OpenAI cuts its last and most important link to China

On July 9, developers in China will lose API access to all OpenAI platforms, the final step in a slow process of cutting off the country’s access to U.S.-based machine learning tools.

Consumer-facing applications like ChatGPT have been unavailable in China more or less from the beginning. But the API is more important than the consumer-facing applications, powering apps to fill out forms, flag problems in internal data, or manage other kinds of automation at scale. Right now, Chinese developers still have access to those tools, and they’re using them — but they’ve got only a week and a half left.

Chinese competitors like Zhipu AI are already using the opportunity to tempt developers over to their platform, and the conventional wisdom is that most apps can transfer from model to model fairly easily. But with Nvidia’s best chips locked out of the country, the Chinese platforms are operating at a severe disadvantage. For at least the next few years, it will be hard for Chinese AI developers to keep pace.

This is mostly the result of U.S. government policy, encouraged by much of the U.S. tech industry. More than with any other consumer technology, artificial intelligence models are being treated as a national security issue , and it’s worth thinking hard about why. Part of it is straightforward pressure from the government, but we’re also witnessing a rising crop of hawkish CEOs who see a great power conflict as baked into the nature of what they’re working on.

Scale AI CEO Alexandr Wang laid out a clear-eyed version of this case in an interview with China Talk earlier this week. “To the degree that you think that AI is a military technology, which it almost certainly is, then the United States government has an imperative to be competitive and frankly, lead on AI,” he said. “They can’t just be passive and let it play out in the private sector.”

Scale AI is in the business of supplying training data for AI models, so it’s natural for Wang to emphasize that side of the conflict. Unless American AI can get better training data than Chinese AI, the thinking goes, we’re at risk of falling behind. I should also note that Scale AI is currently finishing up a four-year, $100 million contract with the U.S. Department of Defense — and would surely like to keep the money coming in. So when Wang worries that the U.S. military isn’t taking the threat seriously enough, there’s a clear financial incentive in play.

OpenAI has encouraged this line of thinking from the beginning, and it’s become a kind of table stakes for U.S. companies dealing in AI. As the walls go back up between the U.S. and Chinese tech sectors, there is a lot of money to be made by companies on the U.S. side of the wall. At the same time, there’s little incentive to question the military value of AI models. If you believe this technology will change the world, then military supremacy can feel like an inevitable part of that change. And if you don’t believe it, why are you working in AI?

Personally, I have my doubts. AI tools have shown their value in speeding up software development and discovering pharmaceuticals , but we don’t have any clear indication it will make much difference in the way we wage war. The safest answer is that we just don’t know what else it will be good for, which makes a big bet on a particular use case seem risky. But whether I believe it or not, many of the most powerful people in the world do believe it — and they’re having a huge impact on the way this technology develops. ▰

Russell Brandom is the U.S. Tech Editor at Rest of World.