Nvidia RTX 3090 Resale Prices Rise on AI Demand

The clearest evidence comes from both enterprise and retail markets. CoreWeave, one of the biggest AI cloud providers, said in its second-quarter 2026 results that older GPU generations continue to “hold their value,” and it extended contracts for A100 accelerators, first launched in 2020, through 2029. The company said older models are priced at levels from prior years or even above them, with capacity essentially sold out.

On the consumer side, the pattern is strongest in Nvidia’s 24GB cards, especially the RTX 3090 and RTX 3090 Ti, which are increasingly useful for local AI work. In the Czech second-hand market, a Founders Edition RTX 3090 was listed at 25,500 crowns, an MSI Suprim at 29,900 crowns and a Gigabyte RTX 3090 Ti at 34,900 crowns. Completed auctions support the pricing: an EVGA FTW3 RTX 3090 sold for 35,755 crowns in June, while another RTX 3090 MSI Suprim X fetched 20,296 crowns in May. A year earlier, comparable RTX 3090 cards on Aukro typically sold for 17,800 to 22,700 crowns.

That price firmness is not broad-based across the graphics-card market. AMD’s Radeon RX 6900 XT, by contrast, is trading around 10,700 crowns on classifieds and sold recently for about 7,900 to 9,000 crowns on Aukro. The difference underscores Nvidia’s moat in AI software. CUDA remains the default development environment for many AI workloads, while AMD is still telling developers how to migrate away from it.

Nvidia itself is reinforcing the idea that GPUs are no longer just standalone chips but part of a broader compute-financing ecosystem. In August, it unveiled partnerships with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR to finance more than $500 billion of AI infrastructure. Nvidia also said the median GPU-hour price rose from about $2.00 in October 2025 to $2.70 in June 2026, a 35% jump in eight months.

For investors, the implication is twofold. The bull case is that Nvidia’s installed base has pricing power beyond first-sale chip revenue, because older GPUs still generate cash flow and extend the life of Nvidia’s platform. The bear case is that the resale market is cyclical and dependent on a supply-demand imbalance that will eventually ease as new capacity comes online. CoreWeave itself is planning to reach at least 8 gigawatts of active capacity by 2030, which suggests the current scarcity premium is meaningful but not permanent.

The broader market signal is that AI demand is starting to commoditize the definition of “new.” In this environment, the value is not only in the latest Blackwell-generation chips but in any sufficiently capable Nvidia hardware that can still be deployed profitably. For owners of RTX 3090-class cards, that leaves a potentially attractive selling window. For Nvidia shareholders, it reinforces the company’s grip on the AI stack — from silicon to software to financing — even as competition and new supply may eventually cap the upside.