A global shortage of GDDR and DRAM memory, supercharged by AI data center demand, has pushed consumer gaming GPU prices 15–50% above MSRP in 2026. NVIDIA's flagship RTX 5090 launched at $1,999 but now sells for ~$3,000 on the street, with individual listings spiking over $5,000. AMD raised GDDR memory kit prices 10% in July 2026, its second such hike in six months, as spot prices for GDDR6 roughly tripled. The shortage stems from fixed-price memory supply contracts that expired at year-end 2025; renewals locked in at dramatically higher rates, hitting manufacturers all at once in January 2026 rather than gradually.
Memory now represents over 80% of a high-end gaming GPU's bill of materials. Both NVIDIA and AMD cut gaming production to prioritize AI accelerators: NVIDIA cut RTX 50-series output 20–40% and shelved the RTX 50 Super refresh indefinitely; AMD raised Radeon RX 9000 prices starting January 2026. Samsung, SK Hynix, and Micron—the three firms making both GDDR and server DRAM—can only allocate so many wafers; hyperscalers building AI clusters will outbid consumer hardware every time. NVIDIA estimates 2026 will be the first year in ~30 years without a new gaming GPU generation, with RTX 60 pushed to 2028.
For practitioners: GPU memory scarcity signals that consumer hardware is now subordinate to AI infrastructure in semiconductor economics. Teams needing compute should treat cloud rental (H100 at $2.19/hr, B200 at $3.50–27/hr depending on provider) as more economical than buying. For vendor infrastructure planning, watchlist memory contract renewals and spot pricing; GDDR7 pricing directly drives accelerator BOM cost and capex models.