The higher prices come as Nvidia and its customers contend with rapidly increasing memory costs. AI accelerators depend heavily on high-bandwidth memory, making changes in HBM pricing particularly important to the overall cost of the systems in which Nvidia's chips are deployed.
Memory prices have risen sharply this year. Analysts projected conventional DRAM contract prices would increase between 58% and 63% quarter over quarter in the second quarter of 2026, after climbing between 90% and 95% in the first quarter. Memory manufacturers have also been directing more production capacity toward HBM and server products.
Supply has tightened alongside those shifts. SK hynix said last year that its entire memory production capacity for 2026 had already been sold, while Samsung and SK hynix increased prices for their 2026 HBM3E supply by nearly 20% before the year began.
The amount of memory required by Nvidia's newest AI hardware makes those increases especially consequential. A Rubin GPU can include as much as 288GB of HBM4, while an NVL72 rack combines 72 GPUs. That puts more than 20TB of HBM into one rack before including memory connected to its Vera CPUs.
Producing HBM also requires roughly four times as much wafer area as an equivalent amount of conventional DRAM, according to the additional report. As AI systems use larger quantities of advanced memory, those components are becoming a significant part of server costs.
The reported increases would push those expenses further onto customers already making substantial investments in AI computing capacity. Because the adjustments depend on the specific processor generation and memory setup, customers will not necessarily see identical changes across Nvidia's server configurations.
Nvidia has already raised prices elsewhere in its hardware portfolio amid higher costs. GeForce graphics card prices increased earlier this month, according to the additional report. The latest reported changes would extend that pricing pressure to AI systems purchased by large data center operators and cloud providers.
The company has not publicly commented on the reported AI server increases, according to Reuters. Nvidia is scheduled to report second-quarter earnings on August 26.
For customers planning large AI deployments, memory configuration is becoming an increasingly important part of the cost calculation. If the reported increases take effect as described, Grace Blackwell and Vera Rubin customers could face substantially higher server prices as rising memory expenses work their way into Nvidia's newest AI infrastructure.
This analysis is based on reporting from cxodigitalpulse.
Image courtesy of NVIDIA.
This article was generated with AI assistance and reviewed for accuracy and quality.
