The AI-Driven DRAM Crisis: Why All Enterprise Hardware Is Getting More Expensive

Demand for memory to train and run AI models is rewriting the global chip market: AI data centers will consume close to 70% of the world's memory chip production this year, leaving barely 30% for PCs, phones, automotive and IoT devices, and pushing DRAM prices up as much as 90% compared to late 2024.

Why AI is eating up memory production

The technical cause is high-bandwidth memory (HBM), the type AI GPUs use to move data at the speed large-model training and inference demand. Producing HBM requires three to four times more wafer capacity than conventional DRAM, so every gigabyte of HBM a manufacturer dedicates to AI is capacity it stops producing for standard consumer memory. NVIDIA is currently the largest buyer of that advanced memory.

Samsung, SK Hynix and Micron: production sold out years in advance

Samsung, SK Hynix and Micron control more than 90% of global DRAM supply. SK Hynix has stated it has already sold out its entire 2026 and 2027 production. Industry estimates suggest global DRAM supply will only cover 60% of demand by the end of 2027, and the sector would need 12% annual production growth when planned growth sits at roughly 7.5%.

The ripple effect reaches laptops, servers and phones

The price hikes don't stay contained to cloud providers buying GPUs. Any equipment carrying DRAM or conventional flash memory — corporate laptops, on-premise servers, sales-team phones — competes for the same manufacturing capacity now prioritized for AI. Industry analysts project the average PC price could rise as much as 8% in 2026 from this factor alone, and several consumer electronics makers have warned they may need to absorb or pass on those costs before year-end.

What it means for a company's IT budget

For IT and finance teams, the practical takeaway is twofold. First, any hardware purchase planned for 2026-2027 — servers, laptops, device fleets — should be budgeted assuming higher memory prices and, in some cases, longer lead times due to component shortages. Second, this same phenomenon is what's propping up the cost of the cloud infrastructure used to run in-house or third-party AI models: cloud providers eventually pass on more expensive memory costs to their compute pricing.