The RTX 5090, which launched at $1,999 in early 2025, now reaches $4,800–$5,000 at major U.S. retailers. ASUS ROG Astral premium variants are listed at $4,829.99 (Best Buy, July 2026), with some AIB models exceeding $5,000. The Founders Edition remains the closest to MSRP but sells out in minutes; supply is severely constrained as NVIDIA production cuts of 20–40% ripple through the consumer channel. The core driver: GDDR7 and GDDR6 memory now accounts for more than 80% of a high-end GPU's bill of materials. A single 16GB GDDR7 module costs over $200—up from roughly $65–$80 in mid-2025—because AI data centers are absorbing ~70% of the world's memory output in 2026 (up from 20–30% in 2022).
Memory price surges began in late 2025 when long-term supply contracts expired; spot-market sourcing then accelerated the climb. AMD's RX 9000 series follows a similar arc. VRAM alone now accounts for ~$820 of the RTX 5090's current street price. Lower-tier cards like the RTX 5080 and RTX 5060 Ti are also experiencing major markups; the RTX 5080 ($799 MSRP) now costs $2,000+. No new consumer GPU generation is coming from NVIDIA in 2026; AMD's next refresh has been pushed to late 2027 or 2028. Gamers are holding onto older hardware rather than upgrading, compressing the upgrade cycle.
For architects: GPU cost inflation is not speculative—it is documented in BOMs and supply contracts. The $1,000+ markup on flagship cards is a structural repricing tied to memory supply, not a bubble. For any infrastructure team budgeting GPU-accelerated workloads, expect continued memory scarcity through 2026 and into 2027. Consumer gaming GPUs are now a secondary market; data center memory demand has repriced the entire category. Long-term contracts for H200 ($30K–$40K+) and next-gen accelerators will likely embed memory cost inflation.