Strategic overview · Demo data · Not for investment use
~92% of sub-5nm global wafer capacity located in Taiwan. No viable alternative source at scale within 5 years. Structural single-point-of-failure for AI chip supply. [est.]
All NVIDIA Hopper/Blackwell and AMD MI300 GPUs require CoWoS-L advanced packaging exclusively from TSMC. No alternative packaging supplier for these configurations at scale. [est.]
SK Hynix supplies ~60% of HBM3e. Samsung and Micron ramping but constrained. NVIDIA H100/H200/B200 all depend on SK Hynix HBM3e. [est.]
Apple consumes ~45% of TSMC N3 capacity. NVIDIA Blackwell on N4P. AMD, MediaTek, Qualcomm competing for N2 allocation. Structural over-subscription risk 2025–2026. [est.]
Microsoft Commits $80B to AI Datacenter Buildout in 2025
Intel 18A Process Node Enters Risk Production Phase
NVIDIA Blackwell GB200 NVL72 Shipments Commence to Hyperscalers
SK Hynix Reports HBM3e Capacity Fully Allocated Through 2025