Why Is the AI Race Creating New Chip Frenemies and Where Does Taiwan Fit in the Power Shift?

Q1. Why are cloud providers building custom silicon now?
Because the scale of AI demand has outgrown any single architecture. AWS says Trainium3 offers over four times the performance and 40 percent better energy efficiency, while still expanding orders for general purpose GPUs. The future is a mix of both systems, not a replacement.

Q2. How is this changing Taiwan’s supply chain outlook?
Whether a rack is filled with GPUs or custom ASICs, most of the chips are produced by TSMC and most of the servers are assembled by Taiwan’s manufacturers. A dual track does not reduce demand. It widens it. Companies across cooling, substrates, and server integration are benefiting from diversified orders.

Q3. What is the biggest new constraint?
Energy. Jensen Huang warned that no country can scale chip fabs, supercomputers, or AI clusters without enough electricity. Memory shortages are also accelerating. Some types of standard memory have jumped 50 to 100 percent in price within a week as production shifts toward HBM for AI systems.

If you want the full analysis, listen to today’s episode.