The Real AI Bottleneck Isn't Talent or Power
It's TSMC

Industry | 5 min read

When policymakers talk about AI risk, they often mention energy usage, safety, or export controls.

But the most important constraint today is much simpler:

There aren't enough advanced chips.

And most of those chips are manufactured by one company: Taiwan Semiconductor Manufacturing Company (TSMC).

Silicon, Not Electricity

Executives from Amazon, Microsoft, Google, and Meta all say they are supply-constrained.
TSMC's own CEO, C.C. Wei, acknowledged that "silicon from TSMC is a bottleneck."

Power plants are being built. Cooling systems are being installed. Data centers are expanding.

But advanced AI chips take years to manufacture capacity for.

Building a new semiconductor fabrication plant (a "fab") takes two to three years before it meaningfully contributes to supply. Decisions made today determine output in 2028 or 2029.

This means AI growth over the next five years depends heavily on choices being made right now in Taiwan.

Why TSMC Is Cautious

From the outside, it may seem obvious: if demand is strong, just build more.

But semiconductor manufacturing is uniquely risky. Almost all costs are upfront capital expenditure. If demand slows, those investments cannot easily be scaled down. Excess capacity can depress pricing power for years.

TSMC plans to spend roughly $52–$56 billion this year on capex, a record for the company. But even that may not fully close the demand gap.

The foundry is rationally cautious. It does not want to "hold the bag" if an AI bubble pops.

Here is the structural tension:

• Cloud companies risk forgone revenue if chips are scarce.
• TSMC risks massive depreciation losses if it overbuilds.

Both sides are acting rationally, but their risk calculations differ.

Why This Matters Beyond Tech

This concentration creates two systemic vulnerabilities:

• Economic bottleneck risk, if AI demand continues to rise and supply lags, global productivity gains could be throttled by a single industrial constraint.
• Geopolitical concentration risk, advanced chip manufacturing is geographically concentrated in Taiwan.

The AI debate often focuses on algorithms.
The more durable issue is manufacturing concentration.

If the world believes AI will meaningfully shape productivity, defense systems, logistics, and medicine, then reliance on a single advanced foundry becomes a strategic issue, not merely a commercial one.

The lesson is uncomfortable but clear:

AI's future is not just a software competition.
It is a manufacturing competition.

And right now, one company sits at the center of it.