Sam Altman warns Samsung of short-term AI demand volatility

Altman's remark to Samsung is less a demand warning than a signal about order-aggregation distortion. When NVIDIA, hyperscalers, and OEMs each independently forecast HBM requirements for the same GPU platform, the resulting "demand" is an accounting artifact, not a physical constraint. The 2024-25 HBM tightness was partially manufactured by this double-counting. The technical consequence is asymmetric. Samsung's HBM3E qualification on NVIDIA's stack trails SK Hynix by roughly two quarters, eroding its leverage in the HBM4 transition. A near-term order pullback would stress the capex recovery cycle at Pyeongtaek P3 and Hwaseong P4. SK Hynix has already locked in the majority of NVIDIA's 2025-26 HBM allocation, while Micron is positioning through AMD's MI400 program—leaving Samsung squeezed in the middle. Geopolitically, Samsung's Xi'an NAND fab represents roughly 20% of global capacity. The structural tension between China-based production and US-based AI customers is intensifying under tightening export controls. Altman's call for "greater supply-chain investment" is, in practice, a request to shift advanced memory capacity toward the US or Japan—significantly raising depreciation costs and forcing a margin-model restructuring. Within 18 months, HBM4 yield ramp is the critical inflection. If Samsung misses NVIDIA's 2025 Q4 qualification window, its HBM share could slide below 25%, handing SK Hynix a de facto oligopoly pricing position above 60% and relegating Samsung to consumer and niche segments.