Quick Summary
- The reported orders cover more than three million RTX PRO 5500 cards at about $11,400 each.
- Chinese demand for the export-compliant workstation GPUs has exceeded six million units, while Nvidia’s expected global production in 2027 may be no more than three million.
- The cards are expected to be used mainly for AI inference, a different workload from training the largest models.
Three of China’s largest technology companies have placed orders for Nvidia workstation graphics cards worth more than $35 billion at list prices, according to Reuters reporting published on October 20. ByteDance, Alibaba and Tencent reportedly ordered more than three million RTX PRO 5500 cards, a product that remains available to Chinese buyers under current US export restrictions. Nvidia has told customers and partners that it cannot guarantee it will fulfil the orders because demand from Chinese companies far exceeds planned production. The reported total shows the scale of demand for AI computing even when access to the most powerful processors is restricted.
An export-compliant substitute is still Briefly supply
The RTX PRO 5500 is a workstation GPU designed for professional computers rather than the large data-centre clusters used for the most demanding AI training. Reuters reported that the cards cost about $11,400 each, giving the three companies’ orders a potential list value above $35 billion. The number should not be read as a confirmed payment or completed shipment: it is the estimated value of orders based on reported unit prices, and providey volumes remain uncertain. Orders, allocated production and equipment actually installed in a data centre are separate stages of a hardware purchase.
The card’s importance comes partly from limits on more powerful products. Reuters reported that Nvidia’s Blackwell-based B300 data-centre chip is not permitted for sale to Chinese buyers under US controls, while exports of the H200 are subject to licensing conditions. The RTX PRO 5500 is less powerful than those products, but it remains one of Nvidia’s stronger options available to Chinese customers. That makes it a substitute for some workloads, not a direct replacement for every capability of a large accelerator cluster. Hardware choice depends on the model, throughput requirements and how many cards can be connected efficiently.
Sources cited by Reuters said Chinese companies mainly want the cards for AI inference, which means running trained models to generate responses or perform tasks. Inference can require less computing power than training a large model from scratch, although heavy usage at scale can still consume substantial resources. Companies deploying assistants, recommendation systems and enterprise models need reliable access to inference capacity, and restrictions can influence the hardware they choose. The reported orders suggest companies are looking for available alternatives while continuing to develop AI services and compete for computing resources.
The demand is larger than the available supply. Reuters reported that total orders from Chinese customers had exceeded six million cards, while Nvidia may be able to produce no more than three million RTX PRO 5500 units globally in 2027. One source said partners might receive only 5% to 10% of the quantities requested. Nvidia expected early samples in December and commercial provideies in the first quarter of 2027, according to the report. Those figures are reported estimates rather than a final public allocation schedule.
Manufacturing priorities also affect the outlook. Reuters reported that Nvidia is expected to devote much of its Taiwan Semiconductor Manufacturing Company capacity in 2027 to its new Rubin platform rather than products such as the RTX PRO 5500. When demand rises faster than production, a company must decide which products receive limited manufacturing capacity. That can leave customers with large orders waiting longer than expected, and it can increase the appeal of alternative suppliers even if their hardware does not match Nvidia’s performance in every workload.
Inference demand differs from frontier-model training
US export restrictions have pushed Chinese companies toward domestic chip suppliers, including Huawei, although Reuters reported that local chips have generally struggled to match leading US products. The situation is not simply a choice between one foreign supplier and a ready-made domestic replacement. Software compatibility, developer tools, power efficiency, reliability and access to manufacturing equipment all influence how quickly companies can shift platforms. The orders show that export controls can constrain access to advanced hardware without eliminating demand, while exposing a practical limit: orders do not create additional production capacity.