TSMC signals 15% 3nm price hike in H2 2026 as AI demand crushes supply

<cite index="23-2">TSMC is planning to raise advanced 3nm process prices by around 15% in the second half of 2026</cite>, marking the chipmaker's second foundry price increase this year. The move signals extraordinary pricing power as AI chip demand continues to vastly outpace manufacturing capacity. <cite index="21-2">Monthly 3nm capacity at Fab 18 has expanded from around 130,000 wafers in early 2026 to 160,000–175,000 in Q2, yet even this ramp-up has failed to ease customer queues</cite>. <cite index="26-1,26-3">A single 3nm wafer currently costs approximately $20,000; a 15% increase pushes that to roughly $23,000 per wafer, with 2nm expected to exceed $30,000</cite>.

<cite index="27-1">TSMC's 3nm capacity remains at high utilization with customer backlog showing no significant easing, driven by the accelerated mass production of NVIDIA's Vera Rubin platform, cloud giants' adoption of 3nm custom AI chips, and continuous orders from AMD and Broadcom</cite>. <cite index="27-3">The demand base has expanded from primarily smartphone SoCs to multiple segments including NVIDIA, AMD, Google, and AWS</cite>. <cite index="26-4">TSMC's pricing authority, combined with 30%-plus revenue growth projections and $52–56 billion in capex commitments, signals a company protecting margins while investing aggressively in expansion</cite>.

For infrastructure teams and chipset architects, TSMC's pricing moves directly impact AI accelerator and custom silicon costs across the AI stack. <cite index="27-3">CEO C.C. Wei admitted supply shortages will persist through 2027 or longer, making advanced-node allocation a strategic constraint</cite>. The willingness of hyperscalers and AI chip makers to absorb mid-double-digit price hikes suggests confidence in their ability to pass costs upstream or absorb them within broader AI capex budgets—a signal of sustained AI infrastructure demand rather than a cooling cycle.