TSMC agrees to price increases, Samsung grows into an alternative
Higher production costs are expected to drive up chip prices for global market leader TSMC. High demand from AI hyperscalers is also contributing.
(Image: Macro photo / Shutterstock.com)
The world's largest contract chip manufacturer TSMC is considering price increases. This could further drive up the prices of processors, graphics cards, and mobile processors, which in turn would affect device prices.
TSMC's CFO Wendell Huang spoke about the potential price increases in an interview with the BBC. He cited higher production costs as the reason, which are likely to have increased especially due to the Iran war. Unspoken, however, is the current market situation: TSMC's modern manufacturing processes are fully booked, including the new 2-nanometer generation (N2, N2P). In addition to Apple, AMD, and cloud hyperscalers in particular have a high interest in the 2 nm processes. Nvidia is currently content with 3 nm for Vera Rubin.
Over 30,000 US dollars for a wafer
Huang promises that prices will not suddenly increase by four or five times and that the adjustments will reflect their value. However, there is a wide range between current prices and a quadrupling.
The entire industry traditionally remains silent about absolute prices. Analysts estimated an N2 wafer at around 30,000 US dollars last year. If a chip is close to the exposure limit of around 850 mm², about 60 to 65 chips fit on a round wafer with a diameter of 300 mm. More than 300 of AMD's compute chiplets for current AI accelerators likely fit on a wafer.
Samsung and Intel benefit
Meanwhile, the other two remaining contract chip manufacturers with current manufacturing processes, Samsung and Intel, are benefiting from the high chip demand. According to reports from South Korea, Samsung's manufacturing division (Samsung Foundry) could return to profitability in the third quarter.
The company had a difficult time for years because, according to consistent reports, the yield of functional chips has been low. At the same time, Samsung lags TSMC technologically, which results in poorer efficiency. The billions in semiconductor profits currently come exclusively from its memory sister company, not from the foundry.
Samsung's current 2 nm process, which is likely just a rebranding of an improved 3 nm process, is said to still have a yield below 70 percent. Nevertheless, low prices could attract customers.
Videos by heise
More capacity thanks to HBM
This year, the 4 nm generation is expected to bring a revenue surge. Samsung is using it to manufacture the new base dies for High Bandwidth Memory HBM4, which for the first time separates the memory cells and the control logic onto different dies. While Samsung's memory division produces the DRAM chips, Samsung Foundry handles the base dies. This allows for production optimizations for both parts. Primarily high-end AI accelerators use HBM.
In addition, Samsung is likely producing the chips for the Groq-3 accelerators in cooperation with Nvidia. Apple is also said to remain interested; however, details are unknown. Nintendo is also reportedly ordering larger quantities, likely 8 nm processors, for the Switch 2.
Intel has reportedly recently landed a large packaging order from Google. Nvidia, among others, is allegedly evaluating Intel's manufacturing processes.
(mma)