Reports that AI system builders are purchasing GeForce RTX 5090 graphics cards in bulk are adding pressure to an already constrained market. Retail listings for Nvidia’s flagship gaming GPU have climbed well beyond its $1,999 launch price, with some models reportedly exceeding $6,500 and approaching $10,000 overseas.
Key takeaways
- RTX 5090 cards are reportedly being bought by the pallet for AI workstations and servers.
- Retail prices have surpassed $5,000 in the United States, with some listings above $6,500.
- The cards offer 32GB of GDDR7 memory, making them cheaper than Nvidia’s professional alternatives.
- Images supporting the reports may be partly edited or AI-generated, so the scale of the purchases remains unverified.
The reports have nevertheless coincided with widespread shortages, making the situation relevant to PC builders and gamers watching the high-end GPU market.
Retail GeForce cards enter AI systems
Images attributed to Hong Kong-based hardware reporting show pallets of boxed RTX 5090 cards being unpacked for multi-GPU AI systems. The products appear to include standard Asus, MSI, and Gigabyte retail models rather than Nvidia’s RTX PRO workstation hardware.
However, several details in the images—including inconsistent model labels and suspicious visual elements—have prompted questions about whether some pictures were generated or digitally altered. The photos should therefore be treated as evidence of a reported trend, not definitive proof of the quantities involved.
Why companies are choosing the RTX 5090
The RTX 5090 includes 32GB of GDDR7 memory and substantial tensor-processing capability. While it lacks the memory capacity and enterprise features of Nvidia’s professional accelerators, it can still be an attractive option for smaller AI deployments when availability matters more than certification or support.
Nvidia’s RTX PRO 6000 workstation card offers 96GB of memory but has reportedly reached around $16,000. Against that price, even a heavily marked-up RTX 5090 may appear economical to companies assembling systems from multiple cards.
Export variants highlight the contrast
Nvidia’s China-specific RTX 5090D and RTX 5090D v2 variants demonstrate that GeForce hardware can be modified to reduce its appeal for certain AI workloads. The standard RTX 5090 is reported to deliver up to 3,352 AI TOPS, while the RTX 5090D is limited to approximately 2,375 AI TOPS.
The cards shown in the reported server images do not appear to be those restricted variants. That has raised questions about why the unrestricted retail model remains readily usable for commercial AI systems, even as gamers struggle to find one near its launch price.
A repeat of the cryptocurrency shortage
The market resembles the 2021 cryptocurrency boom, when miners purchased gaming GPUs in bulk and sharply reduced availability for PC users. Nvidia responded then with Lite Hash Rate products designed to reduce mining performance, but there is no comparable AI-focused limitation on the global RTX 5090.
For ThinkComputers readers building gaming PCs, the immediate result is a shrinking supply of premium cards and inflated pricing. Lower-tier models such as the RTX 5070 Ti may become the more practical choices, although prices across the current generation remain vulnerable to memory constraints and continued AI demand.
Via HKEPC