Nvidia has notified its customers and supply chain partners of price increases exceeding 15% on AI-related GPU products, intensifying cost pressure across an industry already strained by surging demand for compute infrastructure.

The hikes affect both server-grade AI accelerators and consumer graphics cards, and enterprise expenses could climb even higher once downstream markups are factored in. Because Nvidia's accelerators anchor most large-scale AI training and inference clusters — and its largest customers include the major cloud platforms — component-level cost changes at the company tend to propagate across the wider AI infrastructure market.

Memory Costs Drive the Increases

The root cause is high-bandwidth memory, or HBM — the stacked DRAM that feeds data to AI accelerators at high speed. Demand for HBM has surged alongside the AI buildout, and suppliers have responded by raising their prices. HBM output is concentrated among a small number of memory makers — chiefly SK Hynix, Micron and Samsung — and their capacity has been heavily committed to datacenter customers as AI infrastructure spending has scaled. GDDR7 memory modules have reportedly tripled in cost compared with prior generations.

Nvidia designs its GPU dies but relies on those outside suppliers for HBM and GDDR, so memory price moves surface in the GPU kits it sells onward. The company issued notifications to its add-in board partners in both May and July 2026, covering GPU kits that bundle the GPU die itself with the VRAM. The affected lineup spans the latest Blackwell-generation GDDR7 products as well as older GDDR6-based models.

Server GPUs such as the H200 and B200 have seen price increases of up to 15% in early 2026, while consumer cards have recorded more modest wholesale bumps of 5-10%.

Retail Prices Are Rising Even Faster

Median retail prices for RTX 50-series graphics cards climbed by as much as 39% between June and August 2026. The RTX 5070 saw a 36% retail price increase over that window, and the RTX 5060 Ti jumped 39%.

The RTX Pro 6000 Blackwell GPU reached $16,000 by August 2026, up from an initial pre-order price of $7,600 — a 110% increase.

According to analyst estimates, enterprise costs linked to these GPU kit adjustments could ultimately surge by 20-30% once the full system-level impact is taken into account. On the consumer side, retail pricing layers distribution and reseller margins on top of wholesale costs, which is one reason shelf prices can move by more than the underlying component increases.

The Squeeze on AI Builders

Analysts predict continued enterprise cost pressures of 15-20% driven by memory costs alone. For cloud providers and AI-as-a-service companies, these higher hardware costs are expected to flow downstream, making training runs more expensive and pushing up inference costs.

The supply picture adds further context. Memory makers have moved to expand capacity, but new fabrication lines carry long lead times, so output additions arrive over quarters rather than weeks. Meanwhile, memory content per accelerator keeps rising: each successive Nvidia platform has paired its GPUs with larger HBM allocations, increasing the memory share of an AI system's bill of materials from generation to generation.