The memory shortage has mostly been told as a consumer story on this site: DDR5 kits tripling in price, NVIDIA raising GPU kit costs across the GeForce lineup, and a rumored RTX 50 Super refresh that keeps slipping because it depends on the scarcest memory in the industry. The same force is now clearly visible at the opposite end of the market, on the professional Blackwell card that has no cheaper sibling to fall back on.
What changed: RTX PRO 6000 at $13,250
The figure is consistent across several independent outlets. Tom's Hardware reported that NVIDIA's official price for the RTX PRO 6000 Blackwell has risen to $13,250, a 55% increase over its $8,565 launch price from March 2025, in roughly a year, and Wccftech and Gizchina carried the same numbers, with the underlying marketplace listing change traced to iTHome. The one claim that currently rests on a single analysis rather than several is the cost breakdown: Tom's Hardware attributes only about 8% of the increase to higher manufacturing cost and the remaining roughly 92% to supply-and-demand pressure in the GDDR7 market. That split is the sharpest detail in the story and also the least independently corroborated, so it is worth holding a little more loosely than the headline price.
A note on which price is which. The $13,250 is NVIDIA's own marketplace figure; measured against the card's earlier $7,673 pre-order pricing rather than the $8,565 launch, Gizchina puts the rise as high as roughly 73%. Street listings at third-party retailers vary and have at times sat below the marketplace number, so the exact figure you pay depends on where and when you buy. The direction, not the decimal, is the point.
| RTX PRO 6000 Blackwell | Figure |
|---|---|
| Launch price (March 2025) | $8,565 |
| Price (July 2026) | $13,250 |
| Increase | ~55% over ~16 months |
| Share attributed to manufacturing | ~8% |
| Memory | 96 GB GDDR7 ECC, clamshell |
Treat the exact dollar figure as a snapshot with a date on it. Official and street prices on these cards have been moving month to month, and the number that matters for a purchase is the one in front of you when you buy, not the one in a headline from a few weeks earlier.
Why this card is the shortage's clearest signal
The RTX PRO 6000 carries 96 GB of GDDR7 in a clamshell arrangement, which means memory chips mounted on both sides of the board to reach a capacity a single-sided layout cannot. That is the largest VRAM capacity on any discrete graphics card, and it is exactly why this SKU is the most exposed to a GDDR7 squeeze. When the bill of materials is dominated by the most contested component in the industry, the card's price tracks the memory market more tightly than any consumer part does.
There is a second reason the increase sticks here where a gamer might simply buy a cheaper card: for the workloads this card exists to serve, there is no substitute. A studio running large local models, a lab doing path-traced rendering, or an engineer running simulation cannot drop to a 16 GB gaming card and call it equivalent. The capacity is the product. NVIDIA holds a near-monopoly on high-VRAM professional GPUs, AI demand for that memory has no visible ceiling, and buyers who need 96 GB in a workstation have nowhere else to go. Pricing power follows from that position, and the current number reflects it.
NVIDIA is adding VRAM even as it reprices it
The same week's other professional-Blackwell news points in a consistent direction. NVIDIA's RTX PRO 5000 72GB Blackwell is now generally available through partners, a 72 GB card that sits below the 96 GB flagship and 50% above the earlier 48 GB RTX PRO 5000. NVIDIA rates it at 2,142 AI TOPS and positions it for agentic AI development, rendering, and engineering simulation.
Read the two stories together and the strategy is legible. The high-VRAM tier is where memory scarcity bites hardest, so NVIDIA is both charging more for the top card and widening the ladder of capacities beneath it, giving buyers a 72 GB rung to land on when 96 GB is out of budget or supply. It is a rational response to a constrained input, and it is also a reminder that the relief valve here is more product tiers, not lower prices.
The macro picture behind the number
None of this is specific to one card. It is the workstation-visible edge of a market-wide reallocation:
- 3 GB GDDR7 modules, the dense parts high-capacity cards depend on, have been reported at roughly $60 to $70 each against about $20 for standard 2 GB chips, so capacity now scales cost far faster than it used to.
- AI data centers are projected to consume as much as 70% of the world's memory output in 2026, up from roughly 20% to 30% a few years ago, which leaves consumer and workstation devices competing over a shrinking remainder.
- IDC has framed the situation as a potentially structural reallocation rather than a normal cyclical dip, with tight conditions expected to persist toward 2028.
If those framings hold, the RTX PRO 6000's price is not an anomaly waiting to correct. It is what the top of the stack looks like when memory is the binding constraint, and it rhymes with the same pressure our Market Watch tracks on consumer cards every week. It is also the mirror image of the rack-scale AI buildout that is pulling that memory into data centers in the first place.
What it means for workstation and local-AI buyers
For anyone actually costing out a high-VRAM workstation right now, a few practical conclusions follow.
- If the capacity is the requirement, price it as a moving target and buy when the project needs it, not on a prediction that the number drops soon. Every credible framing of this shortage points to persistence, not a near-term correction, so waiting has a real cost and an uncertain payoff.
- If 96 GB is more than the workload truly needs, the newly available 72 GB RTX PRO 5000 or a used previous-generation card (such as prior RTX 6000-class parts) may deliver the working capacity for meaningfully less. Match VRAM to the actual model or scene size rather than buying the flagship for headroom you will not use.
- For local AI specifically, be honest about where the capacity cliffs are. The jump to a 96 GB card is justified by models and context lengths that genuinely need it; below that threshold, a 32 GB consumer flagship like the RTX 5090 covers a large share of local inference at a fraction of the workstation price, with the tradeoff of less memory, no ECC, and a consumer support posture.
- Whatever you choose, plan the purchase around the card that exists at a price you can see today. Pricing this volatile punishes plans built on last month's figure or next quarter's hope.
The consumer side of this shortage gets the headlines because it touches more people. The RTX PRO 6000 at $13,250 is the same story with the volume turned up, on the one card where buyers cannot dodge it by stepping down a tier.
Sources: