Microsoft Is Selling OpenAI Models in China Through a Deal the Creator Itself Refuses to Make
Microsoft is doing business that OpenAI itself refuses to do.
While the creator of GPT models keeps China off its radar over intellectual property concerns and misuse risks, Microsoft went in the opposite direction and quietly became the main supplier of those models to China’s biggest tech companies. Anthropic, another major AI model developer, also stays out of the Chinese market for the same reasons, and its models don’t even show up in Microsoft’s catalog in China.
Seems contradictory, right?
But that’s exactly what’s happening, and the numbers behind this operation are staggering.
The story gained traction after Bloomberg detailed how this arrangement works in practice, revealing that giants like ByteDance, Ant Group, Meituan, and Tencent are buying artificial intelligence models from OpenAI through Microsoft’s Azure infrastructure. Ant Group, for its part, says it develops its own models and that its core products don’t rely on external systems, but it still shows up among the service’s buyers.
ByteDance alone, which has been Microsoft’s largest AI customer in recent years, is on track to spend more than $1 billion per year on these services, according to sources interviewed by Bloomberg.
But what makes this story even more interesting is the unique role Microsoft plays in this scenario — it’s the only American company capable of operating on both sides of this geopolitical equation at the same time.
It’s worth understanding how this works, what risks are involved, and what this move says about the future of artificial intelligence worldwide. 🌐
Receive the best innovation content in your email.
All the news, tips, trends, and resources you're looking for, delivered to your inbox.
By subscribing to the newsletter, you agree to receive communications from Método Viral. We are committed to always protecting and respecting your privacy.
How Microsoft Became the Middleman Between OpenAI and China
To understand what’s going on here, we need to take a step back. OpenAI actively blocks direct access to its GPT models for users and companies located in China. The reasoning is straightforward: the company has serious concerns about intellectual property, the potential misuse of the technology, and the national security implications of transferring advanced AI capabilities to a country that directly competes with the United States on the tech front. This isn’t a small decision — it’s a strategic stance that reflects the current geopolitical landscape.
Microsoft, on the other hand, has a very different relationship with the Chinese market. The company has operated in China for decades, with offices, local partnerships, and robust cloud infrastructure through Azure. This setup allows Microsoft to offer cloud services in compliance with Chinese regulations — something few American companies have managed to build at that scale. And it’s precisely this established presence that opens a path for OpenAI’s artificial intelligence models to reach the market that OpenAI itself decided not to serve directly.
The key to everything lies in the unique contract Microsoft has with OpenAI. This agreement allows Microsoft to set its own terms for selling GPT models overseas, without needing case-by-case authorization from OpenAI. In practice, this means Chinese companies like ByteDance, Ant Group, Meituan, and Tencent access GPT models through Azure OpenAI Service, the enterprise version of OpenAI’s models hosted within Microsoft’s infrastructure. Technically, they’re not contracting OpenAI. They’re contracting Microsoft, which in turn licenses and makes those models available. It’s a layer of intermediation that simultaneously solves the access problem for Chinese companies while keeping OpenAI formally out of the direct equation with China.
One important detail: to limit its exposure, Microsoft doesn’t host OpenAI’s models on Chinese soil. Clients in China access these models over the internet, connecting to data centers located in other countries, like Singapore. This is a relevant technical distinction that Microsoft uses as an argument that it’s taking additional precautions in the operation.
The Numbers That Show How Massive This Market Is
When Bloomberg broke down the details of this operation, one number caught everyone’s attention in the industry: ByteDance, the company behind TikTok, is projected to spend more than $1 billion per year on AI and cloud services through Azure. That’s not just any number. We’re talking about one of the largest cloud computing contracts in the world, and it’s being used to power artificial intelligence products that compete globally with American tools. ByteDance uses these resources to boost its own platforms, including content generation tools, algorithmic recommendation systems, and the development of new AI products.
But ByteDance isn’t alone on that list. Ant Group, the financial arm of the Alibaba ecosystem, Meituan, the delivery and local services giant, and Tencent, the company behind WeChat and one of the biggest tech conglomerates on the planet, are also among the clients using GPT models through Microsoft’s infrastructure. Together, these companies represent a massive slice of the Chinese digital economy, and the fact that all of them are integrating advanced OpenAI AI models into their products speaks volumes about the level of dependency forming around this technology.
Internally, Microsoft doesn’t just accept this growth — it openly celebrates it. According to a transcript of a sales meeting held in July 2025, reviewed by Bloomberg, then-chief commercial officer Judson Althoff told employees that Azure AI revenue in China grew faster than in any other sales territory for the company. Growth was approximately 400% the previous year and nearly tripled in the fiscal year ending June 2025. Althoff described Microsoft as the only company capable of connecting the two great AI hubs of the world: the west coast of the United States and eastern China.
Microsoft president Brad Smith also addressed the topic publicly. In statements to American lawmakers, Smith said the company’s business in China represented approximately 1.5% of total revenue in 2024. That might sound small in percentage terms, but considering Microsoft’s annual revenue exceeds $200 billion, we’re talking about pretty significant figures in absolute terms.
The Behind-the-Scenes Tension Between OpenAI and Microsoft
It’s not all smooth sailing in this relationship. Behind the scenes, OpenAI has been pushing Microsoft to strengthen the mechanisms that prevent Chinese clients from performing so-called distillation. This technique involves using the outputs of a large, powerful model to train a smaller, independent model — essentially transferring knowledge from one system to another without needing the original model afterward. It’s a legitimate concern, because if a Chinese company manages to distill a GPT model and create its own local version, OpenAI’s intellectual property would effectively be replicated outside its control.
Microsoft, for its part, argues that it has automated monitoring mechanisms in place and that it limits model sales to established companies, excluding individual developers. But sources interviewed by Bloomberg indicate that Chinese buyers don’t face significantly more scrutiny than clients in other regions. On top of that, synthetic data generated from the models is extremely hard to trace and police, making effective oversight a real technical challenge.
This tension illustrates a conflict of interest that will likely intensify over time. OpenAI wants to protect its technology. Microsoft wants to maximize the return on its multi-billion-dollar investment in OpenAI. And the Chinese market is right in the middle of this dispute, representing both a massive revenue opportunity and a strategic risk that’s hard to quantify.
The Double Game Microsoft Is Playing
The contradiction becomes even more striking when we look at the flip side. Microsoft isn’t just selling American models to China. It’s also bringing Chinese models to the West.
In January 2025, the company added DeepSeek R1 to Azure AI Foundry, its AI model platform. And more recently, in June 2026, Microsoft confirmed to Axios that it’s testing a fine-tuned, Azure-hosted version of DeepSeek-V4 as a cheaper option for Copilot Cowork, the enterprise agent that currently runs on OpenAI and Anthropic models.
In other words, Microsoft is simultaneously:
- Selling American OpenAI models to Chinese companies
- Selling Chinese DeepSeek models to Western companies
- Pocketing the profit margins on both sides of the operation
This positioning is strategically brilliant from a business standpoint, but it raises deep questions about technological neutrality, national security, and how effective the restrictions governments try to impose on the international flow of AI technology really are. Microsoft is positioning itself as a kind of bridge between the two largest AI ecosystems on the planet, profiting from the movement in both directions. 🤔
What This Move Reveals About the Geopolitics of AI
There’s a fascinating tension at the center of all this. The United States has been applying increasingly strict restrictions on the export of high-performance chips to China, specifically to limit the country’s progress in artificial intelligence. The logic is straightforward: without advanced hardware, training large models becomes much harder. But what the situation involving Microsoft and OpenAI shows is that access to already-trained models is an entirely different door — and it’s still wide open. Chinese companies don’t need to train their own models from scratch if they can access, through APIs and cloud services, models that have already gone through the training process using computational resources they don’t have direct access to.
This point is raising serious questions within the American government itself and among tech security experts. If the logic behind chip restrictions is to prevent China from independently developing advanced AI, allowing Chinese companies to use GPT models through intermediaries could be, at least partially, undermining that goal. Microsoft argues it operates within existing rules and that the models available in China through Azure go through additional filters and restrictions, but critics point out that the line between using a technology and depending on it is getting thinner by the day in this context.
Microsoft’s business in China is already a sensitive topic in Washington, where lawmakers have classified China’s AI advancement as a threat to American industry. OpenAI’s private objections could gain more weight if political pressure increases, creating a scenario where Microsoft might be forced to choose between profits in the Chinese market and maintaining favorable political relationships in the United States.
From Microsoft’s perspective, the position makes strategic sense. The company has a multi-billion-dollar stake in OpenAI and needs to monetize that investment in every way possible. The Chinese tech market is one of the largest in the world, and walking away from it would mean leaving enormous money on the table. Beyond that, Microsoft has a track record of operating in regulated and politically sensitive markets, adapting its products and services to local requirements without abandoning the market entirely. The question is how long this balancing act will be sustainable, especially if American regulations on AI technology exports get tighter.
What This Means for the Future of Global Artificial Intelligence
This story paints a pretty clear picture of how the global artificial intelligence market is organizing itself. On one side, we have companies like OpenAI and Anthropic making decisions based on values, risks, and regulatory pressures to keep their distance from certain markets. On the other, we have cloud infrastructure operating as a separate layer, with its own business rules and the ability to reach where model developers choose not to go. This creates an AI distribution model that’s far less centralized than it appears at first glance — and much more dependent on who controls the infrastructure than on who creates the models.
For Chinese companies, this access to GPT models through Microsoft represents a real competitive advantage in the short term. Integrating advanced AI models into consumer and enterprise products accelerates the development of features that would otherwise take much longer to reach the market. But it also creates a dependency on foreign technology that runs counter to the official policy of technological self-sufficiency that the Chinese government has been promoting. It’s a paradox that the companies themselves seem willing to accept as long as local models haven’t reached the same level of performance.
What stands out from these numbers is that there’s real and growing demand for cutting-edge artificial intelligence models in China, even with local alternatives from Baidu, Alibaba, and other companies in the space. This suggests that, at least for now, GPT models still offer something that local competitors can’t fully replicate — whether in technical capability, performance consistency, or integration with already-established enterprise workflows.
What this episode makes clear is that the battle for leadership in artificial intelligence isn’t being fought only in research labs or the halls of the American Congress. It’s also being decided in cloud contracts, API access policies, and the strategic choices of companies that, like Microsoft, have the ability to operate simultaneously in ecosystems the rest of the world treats as separate. Whoever controls the distribution of artificial intelligence models may hold as much power as whoever creates them — and for now, Microsoft is the only company getting paid by both sides of this equation. 🚀
