OpenAI models still available in China via Azure cloud despite company ban

OpenAI models are still accessible through Microsoft Azure’s cloud in China despite the fact that the company has banned the use of these models in the region. The backdoor access to the models is part of a changing dynamic in China’s tech space, where emerging players hope to fill the gap the ban is poised to leave in the market, even as US-based tech firms look to circumvent growing trade restrictions.

Azure China operates as a joint venture with local company 21Vianet in China, which offers OpenAI’s service, according to an exclusive report by The Information on Monday. Three Azure customers in China also confirmed to the publication that they still have access to OpenAI’s models; two claimed they’ve used OpenAI’s API to train AI models sold to Chinese customers.

Microsoft confirmed to Computerworld Tuesday that Azure regions operated by 21Vianet are physically separated instances from Microsoft’s global cloud, though they are built on the same cloud technical base as its global peers. A company spokesperson said via email that “there has been no change” to its Azure OpenAI service offerings in China, and eligible customers can still receive access “via models deployed in regions outside China.”

Two weeks ago OpenAI sent letters to Chinese users warning it plans to cut off its AI development software and tools starting in July, according to multiple reports, incuding oneby Time magazine. This caused a rush by other China-based AI companies to incentivize developers using OpenAI to switch to their platform. 

“Already we see Baidu, Tencent, Alibaba and many other Chinese companies stepping in with heavy discounts in an attempt to pick up current OpenAI users in China,” said Brad Shimmin, chief analyst, AI and data analytics, at Omdia.

Baidu, for example, has promised free AI model fine-tuning and expert guidance on its flagship Ernie model, along with 50 million free tokens developers can use to query the bot, according to the Time report. Alibaba and Tencent posted ads encouraging the move, while Chinese technology pioneer Kai-fu Lee’s 01.AI is promoting heavy discounts to use its service, Time reported.

Meanwhile, at the World AI Conference in Shanghai last week, another Chinese AI company, SenseTime, unveiled its latest model — SenseNova 5.5; like Baidu, it offered companies 50 million free tokens to use the model, according to a separate report by The Guardian. SenseNova also promised to deploy staff for free to help new clients migrate from OpenAI to SenseTime’s AI tools.

Getting around trade restrictions

Microsoft invested billions of dollars in OpenAI in January 2023 and is closely aligned with the ChatGPT maker, integrating its technology through its own AI chatbot called Copilot, which is hosted on Azure and an integral part of its own products and services.

Microsoft did not provide a motive for allowing access to OpenAI in China through Azure. Shimmin, however, noted that China is a “sizeable market opportunity” for “mega-brands” like Microsoft, Google, Meta and Apple, “one worth the additional cost of establishing sometimes complex operating policies in order to do business in-country.”

For many companies operating within China’s borders, restrictions on technology and other products from US vendors are nothing new given the long-term battle between the two nations over tech supremacy. “Many companies have and are actively circumventing in-house blocks from the government using VPN services,” Shimmin said. 

The US most recently imposed a series of tight restrictions on the export of microprocessors to China. However, US President Joseph R. Biden Jr. made it clear last year that the tech trade war with China extends to other technology, including AI.

A competitive advantage

In addition to OpenAI, a number of US-based AI services aren’t currently operating in China, including Anthropic, which does not support mainland China or Hong Kong, and Amazon Bedrock from AWS, which is only available in the region in Singapore, Japan, and Australia, Shimmin said.

Microsoft’s circumvention of the OpenAI ban “underscores its commitment to the region and to its customers,” Shimmin said. 

It also could help the company maintain its competitive edge and market share, not only in AI but also in China’s lucrative cloud services market, even while keeping its relationship with OpenAI on track, said Stephen Kowski, Field CTO at SlashNext Email Security+.

“By offering continued access to OpenAI models, Microsoft can attract and retain enterprise customers seeking advanced AI capabilities,” he said. “This approach allows Microsoft to balance its partnership with OpenAI and its business interests in China.”

When given the choice to access OpenAI GPT models directly from OpenAI or via Microsoft OpenAI Azure Service, most enterprise customers would likely opt for Microsoft, Shimmin noted, “because they can access GPT without worrying about issues like data leakage or model privacy/security.”

More OpenAI news:

• OpenAI announces new multimodal desktop GPT with new voice and vision capabilities
• OpenAI unveils ‘Model Spec’: A framework for shaping responsible AI
• Is OpenAI’s Sam Altman becoming a liability for Microsoft?