There are new headlines, primarily from The Information, that the Trump administration is hard at work trying to figure out how to legislate bans on the rental of GPU compute to China from places like Vietnam and Singapore. Despite a paywall, a graphic on the piece shows data centers in these outposts beaming valuable data to the Middle Kingdom, which, to the American government, is unacceptable. The question is: can you police another country’s access to remote services?
The Hardware
I’ve already written extensively about the back and firth regarding export controls for Nvidia chips. First, the Chinese could buy H20s. Then they couldn’t. Then they could. As newer chips come on the market, the U.S. relaxes export controls for the old ones. Sometimes, the administration institutes a cap on GPU export numbers to firms like Alibaba.
I think it would be fair to call this a hot mess. But trying to block China from cloud services seems like it would be even harder. I reported last week that despite any Nvidia export bans, which Jensen Huang has historically changing by meeting with Trump, thousands of banned GPUs get smuggled into China, anyway, with switched labels and deceptive packaging.
Hands off That Server!
The Remote Access Security Act, introduced last December, already works to try to curb China’s access to remote services.
“Under current law, bad actors can train AI models by accessing advanced chips under the jurisdiction of the US, and the Bureau of Industry and Security has no authority to require a license. This legislation closes this existing security gap by extending export controls to include remote access scenarios, ensuring that both physical possession and remote access to sensitive technology by foreign advisories face equivalent scrutiny when national security risks are present,” said Dave McCormick, a sponsor.
But, as GPT cutely told me, “bits cross borders more easily than chips,” so I think it’s a fair question to ask: if the U.S. can’t control the hardware, how would officials ever hope to put a serious dent in the sale of international cloud services with both participants beyond American borders?
Roadblocks and Sandbags
There’s an argument to be made that in pursuing these laws, the American government isn’t really trying to completely bar China from signing into remote servers completely, but just to slow the competitor down by attrition, like in the Cold War between the U.S. and the Soviet Union, where eventually, one of the contenders had to revise its imperial ambitions.
That makes more sense, and it’s more like what’s happening with Nvidia. It’s not that the Chinese can’t get any Nvidia GPUs: its more that central command or private parties there can’t just order up the newest Grace Hopper and Vera Rubin architectures willy-nilly, and that these bottlenecks can slow China’s bid for dominance.
As for new rules, here’s more on a House plan to implement another such restrictive piece of legislation.
Here’s testimony from a sponsor, John Moolenaar (R-MI):
“In the AI race, China will buy what it can and steal the rest, which is why it is actively trying to get backdoor access to U.S. data centers and train its AI models via cloud computing. U.S. cloud platforms have a role to play in stopping China’s AI buildup, which fuels its military and surveillance ambitions. This bipartisan commonsense legislation will require them to protect their products and American national security by simply verifying the identity of their users.”
Here’s more from the other sponsor, Josh Gottheimer (D-NJ):
“We can’t let our adversaries — especially China — dodge our export controls by simply renting what they can’t buy. This bill gives American companies the legal clarity they need to do the right thing and report when bad actors are trying to use our own cloud infrastructure to threaten our national security.”
Is it just saber-rattling? Or can China be hobbled, or tamed, in this way?
We’ll see. The Sino-American race is probably not peripheral to the eyes of the world, as other nations also race to create their own national-language and domestically controlled AI platforms.
Keep an eye out.