TSMC's Capacity Gap Is Structural. Plan Around It.

The Warning Is Not Subtle

At TSMC's annual shareholder meeting on June 4, CEO C.C. Wei stated directly that the company will not meet demands from American customers even after U.S. manufacturing facilities become operational. That claim deserves space on every infrastructure roadmap through 2028.

This reflects something beyond typical earnings rhetoric. Wei has returned to the same point repeatedly. At the Semiconductor Industry Association awards in November 2025, he reduced it to three words: "Not enough, not enough, still not enough." He also disclosed that demand for advanced nodes runs roughly three times higher than TSMC's productive capacity. The gap shows no signs of narrowing within any timeline he has disclosed publicly.

Revenue tells the same story. TSMC reported May 2026 net revenue of approximately NT$416.98 billion, up 30.1% from May 2025, with cumulative January-through-May revenue of NT$1,961.80 billion, an increase of 30.0% compared to the same period in 2025. The May figure set an all-time monthly record. Growing revenue 30% while constrained on capacity suggests pricing leverage and tight allocation—not demand recovery. The companies buying are the ones who secured relationships early.

What Arizona Actually Delivers

Arizona is real and moving faster, but the timeline creates a hard problem for anyone deploying in 2027.

The first Arizona fab is operational with volume production on 4-nanometer technology since Q4 2024. The second fab's construction is complete, with tool move-in planned for 2026 and volume production on 3-nanometer technology expected in the second half of 2027. As part of the U.S.-Taiwan trade framework, TSMC envisions building at least four more U.S. chipmaking plants on top of six already planned—a total of 10 facilities with announced investments of $165 billion plus roughly $100 billion in additional capital.

The number is large. It does not solve the immediate problem. Wei stated that the company will not be able to meet demands from its American customers even after its manufacturing facilities get operational in the U.S. Arizona is geopolitical insurance and multi-year capacity. It is not relief for 2026 or 2027 procurement.

Cost structure between Arizona and Taiwan is closer than conventional wisdom holds. Well over two-thirds of wafer cost comes from equipment, narrowing the overall wafer cost difference between Arizona and Taiwan to just under 10%. But the construct masks real complications. Wafers manufactured at Fab 21 in Arizona are currently transported to Taiwan for back-end processes—dicing, testing, and packaging—with some completed chips then returning to the U.S. market, a logistics loop that adds cost and complexity. TSMC plans U.S. packaging facilities to close that gap, but they remain inoperative.

TSMC announced capital expenditures expected to reach $56 billion in 2026. The capex per 1,000 wafers per month at 2nm is substantially higher than at 3nm, and the cost for 1.4nm will be higher still. Each successive node demands more to build and longer to ramp. The financial limit on expansion velocity is rising as demand accelerates.

Whose Hand Wins

TSMC long-term contracts with mainstream companies like Nvidia and Apple consume a large portion of total output, leaving others facing prolonged waits for their semiconductor needs. Nvidia and Broadcom both requested additional production capacity and were both told TSMC could not offer what they wanted. If those two are being rationed, everyone else is operating in genuinely scarce conditions.

TSMC is the sole supplier of advanced processes with meaningful market adoption. Competitors like Intel Foundry and Samsung have alternatives, but throughout the AI cycle, only TSMC's advanced nodes have attracted broad design-in activity.

EUV tool availability, power infrastructure, and access to qualified workforce all act as limiting factors beyond raw fab capacity. None resolve quickly. ASML's EUV tools have their own multi-year backlog. Arizona workforce constraints have been documented since construction began.

Despite the supply-demand gap, Wei added that TSMC will refrain from sudden price increases to maintain stable business relationships. That posture protects incumbents with negotiated rates. It does not help new entrants trying to access the queue at spot prices.

The Planning Implication

This is not a forecasting miss or unexpected shock. Wei told shareholders that AI chip demand will outpace supply for years to come, even as TSMC races to expand manufacturing. The company expressed strong confidence in AI-driven semiconductor demand, expecting full-year 2026 revenue growth to exceed 30% in USD terms.

For infrastructure leaders: TSMC is telling you the constraint holds. Arizona delivers geopolitical diversification but does not materially loosen allocation before 2028 at the earliest. Any 2027 infrastructure plan assuming lead-time compression rests on a false foundation. Companies that locked multi-year agreements in 2024 and 2025 hold valuable positions. Those who did not are competing in a tightening spot market against hyperscalers designed for exactly this squeeze.

One speculation worth naming: if Samsung Foundry or Intel Foundry Services can prove competitive 3nm yields, the current moment offers the most favorable conditions for design-win conversions either has faced. TSMC's constrained allocations are the best business development advantage they possess.

What to Watch

1. TSMC Q2 2026 earnings call (late July): Check for updates on Arizona Fab 2 tool installation and whether the second-half 2027 volume production target holds. Any delay extends the constraint window.

2. Samsung and Intel Foundry design-win announcements: A major AI chip designer moving even a secondary product line to an alternative fab signals the first real release valve on TSMC allocation.

3. Multi-year capacity contract disclosures: Watch for AI chip companies announcing long-term wafer supply agreements. Each lock-in removes allocation blocks from the spot market.

4. ASML EUV shipment data: TSMC's speed at 2nm and 1.4nm depends on EUV tool delivery. ASML's quarterly numbers are the leading indicator for when TSMC's next capacity tranche materializes.

5. U.S. CHIPS Act disbursement pace: TSMC Arizona received $6.6 billion in direct funding from the U.S. Department of Commerce under the CHIPS Act. Monitor for conditions or political pressure that could affect capital timing on remaining Arizona fabs.