For years, there have been reports of China utilizing cloud computing infrastructure, particularly in Singapore and other Southeast Asian nations, to bypass U.S. export bans on advanced AI chips. Since early 2024, the United States has proposed requiring U.S. cloud companies to determine whether their data centers and cloud services were being accessed by companies from adversarial nations to train AI models.
By January 2025, in one of the final export control acts of the Biden administration, the Framework for Artificial Intelligence Diffusion established a framework for validated end-user authorizations under its “trusted datacenter programs,” similar to a “know your customer” (KYC) requirement for cloud operators.
Now, the Remote Access Security Act (RASA), proposed in both the U.S. House of Representatives and the Senate, attempts to further plug the “cloud compute loophole” by authorizing the U.S. government to regulate not only the export of AI capabilities, but also the usage of such capabilities. On the surface, this seems to be a logical extension of the hardware and software export controls already in place, to further prevent U.S. adversaries from accessing U.S. technologies to train their AI models.
But let’s not forget, two years or so is a very long time in the AI time scale, and a lot has happened. Did we learn our lesson, as far as balancing export controls and advancing U.S. technology globally is concerned? Let’s take a quick look back at the rapid back and forth of policy changes in just the past year or so, and the impact on China-U.S. AI competitive reality.
Back in April 2025, the Trump administration tightened export controls on Nvidia’s H20 chips, which had been specifically designed to comply with export restrictions for China, and similar chips from other U.S. companies. Needless to say, the industry pushed back and warned that rather than slowing China’s AI development, the restrictions would instead accelerate the adoption of China’s domestically designed and manufactured alternatives, such as those from Huawei. By late 2025, the Trump administration reversed course to approve H20 exports, and later even the more capable H2000 chips under a licensing regime.
So, did China rejoice and embrace the return of the American chips? No. While Beijing stopped short of banning the import of Nvidia chips, the Chinese authorities delayed, discouraged, or blocked Nvidia purchases by Chinese firms, in favor of grooming its domestic market for its own budding and rapidly improving suppliers such as Huawei and Cambricon. In an interview in May 2026, Nvidia founder Jensen Huang lamented his company’s plunge in the Chinese market share down to “zero.”
This is not to say that export controls on the most advanced AI chips, hardware, software, and manufacturing technologies to China are a mistake. There are strong reasons why these technologies should not be allowed to be exported to foster the AI competitiveness and military capabilities of adversaries. But AI chips and cloud services are not the same. And, second, we must be clear-eyed about the downside effects on any such restrictive policy on U.S. President Donald Trump’s proclaimed goal of exporting the full-stack American AI technology package, detailed in his Executive Order 14320 and later his “America’s AI Action Plan.”
For a long time, U.S. cloud and facility operators have been the world’s most competitive and successful. They dominated as the global suppliers of commercial public cloud services as well as data centers around the world, from the big three cloud providers of Amazon Web Services, Microsoft Azure, and Google Cloud, to data center facility providers like Equinix and Digital Realty. If legislation such as RASA imposes heavy-handed restrictions on them globally – stopping them from providing services to entire categories of legitimate commercial users, instead of clearly-defined targeted bad actors, such as military or prohibited entities – the effects can be devastating.
Rather than denying services to an established black list, if the United States begins to require cloud operators around the world to demand that their customers to “prove your innocence,” the uncertainty or even hostility will drive legitimate users and friendly nations to reduce their dependence on U.S. firms. Instead, these third parties would build their own AI ecosystem around Chinese technologies, or turn toward Chinese cloud providers, such as Huawei Cloud, AliCloud, and others – all eagerly waiting for this windfall to happen. This would be a major setback for the president’s goal to export the American AI technology stack.
When we use export control tools wisely and prudently to limit China from taking advantage of U.S. technology for their own AI development and military applications, we must not forget that dependence on American technology is the last thing China wants anyway. At the end of the day, they do not want any U.S. technology. They want to develop their own indigenous and global firms to displace the American leaders – not only inside China, but in the world, maybe even in the United States too. For this reason, U.S. policy must be designed carefully such that it would never again be taken advantage of by China, and end up helping Beijing to achieve its goal, directly or indirectly.
Beware thinking too much about slowing down our adversaries from overtaking us, lest we take steps that will instead slow ourselves down. It has just happened. Remember that in a real competitive race, the one who runs fastest always win. In the context of global AI competition, it is about keeping U.S. technology in front, maintaining an open global market for our firms to compete freely in, and making sure that the rest of the world will have no reason not to choose American technology over China’s.