Memory: DRAM 1b costs 54% more per square millimetre than TSMC's N2

Manufacturing DRAM should be simpler, and thus cheaper, than producing a leading-edge processor. Figures from the South Korean firm Kernel Insight, reported by the Chinese site IT之家, suggest otherwise. A DRAM 1b wafer costs $0.654 per square millimetre. A wafer on TSMC's N2 process costs $0.424. The gap is 54%.

Two technologies, two levels of complexity

The 1b node refers to a DRAM generation below 10 nanometres, with a density of roughly 0.436 gigabit per square millimetre. These wafers produce DDR5 modules: desktop sticks ranging from 16 to 64 gigabytes, sometimes 128 or even 512, laptop memory, and RDIMM and LRDIMM server modules of up to 256 gigabytes per stick. Servers built with these modules can reach 12 terabytes of memory. Part of this output also supplies the base dies for stacked memory in AI accelerators.

TSMC's N2 process requires double exposure and EUV lithography. In theory, a simpler product should cost less to make. The opposite is happening, driven by a cyclical cause: the widespread shortage of RAM linked to AI server construction.

The effect of AI demand

Major AI server projects absorb both high-bandwidth memory and high-capacity DDR5 modules. Capacity expansion at the three largest memory manufacturers is lagging behind demand. The head of SK Hynix has stated publicly that the AI industry cannot exist without memory, and that his group is accelerating the deployment of its next-generation production. The president of Acer, quoted by IT之家, believes the shortage will not last until 2030, although prices are expected to keep rising before easing in mid-2027.

This price pressure extends across the entire supply chain. In Italy, industry observers note that AI is driving up chip costs, with the bill eventually reaching both businesses and consumers.

A component that is simpler to manufacture can cost more: when AI demand saturates capacity, price no longer depends on technical difficulty but on scarcity.