TSMC commits $29.4B to AI chip capacity, launches Sony sensor venture
What's the deal? Taiwan Semiconductor Manufacturing Co. (TSMC), the world's largest contract chipmaker, will invest $29.4 billion to expand advanced process and packaging production for artificial intelligence (AI) semiconductors. Its board approved the capital budget on August 11, according to Taiwanese media. The money will fund new factories, upgraded processes, and expanded packaging capacity.
Why now? TSMC says demand from AI and high-performance computing customers remains strong, with long-term signals from clients and cloud service providers. In July 2026, it raised its annual equipment investment forecast to $60 billion–$64 billion, up from $52 billion–$56 billion.
What's the endgame? A key focus is advanced packaging. TSMC's "Chip on Wafer on Substrate" (CoWoS) technology, which links graphics processing units and high-bandwidth memory in a single package, is critical for building AI accelerators.
The company currently produces CoWoS at 5.5x reticle size. By 2028 it plans to mass-produce a 14x version integrating roughly 10 large computing dies and 20 HBM stacks per package, and by 2029 aims to reach a 40x System on Wafer design.
The SonyDealroom has a profile for this one. Try Dealroom → angle: TSMC will also form a joint venture with Japan's Sony to develop and produce next-generation image sensors. The two will build Advanced Vision Semiconductor Manufacturing in Kumamoto Prefecture, Japan, with mass production starting in 2029.
Total investment reaches ¥747 billion, with Sony contributing ¥465 billion and TSMC ¥282 billion. Sony will lead sensor technology, planning, and design; TSMC will supply the manufacturing processes. The pair signed a memorandum of understanding in May 2026 and floated expanding into physical AI, including cars and robots.
The signal: TSMC's simultaneous bets on advanced processes and packaging reflect its confidence that AI orders will keep flowing. By pairing chip capacity with a sensor venture, it is positioning itself across both the compute and perception layers of the AI hardware stack.
Read more: Chosun
Image credit: f097653195037
