NVIDIA GPU Purchases: Chinese Companies Buy Over 1 Million in a Year

Chinese technology companies reportedly bought more than 1 million NVIDIA gaming and workstation GPUs in a year to meet their need for AI computing capacity. According to Tom’s Hardware, which reported the Financial Times story based on industry sources on 13 October 2026, these purchases are linked to demand outstripping available supply despite China’s increased domestic chip production. In a report published the same day, Moneycontrol also wrote that Chinese companies were struggling to find enough computing power for their growing AI investments.

Domestic production is growing, but demand is higher

The Financial Times report, as cited by Tom’s Hardware, says China doubled its domestic AI chip production this year. Even so, production is expected to remain at around 3.5 million chips, while demand is reported to have significantly exceeded 5 million units. The report puts the resulting gap at approximately 1.5 million chips. These figures are presented as estimates based on industry sources interviewed by the Financial Times, not as official company statements.

To bridge this gap, NVIDIA gaming and workstation cards that comply with U.S. export controls are reportedly being used for AI workloads. According to the report, most purchases consist of the RTX Pro 6000D, RTX Pro 5000 and RTX 5090. The sources do not provide a detailed breakdown of which companies bought the products, when they did so or in what quantities.

What AI workloads are the GPUs used for?

The report says these cards are not considered to be in the same class as China’s high-end domestic solutions, such as the Huawei Ascend 910C. However, NVIDIA GPUs can be used for simpler AI workloads, including the “prefilling” stage. The purchases therefore do not mean that every type of AI training or inference is being run on these cards.

The Financial Times report, citing industry sources, also identifies NVIDIA’s CUDA software as one reason for choosing the cards. According to the report, CUDA remains a more convenient option for companies than Huawei’s Cann software. This suggests that hardware choices may be influenced not only by chip capacity but also by the existing software ecosystem.

RTX 5090 pricing and the impact on visualization workflows

Tom’s Hardware lists the RTX 5090’s announced retail price as $1,999. According to its report, the card was available from three third-party sellers at prices no lower than $7,199. This pricing and availability information is based on the market observations in Tom’s Hardware’s report; it does not reflect sales conditions in Turkey.

For architecture firms, interior designers and visualization teams, the report is a reminder that GPU demand comes not only from gaming or professional visualization, but also from AI workloads. Growing demand for cards used for AI could put pressure on the pricing and availability of certain GPU models; however, the sources do not measure the impact of these purchases on the Turkish market. Those planning workstation investments should assess the model, memory and software compatibility in relation to their own workflows, and check prices and stock directly before purchasing.