASP falls faster than cost. Traders who only watch bits will donate capital to the upcycle’s last quarter.
HBM for accelerators is a real mix upgrade: tighter qualification, higher ASP, different customers (the GPU-cluster world). It is also still memory, still process, still yield. Treat HBM as a premium SKU inside a cyclical industry, not as a new industry.
China, export rules, and foundry/tool geopolitics now sit on memory the way they sit on logic. Supply can be politically constrained and still oversupplied in the unconstrained regions. Both can be true.
2. What the modern company sells
DRAM and NAND are the two engines. Compute and mobile are demand buckets. Data-center HBM is the new headline bucket. Read pricing commentary before you read TAM slides.
Capex in an upcycle is how downcycles get built. If MU and peers all raise capex together, the lag is a future glut. That lag is the trade.
Customers include device OEMs, hyperscalers, and module houses. Concentration is rising with HBM. Rising concentration is not diversification just because the 10-K lists three end markets. NVIDIA buys high-bandwidth memory for its accelerators; Micron is a supplier, so it shares the AI demand but not the pricing power.
3. Why it still compounds — and what stops it
Process know-how, scale, and oligopoly structure are the moat. When memory prices fall, none of that keeps the shares from dropping hard. It prevents extinction.
What stops an upcycle: capex, a demand pause in servers, or a yield miss on a leading node. What stops the company is rarer. Do not confuse the two.
MU can print record EPS and still be a sell if forward ASP is rolling. Forward ASP is the object.
4. How traders actually use the ticker
MU is a gap name with semiconductor-ETF beta. Implied moves into earnings are large. Size as a cyclical, not as NVDA-lite. MU can move sharply on memory pricing data and guidance, so let the position size calculator assume wide days. Competitor results from Korean memory makers often show up as volume on the stock scanner.
Illustration: $50,000 in the account, $500 at risk, $55 between entry at $980 and the stop, about 9 shares. Write it in the trade plan and check the risk rules before you buy a cycle peak because margins look great.
Pairs vs other memory or vs a SOX basket only with an ASP view.
5. Mistakes, limits, takeaways
Mistakes: marrying MU at cycle peaks; treating HBM as a software multiple; ignoring peer capex. Gross margin, DRAM versus NAND revenue and inventory days are the cycle signals. The guide to financial statements explains them; What Is Stock Trading? covers the basics.
Spot prices move faster than anything written down. Educational only.
Before you trade it
- Memory is still a boom-and-bust business. A premium product line does not change that.
- Watch ASP and peer capex.
- HBM is mix, and it rides the same memory cycle.
- Size for a violent print.
Memory is a cycle, and Micron's best quarters have often come just before the turn. Educational only. Not tax, legal, or a recommendation to buy, sell, or hold MU.