US Begins Approving Nvidia H20 Exports to China After Months of Restrictions

Export Licenses Reopen a Crucial Market

The United States has begun issuing licenses allowing Nvidia to export its H20 artificial intelligence chips to China, reopening access to a market after restrictions imposed in April 2025 threatened billions of dollars in sales. As of August 9, the approvals mark another turn in a dispute involving national security, competition and the computer chips needed to develop modern AI.

Nvidia had warned that restrictions on H20 sales could leave an $8 billion hole in revenue for the quarter ending in July. The company said H20 sales generated $4.6 billion in the first quarter of 2025, when China accounted for 12.5% of its total revenue. Those figures help explain why chief executive Jensen Huang personally pressed President Donald Trump to allow exports to resume.

The Commerce Department's Bureau of Industry and Security, which administers export controls, began approving licenses after a second meeting between Huang and Trump. The approvals followed weeks of delays even after the administration had agreed to change its approach.

The number of licenses, the Chinese companies eligible to receive chips and the total value of authorized shipments have not been disclosed. Nvidia and the Commerce Department have declined to provide detailed comments. Approval gives Nvidia a route back into the market, while the amount of business it can recover depends on the shipments permitted.

The decision reaches beyond one company's sales. Washington wants to limit technology that could support China's military and surveillance systems. Nvidia wants to preserve its position with Chinese customers, who are also being courted by domestic chipmakers. Both governments regard leadership in AI and semiconductors as essential to their economic strength and national security.

Why Chinese Companies Want the H20

Nvidia designed the H20 specifically for China after US export controls prevented it from selling its most advanced AI chips there. The restrictions began in 2022 and became tighter in 2023. They were intended to limit Chinese access to advanced computing technology with possible military or surveillance uses.

The H20 is a graphics processing unit, or GPU. Unlike a traditional computer processor, a GPU can perform large numbers of calculations in parallel, meaning it works on many pieces of a problem at once. That makes it useful for the enormous computing workloads involved in machine learning, image recognition and natural language processing, which helps computers work with human language.

Nvidia reduced certain capabilities, including memory capacity and processing power, to bring the H20 within US export rules. Even with those limits, the chip remains powerful enough to train AI models and run them after training. Training is the process through which a model learns patterns; running a model puts that learning to work on a task.

Tencent, Alibaba and ByteDance have shown strong demand for the H20 as they develop AI systems and compete with international rivals. Access to the chip matters because GPUs support applications ranging from advanced chatbots and recommendation engines to medical diagnostics and cars that drive themselves.

Nvidia was founded in 1993 and first became known for graphics cards used in gaming. Over the past decade, it has become the dominant supplier of chips for AI applications. That transformation has placed a company once associated mainly with computer graphics at the center of technology policy and international competition.

How Restrictions Turned Into Approvals

The Trump administration imposed restrictions on H20 sales in April 2025, following earlier controls introduced under the Biden administration. National security concerns drove the April decision, which threatened to cut Nvidia off from a major market even though the H20 had been designed to meet previous restrictions.

Huang responded with a lobbying campaign that included a White House meeting with Trump. He argued directly for permission to sell the H20 to China. People familiar with the matter described the importance of that discussion:

Huang's Oval Office discussion with Trump influenced the decision to proceed with the export approvals.

The administration's policy shift did not immediately produce export licenses. Nvidia faced weeks of delay while the Commerce Department processed the necessary approvals. The Bureau of Industry and Security began approving licenses only after Huang met Trump a second time.

The sequence shows how the rules changed across several years and then several months:

• 2022: The United States introduced controls restricting exports of advanced AI chips to China.
• 2023: Washington tightened those controls, prompting products designed specifically to meet the rules for China.
• April 2025: The Trump administration restricted H20 sales on national security grounds.
• Following the April restrictions: Huang pressed the administration to permit sales, including through direct meetings with Trump.
• By August 9, 2025: The Bureau of Industry and Security had begun issuing H20 export licenses after weeks of delays.

Nvidia's experience shows the practical difference between a policy announcement and permission to ship a product. A decision to allow some exports still leaves a company dependent on the licenses needed for particular sales.

The Security Argument Behind the Dispute

US security officials and some lawmakers have argued that even chips with reduced capabilities could strengthen China's military and surveillance systems. Their concern is that a product capable of supporting commercial AI development could also provide computing power for sensitive government uses.

A group of security experts set out that objection in a letter to Commerce Secretary Howard Lutnick. They warned that approving H20 sales would be a:

strategic misstep, threatening the US's economic and military advantage in AI.

Nvidia has strongly challenged claims used to argue against its access to China. The company maintains that it should be allowed to sell products that meet US rules, while also denying that its chips contain hidden features permitting remote access or control.

Washington faces competing priorities. Restrictions can limit access to computing technology that officials consider sensitive. They can also reduce revenue for US suppliers and create an opening for competitors outside the United States. The H20 decision reflects the difficulty of pursuing security objectives while preserving the commercial position of American chipmakers.

China has responded to US technology restrictions with trade barriers and efforts to build a semiconductor industry that relies less on foreign suppliers. Those responses make the export debate part of a broader struggle over who supplies the technology and who controls access to it.

Billions in Sales and Inventory Costs

Nvidia's warning of an $8 billion revenue shortfall for the July quarter captured the immediate commercial risk. Export restrictions also left the company facing costs from chips and materials it had already arranged to buy.

The company disclosed a $5.5 billion charge related to unsold inventory and purchase commitments. It later reduced that figure by $1 billion after finding ways to reuse some materials. A charge records a cost against a company's earnings; in this case, the problem involved products and commitments affected by the restrictions.

Nvidia shares fell sharply when the restrictions were announced, reflecting investor concern about future growth. The approval of licenses later brought relief, with the stock rebounding and analysts expressing cautious optimism about the company's prospects in China.

Technology industry analyst Marc Einstein described the financial hit as substantial but manageable for Nvidia. He also suggested that trade negotiations could bring further changes:

While this is certainly a lot of money, this is something Nvidia can bear. As we have seen in the last few days and weeks, this may largely be a negotiating tactic. I wouldn't be surprised to see some exemptions or changes made to tariff policy in the near future, given this not only impacts Nvidia but the entire US semiconductor ecosystem.

Nvidia reported $17 billion in sales to China in fiscal year 2024, about 13% of its total revenue. The company also said China represented 12.5% of revenue in the first quarter of 2025. Both figures show that China accounts for a substantial share of its business, alongside the $4.6 billion generated by H20 sales during that first quarter.

China is also one of the world's largest semiconductor markets. Nvidia's leadership has repeatedly warned that losing access would weaken its competitive position as Chinese manufacturers expand their own production.

Chinese Rivals Seek a Larger Share

Huawei, Cambricon and Hygon are increasing their own AI chip production as Chinese companies seek alternatives to US technology. Huawei has been particularly aggressive in approaching local developers with domestically produced chips, seeking to fill gaps left by American export restrictions.

For Chinese customers, the decision involves more than the capabilities of a particular chip. The April restrictions and subsequent reversal show how access to a foreign supplier can change while companies are planning AI projects. Domestic suppliers are competing for that business as well as for a larger role in China's technology industry.

Analysts at Bernstein project that Nvidia's share of China's AI chip market will fall from 66% in 2024 to 54% in 2025. That is a decline of 12 percentage points. Their forecast still leaves Nvidia with more than half the market, while showing Chinese competitors gaining ground.

Other analysts see substantial opportunities for Nvidia. Needham's N. Quinn Bolton expects strong demand for the H20 and future GPUs designed for China, and forecasts significant revenue growth in the Chinese data center market.

The two assessments describe different parts of the same commercial contest. Nvidia can face strong demand for its products while competitors take a greater share of the market. Renewed export permission helps the company compete for orders, while local manufacturers continue trying to win customers.

Why Hidden Access Has Become a Flashpoint

Chinese regulators have demanded assurances that Nvidia chips contain no hidden features that foreign governments could use to gain access or control. US lawmakers, meanwhile, have pushed for mandatory tracking features on advanced chip exports to help monitor their use in sensitive applications.

A backdoor is a concealed way to access a technology product outside its normal security protections. A kill switch would allow it to be disabled. Nvidia rejects such features, arguing that hidden access would create vulnerabilities and damage confidence in American technology.

David Reber Jr., Nvidia's chief security officer, addressed the issue in a company blog post. Referring to Nvidia GPUs, which are widely used to train and run AI models, he said they:

do not and should not have kill switches and backdoors. Permitting covert access or control would make the technology more vulnerable and would fracture trust in US technology.

Nvidia's position echoes resistance by other US technology companies to government demands for surveillance access. In 2016, Apple refused an FBI demand to create custom software to unlock a shooter's iPhone, arguing that building a backdoor would set a dangerous precedent.

Apple has more recently resisted what it described as a secret UK government order seeking backdoor access to iCloud data. Those disputes, like the debate around Nvidia, involve competing demands for government access, user privacy and confidence in commercial technology.

Nvidia therefore faces scrutiny from both sides of the trade dispute. US officials are concerned about how exported chips might be used, while Chinese regulators want assurances about the security of the products entering their market.

A Technology Divide That Extends Beyond Nvidia

Successive US administrations have tightened controls on semiconductors and other advanced technology to slow China's progress in AI, quantum computing and military modernization. China has pursued greater independence in response, helping drive a gradual separation of the two countries' technology sectors.

Rui Ma, founder of the Tech Buzz China podcast, described how continued restrictions could influence purchasing decisions and supply chains:

I expect the US and China AI semiconductor supply chains to be fully decoupled if restrictions stay in place. It doesn't make any sense for any Chinese customer to be dependent on US chips, especially since there is an oversupply of data centers in China.

Decoupling means reducing the connections and dependence between the two industries. For Nvidia, that could mean losing customers to Chinese suppliers. For Chinese businesses, it would mean relying more heavily on domestic technology rather than products whose availability depends on US export decisions.

The H20 approvals pull in the other direction by allowing some trade to resume. They preserve an opportunity for Nvidia to supply Chinese AI developers while Washington continues to control access through licenses.

What the Decision Changes for Businesses

Nvidia has regained a route into a crucial market after months of uncertainty. It is expected to keep developing versions of its chips for China, adjusting their capabilities to comply with changing US regulations. Chinese technology companies and startups are also likely to keep investing in domestic alternatives to reduce reliance on American suppliers.

For businesses building AI systems, the dispute makes access to computing hardware a planning issue as well as a technical choice. A chip's performance matters, but so does whether a company can obtain it under the rules in force when it needs to buy.

US chipmakers face the same pressure from another direction. They must comply with export controls and respond to security concerns while protecting their sales and position against foreign competitors. The H20 reversal provides commercial relief, although the broader competition between Washington and Beijing continues.

The immediate change is specific: the US government has begun granting licenses for H20 exports after restricting sales in April. The wider contest over AI leadership, domestic chip production and trust in technology remains central to decisions on both sides.

Key Points

• By August 9, 2025, the US Commerce Department had begun approving licenses for Nvidia H20 exports to China.
• The H20 was designed for China to comply with US controls introduced in 2022 and tightened in 2023.
• April restrictions threatened an $8 billion revenue shortfall for Nvidia in the July quarter.
• Jensen Huang's direct discussions with President Donald Trump helped bring about the reversal and subsequent approvals.
• Nvidia disclosed a $5.5 billion inventory and purchase commitment charge, later reduced by $1 billion.
• Bernstein projects Nvidia's Chinese AI chip market share will fall from 66% in 2024 to 54% in 2025 as domestic rivals gain ground.
• Nvidia denies its chips contain backdoors or kill switches, while US and Chinese authorities raise different security concerns.
• The approvals restore a route to sales while the wider US-China competition over AI and semiconductors continues.