Trump’s Illegal AI Chip Export Controls, and Who Can Challenge Them

The Trump administration is taxing chip exports without congressional authorization. In August 2025, President Trump announced he would grant Nvidia export licenses to sell its H20 chips to China, on the condition that the company pay 15 percent of its revenues from those sales to the U.S. government. Trump announced he was extending the approach in December 2025: 25 percent for Nvidia’s advanced H200 chips, with similar arrangements for AMD and Intel. On Jan. 14, Trump formally imposed the 25 percent tax via proclamation. Revenue-for-access is now the administration’s export control policy for advanced artificial intelligence (AI) chips.

This is illegal. The Export Control Reform Act (ECRA) expressly prohibits the Bureau of Industry and Security (BIS) from charging any “fee” in connection with issuing export licenses, and a percentage of sales revenue is a “fee” under ECRA. More fundamentally, conditioning market access on payments to the federal treasury is taxation, a power that belongs exclusively to Congress. The executive cannot impose new taxes without congressional authorization, no matter what it calls the taxes. The revenue sharing conditions may also violate the Constitution’s Export Clause, which prohibits any “Tax or Duty” on exports.

The good news is that companies and customers in the AI supply chain have standing to challenge the revenue sharing conditions. Trump’s statements suggest that the licenses would not have been issued without the payments. Anyone injured by the licenses can therefore trace that injury to the illegal revenue sharing condition. That includes competitor chipmakers now facing increased competition in China, cloud providers who may lose Chinese customers to newly available chips, AI companies competing with the Chinese firms that receive the chips, and U.S. customers whose chip orders may be delayed or more expensive due to the Chinese demand unleashed by the licenses. Depending on their tax laws, some states may have standing as well.

The Revenue Sharing Conditions Violate ECRA

Section 4815(c) of ECRA provides that “[n]o fee may be charged in connection with the submission, processing, or consideration of any application for” an export license. The revenue sharing conditions are “fee[s]” charged in connection with applications for export licenses—the deals apply to the sales that the export licenses authorize. That places the condition outside the BIS’s statutory authority to impose.

The BIS may argue that the structure of the 25 percent charge sidesteps ECRA’s fee bar. The structure has two steps. First, it requires that H200s licensed for sale to China must first be routed from Taiwan, where the chips are made, to the United States, where they will undergo “independent, third-party testing” before export to China. Second, it imposes a 25 percent duty on all H200s imported into the United States not destined for domestic use, with the predictable consequence that the duty falls only on chips moving through the China-license pathway. In Trump’s words, “we’re allowing them to do it [export H200s to China] but the United States is getting 25 percent of the chips in terms of the dollar value.”

On paper, that sequencing separates the payment and the license application, allowing the government to argue that no “fee” is charged “in connection with” the BIS’s consideration of a license. But the separation is formal, not functional: The duty is the price of accessing the licensing channel, and the government has simply relocated the collection point from the BIS’s licensing desk to the customs entry process. The scheme therefore accomplishes indirectly what Congress forbade directly—monetizing export licensing.

Judicial review is available over the BIS’s authority to charge fees. ECRA withdraws review under the Administrative Procedure Act (APA) over the BIS’s actions, but challenges to an agency’s statutory authority are still reviewable when the agency “plainly acts in excess of its delegated powers and contrary to a specific prohibition in the statute that is clear and mandatory.” Section 4815(c) qualifies. It’s a straightforward command that “no fee may be charged” in connection with export licenses. This distinguishes it from the two prior cases that have challenged the BIS’s authority, in which the plaintiffs could not identify a specific statutory command the BIS violated.

The BIS may argue that courts should refrain from judicial review because the matter implicates national security and foreign policy issues. But the challenge to the BIS’s authority concerns only the revenue sharing conditions, not the BIS’s technical determinations. Trump’s own remarks when announcing the deal he struck on H20s—“If I’m going to do that, I want [Nvidia CEO Jensen Huang] to pay us as a country something”—confirm the condition had nothing to do with national security or foreign policy concerns.

The Revenue Sharing Conditions Violate Separation of Powers

Not only are the revenue sharing conditions illegal fees, but because they serve to raise revenue, they are taxes created by the executive without clear congressional authorization. This makes them unconstitutional.

Under the Constitution, the power to raise general revenue through “Taxes, Duties, Imposts and Excises” belongs to Congress. In National Cable Television Association v. United States, the Supreme Court drew a sharp line between cost-based fees for specific services and broader revenue-raising measures that look like taxes, interpreting the Independent Offices Appropriation Act (IOAA) as authorizing agencies to charge only the former.

The 15 percent revenue sharing arrangement is the latter. Trump’s own description of the deal—as a bargain in which he would “do this” (grant the licenses) if the company “pay[s] us as a country something”—frames the condition as a way to raise revenue for the federal government, not to recover the BIS’s costs.

The 25 percent charge also lacks congressional authorization, despite its styling as a tariff. Trump invoked Section 232 of the Trade Expansion Act as the basis for his authority to impose the tariff. That statute authorizes Trump to “adjust the imports of [an] article and its derivatives so that such imports will not threaten to impair the national security” if the commerce secretary determines that “the article is being imported in such quantities or under such circumstances as to threaten to impair the national security.”

But Section 232 authorizes the president to “adjust imports” to protect national security, and Trump’s H200 tariff scheme—consistent with his stated intent—is to extract revenue from chips ultimately destined for export. Trump bases the tariff on the United States’s “dependen[ce] on foreign sources to meet domestic demand for semiconductors” but then exempts virtually all chips imported for domestic consumption: all chips for use in U.S. data centers, for “non-data center” applications, for research and development, for public-sector use, and, for good measure, any other use the Trump administration identifies. The only context in which the tariff would apply is when the chips are imported into the United States before being shipped to China—a process that will exist only because Trump himself mandated it. ​​The Supreme Court has emphasized that in assessing its legality, “the practical operation of the tax, actual or potential, rather than its descriptive label or formal character is determinative.” Here, the practical operation of the tax has nothing to do with regulating true imports.

Because ECRA does not preclude constitutional claims against the BIS, judicial review exists here under Webster v. Doe.

The Revenue Sharing Conditions May Violate the Export Clause

The Export Clause provides that “No Tax or Duty shall be laid on Articles exported from any State.” As the U.S. Court of Appeals for the Fifth Circuit explained in Trafigura Trading LLC v. United States, to determine whether a charge on exports is a tax under the Export Clause, a court must consider whether the charge is based on the quantity or value of the exports and how closely connected the charge is with the government services provided. The revenue sharing conditions fail both tests. They are tied directly to the value of the licensed exports, and as explained above, they bear no relationship to the costs borne by the BIS from reviewing and granting the license applications.

The BIS may argue that the 15 percent revenue sharing condition does not violate the Export Clause because H20 chips are fabricated and shipped from abroad in Taiwan, not from the United States. Some cases do suggest the Export Clause applies more narrowly only to goods shipped from U.S. soil to a foreign country, but the Supreme Court has not squarely ruled on this question.

Meanwhile, the targeted H200 chips will be shipped from the United States to China, making the Export Clause argument stronger. Though styled as an import “tariff,” the relevant constitutional question is whether the exaction so directly and closely bears on the process of exporting that it is “in substance a tax upon exportation.” In Thames & Mersey, the Court invalidated a federal stamp tax on marine insurance policies as applied to export shipments, reasoning that voyage insurance was “an integral part of the exportation” and that taxing such policies “falls upon the exporting process.” Here, the 25 percent charge is structured so that it attaches only when chips are routed through the United States as a required step in the China export process. The tax thus targets exportation in substance even if the collection point is nominally at customs entry, violating the Export Clause.

Many Actors Across the AI Supply Chain Have Standing to Challenge the Revenue Sharing Conditions

Though they are the ones who are taxed, the targeted chip companies are not the only potential plaintiffs. To establish Article III standing, a plaintiff must prove that the BIS’s actions imminently threaten it with a concrete and particularized injury that is fairly traceable to the challenged action and redressable by a judicial ruling. Because Trump’s statements suggest that the licenses were granted because of the payments, one can argue any harm caused by the licenses is traceable to the illegal revenue sharing conditions, and a court decision invalidating the conditions would offer redress. Anybody who suffers competitive or economic injury traceable to the licenses may therefore have standing.

This includes many actors in the AI supply chain. Customer standing covers downstream companies facing increased costs from upstream taxes. Competitor standing covers companies adversarially impacted by agency action because the action benefits their competitors. And courts recognize standing for those suffering economic injury from reduced output or higher prices. Based on these doctrines, below is a non-exhaustive list of potential plaintiffs with strong standing arguments.

Competitors to the Licensed Exporters (Nvidia, AMD, Intel)

Chip Companies Selling Into the Chinese Market    

Companies selling chips in China, such as Huawei, Baidu, and Cambricon, have arguably the most straightforward case for standing: Their competitors, Nvidia, AMD, and Intel, can now sell high-demand AI chips when they otherwise could not.

Cloud Providers Servicing Chinese Customers

Chinese companies rent GPU/AI capacity from cloud providers hosted outside China without violating export controls as an alternative to buying chips. ByteDance, for example, leverages Amazon Web Services (AWS) for inference applications and trains its AI models using U.S.-based cloud services from Oracle. If more licensed chips become available in China, cloud providers could lose business. Cloud providers like AWS, Microsoft Azure, Oracle Cloud, Coreweave, and Google Cloud, therefore, could allege competitive injury from the export licenses.

Cloud services may seem like a different vertical than AI chips, but a plaintiff doesn’t need to operate in the same line of business to be competitively injured. A plaintiff suffers injury in fact sufficient for standing when it “lose[s] sales to rivals, or [is] forced to lower its price or to expend more resources to achieve the same sales, all to the detriment of its bottom line.” Thus, the U.S. Court of Appeals for the District of Columbia Circuit held that oil companies had standing to challenge an Environmental Protection Agency rule lifting restrictions benefiting biofuel producers.

OEMs, Retailers, and Resellers Selling Into the Chinese Market

Original equipment manufacturers (OEMs) compete with chipmakers like Nvidia at the system layer by selling complete, supported servers that integrate Nvidia GPUs. This is an alternative to buying vertically integrated systems (for example, Nvidia’s DGX line) from chipmakers. Domestic OEMs such as Inspur, H3C, Lenovo, and Huawei would therefore have standing. It is unclear to what extent global OEMs Dell, HPE, and Supermicro have offered AI servers in China; Dell’s China-facing website, at least, advertises servers configured with H20s and H200s.

In addition to OEMs, there are smaller retailers and resellers of AI servers that could have standing to sue if they offer AI servers in the Chinese market.

Competitors to the Recipients of the Licensed Chips

AI companies such as OpenAI, Anthropic, Google, Meta, xAI, and Mistral compete globally with the Chinese firms, which are the end purchasers of the licensed sales. Granting export licenses gives DeepSeek more compute, which improves its models and increases pressure on its competitors, including global AI companies. That competitive injury is sufficient for standing.

Chip Customers

Non-Chinese Purchasers of H20s and H200s

Nvidia faces strong Chinese demand for its chips. Because chip supply is constrained, opening Chinese demand by granting export licenses reduces global availability and increases prices outside of China. Until Nvidia released its current generation of Blackwell chips, its H100 and H200 chips were its best-selling AI chips worldwide. U.S. customers range from AI companies like OpenAI to cloud providers such as Microsoft, AWS, Google Cloud, Oracle, CoreWeave, and Cirrascale, to universities with AI research centers. A plaintiff can establish standing by showing it faces increased prices or reduced availability because the revenue sharing conditions enabled the licenses, opening up competing demand.

Non-Chinese Purchasers of Blackwell and Other Latest-Generation Chips

Nvidia is reportedly increasing H200 chip production to fulfill Chinese demand. H200 production is currently limited because Nvidia has focused on producing its latest-generation Blackwell and Rubin chips, which are manufactured using the same TSMC nodes as the H200. Increased H200 production may therefore constrain Nvidia’s ability to fulfill Blackwell and Rubin orders. Nvidia’s Blackwell/Rubin customers may therefore have standing as a result of suffering economic injury through delayed or more expensive orders.

Chinese Purchasers of H20s and H200s

The Supreme Court has held in cases such as General Motors Corp. v. Tracy that customers of taxed suppliers have standing to challenge an upstream tax. The Chinese tech companies purchasing the licensed chips—which include Alibaba, ByteDance, Tencent, and DeepSeek’s parent company, High-Flyer—have a straightforward case of standing to challenge the revenue sharing conditions. Of course, it is highly unlikely they have any interest in suing—they would almost certainly just accept paying an extra 15 or 25 percent for the benefit of getting access to the world’s most powerful chips. Notably, these companies are not competitively injured from the BIS granting the licenses; they are economically injured from the costs of the revenue sharing condition. Redress for them would not mean invalidating the licenses, but getting rid of the fee.

Other Potential Plaintiffs

Other Chip Companies Competing for TSMC Foundry Capacity

TSMC’s manufacturing processes are used to manufacture and package the licensed chips, including Nvidia H20 and H200 chips. The company’s capacity is severely constrained. Any increase in these chips’ production displaces other chip companies relying on TSMC’s foundries. Companies relying on the same TSMC processes as the licensed chips—which reportedly include Apple, Qualcomm, MediaTek, Broadcom, Marvell, and Amazon—may be economically and competitively harmed by reduced fabrication availability in favor of Nvidia, AMD, and Intel. 

Trade Associations

Trade associations have standing if at least one member has standing in its own right, the interests align with the association’s mission, and the relief requested does not inherently require individual participation. Two trade associations in the United States that likely have standing are the Semiconductor Industry Association (SIA), which includes semiconductor companies such as Nvidia, AMD, and Intel, and the Frontier Model Forum, representing U.S. AI companies including OpenAI, Anthropic, and Microsoft. SIA, whose mission is to advance the semiconductor industry’s interests, has an interest in ensuring that the BIS grants export licenses based on valid national-security concerns, not to raise revenue. The Frontier Model Forum, which focuses on public safety and national security risks related to AI, has an interest in limiting Chinese access to powerful chips.

States That Tax Nvidia May Have Standing Based on Lost Corporate Tax Revenue

As companies headquartered in California, Nvidia, AMD, and Intel are subject to California income tax. They are also subject to corporate taxes in other states where they have substantial activities. The Supreme Court has held that a state has Article III standing to challenge a federal regulation if it can show “a direct injury in the form of a loss of specific tax revenues.”

Depending on their specific laws, taxing states may have standing under the theory that the revenue sharing conditions reduce collectible tax revenues. Most states, including California, impose corporate taxes using the sales factor-apportionment method, under which the corporation’s state tax liability looks like this:

As Nvidia has acknowledged, it may not be able to “pass along all or any” of the 15 percent fee on H20s to its customers, and the same presumably applies for the 25 percent fee on H200s. The revenue sharing conditions thus impose a tax that reduces the “company net income” factor, mechanically reducing taxes owed to the state. Like Nvidia’s Chinese customers, note that the states are not injured from the BIS granting the licenses; they are economically injured from the costs of the revenue sharing conditions.

*          *          *

Granting export licenses in exchange for revenue sharing agreements violates ECRA and the Constitution. Well-established standing doctrine shows that a wide range of potential plaintiffs— competitors, U.S.-based customers, OEMs, cloud providers, AI companies, and state governments—can challenge the administration’s actions, and they should. Whatever the geopolitical merits of exporting the licensed chips to China are, export license decisions must be made based on those merits, not to raise revenue. The path to the courthouse is clear. Now someone needs to take it.