AI data centers have absorbed an unprecedented share of the world's memory capacity, creating acute shortages in GDDR6, GDDR7, and HBM. IDC estimates AI data centers now account for roughly 70% of global memory output in 2026, up from 20-30% just a few years ago. The surge has driven GDDR7 prices from $65-80 per 16GB unit in mid-2025 to over $200 by year-end 2025 and into 2026—a tripling in less than nine months. Memory now represents more than 80% of the bill of materials for high-end GPUs, reshaping PC component economics across the board.
The constraint has cascaded into consumer hardware: RTX 5090 retail pricing climbed from a $1,999 MSRP (May 2026) to $4,329 on Amazon by July and to approximately $5,112 in Korean markets by August—2.5x the launch price. AMD followed with a 10-17% price increase in August. NVIDIA has deprioritized consumer gaming GPU production in response; the company has skipped 2026 entirely without a new architecture and shelved the RTX 5080 Super refresh, ending a 27-year streak of annual gaming GPU launches. MSI warned the memory shortage will shrink the PC gaming market by up to 20%.
The constraint extends beyond discrete GPUs: Nintendo raised the Switch 2 to $499 (from $449), Sony raised the PS5 to $649 in a second hike, and Valve hiked the Steam Deck by roughly $240 to a $300 entry point—all citing the same memory shortage. Console manufacturers are competing with AI data center buildouts for DRAM and storage, and AI is bidding supply away. The immediate window for buying pre-August pricing inventory is closing; channel stock is turning over to higher-priced inventory.
For architects and infrastructure planners, this signals a structural shift: GPU and memory procurement for frontier AI has inverted traditional hardware supply curves. Where gaming once set the market volume baseline, AI data center demand now sets the price floor. Inference and fine-tuning workloads on consumer-tier hardware are becoming costlier relative to cloud inference, shifting economics in favor of cloud platforms and away from local deployment. Long-term availability for consumer and SMB tiers remains uncertain pending new memory fabs.