Three of China’s biggest technology companies have placed one of the largest GPU orders in history, and Nvidia has told them it probably cannot fill it. ByteDance, Alibaba and Tencent have ordered more than 3 million RTX PRO 5500 graphics cards with a combined list price above $35 billion, Reuters reported on October 20, citing two people familiar with the matter and a third source. The cards are export-compliant workstation GPUs, the most powerful Nvidia hardware still legally available to Chinese buyers.
The story matters less for the headline number than for what it reveals. US export controls have not dented Chinese demand for Nvidia hardware. They have just redirected it toward whatever Nvidia is still allowed to sell.
The numbers behind the $35 billion order
According to the report, the three companies are paying about $11,400 per card, roughly ₹10 lakh each. Here is the math:
| Metric | Figure |
|---|---|
| Cards ordered by the three companies | 3,000,000+ |
| List price per card | $11,400 (roughly ₹10 lakh) |
| Combined list value | $35 billion+ (roughly ₹3.1 lakh crore) |
| Total RTX PRO 5500 demand in China | 6,000,000+ cards |
| Nvidia’s planned global production in 2027 | No more than 3,000,000 cards |
| Share of requested orders Nvidia may fill | 5% to 10%, per one source |
Put another way, Chinese buyers have ordered more than twice as many cards as Nvidia can build next year, and one source told Reuters that individual partners might receive only 5 to 10 percent of the quantities they asked for.
Why a workstation card, not an AI superchip
The RTX PRO 5500 is a workstation chip built for powerful professional desktop computers, not for the giant data center clusters that run Nvidia’s flagship AI training processors. But it sits in a sweet spot of the export rules. Nvidia’s Blackwell-based B300 data center chip is banned outright from sale to Chinese buyers, and the Hopper-based H200 goes through a case-by-case licensing review. The RTX PRO 5500 slips through because it is classified as a workstation GPU.
The three companies reportedly want the cards mainly for AI inference, the stage where a trained model actually serves users: generating chatbot answers, writing code suggestions, tagging photos. Inference needs far less raw computing power than training a model from scratch, so a card that would be modest for training a frontier model can be perfectly capable of serving one. Nvidia’s Jensen Huang has said the company’s share of China’s AI accelerator market has collapsed from more than 90 percent to nearly zero under the curbs. This order suggests Chinese companies are simply buying whatever slice is still available.
Samples in December, then a factory fight
Nvidia has told customers to expect early samples in December, with commercial shipments starting in the first quarter of 2027, according to the report. After that, the schedule gets complicated. The company is expected to devote much of its manufacturing capacity at TSMC to its next-generation Rubin platform from 2027 onward, which leaves limited room to ramp up RTX PRO 5500 production even as orders keep piling in.
That collision, between booming Chinese demand for export-compliant hardware and a factory line about to be repurposed for Nvidia’s next big thing, is why the fulfillment numbers look so grim.
What this means for the AI hardware race
First, the inference buildout is real. If three companies are willing to put $35 billion on the table for workstation-class cards, it means a real bottleneck in AI is shifting from training giant models to serving them at scale. That is the same logic behind Nvidia’s new open models built to run agents inside corporate data centers and on PCs, and behind models like Meta’s 816-billion-parameter open-weight model, which is designed for exactly this kind of inference deployment.
Second, expect the domestic alternative to get louder. The Reuters report notes that Chinese firms are also exploring homegrown chips, including Huawei’s, which will get a lot more attention if RTX PRO 5500 allocation falls to single-digit percentages.
Third, prices are worth watching. An $11,400 list price for a card in massive demand and limited supply is an invitation to a gray market. Nvidia’s official line is that it cannot guarantee filling the orders. The unofficial line, judging by the numbers, is that a large share of this $35 billion may never turn into hardware.
The broader trend is familiar by now: every time Washington tightens the rules, Chinese buyers find the most powerful chip the rules still allow. Microsoft’s answer was to build its own silicon, the Maia 300, which we covered last week. ByteDance, Alibaba and Tencent are taking the other route: buy everything Nvidia is still permitted to ship, and hope the factory keeps up.
