NVIDIA has almost doubled the list price of the RTX PRO 6000 Blackwell within eighteen months. The card is a well documented example of how heavily conditions in the memory market now feed through to graphics and accelerator cards.
The Price Steps
At launch in early 2025 the card sat at around 8,565 US dollars. In June 2026 NVIDIA raised it to 13,250 US dollars, and in August 2026 to 16,000 US dollars. That is roughly 87 percent above the launch price with no change to the hardware: same GPU, same 96 GB of GDDR7, same 1,792 GB/s of memory bandwidth.
NVIDIA has given no public reason for either increase. Only Workstation Edition prices are published. For the Max-Q Workstation Edition and the Server Edition, NVIDIA lists no separate prices, and offers in the German channel vary widely as a result.
Memory Drives the Price
The 96 GB of GDDR7 are built in a clamshell layout from 32 modules of 3 GB each. That is the largest memory configuration currently fitted to a single card, and it leaves the card unusually exposed: every dollar added to a module works through to the bill of materials 32 times over.
In late July a 3 GB GDDR7 module cost roughly 60 to 70 US dollars. Memory alone therefore accounts for something in the region of 1,900 to 2,200 US dollars, before the GPU, the board, cooling and margin.
Where the Shortage Comes From
Memory manufacturers have shifted capacity to server DRAM and HBM for AI accelerators, where margins are higher. Samsung raised DRAM prices by around 20 percent in the third quarter.
The RTX PRO 6000 is therefore not an isolated case. GeForce cards also went through several rounds of increases in 2026, up to 30 percent at the top end. Market observers expect prices to peak only towards the end of the year.
What This Means for Procurement
Anyone who budgeted for GPU hardware in 2026 should revisit the figures. Buying now means paying the highest price so far and carrying the residual value risk on a card whose price is largely driven by a shortage that may well ease again. Fluctuating lead times come on top of that.
A second point belongs in the same calculation: the Server Edition draws 600 watts, the Max-Q Workstation Edition 300 watts at the same memory capacity. Across a two to three year term that makes a substantial difference to power and cooling, and it also limits how many cards can sensibly run per rack unit.
We are happy to discuss case by case how this price development affects our GPU servers.
