Nvidia AI Server Prices to Rise Over 15% on Memory Costs
Nvidia has told major customers that AI server prices will rise over 15% for early-2027 deliveries, adding at least $5 billion to a 1GW data center's cost.
The price increase covers existing Grace Blackwell systems and the next-generation flagship Vera Rubin line, with the exact hike varying by GPU generation and memory configuration. A 72-GPU Vera Rubin rack currently costs about $7 million; after the rise it could reach roughly $8 million, according to estimates in The Information's report. The server market has already been volatile, with some components fluctuating by 40% in a week, server resellers and GPU cloud vendors said.
Memory costs began climbing sharply this year. TrendForce projected in March that traditional DRAM contract prices would jump 58% to 63% quarter-on-quarter in Q2 2026, while NAND Flash would rise 70% to 75%. In Q3, growth slowed but prices kept rising, with server DRAM contracts expected to be up 13% to 18% sequentially. The research firm also said server DRAM prices are likely to rise quarter after quarter from the second half of 2026 through the second half of 2027.
The supply crunch stems partly from AI chips competing for wafer capacity. HBM uses more wafers than conventional DRAM due to larger dies, more stacking layers, and complex packaging, so when memory makers prioritize HBM, capacity for standard server memory shrinks. TrendForce expects server RDIMM bit supply to grow only 15%-20% in 2027, well behind server CPU shipment growth. Some cloud service providers and server OEMs have already switched from 96GB and 128GB RDIMMs to 32GB or 64GB modules to cut costs. Nvidia itself has reduced SOCAMM memory capacity by half on the next Vera Rubin Superchip because of tight LPDDR5X supply, with suppliers able to meet only about 60% of Nvidia's projected demand.
This is the third round of notable Nvidia-related hardware price increases in 2026. In the consumer market, RTX 50-series graphics cards have risen substantially since summer, with median prices up about 39% for the RTX 5060 Ti 16GB, 36% for the RTX 5070, 27% for the RTX 5060, and 9% for the RTX 5090. Those figures reflect retail prices, which include supply, demand, and channel inventory, not just Nvidia's official pricing. In the professional segment, the RTX PRO 6000 Blackwell with 96GB of memory has gone from about $8,565 in early 2025 to $13,250 in June 2026 and $16,000 in August 2026 on Nvidia's own store.
Beyond memory, the increases are tied to higher system integration complexity, tight supply chains, and competition among cloud providers and AI companies for delivery slots. Nvidia has also moved further into data center infrastructure. On August 21, it announced a minority investment in US data center developer Cloverleaf Infrastructure, reported by the Wall Street Journal to be worth hundreds of millions of dollars, though the amount was not disclosed. Cloverleaf helps secure land, power, and grid connections for data centers and will use Nvidia's DSX platform for site, power, cooling, and compute infrastructure planning. Days earlier, Nvidia said it would invest $1.5 billion in SB Energy, SoftBank's data center arm, and it has partnered with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR to raise more than $500 billion in third-party capital for AI infrastructure. Meanwhile, Amazon, Microsoft, Google, and Meta are all stepping up their own AI chip efforts to reduce dependence on Nvidia.