The TSMC capacity at their 3 nm and 5 nm nodes remains at 100%, and the Taiwanese manufacturer has already begun annual negotiations with clients to adjust tariffs in 2026. According to Taiwanese economic press and analyst estimates, the price review for advanced processes is expected to fall within a range of between 3% and 10%, with variations depending on node, committed volume, and long-term relationship with each customer.
Although TSMC generally does not comment on pricing, the underlying message is clear: structurally higher costs (international expansion in the US and Japan, energy, materials) and the “gigantic” demand for AI and mobile are putting pressure on the equation. The company has already maintained four years of moderate — single-digit — adjustments to sustain investments without jeopardizing customer stability.
What’s behind the increase
- Overwhelming demand. The craze for generative AI and the smartphone upgrade cycle have pushed 3 nm and 5 nm to operate at full capacity.
- Changing mix that shifts the game. The HPC/AI share in the portfolio is increasing and competing for capacity with the historic dominance of the mobile segment. In 2024–2025, 3 nm + 5 nm already accounted for about 60% of quarterly revenue (with 3 nm around 23% and 5 nm approximately 37%).
- Rising Capex and overhead costs. Overseas fabs (US, Japan) are increasing labor, logistics, and compliance costs, requiring visibility of returns.
- Limited competition at the top-end. In advanced nodes, competition is limited, giving negotiating leverage to TSMC — without losing sight of its long-term relationship culture.
Impact on clients… and products
- Major fabless companies (mobile and PC/AI PC). Adjustments of 3%-10% might be partially passed on to the BOM for smartphones and AI PCs in 2026, especially for high-end SoCs.
- Hyperscalers and AI providers. The cost per accelerator and per inference/training will include an additional step, in a year when many centers plan to upgrade to 3 nm or derivatives.
- SMBs and startups. Where volume commitments are absent, price elasticity is lower: access to wafer starts will continue to depend on priorities (volume, roadmap, visibility).
How far could this go?
The range of 3%–10% suggests a cautious adjustment, aligned with recent single-digit increases. Specifically, the “3 nm family” is likely to see at least a low single-digit increase, according to market consensus, while other advanced nodes could pivot based on effective costs, utilization, and long-term commitments made in 2025.
Signals to watch in the coming months
- 2026 contract closures: which nodes and volumes are secured by major clients and under what conditions.
- Production ramp-up pace: evolution of second-generation 3 nm and the first 2 nm (in Taiwan and Japan).
- Competitive response: capacity and yield at Samsung Foundry; progress of Rapidus toward 2 nm (target 2027); ramp-up of Intel Foundry.
- Sustained AI effect: whether HPC/AI demand maintains utilization floors above 90% in 2026.
Strategic outlook
For TSMC, “temperate increases” are a way to finance expansion and maintain peace with clients amid a chronic shortage at the technological frontier. For the broader ecosystem, the message is twofold: plan ahead (capacity, tape-outs, advanced packaging) and optimize costs assuming 2026 won’t bring abundance of 3 nm/5 nm wafers. Those needing 2 nm in the latter half of the decade will have to secure allocations with much greater advance planning.
via: money.udn.com