Nvidia Preps a Third 2026 GPU Price Hike of Up to 30% as the AI Memory Shortage Bites
GDDR6 spot prices have roughly tripled this year as AI data centers eat the world's DRAM supply. The RTX 5070, launched at $549, could climb to $850 — and analysts expect the shortage to last until 2028.
Nvidia is reportedly preparing to raise GeForce RTX prices 20-30% for the third time in 2026, following a fresh DRAM price hike from Samsung. The driver is a global GDDR6/GDDR7 memory shortage — the same AI data center boom that’s straining HBM supply for Nvidia’s own AI accelerators is now cannibalizing consumer GPU memory, and GDDR6 spot prices have roughly tripled this year.
The numbers are already moving. The RTX 5070, which launched in February 2025 at $549, is now selling around $650 at street prices and could climb to roughly $850 under this hike. This round is reported to hit the entire Blackwell lineup, extending the squeeze to mid-range cards like the RTX 5070 that had mostly dodged earlier increases. In China, RTX 5070 Ti listings jumped 21.8% and RTX 5070 listings climbed 16.2% inside a single 24-hour window on July 25-26 — a preview of what board partners globally are bracing for.
AMD is moving the same direction. The company raised the GDDR memory kit prices it charges board partners by roughly 10% effective July 2026 — its second such increase in about six months. Its RX 9070 XT is reportedly facing a comparable 21% price hike at retail.
The root cause sits upstream of both companies: DRAM and GDDR fabs are prioritizing capacity for HBM used in AI data center accelerators, because that’s where the margin is. Consumer GPU memory is getting squeezed out of the supply chain as a side effect, not a target — Samsung, SK Hynix, and Micron are all allocating wafer capacity toward the higher-margin AI memory business, and gaming GPUs are competing for what’s left. Analysts quoted across the coverage expect the shortage to persist until at least 2028, meaning this isn’t a one-quarter blip that resolves once inventory catches up.
For anyone budgeting a GPU purchase or a fleet refresh, the practical read is: prices are going up again before they go down, and “wait for it to normalize” isn’t a near-term strategy. For studios and dev shops running local inference or rendering on consumer-grade Blackwell cards instead of cloud GPU rental, this round of hikes meaningfully changes that cost comparison — cloud rental pricing hasn’t moved anywhere near as fast as street prices on the cards themselves.