Nvidia Halts H200 AI Chip Shipments to China | Zeniteq
Loading
Zeniteq is now available on mobile for iPhone, iPad and Apple Watch.Download nowLearn more
Nvidia Halts H200 AI Chip Shipments to China
Nvidia has stopped licensed Hopper shipments, including H200 processors, and its next-quarter outlook assumes no data-center compute revenue from China.
Nvidia has halted shipments of licensed Hopper-generation data-center chips to China, including H200 processors, and its official fourth-quarter fiscal 2027 guidance assumes no data-center compute revenue from the country.
Interim CEO Colette Kress disclosed the halt on Nvidia's November 17 earnings call. The company had reported $2.4 billion in licensed Hopper sales to China during the quarter ended October 25, 2026, roughly 2.2% of its $107.6 billion worldwide Data Center revenue. Nvidia gave no reason for stopping shipments and has not said whether the halt is permanent.
Nvidia has stopped shipping licensed Hopper-generation data-center chips to China, including its H200 AI processors, interim CEO Colette Kress said on the company’s earnings call. The company’s official outlook also assumes no data-center compute revenue from China in the fourth quarter of fiscal 2027.
The announcement follows $2.4 billion in licensed Hopper sales to China during the quarter ended October 25, 2026. Reuters reported the halt on November 18, following Nvidia’s November 17 earnings release. Kress did not explain the reason for stopping shipments or say whether the halt would be permanent.
The distinction matters: Nvidia has confirmed a shipment halt and excluded China compute revenue from its forecast. It has not announced a permanent withdrawal. For Chinese AI companies, the immediate issue is access to additional hardware through a licensing arrangement that already faced restrictions on both sides of the border.
Nvidia’s Guidance Shows What Has Changed
Kress told investors that “we are no longer shipping licenced Hopper data centre products to China,” according to Reuters. The H200 belongs to that Hopper product family and is used to train and run advanced AI models.
“NVIDIA is not assuming any Data Center compute revenue from China in its outlook.”
That wording is broader than the shipment statement. Kress discussed licensed Hopper products; the written forecast excludes China revenue across the data-center compute category.
The reported sales, official results and forward guidance describe different things:
Measure
Published figure or statement
What it tells readers
Licensed Hopper sales to China in Q3
$2.4 billion, disclosed by Kress and reported by Reuters
China generated revenue from those products during the completed quarter
Nvidia’s worldwide Q3 Data Center revenue
$107.6 billion
The scale of the company’s overall data-center business
China data-center compute revenue assumed in Q4 guidance
The company still forecasts substantial revenue without China compute sales
The $2.4 billion of China Hopper sales equals roughly 2.2% of Nvidia’s reported worldwide Data Center revenue for the completed quarter. That is a calculation from the published figures, not a percentage supplied by Nvidia.
The comparison helps explain why the halt can be significant for Chinese buyers without dominating Nvidia’s overall outlook. It is not a precise measure of China’s share of all Nvidia business: the numerator covers licensed Hopper sales, while the denominator covers the broader Data Center segment.
I’d also avoid treating $2.4 billion as an automatic next-quarter revenue loss. It is a historical sales figure, not a disclosed backlog cancellation or an estimate of what Nvidia otherwise expected to sell. The company’s guidance tells us what it is counting on now, not what a hypothetical unrestricted quarter would have produced.
How the H200 Licensing Arrangement Worked
Washington’s approval of some H200 exports did not create unrestricted access to the chips.
According to Reuters’ account of the arrangement, the United States introduced advanced-chip export controls in 2022 and subsequently tightened them. After a summit with Chinese President Xi Jinping in 2025, President Donald Trump said approved Chinese customers could buy H200 processors.
The United States eased its H200 licensing policy in January 2026, allowing applications to be reviewed individually under security conditions. The arrangement included several separate requirements:
U.S. export licensing: Applications were subject to individual review, rather than general permission to sell freely.
Approved buyers: Washington approved sales to 10 Chinese companies, including Alibaba, Tencent and ByteDance, according to previous Reuters reporting.
A U.S. routing requirement: Eligible chips had to enter the United States before shipment to China.
A tariff: Those eligible chips faced a 25% tariff when imported into the United States.
Chinese import permission: Beijing’s restrictions remained a separate obstacle to delivery into mainland China.
Reuters reports that TSMC manufactures H200 chips for Nvidia in Taiwan. The requirement to route eligible products through the United States therefore added a specific trade condition to the China sales process.
The tariff also needs careful interpretation. A 25% import tariff is a condition attached to the arrangement; it is not evidence that every Chinese customer paid exactly 25% more for a completed system. The reviewed reporting does not establish how that cost was allocated across Nvidia, intermediaries and buyers, or how it affected final transaction prices.
For procurement teams, the practical lesson is straightforward: U.S. approval was one necessary step, not a delivery guarantee. A buyer could be cleared in Washington and still lack permission to bring the hardware into mainland China.
The 600,000 H200 Chips Waiting in Hong Kong
Reuters reported earlier in November that approximately 600,000 H200 chips were being stored in Hong Kong, awaiting shipment to mainland China after Beijing withheld approval for large-scale imports. It also reported, citing people familiar with the matter, that several major Chinese internet companies had each obtained about 10,000 H200 chips in recent months.
Those figures show a constrained supply channel before Nvidia’s newly confirmed halt. They do not describe the same pool of hardware.
The Hong Kong figure concerns inventory reportedly awaiting mainland import approval. The smaller company-level figures concern chips that buyers had obtained. Neither should be added to, or subtracted from, the other without details about the shipments and reporting periods.
It would also be misleading to describe all 600,000 chips as newly stranded by Nvidia’s decision. According to Reuters, they were already waiting because Beijing had withheld approval for large-scale imports. The new halt changes Nvidia’s current shipment position, but the reporting does not establish what will happen to that inventory or when it might move.
This is the most consequential distinction in the available evidence: chips manufactured, chips licensed for export, chips delivered and chips available for computing are different measures.
A processor in storage cannot supply useful AI capacity to a mainland customer waiting to install it. Even delivered chips still need to be integrated into operating systems and data centers before they can support workloads. Inventory is therefore an unreliable shortcut for estimating how much usable compute Chinese labs can access.
There is no supported way to turn the 600,000-chip estimate into a dollar value using the $2.4 billion quarterly sales figure. The sources do not provide matching quantities, configurations, transaction prices or accounting periods.
What the Halt Means for Chinese AI Labs
H200 processors support both training, which develops a model’s learned parameters, and inference, which runs the model to produce answers or other outputs. Restrictions on additional supply can therefore affect both future model development and the capacity available to serve users.
The halt concerns new shipments. The announcement does not say that installed H200 systems will stop operating.
For a Chinese AI company, the likely consequences depend on where its hardware plans stood:
An operator with installed capacity can continue planning around that equipment, while reassessing future expansion.
A buyer waiting on imports faces uncertainty about delivery and deployment schedules.
A lab planning a larger training run may need to adjust its timing, scale or hardware choice if expected processors do not arrive.
A company considering domestic accelerators has a stronger reason to evaluate alternatives, although switching hardware can require software adaptation and performance validation.
These are practical implications of constrained supply, not confirmed outcomes for named model programs. The reviewed sources do not establish how many customers or projects are directly affected.
Reuters reports that Beijing has restricted H200 imports as it seeks to reduce dependence on U.S. technology. AFP’s reporting also describes Chinese authorities steering the domestic technology sector away from procuring H200s.
That creates a tension between industrial policy and near-term compute access. Encouraging domestic chips can strengthen an alternative supply chain, while also requiring buyers to make deployment decisions under tighter hardware constraints.
My view is that delivery certainty deserves as much attention here as a chip’s advertised capability. A procurement plan depends on when usable capacity becomes available. A powerful processor awaiting regulatory clearance cannot meet a scheduled training deadline.
The sources do not establish a resulting change in AI pricing. Limited access could affect infrastructure costs or expansion choices, but there is no basis here to forecast specific increases in cloud GPU rates or model API prices.
Renting Overseas Compute Faces Separate Restrictions
Buying physical chips is only one route to AI computing capacity. Companies can also rent access to processors hosted in overseas data centers, but that route has its own regulatory conditions.
Quartz reported on November 18 that new U.S. rules limit sales of advanced American chips to data centers planning to provide Chinese companies with remote access. Its report also says Kress argued that Nvidia’s unfilled orders for high-end products meant those rules would not affect the company’s top-line revenue.
The available reporting does not provide enough detail to determine which particular remote-access arrangements remain permitted. It does, however, make clear why renting capacity abroad cannot simply be assumed to replace blocked chip deliveries on equivalent terms.
Quartz also reported that Nvidia’s board created a committee covering export restrictions and national security. That is a corporate response to a changing regulatory environment, though it does not establish why Nvidia stopped the licensed Hopper shipments.
What Would Make a Reopening Meaningful?
A resumption announcement would answer only part of the access question. Buyers would still need clarity on several steps before treating future H200 capacity as dependable:
Nvidia’s shipment position: Whether the company resumes sending licensed Hopper products to China.
Export permission: Whether the relevant products and customers qualify under U.S. requirements.
Mainland import approval: Whether Beijing permits the hardware to enter at the required scale.
Delivery and deployment: Whether the processors reach customers and become usable computing capacity.
For Nvidia investors, a future outlook that includes China data-center compute revenue would be a meaningful financial signal. For Chinese AI labs, confirmed delivery schedules would be more directly useful.
Those signals measure different stages of the same supply chain. Nvidia’s current forecast excludes China compute sales; Reuters’ inventory reporting shows why regulatory approval, physical shipments and operational capacity need to be tracked separately.
Frequently Asked Questions
3 questions
1
Is Nvidia’s H200 shipment halt permanent?
Nvidia has not said. Kress confirmed that licensed Hopper shipments to China had stopped but did not provide a restart date or describe the halt as permanent.
2
Are 600,000 H200 chips affected?
Reuters reported approximately 600,000 H200 chips waiting in Hong Kong for mainland import approval before the halt was announced. That is an inventory estimate, not a confirmed count of chips newly affected by Nvidia’s decision.
3
Does Nvidia expect no revenue of any kind from China?
The official guidance is narrower: Nvidia assumes no Data Center compute revenue from China in its fourth-quarter fiscal 2027 outlook. It does not say that every category of China-related revenue will be zero.