Microsoft is expanding its AI business in China by integrating the artificial intelligence models of the Chinese giant DeepSeek into its Azure cloud platform. This move comes in a context of growing tensions between the United States and China on the technological front, with the American company seeking to balance commercial expansion in Asia and the increasing concerns raised by its Western partners.
Summary
Key points
- Microsoft is integrating DeepSeek’s AI models into the Azure platform to strengthen its presence in China.
- DeepSeek’s AI models are significantly cheaper in terms of training and management costs compared to their Western counterparts.
- Microsoft co-founded the Frontier Model Forum with OpenAI, Anthropic and Google to counter the phenomenon of illegal copying of Chinese AI models.
- OpenAI and Anthropic have expressed strong concerns about reverse engineering by Chinese companies.
- The US government has tightened export restrictions and increased security checks on Chinese AI companies, with possible future impacts on Azure.
Microsoft integrates DeepSeek’s AI models on Azure to expand in China
The integration of DeepSeek’s AI models into Azure represents a key strategic step for Microsoft’s AI business in China. Since January 2025, the DeepSeek R1 model has been made available on Azure AI Foundry and GitHub, thus expanding the offering of more than 1,800 models intended for enterprise customers. This choice stems from a strong economic motivation: the models developed by DeepSeek cost significantly less both in the training phase and in terms of operational management, compared to the equivalent Western models, such as those from OpenAI and Anthropic.
Currently, Microsoft is testing the DeepSeek-V4 version for its Copilot Cowork applications, with the goal of reducing inference costs, that is, the expenses related to the operational use of AI models. The adoption of these cheaper models makes it possible to segment workloads, assigning the more expensive models to complex tasks and using DeepSeek for routine or less sophisticated operations.
In practice, Microsoft does not directly market its own proprietary AI models in China, but supports those developed by Chinese companies such as DeepSeek through the Azure platform, which makes it possible to serve global customers without exporting technologies subject to restrictions from the United States.
Collaboration with Western leaders to counter Chinese intellectual theft
Despite leveraging Chinese AI models, Microsoft is also at the forefront with its Western partners in protecting intellectual property. In April 2026, together with OpenAI, Anthropic and Google, it co-founded the Frontier Model Forum, a group that works to counter the so-called “adversarial distillation” practices adopted by some Chinese entities. This process consists of replicating complex models through reverse engineering of their outputs, creating cheaper and more competitive versions to the detriment of the originals.
OpenAI and Anthropic have explicitly expressed concerns regarding security and the protection of intellectual property, welcoming the forum as a joint attempt to tackle the unauthorized spread of AI technology.
US restrictions endanger the use of Chinese models on Azure
A crucial issue for the future, however, concerns US regulatory actions. Since the Trump administration, the US government has tightened control over technological exports and increased security checks targeting Chinese AI companies. These measures include restrictions on advanced chips used for artificial intelligence and on access to the AI models themselves.
These limits are set to heavily affect Microsoft’s AI business in China. A further tightening of the rules could force Microsoft to remove DeepSeek models from Azure, which would eliminate the competitive advantage based on lower costs and could rebalance competition in favor of the more expensive Western models.
A delicate balance between supporting Chinese models and Western partnerships
Microsoft is therefore operating on terrain that requires very careful balancing. On the one hand, it supports the penetration of Chinese models through Azure at a global level, avoiding the direct export of technologies based in the United States. On the other, it maintains strong ties with the main Western AI players, who denounce the risks linked to industrial copy-and-paste practices carried out by some Chinese entities.
This strategy allows Microsoft to offer a diversified portfolio, capable of managing AI services with very different needs. Companies that are more cost-conscious can turn to DeepSeek models for less complex workloads, while more sophisticated and computationally intensive tasks require models from OpenAI or Anthropic.
The issue also has a very significant impact on investors and the global AI market, since the integration of low-cost Chinese models into the same Azure platform that hosts Western models pushes towards segmentation based on costs and complexity, with possible repercussions on the profitability of key partners such as OpenAI.
Content created with the assistance of artificial intelligence and with human editorial review.
