Let's talk about the United States' three "guiding opinions" including the global ban on Huawei's Ascend chips

Source: Dongbuyaqiao Research Institute

Today, BIS announced three "guidances". I have been waiting for a long time but have not seen the specific text. Let me talk about my views based on the existing information.

First of all, BIS is very sloppy. Such an important notice actually has typos: "inference" is written as "interference" (which seems to set the tone that the relevant rules are very unscientific?)

The reason for issuing this notice first is mainly because the "AI Proliferation Rules" officially came into effect on May 15, and American companies are asking: Do we still have to abide by the regulations set by the Biden administration at the time? Can controlled chips still be sold to Tier 2 countries?

BIS clearly stated in this notice: Biden’s AI proliferation rules will be abolished, and BIS Deputy Secretary Kessler has instructed that this rule should no longer be enforced.

Of course, the most important are the three "guidance opinions":

In addition, the U.S. Department of Commerce’s Bureau of Industry and Security announced today actions to strengthen overseas export controls on artificial intelligence chips, including:

• Issued guidance that stipulates that the use of Huawei’s Ascend chips anywhere in the world is a violation of U.S. export control regulations.

• Issued guidance to warn the public of the potential consequences of allowing American artificial intelligence chips to be used for training and reasoning Chinese artificial intelligence models.

• Issued guidance to US companies on how to protect their supply chains from transshipment strategies.

The first "guidance" is indeed quite overbearing, which is equivalent to forcing everyone on a global scale to choose between Huawei H cards and Nvidia N cards. BIS (Bureau of Industry and Security of the US Department of Commerce) directly said: If you use Huawei's Ascend chips, no matter where you are, it is a violation of US export controls.

For third countries outside of China, this statement is more like a pre-warning and a warning - after all, H cards are currently in short supply in China, and there is no surplus to export. So this is more like setting rules and making things clear first.

But for China, this is simply outrageous. You said that Chinese companies using their own domestic chips is also considered "violating US law"? Then can they still be used? If you really say so, does it mean that the United States intends to allow Nvidia to sell N cards in China again? Otherwise, what do you want everyone to use?

Legally speaking, this kind of "guidance" issued by BIS is actually divided into two categories in the US administrative law system:

One ​​is called "interpretive rules", which is to explain how to understand existing laws or regulations; the other is "general statements of policy", which means that law enforcement agencies tell everyone "how I plan to use my law enforcement discretion."

What BIS released this time is actually its understanding of the Export Administration Regulations (EAR) and compliance recommendations, with the purpose of telling companies: "You'd better do what I say, otherwise I may cause trouble for you."

This kind of guidance itself is not a law, nor does it need to go through the "public notice and comment" procedure required by the Federal Procedure Act (APA). It is more like a signal or reminder from law enforcement agencies. If it really wants to be legally binding, it needs to be attached to specific EAR rules, or one day simply write these contents into formal EAR provisions and go through all the rulemaking procedures.

For example, BIS issued a guidance on "deemed reexport" as early as 2013, stating that if you provide U.S. technology to third-country people abroad, it may be regarded as "reexport." However, this opinion was not included in the EAR until 2016. Before that, it was just a "reference for you, my suggestion".

Although the full text of the "Guidance" issued by BIS this time has not been released, judging from the wording ("Using Huawei Ascend chips anywhere in the world violates US export controls"), it may be related to the "Foreign Direct Product Rule" (FDPR) targeting Huawei and the General Prohibition Ten in Section 736.2(b)(10) of the EAR.

EAR Section 736.2(b)(10), if explained in plain language, is: If you know that an export or use behavior has violated U.S. export controls, or is about to violate them,you cannot participate in it.Whether it is selling goods, transshipping, helping to pay, "temporary storage", or providing assistance in other ways, it is considered an act of assistance, and you will also violate export controls.

Since the U.S. government has long accused Huawei of using a "white glove" company to have TSMC manufacture Ascend chips (using U.S. tools), and believes that this violates the FDPR. In this way, anyone who knew that Huawei did not issue this "Guidance" before may still be able to argue that I was not "knowingly aware", but now BIS has clearly stated that using Huawei's Ascend chips violates US export controls, and if you continue to use them, it is obviously "knowingly". Therefore, the "Guidance" actually provides BIS with more sufficient basis for further punishment of some companies that use Huawei's Ascend chips in accordance with violations of US export controls. Looking back at the recent statements of several key figures in US AI policy, it seems that the current Trump administration's strategic containment of China's AI is gradually forming three policy pillars: first, relax regulations at home, increase investment in AI infrastructure, and support local companies to "take the lead" in the AI ​​competition; second, promote the full deployment of the "US technology stack" around the world to ensure that countries use US chips, models and tools instead of China; third, implement more precise and high-pressure export controls against China to block its access to US "neck-blocking" technology.

This first guideline, threatening countries to give up using Huawei chips, clearly reflects the second pillar mentioned above. As David Sacks said in his commentary on AI proliferation rules:

"The United States still has an opportunity to widely deploy the "American technology stack" around the world. While we are still ahead, we must seize this opportunity quickly."

Michael Kratsios, director of the White House Office of Science and Technology Policy, also emphasized:

"This administration must ensure that America's partners and allies use America's excellent technology."

This strategy not only relies on "promoting American technology", but also includes "blocking Chinese technology." Looking around the world, it is China that can truly pose a systemic challenge to AI infrastructure. Therefore, the United States is stepping up efforts to promote its own chips and computing cloud services in third countries, and on the other hand, it is trying to keep China's AI solutions out of the market through export controls, investment restrictions and other means. This is also one of the few issues on which the US government and companies have reached a consensus. Taking Nvidia as an example, Huang Renxun once publicly stated: "Huawei is our biggest competitor." Since the enemy is clear, it is natural to concentrate all policy ammunition on this opponent. You can see that in this "guidance", only the four words "Huawei Ascend" are named, and even the Cambrian period is not mentioned, which shows the clear direction and policy priority of this document. In fact, some American companies have long made similar suggestions to the government: the general idea is that Huawei 5G is late to the United States, and in the end it will have to go through great pains to persuade allies one by one to remove Huawei equipment, so as not to lose again in AI.

For example, AMD CEO Lisa Su recently testified before the Senate Committee on Commerce, Science and Transportation:

The United States should also promote the widespread use of American artificial intelligence technology around the world while ensuring national security. Specifically, the United States needs to develop a clear policy framework through government-business collaboration to avoid international partners turning to alternative technologies due to past restrictions on the spread of artificial intelligence technology.

Microsoft President Brad Smith, who also attended the hearing, also emphasized:

The competition between the United States and China is not only about technological innovation, but also about who can promote technology to other countries faster. The victory or defeat of this part of the competition depends on the first-mover advantage. The United States needs to develop a smart export control strategy to ensure that other countries obtain a reliable and continuous supply of artificial intelligence components and services while ensuring national security. Additionally, we must continue to maintain the international community’s trust in American products, companies, and the country itself.

In its submission to the White House AI Action Plan, Eclipse noted:

“It is critical that leading U.S. companies align with U.S. government policy and seize the initiative to provide viable AI infrastructure solutions to allies. In the 5G competition, it is difficult for the United States to provide a more cost-effective alternative to Huawei products, and U.S. companies are world leaders in the field of AI infrastructure and can provide allies with a repeatable, scalable capability system to build a global network of U.S.-developed AI infrastructure. Unlike the United States in the 5G competition, where it had to struggle to convince allies to remove Huawei equipment, the United States currently has the opportunity to take the lead in providing viable AI solutions to allies. Moreover, the stakes are much higher - unlike 5G telecommunications equipment or dependence on natural resources such as oil, once a country adopts AI infrastructure built on hostile architectures, it is almost impossible to liberate the ideological and social systems that have come to rely on these systems.”

This strategic arrangement actually continues the United States' previous way of controlling semiconductor manufacturing equipment - reaching a "consensus" with its allies (such as the US-Japan-Netherlands trilateral agreement) to build a global technology blockade chain against China. But this is likely to reproduce the inherent defects of the FDPR: it is very difficult for a sovereign country to willingly implement US laws for a long time, especially those targeting China. From the actual effect, in order to achieve the goal of getting these countries to stop using Huawei chips, the United States is likely to still need to negotiate with each country one by one. The complexity, diplomatic costs, and the number of countries involved in this process will far exceed the three-party secret negotiations of the year.

This can't help but remind people of the United States' "Mutual Defense Assistance Control Act of 1951" in the early days of the Cold War, also known as the "Battle Act" (named after its proposer, Republican Congressman Laurence Battle of the United States Congress). The core clause of the bill stipulates that if any country receiving US aid does not impose a strategic material embargo on the Soviet Union and its satellite countries, the United States will suspend its aid to it. The so-called "strategic materials" at that time included not only arms and oil, but also key technological products that could enhance the Soviet Union's military capabilities. This legislation is essentially an attempt by the United States to use aid as a lever to force its allies to completely decouple from the communist camp economically, and to serve its geopolitical and national security goals by controlling trade. This is exactly the same as the United States today using Nvidia GPUs and AI technology stacks as new "strategic materials" to force other countries to "take sides."

But the actual implementation of the "Bartlet Act" has not been smooth. European allies, especially Britain and France, were not willing to completely obey Washington on trade issues with the Soviet Union. After all, in the early days of the Cold War, they still relied on economic exchanges with Eastern Europe to maintain domestic recovery. Faced with the resistance of its allies, the United States once cited the bill to put pressure on it, but ultimately had to make a waiver in its implementation and issued a large number of exemption licenses, which greatly reduced the effectiveness of the policy.

For example, after the Korean War, US Secretary of State Dulles advocated the continuation of a comprehensive embargo on China and North Korea, and urged allies to follow suit. However, Britain and France insisted that the trade policy toward China should be consistent with that toward Eastern Europe, and that "selective decoupling" should not be practiced. Under such pressure, although the United States once relied on the "Battelle Act" to exert pressure, it finally compromised in 1957 and allowed its allies to resume some trade with China and North Korea. The United States itself has maintained a unilateral embargo policy against China until the Nixon administration came to power in 1969, when it gradually relaxed the ban.

Compared with companies in third countries, I think the most embarrassing ones are actually those companies in China that still rely on American technology but must deploy Huawei Ascend chips. For example, some large AI companies, cloud computing platforms, and operators may not have fully achieved "de-Americanization" of technology, and still involve EAR controlled items in the procurement process or supply chain. In this case, if Huawei Ascend chips continue to be used, they may be deemed by the United States to have violated the FDPR and face the risk of being included in the Entity List. Once on the list, even if they completely get rid of American technology in the future, they will encounter many obstacles in business operations, financing, global cooperation, etc. The difficulties may only be felt by experts in compliance/strategy in large companies.

The second "Guidance" is actually about how to prevent American AI chips from being used to help China train large models.

This is not a new issue raised by the Trump administration. Similar restrictions have already begun to be brewed during Biden's term. At that time, two paths were mainly considered:

One ​​is to pass legislation, such as the "Remote Access Security Act", to give the Ministry of Commerce the power to conduct export controls on IaaS cloud services, that is, to prevent American cloud service providers from selling computing power to Chinese users for training cutting-edge models, but this path did not eventually become law.

Another path is the "AI Proliferation Rules" announced later, which directly set limits on the training of large models themselves. As long as the closed-source model is trained using controlled GPUs, servers and other hardware, and the amount of computational operations during training exceeds 10 to the 26th power, even if the model is trained outside the United States, its "model weights" are still controlled items, and exports must comply with U.S. export controls. It also introduced a "red flag warning" - if a U.S. IaaS cloud service provider trains a large model for a subsidiary of a foreign company in the United States in its home country, and the model is later transferred to the customer's foreign parent company, then this behavior may constitute an export, and the U.S. company is at risk of assisting in violations. Therefore, BIS recommends that cloud service providers should find out how customers intend to use the model weights before delivering them, whether they need to apply for a license, and if so, they must get it first.

However, now that Biden's AI Proliferation Rules have been explicitly abolished by the Trump administration, the entire model weight control logic has suddenly lost its specific legal basis. So now we are back to square one: How can the U.S. government prevent American AI chips from being used to train Chinese models? This is an issue that is still hotly debated in the domestic policy circles of the United States.

From the wording of the second "Guidance Opinion", it is more like a "warning" to the public rather than a hard rule - telling you that using American chips to help China train models is a sensitive behavior and may cause trouble, so it is best not to do it. But as to whether it is illegal to do so, whether you will be fined, how much the fine should be, and who will do the fine, none of them are clear. From this perspective, it is unlikely to directly impose clear legal obligations on American cloud service providers. After all, according to the principles of administrative law, if such "Guidance Opinions" want to have mandatory effect, they still have to be attached to specific formal EAR rules, and now there are none.

In contrast, the third "Guidance" seems the least conspicuous. It does not create any new restrictions, but reminds American companies to pay attention to supply chain risks and prevent chips from being transferred. It is more of a "compliance reminder" rather than a legal obligation, but it may actually be used as a risk clue or a basis for post-examination during law enforcement.

The above is only a preliminary analysis based on the limited information currently available. The details still depend on the full text of the "Guidance", which I believe will be released soon.

Attachment: Full text of the Ministry of Commerce’s notice

https://www.bis.gov/press-release/department-commerce-rescinds-biden-era-artificial-intelligence-diffusion-rule-strengthens-chip-related

The Department of Commerce revokes the Biden-era artificial intelligence diffusion rule and strengthens chip-related export controls

Washington, DC – Today, the U.S. Department of Commerce initiated the process to revoke the Biden administration’s artificial intelligence diffusion rule and announced additional measures to strengthen semiconductor export controls worldwide.

The artificial intelligence diffusion rule was issued on January 15, 2025, and its compliance requirements were originally scheduled to take effect on May 15, 2025. These new requirements would have stifled American innovation and imposed burdensome new regulatory requirements on businesses. The AI ​​proliferation rule would also undermine U.S. diplomatic relations with dozens of countries by relegating them to second-class status.

The U.S. Department of Commerce’s Bureau of Industry and Security (BIS) plans to publish a notice in the Federal Register formally confirming the revocation of the rule and will publish a replacement rule in the future.

Jeffrey Kessler, Under Secretary of Commerce for Industry and Security, has instructed enforcement officials at the Bureau of Industry and Security not to enforce the Biden administration's AI proliferation rules, saying:

"The Trump administration will pursue a bold and inclusive strategy to collaborate with trusted foreign countries around the world on AI technology while preventing technology from falling into the hands of our adversaries. At the same time, we do not approve of the Biden administration's practice of imposing its ill-considered and counterproductive AI policies on the American people."

In addition, the U.S. Department of Commerce's Bureau of Industry and Security announced today actions to strengthen export controls on overseas AI chips, including:

1. Issuing guidance stipulating that the use of Huawei's Ascend chips anywhere in the world violates U.S. export controls.

2. Issued guidance warning the public of the potential consequences of allowing U.S. AI chips to be used for training and inference of Chinese AI models.

3. Issued guidance to U.S. companies on how to protect their supply chains from transshipment strategies.

Today’s actions ensure that the United States remains at the forefront of AI innovation and maintains its dominance in the global AI space.