There is no end in sight for the GPU supply problem

The GPU supply crisis shows no signs of abating as AI data centers demand memory at an unprecedented rate. Major GPU manufacturers are redirecting resources to meet this burgeoning demand, leading to a shortage that is adversely affecting the consumer technology sector. Prices are surging beyond inflation levels, with the cost of dynamic random access memory (DRAM) soaring by 400% since 2024, according to a JP Morgan report.

Nvidia's top-tier GeForce RTX 5090 GPU, for instance, is now retailing at $4,920, far exceeding its suggested price of $1,999 when launched in January 2025. This price surge is not limited to the United States; foreign markets are also experiencing steep price hikes. Japan, for example, has seen gaming GPUs priced roughly 41% above their American counterparts.

This shortage has ripple effects across consumer electronics, with the Consumer Price Index for software and accessories increasing by nearly a quarter in the past two years. Major chip producers have already allocated much of their 2027 supplies, focusing on 2028 and beyond. The GPU shortage stems from a memory problem, exacerbated by the sudden surge in data center construction necessitating vast amounts of high-bandwidth memory (HBM).

HBM demand far outpaces current supplies, as AI developers require HBM for both model training and execution. Suppliers are shifting production towards advanced HBM chips like HBM4 and HBM4E, further tightening supplies. This issue is compounded by a highly concentrated market where three companies dominate global DRAM production.

Despite expansions, Micron Technologies, Samsung Electronics, and SK Hynix foresee long-term shortages. Their focus on AI production is causing supply tightness in 2027 and 2028, with some predicting it will take four to five years to achieve adequate wafer supply, with production lagging 20% behind demand until at least 2030.