Chinese and US customers pay Samsung more than Taiwanese ones for the same nodes
New July orders on the 4nm and 5nm lines cost 10-15% more for customers in China and America, against 5-10% for those in Taiwan, as TSMC's capacity fills up.
Original cover art, generated for this story. THE VISSION does not republish third-party press imagery.
- Samsung raised prices on new July orders for its 4nm, 5nm and 8nm foundry processes, with increases of up to 15%.
- Chips on the 4nm SF4 process rose 10-15% for customers in China and the United States but only 5-10% for customers in Taiwan; the 5nm SF5 line rose 10-15% and the older 8nm line by nearly 10%.
- Samsung's SF4 line has run at full capacity since late last year, giving it room to raise prices as AI demand absorbs TSMC's leading-edge capacity.
Samsung raised prices on new foundry orders placed in July across its 4nm, 5nm and 8nm processes, by as much as 15%. The increases are not uniform by customer: chips on the 4-nanometre SF4 process went up 10% to 15% for customers in China and the United States, but 5% to 10% for customers in Taiwan. The 5-nanometre SF5 process rose 10% to 15%, and the ageing 8-nanometre line by close to 10%.
The pricing power is borrowed rather than earned. Demand for AI accelerators has absorbed most of TSMC's leading-edge capacity, and customers who cannot get slots there have fewer alternatives to Samsung — whose SF4 line has run full since late last year. Samsung has lost foundry market share to TSMC over the same period and raised prices regardless, which is what a supply squeeze rather than a competitive win looks like.
The regional spread is the detail with the longest tail. Charging Chinese and American customers more than Taiwanese ones for identical silicon is a pricing structure that tracks geopolitics as much as cost, in a market where second-source options at these nodes are already thin.
For any company taping out on an advanced node, the second source just got more expensive without getting better, which raises the floor cost of every custom AI chip programme currently being pitched as a way to escape Nvidia's margins. A 10-15% foundry increase does not sound like much until it compounds with the memory prices already climbing — the economics of building your own accelerator are moving against the companies that chose that path this year.