|
China is a tough one. It’s a huge market overseen by an authoritarian regime that exercises pretty tight control over media and the exchange of information. That can make it a difficult place to do business for Western companies – including OpenAI and Anthropic, which don’t offer their AI services in the mainland.
That doesn’t mean you can’t get access to frontier Western AI in China, though. Microsoft is happy to sell OpenAI models through its own Azure cloud platform. And it’s making a lot of money doing so, according to a new Bloomberg report.
Its biggest Chinese customer is ByteDance, which is reportedly on track to spend over $1 billion on Microsoft-offered AI and cloud services this year. ByteDance is a huge tech company, and is best known in the West for its TikTok social media platform.
TikTok offers a very helpful illustration of why some Western companies are reluctant to engage with China. In recent years, both the Trump 1 and Biden administrations attempted to ban TikTok – or at least get it divested from ByteDance – due to national security concerns. The fear was that, as a China-based company, ByteDance could be compelled by the CCP to hand over sensitive user information. And by the way – this wasn’t very widely reported during those efforts – but TikTok’s own privacy policy included language allowing the app to collect biometric and background environment data from user videos, potentially turning the app into a useful surveillance and reconnaissance tool.
American AI companies have some of the same concerns. Anthropic, for example, tightened its own terms of use last year in response to concerns that entities from unsupported regions were still finding ways to access its AI. Here’s what the company said about that in a blog post:
Companies subject to control from authoritarian regions like China face legal requirements that can compel them to share data, cooperate with intelligence services, or take other actions that create national security risks. These requirements make it difficult for companies to resist these pressures regardless of where they operate or of the personal preferences of the individuals at those companies. When these entities access our services through subsidiaries, they could use our capabilities to develop applications and services that ultimately serve adversarial military and intelligence services and broader authoritarian objectives. They could also potentially use our models to advance their own AI development through techniques like distillation and to compete globally with trusted technology companies headquartered in the United States and allied countries.
OpenAI has been less vocal about such concerns. But it doesn’t officially offer services in China, and has publicly disclosed multiple instances in which it banned accounts that were suspected to be China-based, state-affiliated “threat actors” who were maliciously using its AI.
The thing is, Microsoft is OpenAI’s biggest investor. It has poured well over $13 billion into the startup, and for a while it was OpenAI’s exclusive cloud provider, though the companies relaxed their partnership somewhat in recent months to allow OpenAI to team up with other cloud and infrastructure providers. Still, Microsoft has plenty of access to OpenAI’s models, and is happy to offer them to partners in China.
This isn’t exactly hypocritical on Microsoft’s part. Here’s part of its general human rights policy:
We believe we can more effectively respect human rights by being present in, rather than absent from, countries with significant human rights challenges. We believe that in the long run, engaging responsibly with people, governments and other stakeholders in difficult environments often holds greater promise for the advancement of human rights.
So, okay. That doesn’t really address the concern about ByteDance sharing user data with the CCP. And it seems like that could plausibly happen, but the data in this case is siloed: Azure in mainland China is a physically separate cloud operated by 21Vianet, a Chinese partner, with data centers located there. This means that, sure, the CCP could demand to see some of this data, but it wouldn’t necessarily get access to, say, American users’ data, which is stored elsewhere and has nothing to do with the arrangement.
The bigger issue here is geopolitical. Earlier this year, OpenAI told US lawmakers that it has reason to believe that DeepSeek – its main Chinese rival – used “model distillation” to replicate capabilities from its own AI models. Distillation is essentially the AI term for reverse engineering – collecting a huge amount of outputs from a given AI model and using them to train your own. High-volume API access would allow exactly that.
This is maybe another reason for OpenAI to further distance itself from Microsoft. It isn’t making any money in China, but its partner is – and it may indirectly be helping its China-based rivals to catch up to OpenAI’s position on the AI frontier.
|
