What's happening
According to TrendForce reporting dated June 22, 2026, TSMC has reduced monthly wafer starts at Fab 15A — its primary 28nm fabrication site — by more than 25% since the beginning of 2026. Output has declined from approximately 200,000 wafers per month earlier this year to roughly 150,000 wafers per month as of June 2026. The reduction represents a deliberate reallocation of manufacturing resources rather than a demand-driven contraction, as TSMC redirects capacity toward its 2nm and 3nm process nodes.
The strategic rationale centers on accelerating production of advanced-node chips targeting AI and high-performance computing applications. TSMC's leading-edge process technologies — including the 3nm node currently in high-volume production and the 2nm node in ramp — serve the company's highest-value customers. TSMC's ticker profile identifies NVIDIA and AMD among its key clients, both of which rely on advanced nodes for their data center GPU and accelerator product lines.
Why it matters for markets
The capacity shift carries direct margin implications for TSMC. Leading-edge nodes command significantly higher average selling prices than mature 28nm processes, meaning a reallocation of fab capacity toward 2nm and 3nm production tends to improve revenue per wafer. TSMC currently carries a market capitalization of approximately $2.43 trillion and a price-to-earnings ratio of 40.1, reflecting investor expectations of sustained growth in advanced-node demand. Any acceleration in leading-edge utilization rates could reinforce the revenue trajectory that underpins that valuation.
For customers dependent on 28nm capacity, the output reduction introduces supply considerations. The 28nm node remains widely used across consumer electronics, automotive, industrial, and communications applications. A sustained reduction of 50,000 wafer starts per month at Fab 15A represents a material tightening of available mature-node capacity from one of the industry's largest foundries. TrendForce notes that the displacement may create spillover demand benefiting alternative foundries including UMC and Vanguard, which operate competing mature-node capacity.
For NVIDIA and AMD, whose AI accelerator and HPC product lines are manufactured on TSMC's most advanced nodes, the capacity reallocation is directionally aligned with their demand profiles. NVIDIA — with a market capitalization of $5.05 trillion and revenue of $253.49 billion — and AMD — with revenue of $37.45 billion — both depend on TSMC's leading-edge processes for their highest-margin data center products. Greater fab capacity dedicated to advanced nodes could support the production volumes required to meet AI infrastructure demand.
Sectors and assets to watch
The primary ticker directly affected is TSM (Taiwan Semiconductor Manufacturing Company), as the operational and financial implications of the capacity shift flow through TSMC's own utilization rates, revenue mix, and margin structure. NVDA (NVIDIA Corporation) and AMD (Advanced Micro Devices) are key TSMC customers whose most strategically important product lines — NVIDIA's data center GPUs and AMD's Instinct accelerators and EPYC processors — are manufactured on the advanced nodes receiving expanded capacity. Any change in TSMC's ability to fulfill leading-edge wafer demand has direct read-through to the production schedules of both companies.
In the mature-node segment, the output reduction at Fab 15A positions UMC and Vanguard as potential beneficiaries of redirected 28nm demand, as customers requiring mature-node capacity may need to qualify alternative suppliers. Neither company is a primary subject of this story, but both have been identified by TrendForce as likely recipients of spillover demand resulting from TSMC's strategic reallocation.
What to watch next
Key developments to monitor include TSMC's official capacity utilization disclosures in upcoming quarterly earnings reports, which will indicate whether the Fab 15A reduction is deepening or stabilizing, and whether advanced-node utilization rates are rising in parallel. Observers should also track whether customers reliant on 28nm supply — particularly in automotive, industrial, and communications sectors — report lead-time extensions or pricing changes consistent with tightening mature-node availability. Statements from NVIDIA and AMD regarding wafer allocation, production ramp timelines for next-generation AI accelerators, and any supply constraints on advanced-node capacity will provide additional signals about how the reallocation is affecting the broader AI chip supply chain.