Nvidia chips may soon be back on the shopping lists of ByteDance and Alibaba, and this time the door could open through a workstation graphics card rather than a data-centre accelerator. On Sunday 27 September 2026, The Information reported that China’s government has signalled it could allow some domestic companies to buy the RTX Pro 5500, a Blackwell-generation card that Nvidia released this month for high-end professional computers. Reuters carried the story the same day and said it could not immediately verify it.

The detail that matters is who is asking the questions. According to The Information, China’s Ministry of Industry and Information Technology (MIIT) recently asked companies including Alibaba Group and ByteDance how many RTX Pro 5500 chips they want to buy and what they would use them for. The ministry has told some Chinese companies that the government intends to approve the purchases. The report also says it is unclear when approvals would come, how many Nvidia chips officials would allow and what standards they would use to decide.

That makes this a story about Beijing, not Washington. For most of 2026, the United States has been willing to license sales of advanced Nvidia chips to approved Chinese buyers, while China’s own regulators have held back the orders. Nvidia’s latest quarterly filing says in plain terms that its licensed H200 sales “were restricted by the PRC government”. The RTX Pro 5500 questionnaire looks like the next step in a rationing system Beijing has been building all year for artificial intelligence hardware.

This article explains what was reported, what the RTX Pro 5500 actually is, and how the H200 saga shaped the process. It also covers why ByteDance and Alibaba have been here before with a workstation card, what Nvidia now earns in China and how domestic chips from Alibaba and Huawei change the calculation. For related background, see our coverage of Huawei’s Ascend 960 launch plan and the US-China tariff cut agreed this week.

What The Information Reported About Nvidia Chips for ByteDance and Alibaba

china bytedance alibaba nvidia chips rtx pro 5500 b workstation tower case with fans

The scoop was written by Qianer Liu, The Information’s Hong Kong-based reporter covering semiconductors and AI in Asia. It was published at 8:00am Pacific time on 27 September under the headline “China Weighs Allowing Purchases of New Nvidia Chips by ByteDance, Alibaba”, and it cites two people familiar with the matter. Most of the article sits behind a paywall, so this summary relies on the two opening paragraphs the publication makes public and on the Reuters and Bloomberg write-ups.

The ministry’s two questions

According to The Information, MIIT, which oversees China’s tech sector, asked the companies two things about their plans for the RTX Pro 5500: how many units they want to buy, and what they will use them for. The ministry then told some of the companies that the government intends to approve the purchases.

That sequence, a request for quantities and uses followed by a signal of approval, is the same shape as a licensing process. A company’s answer effectively becomes its application. It also gives officials the information they would need to set a volume for each buyer, which is exactly what they did with the H200 earlier this year.

Three things the report leaves open

The Information is careful to say what is still unknown. It is unclear when approvals would arrive. It is unclear how many Nvidia chips officials would allow. It is unclear what standards they would use to decide who gets what.

Those three gaps are the whole story for Nvidia shareholders and for the buyers. The H200 went through an approval in January, a US licence round in May and a smaller allowance in July, and still produced shipments worth less than 1% of Nvidia’s data-centre revenue last quarter. An intention to approve Nvidia chips is not an order book.

How the story travelled

Reuters published a short rewrite from Bengaluru, noting that MIIT “told some Chinese companies the government intends to approve the purchases” and that Reuters “could not immediately verify the report.” Bloomberg ran it under the headline “China May Let Alibaba Buy New Nvidia Chips, The Information Says”.

The Information framed the move as a response to demand at home: China’s AI firms “strain for enough computing power to run their chatbots and agents”. The Reuters write-up did not include comment from Nvidia, ByteDance, Alibaba or MIIT.

The RTX Pro 5500: The Nvidia Chips in Question

china bytedance alibaba nvidia chips rtx pro 5500 c clipboard with a lined sheet

The Nvidia chips at the centre of the report are not H200s or a China-specific variant. They are the RTX Pro 5500 Blackwell, a professional graphics card that Nvidia added to its workstation line in mid-September with little fanfare. TechPowerUp described it on 14 September as a card NVIDIA “quietly added” to its pro-visualisation range. Nvidia’s own product page still lists it as “Coming Soon”.

What Nvidia’s spec sheet says

Nvidia’s product page lists 84 GB of GDDR7 memory with error-correcting code, memory bandwidth of 1,398 GB/sec, up to four DisplayPort 2.1b outputs and maximum power consumption of up to 600 W. It carries three ninth-generation NVENC encoders and three sixth-generation NVDEC decoders. The page describes fifth-generation Tensor Cores with FP4 support and “up to 3x the performance of the previous generation”.

TechPowerUp, working from the specification sheet published by board partner PNY, reports that the card uses the GB202 die with 21,760 CUDA cores on a 416-bit memory bus. That is the same CUDA core count as the consumer GeForce RTX 5090, paired with far more memory.

A workstation card built for racks

What sets the RTX Pro 5500 apart from most workstation cards is where it is meant to live. Nvidia says it is “designed for rack-mounted workstation deployment with a choice of air- or liquid-cooled thermal solutions”, so that IT teams can “centralize GPUs into managed racks, share them across users, and scale capacity as demand grows”. The liquid-cooled version comes in what Nvidia calls the RXM form factor.

It also supports Multi-Instance GPU, or MIG. TechPowerUp reports that the card can run as a single 84 GB instance or as two isolated 42 GB instances. In other words, this is a workstation product that behaves a lot like shared server infrastructure, which is why it could matter to companies that run AI services at scale.

How these Nvidia chips compare with the H200

The comparison below uses the figures Nvidia publishes for each product, with the RTX Pro 6000 Blackwell as the step above in the same family. The Zhenwu V900 column uses Alibaba’s own claims as reported by TechNode.

SpecificationRTX Pro 5500 BlackwellRTX Pro 6000 BlackwellH200 (SXM)Alibaba Zhenwu V900
CategoryRack-mounted workstation GPUWorkstation GPUData-centre acceleratorData-centre AI accelerator
Memory84 GB GDDR796 GB GDDR7141 GB216 GB
Memory bandwidth1,398 GB/s1,792 GB/s4.8 TB/sNot published (1,200 GB/s inter-chip)
Maximum powerUp to 600 W600 WUp to 700 WNot published
AvailabilityReleased September 2026, “Coming Soon” on Nvidia’s siteOn saleLicensed for approved China buyers since early 2026Mass production in Q1 2027
Who decides a China saleMIIT approval under discussionNot in the reportUS licence plus Chinese approvalDomestic product

Memory bandwidth is the clearest gap between these Nvidia chips. The RTX Pro 5500’s 1,398 GB/s is 29.1% of the H200’s 4,800 GB/s (1,398 ÷ 4,800), and the RTX Pro 6000’s 1,792 GB/s is 37.3%.

Memory bandwidth, as a share of the H200’s 4,800 GB/s
H200 4,800 GB/s (100%)
RTX Pro 6000 Blackwell 1,792 GB/s (37.3%)
RTX Pro 5500 Blackwell 1,398 GB/s (29.1%)

Why AI companies want these Nvidia chips

For training large models, a card with less than a third of the H200’s memory bandwidth is not a substitute. For running models, the picture is different. Nvidia’s page argues that the RTX Pro 5500’s memory “holds larger models, longer context windows, and multiple models at once—so agentic pipelines run without offloading.” Nvidia markets the card for large language model inference, AI agents, generative AI and computer vision workloads.

That lines up with The Information’s framing: Chinese firms need compute “to run their chatbots and agents”. Serving a chatbot to millions of users is an inference problem, and inference can be spread across many cheaper Nvidia chips in racks. A card that splits into two 42 GB instances also lets one physical GPU serve two workloads.

Why Beijing Now Decides Which Nvidia Chips Get In

china bytedance alibaba nvidia chips rtx pro 5500 d slatted freight crate

For three years, the question of which Nvidia chips China could buy was answered in Washington. In 2026 that changed. The United States eased its rules for Nvidia chips up to the H200, and China’s government became the main barrier to shipments. Reuters put it bluntly in January: “Beijing’s hesitation to allow imports has been the main barrier to shipments.”

Washington opened the door

On 8 December 2025, President Donald Trump announced that the United States would allow the H200 and similar products to be shipped to approved customers in China. The Commerce Department’s Bureau of Industry and Security (BIS) followed on 13 January 2026 with a rule, published in the Federal Register on 15 January, that moved licence reviews for the H200, AMD’s MI325X and similar chips from a presumption of denial to case-by-case review.

The conditions are specific. Exporters must certify that there is sufficient supply in the United States, that production for China will not divert foundry capacity from US customers, that the Chinese buyer has adequate security procedures, and that each item has passed independent third-party testing in the United States. Trump’s deal also gave Washington a 25% share of the revenue.

Beijing closed it in the same week

China moved in the opposite direction. On 7 January, The Information reported that Beijing had asked some tech companies to halt H200 orders and was expected to mandate purchases of domestic AI chips. Reuters then reported that Chinese customs had told agents the H200 was not permitted to enter China, and that officials had instructed companies not to buy the Nvidia chips unless necessary.

On 15 January, Nikkei Asia reported that the central government was drafting rules that would likely regulate the total volume of advanced AI chips local companies could buy. That points to a quota system, not a ban, and quotas need exactly the information MIIT has now requested for the RTX Pro 5500.

Nvidia’s own verdict

Nvidia’s 10-Q for the quarter ended 26 July 2026 describes the result. “The U.S. government, or USG, granted licenses that would allow us to ship small amounts of H200 products to specific China-based customers, but such sales were restricted by the PRC government, and we have been unable to sell all the products for which we have licenses,” it says.

The filing goes further. Nvidia says it was “effectively foreclosed from competing in China’s data center computing/compute market”, and that “the Chinese government has encouraged customers to purchase from our China-based competitors and discouraged customers from purchasing, importing, or using our data center products, including any China-specific product designed to comply with U.S. export controls.”

The timeline so far

DateEventSource
17 Sep 2025Cyberspace Administration of China tells ByteDance and Alibaba to stop testing and cancel orders for the RTX Pro 6000DFinancial Times via Reuters
8 Dec 2025Trump says the H200 can ship to approved customers in ChinaBIS
7 Jan 2026Beijing asks some firms to halt H200 ordersThe Information via Reuters
15 Jan 2026BIS case-by-case rule published in the Federal RegisterFederal Register
28-30 Jan 2026China approves ByteDance, Alibaba and Tencent for more than 400,000 H200s in total, then DeepSeek, all with conditionsReuters
Feb 2026US licences for small amounts of H200 beginNvidia 10-Q
May 2026About ten Chinese firms cleared by the US, capped at 75,000 chips eachReuters
8 Jul 2026Beijing may allow fewer than 200,000 H200s, less than half of what firms requestedThe Information via Reuters
26 Aug 2026Nvidia says China Hopper shipments were under 1% of data-centre revenueNvidia results and 10-Q
22 Sep 2026Alibaba unveils the Zhenwu V900 AI chipTechNode
27 Sep 2026MIIT asks Alibaba and ByteDance about RTX Pro 5500 plansThe Information

The H200 Playbook: How China Rations Nvidia Chips

china bytedance alibaba nvidia chips rtx pro 5500 e four columns stepping down

The best guide to what happens next with the RTX Pro 5500 is what happened with the H200. Each stage of that process produced a headline about China reopening to Nvidia chips, and each was followed by a smaller reality.

January: approval with conditions

On 28 January, Reuters reported that ByteDance, Alibaba and Tencent had been approved to buy more than 400,000 H200 chips in total, with the approvals granted during Nvidia chief executive Jensen Huang’s visit to China. The sources said Beijing was “only granting approvals with conditions”, and that those conditions were still being decided. A fifth source said the licences were “too restrictive” and customers were not converting approvals into purchase orders.

Two days later, Reuters reported that DeepSeek had also received conditional approval, and that the conditions were being set by the National Development and Reform Commission, China’s state planner. One proposal discussed earlier would have required each H200 purchase to be bundled with a set ratio of domestic chips.

May: Washington signs off, buyers wait

In May, Reuters reported that the US had cleared about ten Chinese companies to buy the H200, including Alibaba, Tencent, ByteDance and JD.com, with each capped at 75,000 chips. Lenovo and Foxconn were authorised as distributors. Transactions still froze.

Commerce Secretary Howard Lutnick told a Senate hearing the reason: “the Chinese central government has not let them, as of yet, buy the chips, because they’re trying to keep their investment focused on their own domestic industry.”

July: a smaller allowance

On 8 July, The Information reported that Chinese officials had told Alibaba, ByteDance and DeepSeek they might soon receive permission to buy some H200 chips. Beijing was still deciding the number, which “could amount to fewer than 200,000 in total”, less than half of what the companies had requested earlier in the year, according to the Reuters write-up. Nvidia shares rose 1% on the report.

How many Nvidia chips actually shipped

Nvidia’s August results showed how little came of it. Shipments of Hopper data-centre products to China in the quarter “were less than 1% of Data Center revenue”, which was $89.0 billion. That puts them under $890 million (1% of $89.0 billion). Nvidia also took a $0.4 billion charge in the first half of its fiscal year for excess H200 inventory and purchase obligations, “as the demand for H200 products diminished.”

Each licensed H200 has to pass inspection in the United States before shipment, so it is subject to a 25% tariff on import. Nvidia says it has been unable to pass any of that tariff on to customers.

Two rounds of Nvidia chips approvals side by side

QuestionH200 round (Jan-Jul 2026)RTX Pro 5500 round (Sep 2026)
Chinese agencies namedIndustry and commerce ministries, NDRC, customsMIIT
Companies namedByteDance, Alibaba, Tencent, DeepSeekByteDance, Alibaba
What buyers were askedBuy only when necessary; conditions still being setHow many units, and for what use
Volume400,000+ approved in January; fewer than 200,000 floated in JulyNot reported
US sideLicences with a 75,000-chip cap per company and a 25% cutNot addressed in the report
Outcome so farUnder 1% of Nvidia data-centre revenue; $0.4bn chargeIntention to approve; timing unknown

ByteDance, Alibaba and the Nvidia Chips Beijing Blocked in 2025

china bytedance alibaba nvidia chips rtx pro 5500 f funnel over a jar

This is not the first time ByteDance and Alibaba have lined up for an Nvidia workstation-class card. A year ago, the same two companies were told to cancel their orders for Nvidia chips of this kind.

The RTX Pro 6000D orders

In September 2025, the Financial Times reported that the Cyberspace Administration of China (CAC) had told companies including ByteDance and Alibaba to stop testing the RTX Pro 6000D and cancel their orders. The RTX Pro 6000D was an AI inferencing chip Nvidia designed specifically for the Chinese market. According to the report, as carried by Reuters, several companies had indicated they would order tens of thousands of units and had begun testing and verification work with Nvidia’s server suppliers.

The FT described the ban as stronger than earlier guidance aimed at the H20, the previous China-tailored chip. Reuters had reported earlier that week that the RTX Pro 6000D had seen only lukewarm demand.

The H20 security questions

The earlier guidance followed a dispute about security. Nvidia’s 10-Q says that after US legislative proposals called for mandatory features in Nvidia chips, “China’s government publicly questioned whether our H20 products have built-in vulnerabilities, discouraging customers from purchasing our products.” Nvidia replied that its GPUs “do not include such built-in vulnerabilities”.

Data Center Dynamics reported that the Chinese government issued a directive in August 2025 requiring publicly owned domestic data-centre companies to source more than half of their chips from domestic producers.

The Mellanox antitrust finding

On 15 September 2025, China’s antitrust regulator published a preliminary finding that Nvidia’s compliance with US export controls, “which required us to offer degraded products to the Chinese market, discriminated unfairly against customers in the China market”, in breach of the conditions attached to its Mellanox acquisition. Nvidia’s latest 10-Q still lists that finding as a risk that could lead to financial penalties or restrictions on its business in China.

What is different this time

Three things stand out. The agency asking questions is MIIT, the industrial ministry, rather than the internet regulator that shut down the RTX Pro 6000D. The request is for purchase plans, not a stop order. And the report comes as China’s AI firms, in The Information’s words, “strain for enough computing power”.

None of that guarantees an outcome. The H200 was approved in January and still barely shipped. But for ByteDance and Alibaba, a ministry asking how many Nvidia chips they want is a very different message from a regulator telling them to cancel orders.

What Sales of Nvidia Chips in China Add Up To Now

Nvidia’s China business looks very different from the one it had before the export controls, and workstation Nvidia chips like the RTX Pro 5500 would slot into the part that is still growing.

China revenue doubled, but not from data-centre compute

Nvidia’s 10-Q reports revenue by customer headquarters location. In the quarter ended 26 July 2026, customers headquartered in China, including Hong Kong, accounted for $7,880 million, up from $3,985 million a year earlier, an increase of 98%. Over the first half of the fiscal year, China revenue was $12,430 million, down from $13,644 million.

Because Hopper data-centre shipments to China were under 1% of the $89.0 billion data-centre line, at most about $890 million of that $7.88 billion came from them. Nvidia does not break down the rest. Its geographic figures are based on where each customer is headquartered, not where the products end up, so the China line can include gaming and workstation cards Nvidia is free to ship as well as purchases by China-headquartered companies for use elsewhere.

Nvidia revenue by customer headquarters, quarter ended 26 July 2026 (total $96,221m)
United States $60,074m (62.4%)
Taiwan $26,985m (28.0%)
China including Hong Kong $7,880m (8.2%)
Other $1,282m (1.3%)

The percentages are each region’s revenue divided by the $96,221 million total. China’s 8.2% share is almost unchanged from a year earlier, when $3,985 million of $46,743 million came to 8.5%.

A $108 billion outlook with no China data-centre compute

Nvidia’s outlook for the third quarter of fiscal 2027 is revenue of $108.0 billion, plus or minus 2%. The company states: “NVIDIA is not assuming any Data Center compute revenue from China in its outlook.” Any RTX Pro 5500 sales to Chinese AI firms would therefore be upside against guidance, although not from the data-centre line.

Where a workstation card would book

In Nvidia’s current reporting, workstation sales sit in Edge Computing. That segment brought in $7.2 billion in the second quarter, up 27% from a year earlier. Nvidia said “the increases were driven by strong sales of Blackwell workstations, partially offset by slower consumer PC sales that were tempered by elevated memory and systems prices”.

So if Beijing approves these Nvidia chips in volume, the most visible effect would be in a segment that is already growing, not in the data-centre business where China has been written down to zero.

The costs Nvidia absorbs

The China business has been expensive for Nvidia. It took a $4.5 billion charge in the first quarter of fiscal 2026 for H20 inventory after the US imposed a licence requirement in April 2025, and a further $0.4 billion charge for H200 inventory this year. The 25% tariff on licensed H200s is another cost it says it cannot pass on.

China's Own Chips: The Other Side of the Nvidia Chips Question

Every approval Beijing grants for Nvidia chips is weighed against the domestic industry it wants to build. The week before The Information’s report, Alibaba showed how far that industry has come.

Alibaba’s Zhenwu V900

On 22 September, Alibaba’s chip unit T-Head unveiled the Zhenwu V900 at the Apsara Conference in Hangzhou. According to TechNode, T-Head claims three times the performance of its predecessor, the Zhenwu M890, with 216 GB of memory, 1,200 GB/s of inter-chip bandwidth and native support for FP8 and FP4. More than 1,000 V900 chips can work together as a single system, and Alibaba says a cluster can scale to as many as 500,000 accelerators.

The V900 is expected to enter mass production and go on sale in the first quarter of 2027. T-Head says the Zhenwu series has already served more than 650 enterprise customers.

Memory per chip, as a share of the Zhenwu V900’s 216 GB
Alibaba Zhenwu V900 216 GB (100%)
Nvidia H200 141 GB (65.3%)
RTX Pro 6000 Blackwell 96 GB (44.4%)
RTX Pro 5500 Blackwell 84 GB (38.9%)

Memory capacity is one measure, not a verdict on performance. The shares are each chip’s memory divided by 216 GB, and the V900 figure is Alibaba’s claim for a chip that is not yet in mass production.

Huawei’s Ascend roadmap

Huawei is on a similar timeline. As we reported this month, it pulled its Ascend 960DT forward to the first quarter of 2027, three quarters earlier than its 2025 roadmap, and added an Ascend 980 for 2029. Our Ascend 960 analysis also found that the SuperPoD chip counts Huawei cited this year fall well short of the ones it promised in 2025.

Why imported Nvidia chips still matter until 2027

Both flagship domestic chips reach volume in 2027. Until then, Chinese AI companies need compute now, and imported Nvidia chips are one way to get it. Reuters has described a “growing computing capacity crunch” facing the country’s tech firms. Beijing is also building the data centres to use it, as our report on AI data centres rising in Inner Mongolia showed.

Workstation Nvidia chips like the RTX Pro 5500 fit that gap neatly. It helps AI services run today, while the data-centre accelerators that decide long-term competitiveness remain a domestic priority.

Beijing’s efficiency standards

There is a further lever. Nvidia’s 10-Q says a Chinese government agency has announced an Action Plan that endorses new standards for the compute performance per watt and per unit of memory bandwidth of accelerators used in new and renovated data centres. Nvidia warns that the plan could be applied “in a way that effectively prevents us from being able to design products to meet the new standard”.

Export Rules: Why Workstation Nvidia Chips Take a Different Route

The report says nothing about US approval for RTX Pro 5500 sales to China, and that silence is worth examining. Workstation cards and data-centre Nvidia chips have been treated differently under US controls.

What Nvidia says about workstation GPUs

Nvidia’s 10-Q states that “as of the end of the second quarter of fiscal year 2027, while we were able to ship uncontrolled products to China, such as gaming and workstation GPUs, we were effectively foreclosed from competing in China’s data center computing/compute market.” In other words, some Nvidia chips have kept flowing to China all along, which is part of why its China revenue line never fell to zero.

Whether the RTX Pro 5500 falls into that uncontrolled category is not stated in The Information’s public excerpt, the Reuters write-up or Nvidia’s product page. The same filing notes that US licensing requirements since October 2023 have covered products above performance thresholds, including the RTX 4090, so a workstation label alone does not settle the question.

The limits of the January rule

The BIS rule that opened the H200 route covers “the Nvidia H200 and its equivalents, as well as less advanced chips”, but only if the semiconductors were “commercially available in the United States at the time of publication of this rule”. That publication date was 15 January 2026. The RTX Pro 5500 was announced in September.

That does not mean the card is barred. It may not need a licence at all, or it may be handled under a separate process. It does mean the H200 framework cannot simply be assumed to apply, and anyone modelling RTX Pro 5500 sales to China should watch for a US statement as closely as a Chinese one.

The Remote Access Security Act

Congress is also looking at the cloud. Nvidia’s filing says lawmakers are considering the Remote Access Security Act, which “could prohibit the provision of cloud services to any company with an ultimate parent headquartered in China.” For Chinese firms that rent Nvidia capacity abroad rather than buying hardware at home, that would close another route to Nvidia chips.

Rules that follow the buyer

US licensing requirements also apply to “a party headquartered in, or with an ultimate parent headquartered in” Country Group D5, which includes China, according to Nvidia’s filing. So the relevant question is not only where a chip is shipped, but who ultimately owns the buyer. That matters for ByteDance and Alibaba subsidiaries outside China, and for any reseller dealing with them.

What Happens Next for Nvidia Chips in China

The Information’s report describes an intention to let Nvidia chips in, not a decision. Here are the plausible outcomes and what would signal each one.

Four scenarios

ScenarioWhat it would look likeSignal to watch
Quick, sizeable approvalMIIT approves purchases for several firms within weeksPurchase orders reported; Nvidia comments on China demand
Approval with conditionsVolumes capped per firm, possibly tied to domestic chip purchasesReports of quotas or a domestic-to-import ratio
Slow approval, small volumesRepeats the H200 path: headlines, then a trickleNvidia’s next quarterly report on China revenue
ReversalA regulator halts testing or orders, as with the RTX Pro 6000DCAC or customs guidance to buyers

Signals to watch

The first is purchase orders. In January, Jensen Huang said Nvidia viewed purchase orders as a signal of approval rather than expecting a formal announcement from Beijing, and the same test applies here.

The second is Nvidia’s next set of results. The company has guided to no China data-centre compute revenue, so any change will show up first in Edge Computing commentary.

The third is Washington. A US statement on how the RTX Pro 5500 is classified for China would settle the question the report leaves open. The fourth is Alibaba’s own chip: the more Zhenwu V900 volume Alibaba can promise for 2027, the less pressure Beijing is under to approve imported Nvidia chips beyond a stopgap.

What It Means for Businesses Buying AI Hardware

For most organisations outside China, this story is not about access. It is about what it reveals about the market for Nvidia chips and other AI hardware in the next year.

Workstation GPUs are now AI infrastructure

Nvidia is selling the RTX Pro 5500 for rack deployment and for running agents, and Chinese regulators are treating it as strategic AI hardware. That confirms a shift we covered in our local LLM hardware guide and our guide to choosing a PCIe enterprise AI GPU: professional cards with large memory have become a serious option for running models on premises.

Watch supply and pricing of Nvidia chips

If large Chinese buyers are approved, they would compete for the same Blackwell workstation silicon as everyone else. Nvidia already notes “elevated memory and systems prices” in its consumer PC business. Buyers planning workstation or rack GPU purchases for 2027 should get quotes and lead times early, rather than assuming this year’s prices hold.

Check who owns your customers

Because US licensing follows the ultimate parent as well as the destination, firms that sell, rent or resell GPU capacity should know whether any customer’s parent company is headquartered in China. The rules have changed several times in 2026 and may change again, so compliance checks are worth doing before a contract is signed, not after.

Chinese open models depend on this compute

Nvidia’s 10-Q notes that open-source models originating in China, “such as DeepSeek, Qwen, or Kimi”, are growing in popularity worldwide. Many businesses test or run those models. How much compute their developers can get, from Nvidia chips or from domestic ones, affects how quickly those models improve and how they are licensed. It is worth tracking alongside the models themselves, as our report on China’s AI video industry shows.

Nvidia Chips and China: Frequently Asked Questions

Has China approved sales of the RTX Pro 5500 to ByteDance and Alibaba?

Not yet, according to the report. The Information says MIIT asked Alibaba, ByteDance and other companies how many RTX Pro 5500 chips they want and what for, and told some of them the government intends to approve the purchases. It is unclear when approvals would come or how many Nvidia chips would be allowed.

What is the Nvidia RTX Pro 5500?

It is a Blackwell-generation professional GPU with 84 GB of GDDR7 memory, 1,398 GB/s of memory bandwidth and up to 600 W of power, designed for rack-mounted workstation deployment. Nvidia released it in September 2026, and its product page lists it as coming soon.

Why does China need to approve imports of Nvidia chips?

Beijing has used customs guidance, ministry approvals and conditions set by its state planner to control which Nvidia chips its companies buy. The aim, according to Reuters and US officials, is to balance urgent demand for AI compute against support for China’s domestic chip industry.

Is the RTX Pro 5500 as powerful as the H200?

Not on the measures Nvidia publishes for both. Its 1,398 GB/s of memory bandwidth is 29.1% of the H200’s 4.8 TB/s, and it has 84 GB of memory against the H200’s 141 GB. Nvidia pitches it at running AI models, agents and graphics workloads rather than training the largest models.

How much does Nvidia earn in China now?

Customers headquartered in China, including Hong Kong, accounted for $7.88 billion of Nvidia’s $96.2 billion revenue in the quarter ended 26 July 2026. Hopper data-centre shipments to China were under 1% of data-centre revenue, and Nvidia’s outlook assumes no China data-centre compute revenue.

Has China blocked Nvidia chips for workstations before?

Yes. In September 2025, the Cyberspace Administration of China told ByteDance, Alibaba and other companies to stop testing and cancel orders for the RTX Pro 6000D, a China-specific Nvidia inference chip, according to the Financial Times.

References

China Weighs Allowing Purchases of New Nvidia Chips by ByteDance, Alibaba (The Information)

China weighs allowing ByteDance, Alibaba to buy new Nvidia chips, The Information reports (Reuters)

China May Let Alibaba Buy New Nvidia Chips, The Information Says (Bloomberg)

NVIDIA RTX PRO 5500 Blackwell product page

NVIDIA Unveils RTX PRO 5500 Blackwell Workstation GPU with 84 GB GDDR7 Memory (TechPowerUp)

NVIDIA H200 Tensor Core GPU

NVIDIA Announces Financial Results for Second Quarter Fiscal 2027

NVIDIA Form 10-Q for the quarter ended July 26, 2026

Department of Commerce Revises License Review Policy for Semiconductors Exported to China (BIS)

Revision to License Review Policy for Advanced Computing Commodities (Federal Register)

China gives nod to ByteDance, Alibaba and Tencent to buy Nvidia’s H200 chips (Reuters)

China conditionally approves DeepSeek to buy Nvidia’s H200 chips (Reuters)

China asks tech firms to halt orders for Nvidia’s H200 chips, Information reports (Reuters)

China drafting purchase rules for Nvidia H200 chips, Nikkei Asia reports (Reuters)

U.S. approves Chinese companies to buy Nvidia H200 AI chips (Yahoo Finance)

China plans to let top AI firms buy limited Nvidia H200 chips, the Information reports (Reuters)

China orders tech firms to halt Nvidia AI chip buys, FT reports (Reuters via Malay Mail)

China’s Internet regulator tells Chinese tech companies to stop buying Nvidia AI chips (DCD)

T-Head unveils Zhenwu V900 AI chip in Alibaba’s push to expand its AI infrastructure stack (TechNode)