Memory chipmakers had an exceptional 2026. Micron and SanDisk both posted record revenue and record gross margins as AI demand drained the supply of DRAM and NAND. When a stock has tripled and still trades at a single-digit forward earnings multiple, it looks cheap. That appearance is the central trap of cyclical investing, and memory is one of the most cyclical industries in the market.
This guide explains how to value memory and other deeply cyclical stocks. The short version is simple. One year of earnings tells you very little about what the business is worth. A low price-to-earnings ratio near the top of the cycle is usually a warning rather than a bargain. And a discounted cash flow model only helps if you feed it normalized earnings instead of peak earnings.
Why Memory Is So Cyclical
DRAM and NAND are commodities. One maker's chip is close to interchangeable with another's, so price is set by supply and demand rather than by brand or switching costs. Three forces make that price swing violently:
- Capacity is lumpy. A new fabrication plant costs many billions of dollars and takes years to build. Producers add capacity in large steps, so supply regularly overshoots or undershoots demand.
- Demand is volatile. Memory ships into PCs, phones, cars, data centers, and now AI accelerators. Several of these end markets can soften at once.
- Inventory amplifies both. Buyers stockpile when they fear shortages and stop ordering when they hold too much, which exaggerates the underlying swing.
The result is an earnings curve that looks nothing like a stable business. Within a year, SanDisk's quarterly revenue grew more than threefold, from $1.70 billion to $5.95 billion, while Micron's gross margin jumped from 36.8% to 74.4%. SanDisk's gross margin reached 78.4% in its quarter ended April 2026. Those numbers come from each company's SEC filings, and they are not typos. They are the cycle.
Trap 1: The PE Ratio Lies at the Top of the Cycle
The price-to-earnings ratio compares price to one year of earnings. For a stable company that is reasonable, because next year looks like this year. For a cyclical company it inverts the signal you want.
At the top of the cycle, earnings are at their highest, so the PE ratio is at its lowest. At the bottom, earnings collapse or turn negative, so the PE ratio spikes or stops meaning anything. A memory stock therefore looks cheapest on a PE basis exactly when its earnings are least sustainable.
This is why a single-digit forward PE on a memory name should raise a question rather than settle one. The right question is not "why is it so cheap," but "are these earnings normal, or are they peak earnings that the market already expects to fall." For more on why a multiple is a starting point and not an answer, see .
Trap 2: DCF on Peak Cash Flow Overvalues the Business
A discounted cash flow model projects future free cash flow and discounts it back to today. The danger with a cyclical is the starting point. If you take a peak quarter, annualize it, and grow it at a steady rate forever, you capitalize peak conditions into perpetuity. The model then reports a high intrinsic value that depends entirely on the cycle never turning.
Here is the same business valued two ways, using illustrative figures to show the size of the error:
| Input | Peak-based DCF | Normalized DCF |
|---|---|---|
| Starting free cash flow per share | $30 at the peak | $12 mid-cycle |
| Growth rate, years 1 to 5 | 8% | 4% |
| Terminal growth | 2.5% | 2.5% |
| Discount rate | 11% | 11% |
| Resulting intrinsic value | roughly $430 | roughly $165 |
Same company, same discount rate. The only real difference is whether you started from peak cash flow or through-cycle cash flow, and the intrinsic value moves by more than 2.5 times. For a cyclical, the starting cash flow is the whole ballgame.
How to Normalize Earnings for a Cyclical
Normalizing means estimating what the business earns on average across a full cycle, not at its best or worst moment. A workable process:
- Mid-cycle revenue. Average revenue across a complete cycle, ideally five to seven years, rather than annualizing the latest quarter. For a young company without that history, use the industry's pattern as a guide and acknowledge the wider error bar.
- Mid-cycle margin. Memory gross margins have ranged from the low 20s, and at times below zero, up to the high 70s. Pick a through-cycle average margin rather than the current one.
- Normalized free cash flow. Apply the mid-cycle margin to mid-cycle revenue, then subtract the maintenance capital spending the business needs just to stay competitive. Memory is capital hungry, so this step matters.
- Discount and stress test. Feed the normalized number into the model and pick a discount rate that reflects the risk. The logic for setting that rate by industry is covered in , and the broader two-stage method in .
The output is not a single number. It is a range, and the range is wide. That is honest, and it is the point.
Margin of Safety Matters Most at the Peak
Because your normalized estimate is uncertain and the downside in a downturn is large, you need room to be wrong. Buying a cyclical near the top with no margin of safety means betting that the peak holds. Buying well below your normalized estimate means you can be wrong about the exact mid-cycle figure and still avoid a permanent loss.
This is the same discipline value investors apply everywhere, and it is explained in . For cyclicals it is not optional. The wider the earnings swing, the wider the discount you should demand.
Applying This to Today's Memory Stocks
The two clearest examples in this cycle are Micron and SanDisk. They sit at different points on the risk scale. Micron makes both DRAM and NAND and has exposure to high-bandwidth memory used in AI systems, which gives it some diversification across products. SanDisk is a pure NAND maker that was spun out of Western Digital in 2025, which makes it more concentrated and harder to normalize because it has a shorter standalone history.
Both deserve the same treatment described above. Start from normalized earnings, treat the low forward multiple with suspicion, and demand a margin of safety. The company-by-company detail is in and .
You can run your own normalized case for either name with the . Enter a mid-cycle free cash flow rather than the peak figure, and watch how much the intrinsic value changes.
Frequently Asked Questions
Why is a low PE ratio dangerous for a memory stock? Because the PE ratio uses one year of earnings, and a cyclical company's earnings peak at the top of the cycle. The lowest PE often appears at the moment earnings are least sustainable, so a low multiple can signal peak conditions rather than a bargain.
What does normalized earnings mean? Normalized earnings estimate what a business earns on average across a full cycle, rather than at its best or worst quarter. For a memory company you average revenue and margins over several years instead of annualizing the latest result.
Can you use DCF for cyclical stocks at all? Yes, but the starting cash flow must be a mid-cycle figure, not a peak one. A DCF built on peak free cash flow assumes the peak lasts forever and badly overstates intrinsic value. Built on normalized cash flow, DCF works and produces a sensible, if wide, range.
How big a margin of safety should I use for memory stocks? There is no single number, but the wider the earnings swing, the larger the discount to intrinsic value you should require. Deeply cyclical names justify a bigger margin of safety than stable businesses because the downside in a downturn is larger.
Ready to test a normalized case? Open the and compare a peak input against a mid-cycle input for any memory stock.
