Why does Nvidia now sell almost no data-center chips in China?
Nvidia did not lose China to one rule. Washington set the first licence requirement in August 2022 against A100-level performance, and each time Nvidia engineered a part that sat under the line, the line moved: the October 2023 rules removed the interconnect test the A800 and H800 were built around, and the H20 built to fit the new thresholds itself needed a licence from April 2025. Commercially it was already struggling, selling at a discount to Huawei's competing part in 2024. The H20 restriction then produced a US$4.5 billion charge and US$2.5 billion in unshipped revenue. Although Nvidia said in July 2025 that Washington had assured it licences would be granted and it hoped deliveries would resume, it recorded no H20 sales to China-based customers that quarter and assumed none in its next-quarter outlook. Beijing then closed its own side: a summons over alleged backdoors, a preliminary finding that Nvidia had violated the Anti-Monopoly Law, a rule that new data-centre projects taking state funds buy only domestic chips, and customs agents told the H200 was not permitted - days after Washington had approved its export. By mid-2026 China Hopper shipments were under 1 percent of data-center revenue, and in September DeepSeek was planning a 160,000-chip Huawei cluster. One route around the rules is documented through February 2026: servers carrying Nvidia Blackwell chips reached Chinese customers through a US subsidiary that is not itself on the Entity List.
Trace the chain
Build to the line, watch it move
- 1A100 and H100 licensedRegulatory
US imposes a license requirement on Nvidia A100 and H100 exports to China
Nvidia disclosed that the U.S. government had imposed, effective immediately, a license requirement for exports of its A100 and forthcoming H100 chips to China, including Hong Kong, and Russia, along with systems containing them and any future chip at A100-level performance. Nvidia said it did not sell to Russia and that the rule could affect H100 development and A100 support.
Nvidia Form 8-K ·
On August 26, 2022, the U.S. government, or USG, informed NVIDIA Corporation, or the Company, that the USG has imposed a new license requirement, effective immediately, for any future export to China (including Hong…
Why it matters here. A licence requirement pegged to A100-level performance invited a part engineered to sit just under it, which Nvidia unveiled two months later.
US Export Controls on Nvidia AI Chips to China: A100 to H20 to H200
- 2The A800 workaroundUpdate
Nvidia unveils the A800, a slowed A100 built to sit under the October rules
Two months after the U.S. cut off the A100 and H100, Nvidia offered the A800 as an alternative for Chinese customers, saying it met the government's test for reduced export control and could not be programmed to exceed it. The part had gone into production in the third quarter. Nvidia had previously said the ban could affect as much as US$400 million of third-quarter sales to China.
TechCrunch ·
The A800 meets the U.S. government's clear test for reduced export control and cannot be programmed to exceed it.
Why it matters here. Nvidia unveiled an A800 that met the government's reduced-control test two months after the licence requirement, and Chinese internet firms later ordered about 100,000 of them.
US Export Controls on Nvidia AI Chips to China: A100 to H20 to H200
- 3Orders before the rulesReport
China's internet giants order US$5 billion of Nvidia chips before feared new controls
The Financial Times reported that Baidu, ByteDance, Tencent and Alibaba had ordered about 100,000 A800 processors worth US$1 billion for delivery that year, and a further US$4 billion of GPUs for 2024, rushing to secure supply amid fears of further U.S. export controls. Demand for the compliant part was intact two months before the October 2023 rules removed the test it was designed around.
Financial Times, via Malay Mail ·
Baidu, TikTok-owner ByteDance, Tencent and Alibaba have made orders worth US$1 billion to acquire about 100,000 A800 processors from the US chipmaker to be delivered this year, the FT reported, citing multiple people…
Why it matters here. Orders for about 100,000 A800s and four billion dollars more for the following year show the workaround was working commercially, which is what the next rule took away.
US Export Controls on Nvidia AI Chips to China: A100 to H20 to H200
- 4The test removedRegulatory
BIS tightens the chip rules, dropping the interconnect-bandwidth test that A800 and H800 were built around
BIS released updated rules on advanced computing chips and chipmaking equipment. Among the changes it removed interconnect bandwidth as a parameter for identifying restricted chips, which brought Nvidia's China-specific A800 and H800 within the controls, and extended license requirements to more countries. Nvidia's same-day filing listed A100, A800, H100, H800, L40, L40S and RTX 4090 as covered.
Bureau of Industry and Security ·
Based on public comments, recent technological developments, and analysis of the prior rule's national security impact, the AC/S IFR removes 'interconnect bandwidth' as a parameter for identifying restricted chips.
Why it matters here. Dropping the interconnect-bandwidth test closed the A800 and H800, so Nvidia had to design again to whatever the new thresholds allowed.
US Export Controls on Nvidia AI Chips to China: A100 to H20 to H200
- 5H20 built to fitReport
Reports say Nvidia is preparing H20, L20 and L2 chips that fit under the new China limits
The Register, citing Chinese outlet Chinastarmarket and supply chain sources, reported that Nvidia was working on three GPUs for China, the H20, L20 and L2, engineered to stay below the performance thresholds in the October 2023 rules, with an announcement possible as early as November 16.
The Register ·
According to Chinastarmarket, which cites supply chain sources, the GPUs include the H20, L20, and L2, and could be announced as early as November 16
Why it matters here. Three GPUs engineered to stay below the October 2023 thresholds were being worked on, with an announcement reported as possible within weeks, but the H20's launch was pushed into the following year.
US Export Controls on Nvidia AI Chips to China: A100 to H20 to H200
- 6H20 delayedDelay
Nvidia delays the H20 launch in China to the first quarter of 2024
Reuters reported, citing two people familiar with the matter, that Nvidia had told Chinese customers it was delaying the launch of the export-compliant H20 until the first quarter of 2024, because server makers were having trouble integrating the chip into their systems. Nvidia was also preparing the L20 and L2 for the same market. Its shares closed down about 1.9 percent.
CNBC ·
Reuters, citing two sources familiar with the matter, reported that Nvidia told Chinese customers that it is delaying the launch of an AI chip that is designed to comply with U.S. export rules until the first quarter…
Why it matters here. Server makers could not integrate the chip in time, so the next word on the compliant parts was Huang saying they were out with customers as samples.
US Export Controls on Nvidia AI Chips to China: A100 to H20 to H200
- 7Sampling without a licenceUpdate
Huang says Nvidia is sampling two China chips that comply without a licence
After Nvidia's quarterly results, Jensen Huang told Reuters the company was sampling two new China-market AI chips with customers and that both complied with the rules without needing a licence. He did not name the chips or the customers. The move was Nvidia's attempt to defend a market position that the October 2023 controls had put at risk.
Reuters, via Yahoo ·
We're sampling it with customers now. Both of them comply with the regulation without a license. We're looking forward to customer feedback on it," Huang told Reuters on Wednesday in an interview after releasing…
Why it matters here. Huang's claim was that both parts complied without needing a licence; what that did not settle was the price, and the H20 went on to sell at more than a 10 percent discount to Huawei's part amid oversupply.
US Export Controls on Nvidia AI Chips to China: A100 to H20 to H200
- 8Below Huawei's priceReport
H20 sells below Huawei's chip as supply outruns Chinese demand
Reuters reported that Nvidia had cut H20 prices in China, where oversupply left the chip selling at more than a 10 percent discount to Huawei's competing part. The first export-compliant chip Nvidia built for China after the October 2023 rules was therefore competing on price against the domestic alternative it was meant to hold off.
TrendForce, citing Reuters ·
an oversupply caused the chip to be priced lower than its rival, Huawei, in some cases even at an over 10% discount
Why it matters here. Discounting the compliant chip below Huawei's part meant it was already losing on commercial terms before Washington required a licence for it too.
US Export Controls on Nvidia AI Chips to China: A100 to H20 to H200
- 9H20 needs a licenceRegulatory
US requires a license for H20 exports to China; Nvidia flags up to US$5.5 billion in charges
Nvidia disclosed that on April 9 the government required a license for exports of H20 and any chip with its memory or interconnect bandwidth to China, Hong Kong, Macau and D:5 countries, and on April 14 said the requirement would apply indefinitely. Nvidia expected up to about US$5.5 billion of first-quarter charges for H20 inventory, purchase commitments and reserves.
Nvidia Form 8-K ·
On April 9, 2025, the U.S. government, or USG, informed NVIDIA Corporation, or the Company, that the USG requires a license for export to China (including Hong Kong and Macau) and D:5 countries, or to companies…
Why it matters here. A licence requirement applied indefinitely to the H20 turned an export question into a balance-sheet one.
US Export Controls on Nvidia AI Chips to China: A100 to H20 to H200
- 10H20 losses bookedFinancial
Nvidia takes a US$4.5 billion H20 charge, loses US$2.5 billion in shipments, expects US$8 billion hit next quarter
In first-quarter fiscal 2026 results Nvidia recorded a US$4.5 billion charge for H20 excess inventory and purchase obligations, said it could not ship a further US$2.5 billion of H20 revenue, and guided second-quarter revenue of US$45 billion reflecting about US$8 billion of lost H20 revenue.
Nvidia Newsroom ·
NVIDIA incurred a $4.5 billion charge in the first quarter of fiscal 2026 associated with H20 excess inventory and purchase obligations as the demand for H20 diminished.
Why it matters here. US$4.5 billion of charges and US$2.5 billion of unshipped revenue made the cost explicit, and Washington then said the licences would be granted after all.
US Export Controls on Nvidia AI Chips to China: A100 to H20 to H200
- 11US signals H20 resumptionAnnouncement
Nvidia says the US has assured it H20 licenses will be granted and sales to China will resume
After Jensen Huang met officials in Washington and travelled to Beijing, Nvidia said the U.S. government had assured it that H20 export licenses would be granted and that it hoped to start deliveries soon. It also announced an RTX Pro GPU aimed at the China market.
Nvidia Blog ·
The U.S. government has assured NVIDIA that licenses will be granted, and NVIDIA hopes to start deliveries soon.
Why it matters here. Although Washington assured Nvidia that H20 licences would be granted and it hoped to ship soon, Nvidia's next quarterly results recorded no H20 sales to China-based customers and assumed none in the following quarter.
US Export Controls on Nvidia AI Chips to China: A100 to H20 to H200
- 12H20 sales stayed zeroFinancial
No H20 sales to China in the quarter; Nvidia assumes none in its outlook
Second-quarter fiscal 2026 results showed no H20 sales to China-based customers, a US$180 million release of previously reserved H20 inventory tied to about US$650 million of sales to a customer outside China, and a third-quarter outlook that assumed no H20 shipments to China.
Nvidia Newsroom ·
There were no H20 sales to China-based customers in the second quarter.
Why it matters here. Nvidia recorded no H20 sales to China-based customers and assumed none in the following quarter, so the promised US licence resumption had not restarted trade before China raised security objections.
US Export Controls on Nvidia AI Chips to China: A100 to H20 to H200
Beijing closes its sideRuns from 2025, alongside the strands above, not after them
- 13Backdoor summonsRegulatory
China's cyberspace regulator summons Nvidia over alleged H20 backdoor risks
The Cyberspace Administration of China said it had interviewed Nvidia and asked it to explain security risks from backdoor vulnerabilities in H20 chips sold to China and to submit supporting material. Nvidia said its chips have no backdoors that would allow remote access or control.
The Register ·
In order to maintain the network security and data security of Chinese users, the State Internet Information Office interviewed Nvidia on July 31, 2025, and asked Nvidia to explain the security risks of the backdoor…
Why it matters here. China’s regulator required Nvidia to explain alleged H20 backdoor risks and submit material, and Nvidia next publicly denied that its chips contain backdoors, kill switches or spyware.
US Export Controls on Nvidia AI Chips to China: A100 to H20 to H200
- 14Nvidia denies backdoorsResponse
NVIDIA denies its chips contain backdoors, kill switches or spyware
NVIDIA said its chips contain no backdoors, kill switches, or spyware. The statement addressed allegations concerning NVIDIA chips.
NVIDIA Blog ·
There are no back doors in NVIDIA chips. No kill switches. No spyware.
Why it matters here. The denial addressed the security allegation but not the competition one, and the market regulator's finding came next.
US Export Controls on Nvidia AI Chips to China: A100 to H20 to H200
- 15Antitrust findingLegal
China's SAMR says a preliminary probe found Nvidia violated the Anti-Monopoly Law
The State Administration for Market Regulation said its preliminary investigation found Nvidia had violated China's Anti-Monopoly Law and the conditions attached to its 2020 approval of the Mellanox acquisition, and escalated the case to a formal investigation.
The Register ·
Recently, after preliminary investigation, Nvidia violated the Anti-Monopoly Law of the People's Republic of China.
Why it matters here. Escalating to a formal investigation kept Nvidia's China position open as a legal question while procurement rules began closing it as a practical one.
US Export Controls on Nvidia AI Chips to China: A100 to H20 to H200
- 16State money buys domesticRegulatory
Beijing requires state-funded data centres to use domestic AI chips
Reuters reported that Beijing had issued guidance requiring any new data centre project that had taken state funds to use only domestically developed AI chips, and that projects less than 30 percent complete were told to remove foreign chips already installed or cancel planned purchases. More advanced projects would be assessed case by case. The rule reached Nvidia's H20, the part built specifically to satisfy the U.S. rules, and applied to a sector that had taken more than US$100 billion of state money.
Reuters, via InvestingLive ·
It is reported that Beijing has issued a guidance that requires new data centre projects that have received any state funds to only use domestically developed AI chips.
Why it matters here. Requiring new state-funded projects to use only domestically developed chips reached the H20 itself, and the next directive stopped the H200 at the border.
US Export Controls on Nvidia AI Chips to China: A100 to H20 to H200
- 17Customs turns H200 backRegulatory
Chinese customs agents are told H200 chips may not enter the country
Reuters reported that Chinese customs authorities had told agents the H200 was not permitted to enter China, and that officials had summoned domestic technology companies on January 13 and instructed them not to buy the chip unless necessary. One person said the wording was severe enough to amount to a ban for now. The directive came days after the Trump administration formally approved H200 exports to China with conditions, and the authorities gave no reasons and no indication whether it was formal or temporary.
Reuters, via Rappler ·
Chinese customs authorities told customs agents this week that Nvidia's H200 artificial intelligence chips are not permitted to enter China, according to three people briefed on the matter. Chinese government officials…
Why it matters here. Chinese customs' instruction that H200 was not permitted to enter, despite Washington's conditional approval days earlier, helps explain why Nvidia's China Hopper shipments were later below 1% of data-center revenue.
US Export Controls on Nvidia AI Chips to China: A100 to H20 to H200
What is left
- 18Under one percentFinancial
Nvidia reports China Hopper shipments were under 1% of data center revenue
Nvidia reported that Data Center Hopper shipments to China were less than 1% of Data Center revenue in the second quarter of fiscal 2027. The quarter ended July 26, 2026.
U.S. Securities and Exchange Commission, Nvidia Form 10-Q ·
Shipments of Data Center Hopper products to China during the second quarter of fiscal year 2027 were less than 1% of Data Center revenue.
Why it matters here. China Hopper shipments below 1 percent of data-center revenue is what the two-sided closure adds up to, and the next entry shows one buyer planning its cluster on domestic accelerators instead.
US Export Controls on Nvidia AI Chips to China: A100 to H20 to H200
- 19DeepSeek picks HuaweiReport
DeepSeek plans a 160,000-chip Huawei cluster instead of Nvidia parts
Bloomberg reported that DeepSeek intended to deploy at least 160,000 of Huawei's next-generation Ascend 950DT accelerators at a data centre it was building in Inner Mongolia, using them to operate its models, which would be one of the largest known clusters of Huawei AI chips. People familiar with the matter said the timeline depended on Huawei's production capacity and that DeepSeek did not currently plan to use the chips for training. DeepSeek did not respond to requests for comment and Huawei had no immediate comment.
Bloomberg, via Yahoo Finance ·
DeepSeek plans to deploy at least 160,000 of Huawei Technologies Co's top accelerators at a massive data centre it's building in Inner Mongolia, which could create one of the largest known clusters of Huawei artificial…
Why it matters here. A plan to put DeepSeek's model serving on at least 160,000 Huawei accelerators is prospective demand pointed away from Nvidia, which leaves what still crosses the border arriving another way.
US Export Controls on Nvidia AI Chips to China: A100 to H20 to H200
What pushes back
- 20The diversion route
NYT: Inspur's US subsidiary shipped $5.6 billion of technology to Southeast Asia, including Blackwell servers
The New York Times reported that Aivres, a California server maker, had exported at least US$5.6 billion of advanced technology to Southeast Asia between April 2024 and February 2026, including more than US$3 billion of computers carrying Nvidia's Blackwell chips, and that those servers ultimately served Chinese customers including ByteDance and Alibaba. Aivres is Inspur Group's US subsidiary and is not itself on the Entity List, though Inspur was added in 2023. Nvidia spokesman John Rizzo said the company does not support product diversion and sells only to well-known partners that work with Nvidia to comply with export control rules. Federal officials had begun looking into the subsidiary's business.
Asia Times ·
The New York Times reported on Sunday that Aivres, a California-based server maker, exported at least $5.6 billion in advanced technology to Southeast Asia between April 2024 and February 2026, including more than $3…
What it pushes against. This pushes against a clean zero: Blackwell-equipped servers ultimately served Chinese customers through Aivres, a US subsidiary not itself on the Entity List, even though Nvidia says it does not support diversion.
US Export Controls on Nvidia AI Chips to China: A100 to H20 to H200
Changes to this chain
- Proposal #147 accepted (verify:v23): why of step 11 -> Although Washington assured Nvidia that H20 licences would be granted and it hoped to ship soon, Nvidia's next quarterly results recorded no H20 sales to China-based customers and assumed none in the following quarter.
- Proposal #127 accepted (publish:nvidia-acquisitions): insert after 11: [cause] H20 sales stayed zero -> nvidia-china-export-controls#55: Nvidia recorded no H20 sales to China-based customers and assumed none in the following quarter, so the promised US licence resumption had not restarted trade before China raised security objections.; answer: Nvidia did not lose China to one rule. Washington set the first licence requirement in August 2022 against A100-level performance, and each time Nvidia engineered a part that sat under the line, the line moved: the October 2023 rules removed the interconnect test the A800 and H800 were built around, and the H20 built to fit the new thresholds itself needed a licence from April 2025. Commercially it was already struggling, selling at a discount to Huawei's competing part in 2024. The H20 restriction then produced a US$4.5 billion charge and US$2.5 billion in unshipped revenue. Although Nvidia said in July 2025 that Washington had assured it licences would be granted and it hoped deliveries would resume, it recorded no H20 sales to China-based customers that quarter and assumed none in its next-quarter outlook. Beijing then closed its own side: a summons over alleged backdoors, a preliminary finding that Nvidia had violated the Anti-Monopoly Law, a rule that new data-centre projects taking state funds buy only domestic chips, and customs agents told the H200 was not permitted - days after Washington had approved its export. By mid-2026 China Hopper shipments were under 1 percent of data-center revenue, and in September DeepSeek was planning a 160,000-chip Huawei cluster. One route around the rules is documented through February 2026: servers carrying Nvidia Blackwell chips reached Chinese customers through a US subsidiary that is not itself on the Entity List.
- Proposal #111 accepted (verify:v21): why of step 12 -> China’s regulator required Nvidia to explain alleged H20 backdoor risks and submit material, and Nvidia next publicly denied that its chips contain backdoors, kill switches or spyware.
- Step 18 why edited by the editor: Do not claim none of it is installed
- Proposal #110 accepted (verify:v19): why of step 2 -> Nvidia unveiled an A800 that met the government's reduced-control test two months after the licence requirement, and Chinese internet firms later ordered about 100,000 of them.
- Proposal #108 rejected (verify:v13)
- Step 11 why edited by the editor: An assurance and a hope are not a settled outcome
- Step 9 why edited by the editor: Drop a premise the adjacent entries do not carry
- Proposal #109 rejected (verify:v17)
- Step 18 why edited by the editor: The entry reports a plan, not completed sourcing
- Step 11 why edited by the editor: The assurance closes the Washington strand
- Step 10 why edited by the editor: Bridge to the US assurance
- Step 9 why edited by the editor: Bridge to the charge that follows
- Proposal #107 accepted (publish:amd-china-export-controls): insert after 9: [cause] H20 losses booked -> nvidia-china-export-controls#50: The US$4.5 billion charge and US$2.5 billion in unshipped H20 revenue made the sales interruption concrete before Nvidia said it hoped deliveries would restart.; insert after 10: [cause] US signals H20 resumption -> nvidia-china-export-controls#51: Nvidia said it hoped to restart H20 deliveries after the US assurance, so China's regulator became the next recorded obstacle to H20 sales.; answer: Nvidia did not lose China to one rule. Washington set the first licence requirement in August 2022 against A100-level performance, and each time Nvidia engineered a part that sat under the line, the line moved: the October 2023 rules removed the interconnect test the A800 and H800 were built around, and the H20 built to fit the new thresholds itself needed a licence from April 2025. Commercially it was already struggling, selling at a discount to Huawei's competing part in 2024. The H20 restriction then produced a US$4.5 billion charge and US$2.5 billion in unshipped revenue; although Nvidia said in July 2025 that Washington had assured it licences would be granted and it hoped deliveries would resume, Beijing then closed its own side: a summons over alleged backdoors, a preliminary finding that Nvidia had violated the Anti-Monopoly Law, a rule that new data-centre projects taking state funds buy only domestic chips, and customs agents told the H200 was not permitted - days after Washington had approved its export. By mid-2026 China Hopper shipments were under 1 percent of data-center revenue, and in September DeepSeek was planning a 160,000-chip Huawei cluster. One route around the rules is documented through February 2026: servers carrying Nvidia Blackwell chips reached Chinese customers through a US subsidiary that is not itself on the Entity List.
- Added the A800 and the pre-rules order rush as steps 2 and 3: the chain asserted that Nvidia kept building parts under the line without ever showing one, or showing that it sold.
- Proposal #91 rejected (verify:v10)
- Step 5 why edited by the editor: Name the commercial evidence the next entry records
- Step 3 why edited by the editor: An announcement date is not a delivery schedule
- Proposal #90 rejected (verify:v9)
- Step 14 why edited by the editor: The entry reports a plan, not a deployment
- Step 13 why edited by the editor: Do not generalise one reported plan into buyers at large
- Step 11 why edited by the editor: Drop the successor characterisation neither entry makes
- Proposal #89 rejected (verify:v6)
- Step 14 why edited by the editor: Stay inside what the entry states
- Step 7 why edited by the editor: Name the next recorded consequence
- Proposal #88 rejected (verify:v4)
- Step 14 why edited by the editor: Drop the counterfactual about rules changing
- Step 3 why edited by the editor: The source says an announcement was possible, not that the chips were ready
- Proposal #87 rejected (verify:v2)
- Proposal #86 accepted (verify:v1): why of step 12 -> Chinese customs' instruction that H200 was not permitted to enter, despite Washington's conditional approval days earlier, helps explain why Nvidia's China Hopper shipments were later below 1% of data-center revenue.; why of step 15 -> This pushes against a clean zero: Blackwell-equipped servers ultimately served Chinese customers through Aivres, a US subsidiary not itself on the Entity List, even though Nvidia says it does not support diversion.
Timelines in this chain
- US Export Controls on Nvidia AI Chips to China: A100 to H20 to H200
- US Export Controls on AMD's AI Chips in China: From the MI250 License to MI308 and MI325X
- Nvidia's Dealmaking: From Mellanox and the Failed Arm Bid to Groq and Hugging Face
A chain is written by the editor from entries already published on the timelines above, each of which was read against its source before publication. The chain changes when those entries do. Data and terms.