Trend shift
Memory-price reversal raises AI compute costs
Stanford DAM data and a media analysis put DDR5 at roughly $11.41-$13.28 per GB, near normalized 2007 levels. Memory may be returning as a material cost constraint in AI infrastructure.
Memory pricing may be moving from a background variable to an explicit constraint in AI infrastructure. A Tom's Hardware analysis based on Stanford DAM Project data puts DDR5 at roughly $11.41-$13.28 per GB and compares that level with normalized 2007 pricing. That does not yet establish a lasting shortage, but it means server-cost models can no longer assume memory will steadily get cheaper.
The long deflation curve has reversed
The Stanford DAM Project maintains public price series for DRAM, NAND and storage, creating a basis for comparisons across cycles. Tom's Hardware reports DDR5 pricing of roughly $11.41-$13.28 per GB, while Daniel Lemire compares the level with normalized 2007 conditions. The observable change is not merely a higher quote for one component: per-unit commodity-memory costs have departed from the assumption of persistent decline.
AI servers increase memory's cost weight
AI infrastructure purchases involve more than accelerators. Training, inference, data caching and model serving all rely on DRAM, storage and high-bandwidth memory tiers. As configurations become denser, memory prices have more influence over total server cost and deployment timing. The public data do not quantify AI demand's causal contribution, but GPU availability need not translate proportionally into lower compute costs when memory no longer deflates.
Procurement may shift from chips to systems
Cloud operators, server integrators and enterprise buyers may consequently pay more attention to memory capacity, bandwidth configuration, inventory cycles and workload utilization rather than GPU pricing alone. The strongest countercase is that a long-run series and a media estimate do not establish durable high prices or prove AI demand caused them. Supply expansion or softer demand could still return memory pricing to a more typical cyclical decline.
What to watch next
Watch whether subsequent Stanford DAM Project data show DDR5 pricing continuing to rise or staying elevated; whether memory makers report longer lead times, price increases or inventory changes; and whether cloud and server vendors identify memory configuration, supply or cost as an AI deployment constraint. Persistent prices alongside company disclosures would strengthen the claim, while a clear reversal would weaken it.