ByteDance, Alibaba and Tencent have ordered over three million RTX PRO 5500 cards, with a collective sticker value above $35 billion, Reuters reported on October 20, citing unnamed sources. The report describes the cards as export-compliant. Nvidia cannot guarantee full deliveries; one source said partners might receive only 5–10% of requested quantities. Sources expect commercial shipments in early 2027. The buyers did not respond to Reuters. Nvidia’s spokesperson criticized restrictions on U.S. suppliers.
That distinction changes the meaning of the headline number. An order is not a completed sale, a delivery or an operating AI service. The reported sum describes potential demand—not revenue Nvidia has already collected. The immediate contest is over access to machines, not just access to money.
Why the RTX PRO 5500 matters
Nvidia’s own product page describes a Blackwell workstation GPU for rack-mounted deployments, with 84 GB of GDDR7 memory and support for language-model inference, agents, simulation and graphics. Its published specifications remain preliminary, and the page still labels it as coming soon.
This is not simply a gaming card being marketed under an AI label. Nvidia explicitly targets professional AI workloads and offers partitioning into two isolated GPU instances, each with its own resources. Those are manufacturer-described capabilities, not independent evidence of how a Chinese buyer’s production service will perform.
For readers following the AI race, that is an important distinction: the ability to run models does not establish a universal replacement for a large training cluster. Workload fit, software, memory and deployment design still matter. A model generating answers and a team training its successor are not asking the hardware to do the same job.
Export access is not an on-off switch
The regulatory background also needs precision. In its January 13, 2026 announcement, the U.S. Bureau of Industry and Security described case-by-case licensing for H200, AMD MI325X and similar processors—not unrestricted sales. Conditions included protecting production capacity available to U.S. customers, buyer compliance procedures and independent testing in the United States. That historical announcement is context, not a new October approval for these orders.
Our reading: policy permission and physical availability are separate bottlenecks. A product can fit a permitted route and still fail to arrive in the volume a buyer wants. Conversely, having a budget says little about whether the right computing capacity will be available when a service needs it.
The next meaningful number is therefore not another headline order total. It is how much hardware actually ships, reaches customers and becomes useful capacity. Until then, the gap between wanting Nvidia chips and operating them remains the story.
