🧠 Food for thought

Implications, context, and why it matters.

AI profits are increasingly driven by a shift in customer mix

• More AI-related profit comes from a new revenue mix. High-performance computing (HPC), a category that includes AI chips and data center processors, now brings in over half of quarterly revenue, while smartphones contribute less 1.
• This AI-leaning lineup earns higher margins than past cycles. Gross margins run above 62% and operating margins sit above 54%, rather than matching TSMC’s historical cyclical average 2.
• TSMC’s process lead supports steep pricing. A 3-nanometer wafer costs about US$25,000 to US$27,000, and next-generation 2-nanometer wafers cost over US$30,000 1.

The AI boom runs through constrained advanced manufacturing and packaging capacity

• TSMC sits at the center of the AI buildout and limits supply. Its business has moved away from consumer spending swings toward infrastructure demand from hyperscalers, large cloud computing companies such as Amazon Web Services and Microsoft Azure, plus governments 2.
• Advanced packaging faces tight supply. Nvidia has booked 800,000 to 850,000 wafers for 2026, which tops half of TSMC’s projected CoWoS capacity by the end of 2026. CoWoS is an advanced chip-packaging technology used to connect processors and memory more efficiently for AI workloads 1.
• Global expansion lowers concentration risk yet weighs on profit. TSMC said overseas fabs, or chip manufacturing plants, will cut gross margin by about 2% to 3% over the next several years due to higher operating costs and ramp inefficiencies 1.