The Geopolitical Debates Over Controlling Cloud Compute
If U.S. policymakers continue down the path of restricting China’s access to frontier AI, they will eventually have to implement some sort of restriction on cloud access.
This article was published at the Carnegie Endowment for International Peace. Read it HERE.
Much of the discourse on export controls has focused on restricting the sale of physical semiconductors. But a more understated loophole is allowing Chinese customers to remotely access the same restricted U.S. chips through cloud service providers. Now, Congress is trying to close that loophole.
In a new article for the Carnegie Endowment for International Peace, I discuss the logic behind these “cloud controls” and why they're more complicated than they seem.
Here are my 4 key takeaways:
1. THE LOOPHOLES: Chinese companies can legally lease compute either from third-country or U.S.-based clouds. Because the physical chips are housed in licensed data centers, it doesn't count as an "export" subject to U.S. restrictions.
2. THE LOGIC BEHIND CLOUD CONTROLS: Cutting off physical chips to China for national security purposes is undermined by remote access. It also weakens the leverage that the U.S has in physical chip license negotiations and undercuts American R&D expenses into model development.
3. THE CONS OF CLOUD CONTROLS: The compliance burden of cloud controls may push foreign-owned data centers to seek more convenient Chinese alternatives, not to mention the diplomatic objections to U.S. extraterritorial authority. Remote access may also provide the US with an intelligence window into China's AI development.
4. OPEN QUESTIONS: Would cloud controls impose restrictions on all foreign entities or just Chinese renters? And what cloud services would be restricted? Each answer has to reconcile security concerns with the practical and strategic costs of regulating a transnational service.