When data security drives AI preference

AI evolves rapidly, but some firms opt to wait and see until their ideal solutions emerge

Featured faculty: Jiang Wenxi and Gao Zhenyu
Written by Putro Harnowo

The launch of ChatGPT in late 2022 by OpenAI was a turning point for artificial intelligence (AI) worldwide. While the chatbot was not officially available in China, enterprise partnerships with Microsoft Azure and workarounds such as virtual private networks and third-party proxies have helped, to some extent, the ChatGPT moment reach the country.

However, that access closed when OpenAI decided to withdraw completely from the country in 2024. Another AI giant, Anthropic, released Claude in 2023 but never made it available to the Chinese market. Although it seems that American tech firms are pulling away, Chinese firms are equally hesitant to rely on foreign AI technology.

No wonder that DeepSeek’s debut in early 2025 sent another shockwave, the DeepSeek moment, and dramatically jump-started China’s AI adoption. “Technology adoption is not purely economic, but one deeply intertwined with strategic priorities,” says Jiang Wenxi, Professor of Finance at the Chinese University of Hong Kong (CUHK) Business School.

In a paper titled AI sovereignty, Professors Jiang and Gao Zhenyu, Associate Professor in the same department, in collaboration with Fudan University’s Ren Haohan, Wang Kemin, and Wu Yuezhi, examine more than 28,000 detailed dialogues between Chinese listed firms and their investors from January 2022 to June 2025 on investor interaction platforms. Investors frequently inquire whether firms use or plan to use specific AI models.

As expected, ChatGPT and DeepSeek were discussed far more frequently than other models. The launch of both also triggered notable adoption spikes, and a consistent pattern appears: Many Chinese firms were hesitant to use AI at first, but once DeepSeek became available, they accelerated.

From an economic point of view, DeepSeek is significantly cheaper than ChatGPT, so this could be one of the main reasons for avoiding foreign AI models, but the study notes that cost has never been a problem. Rather, there is a bigger concern.

Data is the lifeblood of AI

Firms that are consistently reluctant to adopt foreign AI but show a greater preference for domestic AI often use data-related terms and keywords in their annual reports. This suggests that the data is highly valuable to them and any leaks or breaches will result in dire consequences.

AI tools often require sending data to external servers, which raises the risk of sensitive data leakage. Since Chinese firms handling critical data are subject to regulations on data handling and processing, sending data to a foreign AI’s servers could pose a compliance risk.

“Priority of safeguarding data security makes firms wary of foreign AI models that could inadvertently funnel proprietary or sensitive data out of China,” says Professor Jiang.

Data is a fundamental resource for training AI models. Without vast amounts of data, it would be impossible for AI models to perform analysis and generate outputs and predictions. Businesses also increasingly see data as a critical asset, much like oil or other resources, that can give them a technological edge.

Therefore, China has issued a series of laws, including the Cybersecurity Law (2017), the Data Security Law (2021), and the Personal Information Protection Law (2021), that provide a comprehensive framework for handling data. These laws require data to be stored on servers physically located within the country.

Cybersecurity Law laid the foundation for data protection and set broad rules for network security and critical information infrastructure. Data Security Law imposes strict rules on data collection, storage, use, and cross-border transfer. The Personal Information Protection Law dictates how personal data must be handled, including requirements for transferring outside of China.

Similar to many other countries, China’s data protection laws borrow heavily from the EU’s General Data Protection Regulation. While the EU does not impose a blanket requirement to keep all data locally, it sets very strict rules on how its citizens’ data must be protected, regardless of where the company processing the data is located.

“Technology adoption is not purely economic, but one deeply intertwined with strategic priorities.”

Professor Jiang Wenxi

More than just supporting local technology

China produces more top AI researchers and talent, but the US leads in AI models and the critical chips for AI training. The discussions captured on investor interaction platforms also acknowledge this. Investors understand that a preference for domestic AI models may place Chinese firms at a competitive disadvantage.

Regardless, Chinese firms are keen to use domestic AI models to avoid compliance risks. “Although the best Chinese AI models still underperform the US ones, the gap is quite small and may not be significant when applied to specific business scenarios,” Professor Jiang adds.

Further analysis indicates that Chinese firms do not completely ignore foreign models but strategically limit their use to low-risk domains, such as customer service, translation, and other non-critical functions.

Moreover, the market supports this approach. When Chinese firms, especially those in strategic national sectors, announced their AI adoption, their stock prices rose significantly. Conversely, when these firms said they were using foreign AI, their stock performed worse or at least stayed unchanged.

The future of AI ecosystems

AI has become a foundational utility, much like the internet, and has driven exponential investment in China and the US, the dominant players in AI development. Alas, the two superpowers have locked in a head-to-head competition for years.

Most recently, the US has banned the sale of its advanced AI processors to Chinese firms and their subsidiaries, while China intervened in the acquisition of a Chinese-founded AI firm, Manus, to prevent the outflow of proprietary technology and AI talent to the US.

The available data and privacy frameworks determine how AI models are trained, which then shapes the next wave of AI innovations. Since data security influences AI adoption and development, this loop may lead to AI models from two dominant powers evolving differently.

“The development of AI technology may evolve into two separate approaches and ecosystems,” Professor Jiang says. “However, Chinese models are open source, so firms can fine-tune them locally to meet specific needs with greater data control for developers to adopt. This gives some hope that the separation might not be exacerbated.”

Ultimately, with the growing calls for global AI governance to safely advance the technology, collaborative frameworks may emerge to mitigate the risks of a fragmented ecosystem and ensure that the benefits of AI technologies are shared universally.