John MACDOUGALL
Co-founder, President and CEO of US technology company Nvidia, Jensen Huang delivers his keynote speech next to computer components during the Nvidia GTC Berlin 2026 Artificial Intelligence conference, on October 21, 2026 in Berlin.

Spot quotes for Nvidia's (NASDAQ: NVDA) flagship RTX 5090 gaming graphics card at computer malls in Shenzhen's Huaqiangbei have climbed to 51,000-55,000 yuan per card, or about HK$56,000-HK$60,000. When reporters from the National Business Daily visited several retailers on October 25, some said the price had still been in the 40,000-yuan range last month, meaning it had risen by nearly 10,000 yuan in just over a month. A visit by 21st Century Business Herald on October 23 found wholesale-channel quotes of 55,000-60,000 yuan. The RTX 5090 launched at around 15,000-16,000 yuan, putting current prices at more than three times the original level. The next-tier RTX 5080 has also risen from a launch price of about 8,000 yuan to 16,000-17,000 yuan, nearly doubling.

Nvidia denies production halt, but market keeps bidding up prices

The latest surge was triggered by a rumour Nvidia has publicly denied. On October 10, two hardware tipsters separately claimed on social media that Nvidia had stopped supplying the GB202 core chip used in the RTX 5090 to GeForce gaming cards, with all new output redirected to RTX Pro professional cards. One of them posted what appeared to be an internal notice from a graphics card maker, saying the halt also covered the mainland-specific RTX 5090D. German hardware publication PC Games Hardware subsequently cited wholesale and retail-channel sources on October 15 as saying Nvidia had stopped allocating new GB202 chips to gaming cards.

Jeff Fisher, Nvidia's senior vice-president in charge of its GeForce business, said in Taipei on October 16 that the RTX 5090 had not been discontinued and that the company would continue shipping it. He acknowledged, however, that capacity across the supply chain was constrained. He did not say when supplies would return to normal or provide shipment figures.

The two accounts may not necessarily conflict. Even if Nvidia has stopped allocating new chips, board makers can continue shipping chips and finished products already in their inventory. The question is how long those stocks will last. Huaqiangbei retailers generally said it was difficult to obtain new supplies upstream and that most products currently circulating were from their own inventories. But when asked whether they had received formal discontinuation notices from distributors, the retailers could not provide evidence.

Same chip, nearly sixfold price gap

There is a commercial rationale behind the rumour. GB202 is not exclusive to gaming cards. Nvidia's RTX Pro 6000, RTX Pro 5500 and RTX Pro 5000 professional workstation graphics cards all use the same chip. The RTX Pro 5500 has the same 21,760 CUDA cores as the RTX 5090, but comes with 84GB of GDDR7 memory, about 2.6 times the 32GB on the RTX 5090.

Reuters, citing people familiar with the matter, reported that the RTX Pro 5500 sells for about US$11,400, or approximately HK$89,000, per card. The RTX 5090 has a US recommended retail price of US$1,999, or about HK$15,600. The same chip therefore commands about 5.7 times more when sold to enterprises than to gamers.

More than US$35 billion in orders, most may not be fulfilled

How strong is enterprise demand? Reuters reported on October 20, citing two people familiar with the matter, that ByteDance, Alibaba (9988.HK) and Tencent (0700.HK) had jointly ordered more than 3 million RTX Pro 5500 cards, worth over US$35 billion, or about HK$273 billion, at list prices. The RTX Pro 5500 is a workstation graphics card Nvidia developed to comply with US export controls. Although it is far less powerful than data-centre chips such as the H200 and B300, it remains one of the most powerful Nvidia products currently available to mainland buyers. The three companies mainly intend to use the cards for AI inference, or running models that have already been trained.

However, the size of the orders far exceeds Nvidia's production capacity for several reasons:

  • Orders and capacity: The same report said total orders from mainland customers had exceeded 6 million cards, while Nvidia could produce no more than about 3 million globally in 2027.
  • Delivery ratio: Nvidia has told customers and partners that it cannot guarantee full delivery. One source said partners might receive only 5%-10% of the quantities ordered. Based on the three companies' combined order of 3 million cards, that would amount to about 150,000-300,000 units.
  • Timetable: Nvidia expects to provide samples in December, with commercial shipments beginning in the first quarter of 2027.
  • Future capacity: Another source said Nvidia would allocate most of its TSMC (2330.TW) capacity to its next-generation Rubin data-centre platform from 2027.

Nvidia and the three companies have not publicly confirmed the details of the orders.

The three technology giants are not the only ones in the queue. Shenzhen-listed Tanyang Technology (A-share code: 300872.SZ) announced on October 13 that it would buy 20,000 RTX Pro 5500 cards from an unnamed counterparty. The company also warned of risks including delays in delivery by instalment and receiving fewer cards than expected, with the seller not necessarily liable for breach of contract.

In other words, those large orders will not begin to be delivered until next year, so the current price surge in Huaqiangbei is not simply the result of the orders having already "taken away" capacity. But it clearly shows that enterprise demand for high-end chips such as GB202 is rising sharply, leaving gaming cards further down the allocation list.

More than one factor behind inflated prices

It would also be incomplete to attribute current prices entirely to large companies competing for supplies. Other factors include:

  • AI applications: During its visit, the National Business Daily found that retailers said most current RTX 5090 buyers were purchasing them for AI applications. Many small and medium-sized enterprises and developers use the cards to run open-source large language models locally or generate AI images, with the 32GB of memory being a key advantage. Some gamers are still buying them for gaming.
  • Memory costs: A sharp rise in memory prices earlier this year pushed up graphics card costs. During a visit in August, 21st Century Business Herald reported that the RTX 5090 had already risen from about 20,000 yuan to roughly 35,000 yuan. Retailers attributed the increase to higher graphics memory costs and factories withholding stock.
  • Overseas buyers: An October report by the same publication said overseas buyers from India, Russia, France, Italy and Brazil had been making intensive purchases of graphics cards and other components in Huaqiangbei since September.

The rumour remains unverified, but prices have already been driven higher.

Gaming business becoming increasingly "small" in Nvidia's accounts

Nvidia's revenue mix helps explain its priorities. In the second quarter of fiscal 2027, ended July 26, 2026, Nvidia reported revenue of US$96.2 billion, up 106% year on year. Data-centre revenue came to US$89 billion, up 117% and accounting for about 92% of the total. According to overseas media reports, Nvidia no longer separately disclosed gaming revenue in those results. The company expects third-quarter revenue of about US$108 billion, plus or minus 2%, a forecast that does not include any data-centre computing revenue from China.

What to watch next

First, Nvidia's third-quarter results. As gaming revenue is no longer reported separately, it will be difficult to directly compare the performance of the gaming and data-centre businesses. A more useful indicator will be whether management comments on GeForce supplies and the allocation of GB202 chips.

Second, actual deliveries of the RTX Pro 5500. The number of samples supplied in December and commercial shipments in the first quarter of next year will show how many GB202 chips Nvidia ultimately allocates to professional cards.

Third, retail restocking. Only if multiple board makers resume stable shipments and retailers return to normal restocking will Nvidia's claim that it is continuing to ship the RTX 5090 be shown to have reached the market. Otherwise, high prices in Huaqiangbei could persist. There is also an unconfirmed rumour that Nvidia may launch a 24GB version of the RTX 5080 in the first quarter of 2027 to replace its flagship position.

Fourth, progress by mainland Chinese GPU makers. The large-scale purchases of Nvidia products that comply with export controls are a transitional arrangement. The prices of domestic products such as Huawei's Ascend 950PR are already close to that of the RTX Pro 5500. It is worth noting that when reports emerged in late September that Beijing might approve companies' purchases of the RTX Pro 5500, mainland semiconductor shares fell the following day. This suggests the market also sees Nvidia's return as competitive pressure on domestic manufacturers, rather than merely an opportunity for substitution.