2028 is the earliest point when the industry’s competing expansion plans could simultaneously translate into meaningful production, according to Counterpoint Research Director MS Hwang."The point at which all of these expansion plans will simultaneously contribute to actual production output is expected to be 2028 at the earliest," Hwang told Benzinga in an exclusive interview.The timeline matters because demand is already outrunning supply. AI is no longer consuming only high-bandwidth memory attached to GPUs. The next wave of infrastructure spending is also pulling conventional DRAM into the shortage.AI Is Turning Conventional DRAM Into A BottleneckSuccessive waves of AI adoption are driving demand for both conventional DRAM used alongside CPUs and HBM used by AI accelerators. Counterpoint expects pent-up demand from agentic AI and AI server CPU growth to push conventional DRAM prices even higher.Prices already surged sequentially during the second quarter:
Samsung emerged as the biggest beneficiary. The South Korean giant reclaimed the global DRAM crown with 39% revenue share, returning to levels last seen in 2024.
SK hynix dropped from 39% a year earlier to 26%.
Micron reached 25%, putting it within one percentage point of second place.
