NVIDIA's RTX 5090 flagship GPU, originally priced at $1,999, is now exceeding $5,000 in parts of Asia and above $4,200 in US retail—a 114.5% increase driven not by GPU silicon scarcity but by exploding memory costs. AI data centers have diverted the majority of global DRAM and HBM supply toward training and inference workloads, with memory now accounting for more than 80% of a gaming GPU's bill of materials. NVIDIA has voluntarily cut RTX 50-series consumer production by 30-40% in the first half of 2026, redirecting memory supply toward higher-margin data center products.
IDC forecasts AI data centers will consume up to 70% of global memory output in 2026, up from 20-30% in 2022. Samsung, SK Hynix, and Micron have all pivoted production toward high-bandwidth memory for servers, leaving consumer GDDR6/GDDR7 supply rationed. AMD followed NVIDIA with price hikes of 10-15% on Radeon RX 9000-series cards effective August 2026. Gigabyte announced 20-40% price increases on GPU orders effective August 1, following manufacturer price revisions.
This shortage structurally differs from the 2021 crypto-driven crisis: AI infrastructure capex is backed by trillion-dollar corporate balance sheets and growing, not speculative. Microsoft, Google, Amazon, Meta and Oracle collectively plan $660-690 billion in capex for 2026, the majority for AI compute. Console makers and handheld vendors (Nintendo, Sony, Valve) have already raised prices mid-generation—a rare signal in healthy markets—citing DRAM constraints. Analysts project no relief until late 2027 at earliest, with the RTX 60 series pushed from late 2027 to 2028.