America is making a critical mistake in its AI strategy—it is overreaching in playing defense but doing the opposite in playing offense. While Washington is busy building regulatory walls to restrict access to AI compute power (a term encompassing computational resources used to train and run AI models), China is circumventing these barriers and strategically developing asymmetric advantages.

Recognizing that it will take time for China to match the US in high-performance AI compute, Chinese companies have focused on developing more efficient, less expensive AI technologies. They are also seeking to build dominant positions in open-source AI, cloud infrastructure, and global data ecosystems. These strategies enable China to offer cheaper, unrestricted AI access to countries frustrated by US policies, embedding itself into emerging markets in ways that will be difficult to dislodge. This is not just an AI competition—it is a battle for control over the global digital infrastructure today and in the future.

The Framework for AI Diffusion is a well-intended defensive policy that aims to secure US AI leadership by controlling the outflow of US technology, but in such a breathtakingly fast-changing innovation environment, its solution is already stale and fails to account for how China is adapting. 

By going beyond what is necessary to ensure its security, arbitrary US restrictions cede ground to China, creating an opening for Chinese AI firms to dominate AI adoption in the majority of nations. If the US wants to stay ahead, it must not overdo building high regulatory walls and not underdo running faster to stay ahead of China.

What’s Changing?

In October 2022, the US issued export controls (expanded in 2023 and 2024) restricting China’s access to AI semiconductors and manufacturing equipment. The 2025 expansion of these restrictions includes:

• Global licensing policy for exporting advanced chips used in data centers. 
• Limits on AI closed model weights—the critical parameters that power AI models.
• Export quotas on 140+ countries. 
• Universal Validated End-User (UVEU) and National Validated End-User (NVEU) designations, allowing trusted data centers to receive AI-related exports without individual licenses, but under strict compliance requirements (including prohibition against extending compute power to restricted countries).
• Geographic restrictions on compute power placement.

Countries are now classified into three tiers:

• Trusted Countries: The US and 18 close allies with unrestricted AI chip access, but limitations on cloud providers.
• Intermediate Countries: Face strict quotas, while cloud providers from trusted nations face limits on compute capacity placement.
• Restricted Countries: Includes China, Russia, Iran, and North Korea. They are effectively blocked from AI chip access.

Policy Undermines Allies’ Trust, Encourages Others to Hedge

This policy alienates key allies that the US relies on for defense, intelligence, and trade. To wit, 22 NATO members, including Poland and Greece, which the US has pledged to defend (along with Israel), are not classified as trusted countries. Poland’s Minister of Digital Affairs Krzysztof Gawkowski said, “This is a bad signal for Poland as it shows that the US administration does not understand that it has a loyal ally," Noach Hacker, minister of economic affairs at the Israeli Embassy to the US, said, “Excluding Israel from the recent regulatory framework as a close ally undermines [our] collaboration and misrepresents the strategic and economic reality.”

The blunt quota system applies equally to 140+ countries, treating disparate countries such as India and Tuvalu the same. This discourages key nations that hedge between the US and China—including India, Indonesia, Brazil, Turkey, Saudi Arabia, the UAE, Vietnam, and Singapore. Instead of pulling them closer, these restrictions could push them toward China. 

There must be a path for allies to become classified as a trusted country. The path should include incentives to hedging countries to nudge them America’s way.

Quotas and Geographic Compute Power Caps Open Door to China

The 2022 restrictions aimed to block China from acquiring AI chips for military applications. The 2025 rules seek to prevent China or other restricted countries from accessing AI compute power by implementing global licensing of advanced chips, restrictions on the provision of services by validated cloud providers, and AI closed model weights. These sound steps significantly decrease the likelihood of restricted countries benefiting from US compute power.  

It is not clear that the incremental security benefits of quotas and compute power caps outweigh their unintended negative consequences. Proceeding with these arbitrary restrictions could have a profound impact on the way AI transforms the world’s future. 

Overly restricting US industry could kill the geese whose eggs the rule seeks to protect. 

Raising the administrative hurdles for US firms to serve emerging markets and limiting their options for aiding a host country’s development could facilitate China dominating the digital infrastructure in most of the world’s nations. This would be a serious security risk. It would also put most of the world’s population in digital ecosystems designed to control citizens rather than empowering them. Much is at stake. Is the juice worth the squeeze?

Adding a quota-based AI allocation system forces companies to navigate a chaotic approval process, especially with the lack of a real-time, government-run tracking system of global compute power and the inherent need for the government to decide between competing firms if the quota is breached. Quotas risk discouraging companies from the US and allied countries from pursuing opportunities in the intermediate countries, especially in emerging markets. 

This would continue a pattern of over-regulation that has opened the door for expanded Chinese influence in Latin America, in addition to China becoming the leading economic partner to Africa. 

China is itching to extend its influence in AI. In its recently announced AI Capacity-Building Action Plan, China seeks to implement the UN's AI capacity-building resolutions, stating that “China is ready to establish an international cooperation platform for AI capacity-building.”

The restrictions on placement of compute power are equally troubling. New rules force US cloud providers (AWS, Microsoft, Google, Oracle) to keep at least 50% of their compute power inside the US while capping compute installations in individual intermediate countries at 7%. The desire to keep more data centers in the US is understandable but best achieved by making the country the best place to do so, including by ensure adequate energy supply and talent.

Similarly, trusted non-US cloud providers can only place 25% of their compute power in intermediate countries. Since 17 EU countries are classified as intermediate countries (much to the chagrin of the European Union),  a country such as France would be constrained in serving its fellow EU members, not to mention the rest of the world.

These geographic restrictions risk driving intermediate nations toward alternative suppliers that would allow them the economic benefits of data centers in their own countries, something Chinese firms are happy to accommodate. By prioritizing domestic resource concentration over AI leadership, these rules make it harder for US companies to compete globally.  Meanwhile, China will proceed with its Digital Silk Road, bringing to bear the full weight of the government to bundle multiple different providers in a package sweetened with enough subsidies to distract a buyer from any technology shortcomings.

The more sales US companies forfeit to Chinese competitors, the fewer funds they have and the more funds Chinese competitors have available to invest in research to get ahead. 

It is risky to assume that the US can maintain its current leadership position over the long term if its companies are only unfettered in doing business in 18 other countries while China has access to the whole world.

Ignoring China’s Alternative AI Strategy

The regulations reflect an outdated assumption: that US AI compute power is irreplaceable, and that blocking access to advanced computer chips and cloud providers will keep China contained. The reality is China is taking the exact opposite approach—embracing open-source AI while the US moves toward closed, tightly controlled AI systems.

For instance, China’s DeepSeek-R1 model  rivals OpenAI’s o1 model while being more efficient, open-source, and freely available. 

Chinese cloud providers are also expanding in emerging markets, seeking to embed themselves as the default AI provider for regions that Washington has ignored.

Imposing excessive US restrictions on intermediate countries will make Chinese AI solutions more attractive. The US must recognize that maintaining AI leadership is not just about blocking adversaries—it’s also about ensuring allies and partners have reliable access to superior technology that they can deploy at scale, instead of being forced to resort to a competing product.

The Data Factor: A Growing Strategic Blind Spot

Compute power is only one part of the AI equation—data is equally important. If the US forces intermediate countries to rely on Chinese cloud providers, it risks giving China unparalleled access to global AI training data.

China’s national security and cyber laws grant its government broad oversight over companies such as Alibaba Cloud and Huawei Cloud, meaning foreign data stored on their servers may not be fully private. This could give China a huge advantage in developing customized AI models that are optimized for different countries, allowing it to entrench its AI dominance in emerging markets.

The US made the same mistake with telecommunications by restricting Huawei in allied nations but allowing it to dominate elsewhere. Huawei has been able to dominate global markets in part because the US does not have an equivalent alternative. The US now leads in this future-changing technology and has the market share. However, America could lose its AI lead if China gains a significant data advantage.

China’s Expanding Control Over Critical Infrastructure

If the US does not compete aggressively for intermediate markets’ AI and telecommunications business, America and its allies may find themselves in a world where China controls data centers and telecommunications networks in the rest of the world—creating massive leverage in times of geopolitical tensions. 

While the US overreaches in building regulatory walls, China is encircling its competitors—laying the groundwork for a future where it controls chokepoints around the world. 

China has already demonstrated its forethought, planning and deliberating sustained execution to dominate markets. It has done the same in telecom, critical minerals, solar panels, electric vehicles, and more.  Sustaining America’s AI lead will require more thoughtful restrictions and more energetic effort to keep innovating.

A Smarter Approach: Blocking Access and Running Faster 

In addition to calibrated restrictions on who gets access to AI compute to prevent benefiting authoritarian militaries, the US must run faster in AI development. It must ensure an AI diffusion policy advances security without needlessly sapping free-market innovation and economic competitiveness.

As the administration of President Donald Trump crafts a comprehensive AI action plan over the next 180 days to enhance AI capabilities, the following suggestions should be considered:

• Keep licensing requirements, restrictions on UVEUs, NVEUs and AI model weights, but remove artificial caps and geographic restrictions on compute power, perhaps replacing them with reporting requirements to guide government responses to license applications. 
• Establish a path for intermediate countries to become trusted partners so that key nations can transition toward greater alignment with US AI policies.

Run faster.

• Ensure America is the most attractive place for data centers by streamlining permitting and providing access to affordable energy.
• Fully fund the research initiatives authorized under the CHIPS and Science Act to accelerate AI advancements in the US.
• Ensure universities and startups have access to high-performance AI compute so that the next generation of AI leaders can emerge domestically.
• Invest heavily in STEM education to create a workforce capable of maintaining AI leadership.
• Continue attracting the world’s top AI talent, ensuring that the best minds in AI research choose to develop cutting-edge models in the US.

Final Thoughts: The Risk of Losing by Playing the Wrong Game

The US is playing defense while China plays offense—and the longer America waits to adjust, the greater the risk it will lose its advantage. 

AI is not just about who builds the most powerful chips; it’s about who controls the global AI ecosystem. This is a race America must win. 

The AI diffusion rule as currently written creates a regulatory mess. It discourages key allies and hedging nations, while hampering the ability to meet emerging market needs, thereby handing China an economic advantage.

The national security goal of restricting China’s AI progress can be achieved without harming the US economy. If these rules are not adjusted, expect America’s tech dominance to erode—one misguided regulation at a time.

Article Co-Sponsored By: 

Image Credit

Boston Global Forum