What's happening
Taiwan Semiconductor Manufacturing Company (TSM) has reduced monthly wafer starts at Fab 15A — its primary 28nm production site — by more than 25% since the beginning of 2026, cutting output from approximately 200,000 wafers per month to around 150,000. The freed capacity is being redirected toward advanced 2nm and 3nm process nodes, which are in high demand from AI chip designers and high-performance computing customers. TSMC CEO CC Wei has stated he would 'like' to raise prices, citing persistent AI chip shortages that he expects to last years, with the company's advanced fabs running at full capacity. CFO Wendell Huang separately confirmed to the BBC that rising costs are a factor, while ruling out sudden large price increases.
The pricing trajectory for leading-edge nodes is already steep and moving higher. Three-nanometer wafers are currently priced at approximately $20,000 each. Two-nanometer wafers — the next generation now entering volume production — are expected to be priced at least 50% higher, exceeding $30,000 per wafer. TSMC is implementing 5–10% price hikes on advanced nodes below 5nm starting in 2026, with a potential additional 5–10% increase planned for 2027. Market research firm TrendForce has separately reported that a price increase of up to 15% on 3nm wafers is under consideration for the second half of 2026.
Why it matters for markets
The combination of reduced mature-node output and rising advanced-node pricing creates a dual cost and supply pressure for TSMC's largest fabless customers. Nvidia (NVDA), with a market capitalization of $4.82 trillion and annual revenue of $253.49 billion, manufactures its data center GPU lineup — including the A100 and H100 accelerators that underpin its AI computing franchise — on TSMC's most advanced nodes. AMD (AMD), with a market cap of $847.49 billion and revenue of $37.45 billion, similarly relies on TSMC's leading-edge processes for its EPYC server CPUs and Instinct AI accelerators. A 5–10% increase in per-wafer costs at the 3nm node, on top of a base price of approximately $20,000 per wafer, translates directly into higher bill-of-materials costs for any chip produced at that node. If the 2nm transition proceeds at the anticipated price point above $30,000 per wafer — more than 50% above current 3nm pricing — the cost impact on next-generation AI chip programs would be material.
For TSMC itself, the capacity reallocation and price increases are structured to improve margin realization on its most advanced processes. TSMC's ticker profile reflects a P/E ratio of 38.0 against a revenue base reported at $4.10 trillion, and the company's 52-week price range of $220.80 to $476.79 illustrates the degree to which investor valuation has tracked the AI infrastructure buildout. Higher per-wafer revenue on 2nm and 3nm nodes, combined with reduced allocation to lower-margin 28nm production, is consistent with a strategy of margin expansion at the leading edge. The 25%-plus reduction in 28nm wafer starts also signals a structural, rather than cyclical, shift in TSMC's capacity priorities — one that could tighten supply for automotive, industrial, and consumer electronics segments that rely on mature process nodes.
Sectors and assets to watch
The most directly affected publicly traded companies are TSMC (TSM), Nvidia (NVDA), and AMD (AMD). TSMC is the sole manufacturer of the world's most advanced logic chips at scale, and its pricing and capacity decisions set the cost floor for the entire fabless semiconductor industry. Nvidia's data center GPU business — the primary driver of its $253.49 billion revenue base — is manufactured on TSMC's advanced nodes, meaning any sustained increase in wafer costs will affect the cost structure of products such as the H100 and its successors. AMD's EPYC and Instinct product lines, which compete directly in the data center and AI accelerator markets, face the same input cost dynamics. Apple and Qualcomm, both named TSMC customers in the company's ticker profile, are also significant consumers of advanced node capacity, though they are not the primary focus of this story.
Beyond the named AI chip designers, the 25%-plus reduction in 28nm wafer starts at Fab 15A has implications for the broader mature-node supply chain. Automotive semiconductor suppliers, industrial chip makers, and consumer electronics manufacturers that source 28nm capacity from TSMC could face tighter availability as the foundry prioritizes its leading-edge roadmap. This dynamic may benefit competing mature-node foundries, though TSMC remains the dominant provider of advanced process capacity with no direct peer at the 2nm and 3nm nodes.
What to watch next
Key developments to monitor include whether TSMC formally announces the 5–10% advanced-node price increase for the second half of 2026 and whether the TrendForce-reported potential 15% increase on 3nm wafers is confirmed or implemented. Earnings calls and investor days from Nvidia and AMD will be important venues for management commentary on wafer cost trends and whether higher input costs are being absorbed, passed through to customers, or offset by volume commitments. The pace of TSMC's 2nm ramp — and the extent to which wafer pricing at that node settles above or below the $30,000 threshold — will determine the magnitude of cost pressure on next-generation AI chip programs. Any further statements from CEO CC Wei or CFO Wendell Huang on the timeline and scale of additional 2027 price adjustments will also be closely watched by the industry.