GPU supply shortage shows no sign of ending

The ongoing shortage of graphics processing units is being driven by a memory bottleneck, analysts say. A shift from DDR4 to DDR5 formats has already slowed chip supplies, and the rapid build‑out of AI data centres has further strained the market for high‑bandwidth memory, a specialised DRAM type needed for training and running AI models.

High‑bandwidth memory competes for the same wafer capacity as conventional DRAM, and manufacturers have been re‑orienting production toward the more profitable HBM chips, which sell for about 30 percent more than traditional DRAM. The transition to newer HBM generations, such as HBM4 and HBM4E, is also limiting available stock as factories re‑tool for the upgraded technology.

A Samsung executive told Reuters in September that HBM could account for nearly 30 percent of total DRAM production in 2027, up from 20 percent the year before, and some expect the share to rise to almost 40 percent by 2029. AI data centres are projected to consume roughly 70 percent of global DRAM supplies this year.

The market is highly concentrated, with Micron Technologies, Samsung Electronics and SK Hynix together responsible for about 90 percent of worldwide DRAM output. All three firms anticipate long‑term shortages despite expanding capacity, and they are shifting focus toward data‑centre demand at the expense of consumer GPUs; Micron, for example, stopped direct‑to‑consumer sales last year, prompting continued RAM price increases.