Semicon Briefing

Samsung is starting test production at its Taylor fab and could challenge TSMC's 2nm leadership as the second company after Intel (18A). The move signals a genuine three-way constellation in sub-3nm foundry competition, fundamentally pressuring TSMC's pricing power and customer lock-in.

Samsung's $17 billion fab in Taylor, Texas is entering the equipment installation phase and is advertising 183 technical positions – a concrete milestone for U.S. manufacturing capacity beyond TSMC. The parallel to the TSMC Arizona cluster shows that the CHIPS Act strategy is actually pulling production facilities to the U.S.

Washington shifted to a performance-based threshold approach for chip export controls, but Chinese buyers continue to avoid NVIDIA H200 – apparently due to loss of trust and preference for domestic alternatives. This withdrawal undermines NVIDIA's near-term China revenue expectations despite formal relaxation.

Prediction markets value NVIDIA's status as the world's largest company as of April 30 at 97% – at $3 million volume. This reflects the consensus that AI chip demand experiences no structural dent in the near term, despite Iran conflict and oil price volatility.

TSMC is planning a SoIC monthly capacity of 10,000–15,000 wafers in 2026 for NVIDIA's upcoming Rubin Ultra and Feynman generations; Besi, Applied Materials, and TEL are considered primary beneficiaries. The capacity planning shows that advanced packaging is becoming the new bottleneck in the AI chip stack.

ams-OSRAM reports measurable operational synergies from the 2020 merger for the first time, after the company divested its sensors division (to Infineon) and lighting division (to Ushio). The strategic refocus on digital photonics and automotive lighting systems shows that the restructuring process is entering the revenue phase.