TSMC's North America chief receives roughly ten calls per day from customers asking about capacity, a level of demand pressure he says he has not seen as routine practice. His account, reported by BigGo Finance, gives a rare first-hand measure of how hard customers are pushing for wafer supply. Analysts cited alongside his remarks warn that the pace of AI expansion could run into a manufacturing bottleneck.

The number matters because it comes from the person who handles TSMC's largest customer base. North America accounts for the bulk of TSMC revenue, and it is where most AI chip designers sit. Ten capacity inquiries a day is not a forecast or a projection; it is a direct observation of customer behavior at the top of the supply chain.

What is driving the calls?

AI is the short answer. Training and inference chips for large models require advanced-node wafers in volume, and TSMC is the dominant manufacturer at those nodes. Every customer building an AI accelerator, a networking processor, or a custom silicon part needs TSMC allocation to ship product.

The call volume signals that customers do not assume supply will be there when they need it. Instead, they are working the phones to secure lines, confirm schedules, and lock in future output. That behavior typically appears when buyers expect demand to outrun capacity.

The takeaway: the AI boom has moved from forecast to procurement pressure, and it is landing directly on TSMC's sales organization.

Where could the bottleneck appear?

The analysts cited in the report frame the risk in manufacturing terms. AI expansion could hit a ceiling not because demand fades, but because the physical supply chain cannot expand fast enough.

Advanced logic is one pressure point. Leading-edge nodes already run near full utilization, and adding capacity means new tools, new cleanroom space, and long equipment lead times. Advanced packaging is a second pressure point; AI chips depend on high-density packaging to link logic and memory, and that segment has less installed capacity than wafer fabrication.

Neither TSMC nor the analysts quoted specific capacity figures or dates for when a bottleneck might bind. The warning is directional rather than scheduled, and it rests on the gap between demand growth rates and the pace at which manufacturing capacity can be added.

The takeaway: the constraint story is about expansion speed, not current output; the question is whether fabs and packaging lines can scale before demand collides with their limits.

Why the pressure concentrates on one supplier

TSMC's position amplifies the effect of these calls. The company manufactures the majority of the world's advanced logic, and AI chip designers have few alternatives at the leading edge. When ten customers a day call one supplier, allocation decisions at that supplier effectively set the shipping schedules of the AI industry.

That concentration also explains the analysts' concern. A bottleneck at a single dominant manufacturer does not spread pain evenly; it decides which products reach market and which slip. Customers with stronger relationships and larger commitments tend to secure supply first, leaving smaller players to wait.

The takeaway: capacity calls at TSMC are a leading indicator for the whole AI hardware market, because one company's allocation decisions ripple through everyone's product plans.

What comes next

For now, the ten-calls-a-day figure stands as the clearest public signal of supply tightness from inside TSMC. The analysts' bottleneck warning remains a risk scenario rather than a confirmed event, and no capacity shortfall has been declared.

The commercial question for the coming quarters is whether TSMC can convert this demand pressure into expanded output faster than AI orders accumulate — a race that will show up first in allocation decisions, lead times, and pricing at the leading edge.