StockResearch

AI Memory

Storage-system vendors sit near the top of their own 5-year EV/Sales range (98th percentile); memory suppliers sit in the upper half of theirs (72nd).

10 companies · data as of Oct 19, 2026

Tap a layer to see its companies.

Open these 8 in the screener

Equal-weight basket +728% vs SMH +290% since Oct 2023

Hypothetical equal-weight index, price only. Not a fund.

The thesis

AI servers need far more memory than ordinary ones, and since late 2025 demand has outrun supply, pushing memory prices and revenue to records. The stocks moved faster than filed revenue: most members sit near the top of their own five-year EV/Sales range. The open question is when new factories end the shortage.

By Jake, founder of StockResearch.app · Oct 5, 2026 · Opinion, not investment advice

Why AI ran short of memory

Every AI accelerator sits next to stacks of high-bandwidth memory (HBM), and every AI server needs more ordinary DRAM and flash than the servers before it. Only three companies make HBM in volume: SK hynix, Samsung and Micron. HBM also uses more factory space per bit than ordinary DRAM, so each shift toward it shrinks supply of everything else.

Memory became one of the largest costs in AI. A Google supply-chain director said memory is now more than 75% of an AI server’s hardware cost (TrendForce). Microsoft said about $25 billion of its record capital budget reflects higher component prices, chiefly memory (Tom’s Hardware). Even Zscaler, a security company, expects higher capex because of memory prices (Zscaler Q4 call).

The sellers posted record results. Micron’s September quarter revenue was $54.2 billion, up 379% from a year earlier, and it has agreed sales for most of its 2027 HBM supply at higher prices (Micron). SK hynix earned a 76% operating margin in its June quarter (SK hynix 6-K). TrendForce expects contract prices to rise again this quarter: 10% to 15% for ordinary DRAM and 15% to 20% for NAND flash (TrendForce).

Why this page ranks on sales, not earnings

Memory earnings swing harder than almost any other industry’s, which makes trailing P/E misleading at both ends. In August 2018, near the 2018 memory-price peak, Micron’s P/E was 5.2, the lowest reading of its previous five years. The stock then fell about a third as memory prices dropped. In June 2023, near the bottom, its P/E sat at the 98th percentile because earnings had collapsed; the stock rose about 75% over the next eleven months.

EV/Sales still has a reading when a company loses money, and it moves less than P/E, so it is the headline here. It does not time cycles either: revenue swells at a peak too. Switch the chart to P/E to see the gap. When we wrote this (October 5, 2026), Micron sat at the 59th percentile of its own P/E range and the 96th on EV/Sales.

Our figures use filed quarters only. Micron’s record September quarter enters when its annual report is filed, which on our estimate pulls its EV/Sales toward 9 times, still around the 86th percentile.

What the market already priced

When we wrote this, every ranked member except SanDisk sat above the 80th percentile of its own five-year EV/Sales range. Everpure and NetApp were at the top of theirs, FormFactor at the 97th, Seagate at the 92nd and Rambus at the 90th. SanDisk, at the 52nd, is ranked against less than a year of history and its figures are under data review. Western Digital’s 84th mixes in years when it still owned the flash business, so read it loosely.

Forward multiples tell a different story. On analysts’ estimates, Micron traded near 6 times forward earnings and SK hynix and Samsung near 4. Investors are paying up for revenue already reported while doubting that record profits last. SK hynix and Samsung file in Korea, so they show a current value without a rank; US investors have been able to buy SK hynix on Nasdaq (SKHY) since July 2026.

How memory cycles end

Shortages end when supply catches up, and supply is coming. Micron plans to spend about $25 billion on capex in the first half of fiscal 2027 alone, nearly its full fiscal 2026 total, with its first new Idaho fab starting output in mid-2027 (Micron prepared remarks). SK hynix plans its first Yongin cleanroom in February 2027, and China’s CXMT and YMTC target new output in 2027 (KrASIA). On October 2, Toshiba said it would double hard-drive capacity, and Seagate and Western Digital fell about 11% that day (ZeroHedge).

Demand outside AI is already giving way. Gartner expects PC shipments to fall 10.4% and smartphones 8.4% in 2026 as memory costs raise prices (Gartner). The last downturn shows the stakes: Micron’s revenue halved in fiscal 2023 and it lost $5.8 billion (Micron). Micron argues this time is different, expecting conditions to be “much tighter in calendar 2027 and 2028 than they were in 2026” (Micron prepared remarks). Deloitte expects major new supply only in 2029 or 2030 (Deloitte). The AI Infrastructure theme tracks the chip and equipment buyers on the other side of these orders, and the AI Power theme the electricity they run on.

What would change the view

  • Contract prices for ordinary DRAM or NAND falling quarter over quarter.
  • Large cloud companies cutting 2027 data-center spending plans.
  • HBM supply deals for 2028 signed at lower prices than 2027.
  • CXMT, YMTC or the new Korean and Idaho fabs shipping volume ahead of schedule.
  • Toshiba’s added hard-drive capacity pulling drive prices down.

Questions

What are AI memory stocks?

Companies that make or sell the memory and storage AI data centers need: DRAM and high-bandwidth memory (HBM), NAND flash, hard drives, storage systems, and the test and interface chips memory makers buy. This page groups 10 of them into 5 layers and shows where each one’s valuation sits against its own five-year history.

Is this a list of stocks to buy?

No. It is a research page. The basket index is a hypothetical equal-weight average that shows how the group moved; it is not a fund, a model portfolio or a recommendation. Nothing on this page is investment advice.

What is HBM and why does AI need it?

High-bandwidth memory is DRAM stacked in layers and placed right beside an AI chip, so the chip can read model data fast enough to stay busy. Only SK hynix, Samsung and Micron make it, and each new generation of AI accelerator carries more of it.

Why rank memory stocks on EV/Sales instead of P/E?

Memory profits swing from losses to records within a few years. P/E looks lowest near the top of a cycle, when earnings are at a peak, and is missing or very high near the bottom. EV/Sales keeps a reading through losses and moves less. The chart can switch to P/E.

Why are SK hynix and Samsung not ranked?

Both report in Korea rather than filing quarterly reports with the SEC, so we cannot build their weekly five-year history. They show a current value from a market-data snapshot instead. SK hynix has also traded on Nasdaq as SKHY since July 2026.

Why compare the basket with SMH?

The closest fund, Roundhill’s Memory ETF (DRAM), only began trading in April 2026, which is too short for a multi-year comparison. SMH is the most traded US semiconductor ETF and holds Micron, so the gap shows how memory did against the chip sector as a whole.

How do memory cycles usually end?

With new supply. High prices fund new factories, which take two to three years to build, and prices fall once they ship. In the last downturn Micron’s revenue halved in fiscal 2023. The current question is whether AI demand absorbs the fabs opening from 2027.

Computed in-house from SEC EDGAR filings and weekly Yahoo Finance prices · updated Oct 19, 2026.

Research, not investment advice. The basket is a hypothetical index, not a fund.