Rapidus courts US AI customers ahead of planned 2H27 2nm mass production

Rapidus's 2H27 2nm mass production target is less a technology milestone than a geopolitical positioning play—the third pole in America's effort to diversify AI compute supply chains away from Taiwan, China. At the technical level, 2nm mandates High-NA EUV and GAA transistor architecture; without securing ASML equipment capacity by 2026, yield ramp will stretch beyond 18 months. More critically, the AI bottleneck has shifted to advanced packaging (CoWoS) and HBM bandwidth. A 2nm logic die without a mature back-end ecosystem offers limited value to NVIDIA or AMD. The "US customers first" strategy is a precise arbitrage of CHIPS Act subsidies, but it also locks Rapidus's capacity allocation to export-control lists. Any technology spillover toward China becomes an immediate compliance trigger. On competition: TSMC's N2 has already locked Apple and NVIDIA for 2025-26; Samsung's 2nm GAA secured Qualcomm's flagship SoC; Intel 18A is poaching foundry work. Rapidus's real window isn't substitution—it's the incremental long tail of AI inference ASICs. 12-24 month call: expect 1-2 design wins by Q4 2026, but equipment delivery delays and talent gaps will compress the ramp. Q2 2027 is the kill line—if yield hasn't crossed 85%, the 2nm narrative downgrades from strategic backup to policy exhibit.