U.S. wants Asia to use its AI, but China dominates cheaper models

BEIJING — The artificial intelligence race between the U.S. and China is heating up in the world’s largest continent: Asia.

“The American strategy is to stop China from becoming the leading AI supplier for the rest of Asia … and frankly the whole world,” said Gary Dvorchak, managing director at The Blueshirt Group. While China’s alternatives are cheaper, he pointed out the U.S. currently offers a more complete solution from chips to AI models.

But the U.S. sales challenge was apparent at the Asia-Pacific Economic Cooperation “Digital Weeks” in the southwestern Chinese city of Chengdu this month.

The U.S. left few public traces of its involvement in the event, despite a U.S. official and a U.S. business representative to APEC both highlighting AI in promoting the Chengdu event earlier this year. The subdued U.S. presence comes after Anthropic flip-flopped on its Fable AI model release due to abrupt U.S. policy changes, and new Chinese AI models have recently launched similar capabilities for far less.

In contrast, last summer at the first APEC AI meeting in South Korea, Michael Kratsios, President Donald Trump’s chief science and technology policy advisor, highlighted the U.S. AI Action Plan and the establishment of the American AI Exports Program, according to a White House transcript of his remarks.

However, earlier this month Politico cited three former officials in reporting the Commerce Department has so far received a less-than-expected 78 applications for the American AI Exports Program.

And on July 24 at an APEC High-level Forum on AI organized by China’s cybersecurity regulator, Bill Guidera, deputy under secretary for innovation and engagement at the U.S. Department of Commerce, still focused on the AI exports program, according to materials reviewed by CNBC.

He called broadly for Asia-Pacific partnerships, noting buyers can acquire a full U.S. tech stack or just portions through the exports program. “It is the brilliant design that shows the strength, security and capability of U.S. AI,” Guidera said.

The Commerce Department’s International Trade Administration confirmed in a July 29 social media post that Guidera spoke in Chengdu. When asked about the Politico report, an ITA spokesperson said the volume of applications “exceeded our expectations.” The White House did not respond to a CNBC request for comment.

American business showcases were also limited. Google and Meta were the only U.S. companies that CNBC noted had booths at APEC as of July 23, among booths for Thailand and China, which were mostly Chengdu-based companies. The two U.S. companies respectively emphasized molecular AI system AlphaFold and AI applications for small businesses, rather than large language models.

Google’s government affairs vice president, Wilson L. White, only made passing references to Gemini in a speech on July 24, while Tencent Vice President Cai Guangzhong took the stage after White to emphasize growing adoption of its Hunyuan LLM and a cloud project in Thailand.

China is hosting APEC this year, which comes at a critical moment of U.S.-China tensions and tech rivalry.

Beijing has doubled down on the opportunity to emphasize its AI capabilities, which are mostly open-source versus largely closed U.S. models.

Chinese President Xi Jinping announced at the World AI Conference in Shanghai on July 17 that China would provide developing countries with 5,000 opportunities in AI training and seminars, while developing AI application cooperation centers with Southeast Asia and other regions.

Beijing then sent a high-ranking official, Vice Premier Zhang Guoqing, to advocate for developing tech standards with other Asia Pacific nations at the minister-level APEC Digital Weeks on July 23. Later that day, the 21 member economies, including the U.S., agreed to back open-source AI with “strong security.”

“The ‘endorsement’ of open-source models with strong security assurance gives China’s open-weight strategy greater regional legitimacy, especially across emerging Asian economies where deployment cost and technological sovereignty are major considerations,” said Wei Sun, principal analyst, artificial intelligence, Counterpoint Research.

With more than 1,300 living languages in Southeast Asia alone, just using a U.S. or Chinese AI model isn’t as straightforward as it looks.

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

Ting said entities in Southeast Asia tend to use Nvidia chips, especially for AI training, but may use other chips for running models. He noted Votee’s open-source model for Cantonese speakers was developed partly using Alibaba’s open-source Qwen model.

AI’s ability to generate economic returns remains critical regardless of origin.

“While the U.S. and China are fiercely competing in AI technology and diplomacy through distinct approaches, they ultimately cannot fully decouple from one another,” said Yue Su, principal economist at the Economist Intelligence Unit.

She pointed out the light U.S. presence at APEC wasn’t that surprising given other events, such as a San Francisco AI Summit on July 24.

South Korea’s tech ministry organized the event, where President Lee Jae-myung sought to build on Korean chip and AI megaprojects by meeting with U.S. frontier AI model leaders Sam Altman of OpenAI, Dario Amodei of Anthropic and Jensen Huang of Nvidia.

—CNBC’s Jenny Lee contributed to this report.