Memory costs more per millimetre than logic. The patent war over DDR5 and HBM

Kernel Insight, a Korean analytics firm, calculated that producing 1 mm² of DRAM wafer in the 1b node costs $0.654. A wafer in TSMC's N2 process costs $0.424 per mm². Memory is a simpler product to build, with no multiple patterning or extreme ultraviolet lithography. Yet it is 54 percent more expensive per unit of area than the most advanced logic. This reverses the normal relationship.

The cause is structural, not technical. Demand for AI servers is draining manufacturing capacity. As the Chinese outlet 钛媒体 reports, one AI server uses 8 to 10 times more DRAM than a traditional server, and more than three times as much NAND. Samsung, SK hynix and Micron have shifted more than 70 percent of their new capacity in advanced processes to higher-margin products, namely HBM and server memory, taking it away from consumer DDR5. The 1b node also produces the base die for HBM3E stacks used in AI accelerators.

The result is prices that refuse to return to normal. The CEO of SK hynix repeats that the AI industry does not exist without memory. Acer chairman Chen Wei forecasts that prices will rise until mid-2027, although the shortage will not last until 2030.

On the other side is demand that can vanish faster than it appeared. At the launch of V4.1-Flash, DeepSeek announced that its new architecture and context cache compression reduce HBM requirements to a quarter and SSD requirements to an eighth of previous values. For memory suppliers, this is a warning: part of current demand stems from model inefficiency, not from the number of queries.

Law is entering the game as well. On 23 September, the US International Trade Commission opened investigation 337-TA-1523 into DRAM memory and products containing it. Netlist filed the complaint, naming Micron, Hewlett Packard Enterprise, Lenovo and Supermicro. The company has a history of wins. SK hynix paid it $40 million in 2021. A jury awarded $303 million against Samsung in 2023 and $118 million a year later, and $445 million from Micron in 2024. Netlist had previously sued over DDR5 and HBM modules, and since July, investigation 337-TA-1511 has been ongoing, with Nvidia, Broadcom and Google among the defendants. Each of these cases is a cost that sooner or later lands in the price of memory.

The patent war is the second, less visible mechanism pushing prices up. Settlements and verdicts, counted in hundreds of millions of dollars, are booked as operating costs. These costs sooner or later reach the memory price list for end customers. Combined with a tight production balance, this means the availability of modules depends not only on demand from data centres but also on legal risk. For a buyer of AI servers, this is one more argument to sign long-term contracts and secure supply before prices climb again.