The global RAM market is experiencing an unprecedented crisis caused by the explosive demand for chips for artificial intelligence. DRAM prices have increased by 171,8% over the year, exceeding even the growth rate of the value of gold. A 64 GB DDR5 memory kit already costs over $500 — more expensive than the PlayStation 5 game console. We figured out what the DRAM shortage threatens for ordinary users, who is to blame for this and how long the crisis will last.
Current market situation
The three largest memory manufacturers Samsung, SK hynix and Micron control about 70% of the global DRAM market. After an oversupply of RAM in early 2023, manufacturers deliberately reduced production and by the end of 2024 the excess inventory was absorbed by the market. The manufacturers themselves reoriented production capacity to produce more profitable HBM and DDR5 memory. In addition, they simultaneously reduced production of the old DDR4, which led to a shortage of DDR4, despite the fact that many customers still use it.
This has led to manufacturers raising prices for DRAM. Samsung raised prices for individual memory chips by up to 60% in November 2025 compared to September. The contract price of a 32GB DDR5 module increased from $149 in September to $239 in November. Prices for 16GB and 128GB DDR5 modules jumped by about 50% to $135 and $1,194, respectively.
Artificial intelligence as the main culprit
The main reason for the shortage is the unprecedented race among tech giants to build out AI infrastructure. AI data centers are consuming huge amounts of memory and crowding out the consumer market.
Nvidia’s financial report clearly demonstrates this trend — almost 90% of the company’s earnings came from the data center segment that uses Nvidia chips.
«Over the past six months, the demand for computing power has increased dramatically. Artificial intelligence has reached a point where everyone wants to use it. We are seeing two exponential trends at the same time,» said Nvidia CEO Jensen Huang.
Spending on artificial intelligence is expected to reach about $2 trillion by 2026. And Huang predicts that investment in AI infrastructure will reach $3–4 trillion by the end of the decade.
Many new AI GPU clusters are using specialized high-bandwidth memory (HBM) stacks, which in turn reduces the available manufacturing capacity for standard DRAM. In fact, memory manufacturers have reallocated more than three times the wafer capacity for HBM chips compared to conventional DRAM, thereby reducing the supply of conventional memory. The result is a decline in the number of DDR4/DDR5 chips on the market, just as hyperscalers are rushing to buy more memory for both AI and traditional servers.
The hype has spread beyond DDR4/DDR5 modules. Currently, along with DRAM, NAND flash memory (storage) and even hard drives are also in short supply.