Samsung Chip Price Increases Show AI Foundry Capacity Tightening
Samsung raised prices on advanced foundry nodes as AI demand tightened 4nm, 5nm and mature-node capacity tied to HBM base dies, custom processors and supporting chips.

Samsung's July price increases for advanced foundry work show AI demand moving from headline accelerator supply into the manufacturing capacity behind memory base dies, custom processors and supporting chips, Tech Wire Asia reported.
New orders on Samsung Electronics' 4-nanometre process rose by as much as 15% in July.
Reuters' sources placed the SF4 increase at 10% to 15% for China and US buyers, and at 5% to 10% for Taiwan-based customers.
The price movement was not limited to one node.
Samsung's 5nm process also rose 10% to 15%, and its 8nm process increased by close to 10%.
Samsung declined to discuss operational pricing details, but the changes followed earlier advanced-node increases that TrendForce linked to customer notices in the fourth quarter of 2025.
AI demand is tightening Samsung's capacity through two related channels.
The Pyeongtaek SF4 line is used for external logic-chip orders, including work for Qualcomm, and for high-bandwidth-memory base dies.
Samsung's HBM4 uses a logic base die made on the company's 4nm process, connecting memory growth directly to foundry allocation rather than leaving it as a separate packaging story.
Samsung tied its improved second-quarter foundry performance, before incentive-related provisions, to HBM base-die demand and US customer orders.
The second-half plan adds more 4nm language-processing-unit chip and HBM base-die sales while the company ramps its second-generation 2nm process for AI and high-performance-computing design wins.
TrendForce's 2026 outlook shows why customers are encountering less spare capacity: worldwide foundry sales are projected at roughly $218.8 billion, a 24.8% annual increase, as AI GPUs and custom silicon absorb 5nm, 4nm and newer production.
TSMC's 7nm-and-below processes supplied 77% of its second-quarter wafer revenue, including 33% from 5nm and 30% from 3nm.
Mature processes are also being pulled into the AI buildout.
Power-management chips, interposers, programmable logic and optical-communications components rely on older 8-inch and selected 12-inch manufacturing, and TrendForce's data put average utilisation across the world's 10 largest 8-inch foundries at 88% in 2026, with a second-half move toward 90%.
Average 8-inch foundry prices increased 5% to 15% from the first quarter to the second.
Samsung remains much smaller than TSMC in contract chipmaking, but the July price increases give it more room to monetise nodes that serve both outside customers and its own memory stack.
TrendForce put Samsung Foundry at just over $3.2 billion of first-quarter 2026 revenue, leaving its global share at 6.5%; TSMC held 72%.
Two large customer tracks show where Samsung is trying to turn that capacity pressure into longer contracts.
A July memorandum of understanding with Broadcom covers memory, foundry services and advanced packaging, and the companies value the planned five-year collaboration through 2030 above $200 billion.
A separate 22.8 trillion won semiconductor manufacturing agreement runs through the end of 2033, and Elon Musk has identified Samsung's Taylor, Texas, plant as the manufacturing site for Tesla's next-generation AI6 processor.




















