SK hynix projected on 14 October that AI infrastructure performance would grow roughly 20-fold by 2030 as demand for inference rises, Yonhap reported. Ahn Hyun, the chipmaker’s head of development, gave the projection in a keynote speech at the 2026 Semiconductor Exhibition in southern Seoul.
Ahn said inference demand is growing along three dimensions at once: the complexity of tasks, the amount of information processed in each instance and the number of requests. Together, those changes are increasing demand for AI infrastructure, he said.
Growing capacity requirements for key-value caches are also affecting memory use. The caches store information that AI models have already processed so it can be reused. Their data must be distributed across high-bandwidth memory, dynamic random-access memory and solid-state drives. SK hynix is pursuing 3D stacked DRAM and high-bandwidth flash while improving its existing HBM products, Ahn said.
