Research Archive

Bottom line. The memory cycle re-accelerated post-Q1-2026 earnings. Hyperscaler capex updates reaffirmed multi-year HBM tightness and pulled forward the inflection we had flagged in the April monthly review. We upgrade Wave 2 (Memory) back to active status alongside Waves 3, 4, and 7. Conviction holds at A for MU and SNDK; A- for SK Hynix (SKHF) on listing thinness; B+ for Samsung-electronics-equivalent exposure given the diversification drag.

What changed since the April note. April's call (monthly review, 2026-04-15) graded Memory at A on supply-side data but flagged Q1 prints as the binary gate. Q1 prints cleared the bar: HBM3E pricing held at the high end of guidance, HBM4 sampling commentary advanced, and NAND ASPs stabilized faster than the bear case modeled. The "HBM commoditizes in 2026" risk we explicitly tracked has been pushed out — current evidence points to tight HBM through 2027 with the commoditization window opening in 2028 as new fab capacity comes online.

Demand side. Hyperscaler capex guidance for 2026 is up vs. the January reset; AI training cluster builds are HBM-bound at the architectural level (one H100/H200/Blackwell SKU consumes 5–8x the HBM of a 2023-vintage A100-class node). NVIDIA's roadmap shipping cadence multiplies per-rack HBM content faster than wafer capacity is being added. The single number that matters: HBM wafer share at the leading-edge memory fabs is rising every quarter, displacing commodity DRAM — and the displacement is the bull case for ASP on the commodity side too.

Positioning — Micron (MU). Last 222.53→704.68 reference range (NVDA→MU intraday print 2026-05-19) understates how distorted the relative-performance picture has become; MU is the cleanest HBM-leverage US-listed name and Q1 print validated the capex / mix-shift story. Conviction A. Catalyst watch: HBM4 qualification announcements, Hynix relative pricing disclosures.

Positioning — SanDisk (SNDK). Post-split standalone NAND-pure-play. The bear thesis (NAND ASPs collapse in over-supply) has not played out; ASPs stabilized in Q1 and the SSD demand pull from enterprise / AI training datasets is sustaining throughput. Conviction A. Catalyst watch: enterprise SSD bit-shipment data, hyperscaler SSD RFQs.

Positioning — SK Hynix (SKHF). The single largest HBM3E supplier to NVIDIA. The thesis here has always been: own the supplier with the longest qualification lead vs. NVIDIA's roadmap. That has not changed. Listing thinness on US ADR keeps conviction at A- rather than A. Catalyst watch: HBM4 customer-qualification cadence, KOSPI primary-listing fundamentals.

Positioning — Samsung (no clean US listing). Conglomerate diversification continues to dilute the HBM story for US investors; we monitor as part of the supply picture but do not carry it as a named conviction position.

CXL transition assessment. The pre-consensus call (Wave 6, "CXL is the next wall once HBM commoditizes") remains intact and the timing window is now 2027–2028 rather than late 2026. ALAB is the cleanest pure-play and printed a single-session +15.9% on 2026-05-19 against the broader market on what we read as a specific accumulation catalyst. The thesis is still pre-consensus by the standard we use (sell-side coverage thin, no mainstream macro narrative). Conviction A-, accumulation-phase rather than catalyst-driven.

What would invalidate this. (1) HBM3E pricing breaks the high-end band over two consecutive quarters; (2) hyperscaler capex guidance is cut more than 10% in a single update; (3) a new-fab capacity announcement pulls HBM commoditization back into 2026; (4) NAND ASPs roll over for two consecutive months on inventory data. None of these are currently signaling.

Data-source note. Live prices in this review (MU 704.68, SNDK 1362.39, ALAB 249.83) were pulled from Finnhub on 2026-05-19. YTD-2026 percentage prints are not included in this note because the current Finnhub tier does not include historical candle data; the master thesis carries the same caveat with a follow-up ticket to resolve the YTD baseline source. We chose to publish the qualitative review on schedule rather than delay it for the YTD column.