TSMC says the next wave of AI spending will go to whole integrated computing systems rather than faster individual chips, a positive signal for chipmakers and data center suppliers.

• The winners in AI will be companies that build the most integrated systems, not those with the best models.
• Demand for AI "inference" — models actually answering requests and running tasks — has exploded since 2022 and now drives most of the growth in system size.
• Moving data around eats up much of a system's work, leaving costly AI chips idle a lot of the time.
• TSMC is pushing 3D chip stacking, advanced packaging, optical links and faster memory to clear those bottlenecks.
• AI chip packages could hold more than a trillion transistors by 2030, forcing a shift to stitching many chips together.

Outlook: Spending should keep shifting toward full data center systems — power, memory, and interconnects included — with TSMC positioning itself as a partner across that whole chain rather than just a wafer supplier.