The competition in artificial intelligence between the U.S. and China is escalating across Asia, highlighting the complex dynamics of the tech landscape. According to Gary Dvorchak, managing director at The Blueshirt Group, the U.S. aims to prevent China from establishing itself as the premier AI provider in the region and beyond. Although China's AI alternatives are often more affordable, Dvorchak notes that the U.S. still delivers a more comprehensive product range, from processors to AI models.
At the recent APEC Digital Weeks hosted in Chengdu, China, on July 23, 2026, the diminished U.S. involvement was striking. Despite earlier commitments from U.S. officials and business representatives regarding AI's significance in promoting the event, the U.S. left a minimal imprint. This reduced presence follows Anthropic’s last-minute decision to delay the release of its Fable AI model due to unexpected U.S. policy shifts, while several new Chinese models have emerged, offering comparable functionalities at a lower cost. In contrast, last summer at the inaugural APEC AI meeting in South Korea, top advisor Michael Kratsios outlined the U.S. AI Action Plan and the introduction of the American AI Exports Program, as recorded by the White House.
Despite recent developments, Politico reported earlier this month that the Commerce Department has received only 78 applications for the American AI Exports Program—far fewer than anticipated. During the APEC High-level Forum on AI on July 24, Bill Guidera from the U.S. Department of Commerce reiterated the importance of the exports program, advocating for stronger partnerships in the Asia-Pacific region. He emphasized that buyers could opt for complete U.S. tech solutions or select components, stating, "It is the brilliant design that shows the strength, security, and capability of U.S. AI." The International Trade Administration confirmed Guidera's participation at Chengdu on July 29. An ITA representative remarked that the number of applications had surpassed expectations, though the White House did not provide comments on this matter. The American corporate representation was limited, with only Google and Meta maintaining booths at the event, overshadowed by numerous local companies from Thailand and China. Both U.S. firms focused on specific applications, such as Google’s AlphaFold for molecular AI and Meta's tools for small businesses, rather than large language models. Notably, while Google's VP Wilson L. White briefly mentioned Gemini, Tencent's VP Cai Guangzhong highlighted the increasing adoption of its Hunyuan LLM and cloud initiatives in Thailand shortly after.
China's agency in AI diplomacy is particularly salient this year, coinciding with heightened U.S.-China tensions. President Xi Jinping recently announced at the World AI Conference in Shanghai that China would offer 5,000 training opportunities in AI for developing nations, along with plans to establish AI cooperation centers in Southeast Asia and elsewhere. During APEC Digital Weeks, Vice Premier Zhang Guoqing advocated for collaborative technology standards within the region. Furthermore, the 21 member economies at this event agreed to support open-source AI, which, according to analysts, bolsters China's strategy by enhancing regional credibility especially among developing Asian economies that prioritize affordability and technological independence.
Looking beyond a binary U.S.-China competition, analysts like Wei Sun from Counterpoint Research suggest a more integrated approach to AI technology in Asia. With over 1,300 languages spoken across Southeast Asia, relying solely on either American or Chinese models presents challenges. Government investments in localized AI solutions have already reached billions, as noted by the CEO of Votee AI, Pak-Sun Ting. His company is actively engaged with several Southeast Asian governments, generating over $10 million in annual revenue. Ting highlighted that while Nvidia chips are prevalent for AI training, varied chips are employed for running models, with Votee’s open-source Cantonese model being built partly on Alibaba's Qwen framework.
