RAMageddon Stock Digest

Undervalued memory & storage plays · 24 Jun 2026 · personal research, Bangkok
⚠ Not financial advice. Prices/multiples are approximate live snapshots (23–24 Jun 2026) and already reflect the big 2026 run-up. Memory is deeply cyclical — read the caveats at the bottom.
The setup. DDR5 contract prices tripled–quadrupled since Oct 2025. TrendForce: DRAM +90–95% QoQ in Q1'26, +58–63% in Q2'26; NAND +55–60%. The cause is structural, not transient — the big-3 reallocated wafers to high-margin HBM for AI; AI absorbs ~20% of DRAM by end-2026 while capacity grows only ~16%/yr. HBM is sold out through 2027–2028. BofA calls 2026 a "supercycle like the 1990s."

The key insight. The obvious makers already ripped (Micron +273% YTD, SanDisk +623%). But earnings are exploding faster than price, so the producers trade at low forward P/E — the opposite of overpriced. The genuinely expensive corner is the test/equipment names (40–50x). "Undervalued vs the catalyst" points back at the producers on forward math, plus a few overlooked suppliers.

Tier 1 — HBM/DRAM leaders, cheap on FORWARD earnings

SK Hynix000660.KS · Korea
Price ≈ ₩2,555,000Fwd P/E ≈ 6.2xMkt cap ≈ $1.3T

Undisputed HBM leader (~57% share), capacity fully booked through 2027–2028. At ~6x forward earnings it's priced as if the cycle ends tomorrow — yet the supply thesis says it runs 2+ more years. Purest single-name exposure to the AI-memory bottleneck.

Samsung Electronics005930.KS · Korea
Price ≈ ₩310,000Fwd P/E ≈ 5.3xMkt cap ≈ $1.5T

The laggard catching up — trailed Hynix on HBM qualification, so it re-rated later and sits at the cheapest forward multiple here (~5x). Diversification (foundry, phones, displays) is a margin of safety, plus optionality if HBM4 wins more NVIDIA share.

MicronMU · US NASDAQ
Price ≈ $1,079Fwd P/E ≈ 8.9xtrailing 51xMkt cap ≈ $1.2T

Cleanest, most liquid US way to own the cycle. Up ~273% YTD, but at <9x forward the multiple compressed as estimates raced higher. Only Western firm with HBM + DRAM + NAND under one roof. The "obvious" pick that's paradoxically not overpriced on forward math.

Tier 2 — NAND / storage price-inflation

SanDiskSNDK · US NASDAQ
Price ≈ $1,983Fwd P/E ≈ 10.8xtrailing 68xMkt cap ≈ $294B

Pure-play NAND/storage spun out of WD, levered straight to NAND ASPs (+55–75% forecast 2026) and enterprise-SSD demand (+86% QoQ in Q1'26). Extreme momentum (+623% YTD) = high risk, but ~11x forward isn't rich if NAND stays sold out.

Silicon MotionSIMO · US ADR
Price ≈ $317Fwd P/E ≈ 30xtrailing 63xMkt cap ≈ $11B

Leading NAND-flash controller designer — picks-and-shovels for every SSD/UFS module. Rides SSD volume + content growth; BofA target lifted to $450. Risk: extreme NAND prices can crimp module-maker demand. Less of a slam-dunk than the producers.

Tier 3 — Overlooked second-derivative suppliers (less obvious)

Penguin Solutions★ best reasonable-valuation pickPENG · US NASDAQ
Price ≈ $68Fwd P/E ≈ 24xtrailing 98xMkt cap ≈ $3.6B

AI-memory integrator (Integrated Memory, MemoryAI, CXL, AI-factory deployments) sitting in the inference memory bottleneck — Q1'26 sales $343M, FY26 guide raised to high end. Not a household name, so it hasn't been bid to the moon. $108 street target (~31x FY27 EPS $3.51) = ~58% upside. The most genuinely reasonably-valued growth name here.

Photronics★ cheapest multiple on the boardPLAB · US NASDAQ
Price ≈ $32Trailing P/E ≈ 12x · Fwd ≈ 16xMkt cap ≈ $1.9B

Largest independent photomask maker — every new memory node and capacity expansion needs masks. Deepest-value, most overlooked supplier here (low-teens P/E vs 40–50x for other equipment). Lower beta to the mania, but a steady toll-taker on the capex wave Nanya/Winbond/Micron are now unleashing.

Tier 4 — Taiwan momentum (high beta, already extended)

Nanya 2408.TW — fwd ~7.8x, revenue +649%, repeated limit-up (NT$454). DDR4/legacy-DRAM squeeze beneficiary.  ·  Winbond 2344.TW — fwd ~7.2x (NT$211), niche DRAM/flash; MS target NT$100→222. Both cheap on forward but have run vertically — treat as trades, not value entries.

Avoid as "value" right now (already richly re-rated)

FormFactor (FORM) fwd ~51x · Lam 47x · Applied 36x · KLA 49x · ACM Research 44x · Onto 32x · Camtek 38x · Teradyne 45x · Advantest very high — good businesses, none "undervalued." Aehr (AEHR) fwd ~688x and Netlist (NLST) no earnings ($2.68, an IP-litigation lottery ticket) — speculative, not value.

How to read this

  1. Cyclicality / value trap: memory's lowest P/Es usually appear at the cycle peak, right before earnings roll over. The bull case rests entirely on the shortage persisting into 2027–2028. If AI capex blinks or capacity catches up, these de-rate hard and fast.
  2. The run already happened: most of the $5→$40 DDR5 move is in the rear-view. You're betting the next leg (HBM4, sustained tightness), not the move that already occurred.
  3. My ranking for undervalued + actionable + not-obvious: PENG and PLAB risk-adjusted; SK Hynix / Samsung for pure cyclical torque at single-digit forward multiples if you can stomach volatility; MU as the liquid US proxy.
  4. Access from Bangkok: US names trade via any international broker (IBKR etc.). Korea/Taiwan need KRX/TWSE access or use ADRs. The Roundhill Memory ETF (DRAM) bundles Micron/SanDisk/Samsung/SK Hynix in one US ticker if you'd rather own the basket.

Multiples move daily — re-check before acting. Snapshot data via yfinance, 23–24 Jun 2026. Sources: TrendForce, BofA, Morgan Stanley, CNBC, company filings.