China’s GPU market has recently seen pronounced price increases. According to an October 25 report by National Business Daily, spot quotes for the RTX 5090 at Huaqiangbei in Shenzhen reached RMB 51,000 to RMB 55,000, up nearly RMB 10,000 from more than a month earlier. Some channel quotes for the RTX 5080 also reached RMB 16,000 to RMB 17,000.
Some mid-range graphics cards have been affected as well, and the graphics card market has shown a “one price per day” pattern.
Behind these increases, AI demand is now influencing the GPU market. The price rise is driven by multiple factors, including tight supply, market expectations and AI computing demand.
The market had previously circulated rumors that the RTX 5090 had been discontinued. Although Nvidia has publicly denied the claim, some channels remain concerned about future supply.
At the same time, applications such as large language model (LLM) deployment, AI image generation and model inference have increased demand for high-performance GPUs, and some consumer-grade graphics cards are now being used for AI computing tasks.
Changes in supply and demand, combined with channel price volatility, have further raised the cost of graphics card procurement.
As GPU procurement costs rise, the advantages of cloud computing power rental are becoming more evident. A single high-performance GPU can cost tens of thousands of RMB, and multi-GPU deployment also requires servers, power and ongoing maintenance.
For AI development teams, computing tasks are often phased. Model deployment, fine-tuning and inference testing require different GPU configurations and usage durations, so direct hardware purchases can create high upfront investment.
Cloud GPU rental has therefore become a more flexible option. Developers can obtain computing resources based on project requirements and arrange costs by usage period, without making a one-time purchase of expensive equipment.
According to Mysteel data from October 26, the monthly rental price index for a single RTX 5090 32G card was about RMB 2.81 per card per hour, and the price remained stable that day. The increase in physical graphics card prices has not yet been fully transmitted to the rental market.
This offers a new cost option for teams with short-term or phased computing needs.
StarWar Cloud Intelligent Computing provides cloud GPU computing services for AI developers and enterprises, supporting application scenarios such as model deployment, fine-tuning and inference testing. It helps users flexibly configure computing resources based on actual tasks and reduce upfront hardware procurement pressure.
The GPU price surge is changing how AI developers choose computing power. As graphics card prices continue to climb, more developers are recalculating the costs of hardware procurement versus cloud rental.
Long-term, high-utilization tasks may still be suitable for owned equipment, while short-term experiments, model testing and projects with rapidly changing requirements can use cloud computing power to reduce fixed investment.
China’s graphics card price surge has created new development opportunities for the GPU rental market. Whether market demand can further translate into business growth will still depend on rental prices, resource supply and service experience.
Computing power can be purchased or rented on demand. Making every unit of compute investment better match actual demand is becoming an important choice in AI development.