← All commentary
2026-08-27

Serenity chases the "legacy legacy" memory bottleneck, opening a stake in Taiwan's ESMT on DDR2/DDR3 price hikes

Serenity, the stock-picker who builds his portfolio by hunting for upstream "chokepoints" — critical suppliers with few substitutes that the whole AI buildout leans on — spent today pushing his long-running memory thesis into an unusual corner: the oldest, cheapest DRAM chips still in production.

A new name: ESMT. His main disclosure was a fresh position in ESMT (Taiwan ticker 3006), a roughly $2.5B fabless memory-chip designer that buys wafer capacity from foundry PSMC and specializes in older DDR2 and DDR3 DRAM. Serenity laid out a striking earnings ramp: he cited a jump in monthly net income from around $16M in January to roughly $110M in July (a company-reported figure, with intermediate months estimated from quarterly totals), which on an annualized basis he pegs at under a 2x price-to-earnings ratio. He argues the driver isn't just cheap old inventory being sold into a hot market, but a widening gap between wafer costs and rising legacy DRAM prices — and that supply stays tight as legacy makers like Winbond exit certain DDR2 lines. He explicitly likened the setup to the earnings inflection at Sandisk (SNDK), the US NAND-flash maker he treats as the template for his "supercycle" thesis: a structural, AI-driven memory shortage that lets producers hike prices repeatedly. He also mentioned starting a position in Etron, another Taiwanese legacy-memory name, framing this as a "legacy legacy" cascade down the memory stack.

NVIDIA's read-through. Serenity tied the move to NVIDIA (NVDA) earnings, which he uses as a demand signal rather than a holding — he front-runs NVIDIA's suppliers rather than owning the crowded mega-cap itself. He noted NVIDIA's forward purchase commitments swelled dramatically, driven heavily by memory procurement, and pointed to management's comments about extreme memory pricing as confirmation that high-end demand is structural, not a cycle blip.

Trimming the winners. Notably, he said he has scaled back much of the memory book that made his first half of 2026 — a basket that spanned Micron (MU), the US DRAM/HBM maker he has called a potential "next NVIDIA" and holds at a 10% weight; Sandisk; controller near-duopolist Silicon Motion (SIMO), his backdoor NAND play; plus Phison, Nanya, Macronix and Winbond. His logic: names like Micron have already re-rated hard (he cited a move from roughly $300 to over $1,000), so the biggest "price discovery" is behind them and it's now a waiting game for operating income to catch up. The exceptions he's keeping are his Samsung and SK Hynix exposure — expressed through the South Korea ETF EWY, which he treats as a concentrated, leveraged bet on the two Korean memory giants via long-dated call options.

The through-line: with the front-end memory rally maturing, Serenity is rotating toward the least-followed, deepest-value end of the shortage — and openly asked whether the market simply hasn't noticed ESMT yet, or whether he's early on a thesis with a catch he can't see.

*Derived commentary, not investment advice.*

Generated by the claude-opus-4-8 pipeline. Derived content; not investment advice.