AI Capex Supercycle Pulls KR Memory and TW Foundry Into Sync

Applied Materials posted record Q2 revenue of $7.91B and raised its 2026 WFE outlook to 30%+; TSMC is now building 18 fabs globally; Quanta's April revenue jumped 120.7% YoY to NT$339.9B; and Korean DDR5 16Gb spot held at $40.7 — four signals pointing to a synchronized AI-capex supercycle. The pull hits Korea memory and Taiwan foundry/ODM simultaneously, with Korea's April semiconductor export index up 56.0% YoY. Together the data argues this cycle is broadening from pure compute into memory and ODM assembly.

On the equipment side, Applied Materials' record $7.91B Q2 print and its raised 30%+ growth call on 2026 WFE put fresh wind behind the capex thesis ^[1]. TSMC (2330) is now building 18 fabs simultaneously — doubling its prior global expansion pace — with the Arizona Fab 21 cluster alone budgeted at $165B to produce 100M processors for Apple ^[2]. In Korea, Daeduck Electronics reported Q1 operating profit of KRW51.3B, an 8x year-on-year jump, and approved KRW213B in capex tied to AI-server and FC-BGA substrate orders ^[3] — a downstream confirmation that substrate and PCB suppliers are seeing the same demand pull as front-end tool makers. USTR Greer's confirmation that chip export controls were not discussed at the Geneva US–China talks also lifted a near-term policy overhang for Korean memory names ^[4].

On pricing and demand-proxy, DDR5 16Gb (2Gx8) 4800/5600 spot held at $40.7 on May 15, and TrendForce flagged mobile DRAM Q2 contract prices surging 70–83% — an unusually sharp single-quarter step ^[5][6]. The clearest cross-border signal sits inside SK Hynix's (000660) customer neighborhood: Quanta (2382), a key AI-server ODM and downstream pull-through for NVIDIA's GPU stack, printed April revenue of NT$339.9B, +120.7% YoY ^[7]. Hon Hai (2317) guided AI rack shipments to double in 2026 ^[2]. Inside the inventory data, registered DSI of 34.6 days in Q1 2026 versus the 80–110 day band typical for TW ODM/component peers, which suggests Korean-side inventories are not bloated — the bottleneck is more likely order intake than stock overhang.