US wants to use AI in Asia, but China has monopoly on cheap models

BEIJING — The artificial intelligence race between the United States and China is intensifying in Asia, the world’s largest continent.

“The U.S. strategy is to prevent China from becoming a major AI supplier to the rest of Asia and, frankly, the entire world,” said Gary Dvorchak, managing director at Blueshirt Group. He noted that while Chinese alternatives are cheaper, the United States currently offers more complete solutions, from chips to AI models.

However, the Asia-Pacific Economic Cooperation Conference’s Digital Week held this month in Chengdu, a city in southwestern China, revealed the US’ sales challenges.

Although U.S. officials and U.S. business representatives in APEC highlighted AI in promoting the Chengdu event earlier this year, the U.S. has left little public trace of its involvement in the event. The US presence has slowed since Anthropic upended the release of its Fable AI model due to a sudden policy change in the US, and recently new AI models from China have been launched with similar features at much cheaper prices.

In contrast, at the first APEC AI Conference in South Korea last summer, President Donald Trump’s chief science and technology policy advisor Michael Crasios emphasized the creation of a U.S. AI Action Plan and a U.S. AI Export Program, according to a White House transcript.

However, Politico cited three former officials earlier this month when it reported that the Commerce Department had received a lower-than-expected 78 applications for the U.S. AI export program so far.

And on July 24, at the APEC High-Level Forum on AI hosted by China’s cybersecurity regulator, Bill Guidera, the U.S. Department of Commerce’s undersecretary for innovation and engagement, remained focused on AI export programs, according to documents reviewed by CNBC.

He called for broader partnerships in the Asia-Pacific region, noting that buyers could acquire all or part of the U.S. technology stack through export programs. “This is a great design that demonstrates the strength, safety, and capability of American AI,” Guidera said.

The Department of International Trade of the Ministry of Commerce confirmed in a social media post on July 29 that Mr. Guidera had made the remarks in Chengdu. Asked about the Politico report, an ITA spokesperson said the number of applications “exceeded our expectations.” The White House did not respond to CNBC’s request for comment.

American business showcases were also limited. As of July 23, CNBC noted that Google and Meta were the only US companies with booths in APEC, while most of the booths in Thailand and China were based in Chengdu. The two US companies focused on their molecular AI system AlphaFold and AI applications for small and medium-sized businesses, respectively, rather than large-scale language models.

Wilson L. White, Google’s vice president of government affairs, only mentioned Gemini in a July 24 speech. tencent Mr. White was followed by Vice President Tsai Guangzhong, who highlighted Hunyuan LLM’s expansion in adoption and cloud projects in Thailand.

China will host APEC this year at a critical time of tensions and technological competition between the United States and China.

The Chinese government is taking further opportunities to highlight its AI capabilities, which are mostly open source versus the mostly closed US model.

At the World AI Conference held in Shanghai on July 17, Chinese President Xi Jinping announced that his country would provide 5,000 AI training and seminar opportunities to developing countries, as well as establish an AI application cooperation center with Southeast Asia and other countries.

Subsequently, the Chinese government dispatched a senior official, Vice Premier Zhang Guoqing, to advocate the development of technology standards with other Asia-Pacific countries at the ministerial-level APEC Digital Week on July 23. Later that day, 21 member states, including the United States, agreed to support open source AI with “strong security.”

“‘Supporting’ an open source model with strong security guarantees gives regional legitimacy to China’s open weight strategy, especially in emerging economies in Asia where implementation costs and technology sovereignty are key considerations,” said Wei Sun, principal analyst for artificial intelligence at Counterpoint Research.

But rather than the world becoming divided into U.S. and Chinese AI domains, Sun expects a combination of technologies, especially in Asia.

There are over 1,300 languages ​​in Southeast Asia alone, so just using an AI model from the US or China is not as easy as it seems.

Governments in Asia and elsewhere are spending “billions of dollars” on AI systems tailored to local languages, according to privately funded startup Votee AI. CEO Pak-Sun Ting said the startup is already generating annual revenue of well over $10 million, working with at least five governments, including two in Southeast Asia.

Ting said companies in Southeast Asia are particularly likely to use Nvidia chips for AI training, but may use other chips to run models. He pointed out that Votee’s open-source model for Cantonese speakers was developed partially using Alibaba’s open-source Qwen model.

AI’s ability to generate economic benefits remains important regardless of its origins.

“The United States and China are fiercely competing in AI technology and diplomacy through different approaches, but ultimately they cannot be completely separated from each other,” said Yue Su, chief economist at the Economist Intelligence Unit.

He noted that the weak US presence in APEC is not too surprising given other events such as the San Francisco AI Summit on July 24.

The event was hosted by South Korea’s Ministry of Technology, and President Lee Jae-myung met with American frontier AI model leaders Sam Altman of OpenAI, Anthropic’s Dario Amodei, and NVIDIA’s Jensen Hwang to develop South Korea’s chip and AI megaprojects.

—CNBC’s Jenny Lee contributed to this report.