Taiwan Semiconductor Manufacturing Company (TSMC) has told customers to expect 3-6% wafer-out price increases from January 2027, according to a Tech Times report and industry sources cited by Benzinga. For AI accelerator customers ordering above their pre-committed volumes, an additional 10-15% high-performance-computing surcharge will push effective price increases toward 21%.
Where the increase lands
The steepest adjustments hit the leading edge: TSMC 2nm and 3nm gate-all-around nodes, where structural supply shortages persist. Mature nodes in the 28-45 nm range see the lower end of the range or individual negotiation, particularly for automotive and industrial buyers. Standard capacity-commitment holders take the base 3-6% increase.
The HPC surcharge is the story
The 10-15% HPC overflow surcharge is the notable move. In practice, an AI accelerator order that already carries a 6% base wafer-out increase plus a 15% HPC surcharge reaches an effective all-in cost increase near 21%. TSMC order book is full through 2030 and it has effectively lost competitive pricing pressure at the leading edge.
Competitors already moved
Samsung Electronics raised its own advanced-node foundry pricing 10-15% in August 2026, and Taiwan UMC, Powerchip and VIS have announced their own 2027 increases. Neither Intel Foundry nor Samsung offers 2nm capacity at TSMC scale, so hyperscalers and AI-chip designers have limited leverage to walk. The move sets up 2027 as a year where AI margins depend as much on foundry contract terms as on model performance - a change already reflected in AMD ballooning data-center revenue and NVIDIA Rubin roadmap.
Knock-on for AI infrastructure
The hike compounds pressure on hyperscale AI economics just as Oracle issued a force-majeure notice on its New Mexico Project Jupiter campus and Firmus Technologies kicks off a A$7B ASX IPO on the back of AI data-center demand.
Reporting based on coverage from Tech Times and Benzinga.