Asian firms divided on Chinese AI as geopolitics bite

When Kien Ta pitched an AI solution to a Japanese healthcare firm, he suggested they use Qwen. Ta, the vice president of Vietnam-based software firm Rabiloo, thought that Alibaba’s open-source model was the best at Japanese language among the LLMs available.

But the client was quick to say no. They wanted Meta’s Llama instead. Ta never asked for the reason behind the preference, but he surmised they were only being “careful.”

“They are part of the regulated industry after all,” he adds.

As geopolitical tensions between the US and China continue to heat up, some enterprises in Asia are becoming more vigilant about which AI solutions to tap into.

Since 2017, the US has increasingly challenged China’s development of AI, implementing measures like bans on crucial chips for AI development. Under new rules issued earlier this year, companies with ties to Chinese firms can even lose access to high-end AI chips.

The concern, among other things, is that Chinese laws require local tech firms to cooperate with the government regarding issues of national security, which could then require them to allow authorities to access their stored data.

BT in your inbox

Start and end each day with the latest news stories and analyses delivered straight to your inbox.

Sign Up

Sign Up

In Japan, major companies, from SoftBank to Toyota, have banned the use of Chinese LLM DeepSeek inside company premises and have asked employees not to use it on company devices. One Toyota official reportedly said that the automaker had “concerns from an information security standpoint.”

The Chinese government has stressed that it will “never ask any company or individual to collect or store data against laws.”

The pronouncement hasn’t moved the needle much, though.

In fact, the use of Chinese AI has become a sensitive subject in the business community. In the course of reporting for this story, Tech in Asia encountered some enterprises who refused to even talk about their partnerships with AI companies from the country.

Some Asian firms, especially those with the goal of going global, are making sure their tech stacks do not include Chinese LLMs once they enter the US.

“These companies know they will face some level of uncertainty and scrutiny regarding regulation at some point regarding what AI supplier they use,” says James Pang Yan, co-director of the Business Analytics Center of the National University of Singapore (NUS) Business School.

“It will not be immediate, but long-term, the uncertainty and scrutiny will be there.”

Overall, experts and stakeholders interviewed by Tech in Asia believe that companies in the region – especially those only catering to clients in Asia – aren’t taking a hardline stance on the geopolitical divide.

For instance, in the Philippines, businesses aren’t currently forced to choose an AI supplier based on their origin, Amil Azurin, president of Philippine-based data center firm ePLDT, tells Tech in Asia.

But he believes the local government would require them to have an AI partner that “can have their data processed locally” to ensure data sovereignty, he adds.

Closed source vs. open source

The question on which LLM to use for a specific use case often begins with whether an enterprise wants the solution to be open- or closed-source.

For Thailand-based Amity Solutions, about 80% of its use cases for clients still use closed-source models like OpenAI’s GPT, Gemini, or Claude. The startup prefers these, as the models have achieved “economies of scale.”

“If you [go from] deploying the solutions built from these closed-source LLMs on your own private cloud to using an open-source LLM, that could be very expensive because it demands massive amounts of computing power,” Touchapon Kraisingkorn, group CTO and head of AI Labs at Amity Solutions, tells Tech in Asia.

The company counts large enterprises in Thailand as its clients, including the country’s largest telecommunications firm True Digital, retail chain Lotus, and listed Kasikornbank.

But for larger businesses that have established their AI requirements and are looking for more security for their proprietary data, open-source models are more attractive options, says Yan of NUS Business School. This allows them to choose how the model would process their data and where that information is stored.

Chinese players have long dominated the open-source LLM space, with Qwen and DeepSeek ranking in leaderboards as some of the most reliable open-source solutions.

OpenAI in August introduced its own open-source model, GPT OSS. But before that, only Meta’s Llama was among Western-developed LLMs that had tried to penetrate that market.

Experts have long said that offering open source is “strategic” for Chinese AI companies. It’s a way to increase adoption with a lower barrier of entry, as it allows even ordinary coders to build on top of the LLMs.

Doing so has allowed the likes of DeepSeek and Qwen to develop a strong following among the developer community.

Concerns muddle Chinese AI expansion

But enterprises still choose models based on where they come from.

Jerry Ye, CEO of Whale AI, says that some of his clients tell him they want no Chinese LLM to be included in their AI solutions.

Still, others specifically request DeepSeek or Qwen, as these are known for their reasoning and language processing capabilities.

Ye also has clients that serve both the US and Asia, and they choose to have different AI stacks for each market. He believes, however, that such a stance comes out of concerns on regulatory compliance more than geopolitical issues.

“Especially if the company is in a health or financial sector, where there are strict guidelines on how they process data – they really are more careful,” he adds.

Functionality and cost > source concerns

For most enterprises, though, an LLM’s functionality and cost are the two most important factors.

Sometimes, the question about which AI solution to choose is as simple as determining whether it would work with the firm’s existing cloud partner, for instance.

“Running an AI solution independently, outside of an entire organization’s cloud infrastructure, is possible,” Rabiloo’s Ta says. “But it could get costly and complicated, fast.”

Despite the stigma or security concerns involved in choosing an AI partner, Ming Yii Lai, a senior consultant at Daxue Consulting, believes that most businesses are “ultimately pragmatic.”

“The main criteria in the end would be the AI services that provide clear, sustainable value, practical business fit, and strong local support,” she adds.

Question of sovereignty

If there’s a common concern for companies, however, it would be on policy. No country in Southeast Asia has passed a definitive law on AI.

Most have taken a “soft” regulatory approach. Instead of implementing penalties, Singapore, for instance, has opted to install a framework that suggests ways on how enterprises could adopt AI.

But a number of governments in the region are seeking AI data sovereignty, particularly in how LLMs process and store the information of their citizens.

Such policies pressure AI companies to either establish data centers in the region or seek local partners to compete. This could be more difficult to do for Chinese AI companies compared to their Western counterparts.

In the past decade, Google, Microsoft, and AWS have all established key partnerships with governments in Southeast Asia. Alibaba is making inroads on this front, as it has announced data center investments in Malaysia, Singapore, and the Philippines.

For Hanno Stegmann, managing partner and director of BCG X, Chinese AI companies could make more inroads in Asia if they address enterprises’ data-related concerns “proactively”.

“I think a lot of doors are not opened for these [Chinese AI companies] because they are not answering data-related concerns proactively,” he says. “If they say first thing that this isn’t an issue, and that the data processed by these companies are not sent back to Beijing [that would help],” he says.

DeepSeek and Qwen have not issued statements regarding the security concerns against them, though both have made their data privacy policies public. Tech In Asia