Trump's Delayed Xi Meeting May Blunt U.S. AI Chip Leverage

President Donald Trump has officially postponed his visit to China, citing the war with Iran, drawing attention to America's complicated calculations regarding China's technology and capital power. The Trump administration has taken the position of curbing China's artificial intelligence (AI) ambitions, led by export controls on cutting-edge chips such as Nvidia's graphics processing units (GPUs). But Wall Street and Washington insiders do not see that goal as easily achievable. Experts say China holds a clear lead over the United States in nearly every manufacturing sector, raising the likelihood that Beijing could seize dominance in physical AI — technology that merges AI with the physical world — through efficient data training. Concerns are also mounting that while China accelerates self-sufficiency in AI, the U.S. has too readily ceded critical supply chains including rare earth minerals.

Trump to Visit Beijing May 14–15; 'China Open-Source Ecosystem' Threat Debate Heats Up in U.S.

White House Press Secretary Karoline Leavitt said at a briefing on Friday that "we are pleased to announce that the long-awaited meeting between President Trump and President Xi will take place May 14–15." Leavitt added that "President Trump and First Lady Melania Trump will host President Xi and Madam Peng Liyuan for a return visit to Washington, D.C. at a date to be announced later this year."

Trump's China visit was originally scheduled for March 31 through April 2. But after Trump launched airstrikes against Iran on March 28 and the war appeared likely to drag on longer than expected, the trip was pushed back by roughly six weeks. Trump had said on April 16 that he "asked for a postponement of about a month," then three days later on April 19 said the visit had been "delayed by about a month and a half."

Diplomatic circles and Wall Street expect the U.S. and China to extend a trade war truce by up to one year at the summit. Trump and Xi held their first bilateral summit in six years in Busan on October 30 last year on the sidelines of the Asia-Pacific Economic Cooperation (APEC) leaders' meeting in Gyeongju, North Gyeongsang Province. Extending the truce agreed upon at that time by one year would buy Trump political time through the November 3 U.S. midterm elections.

Wall Street and diplomatic observers forecast Trump will ask China to expand purchases of American soybeans and buy U.S. oil and gas. He is also expected to use tariff reductions as leverage to demand improvements in China's unfair trade practices, further market opening, and the lifting of additional export controls on critical strategic minerals including rare earths. Although the Supreme Court has effectively neutralized reciprocal tariffs and fentanyl tariffs, observers say Trump will use tariffs authorized under Section 301 of the Trade Act, for which investigations have recently begun, as a bargaining weapon. Other negotiating tools Trump may deploy include cutting off energy to anti-American nations and restricting exports of American AI chips.

As advanced AI chip export restrictions are regarded as America's most powerful trade weapon, the U.S.-China Economic and Security Review Commission (USCC), a congressional advisory body, issued a report on Wednesday diagnosing that "China is building market dominance through open-source AI even as the U.S. restricts access to semiconductors." The report found that as U.S. export controls made it harder for China to obtain high-performance AI chips, Beijing has leveraged the open-source ecosystem to independently develop technology that runs efficiently despite lower computing power. The report also noted that China is narrowing the technology gap using massive volumes of real-world data generated in industrial settings, including manufacturing processes and robotic movements. "China is generating data across diverse fields to improve AI models in order to advance its manufacturing base, logistics networks, and robotics," the report said. "Chinese laboratories have narrowed the performance gap with leading Western large language models (LLMs) through an open ecosystem."

The USCC further warned that China's innovation through open source could undermine the position of U.S. closed-model AI companies — OpenAI, maker of ChatGPT, and Anthropic, maker of Claude — which have invested billions of dollars. "The proliferation of open models creates an alternative pathway to AI leadership," the report said. "As the AI frontier shifts from LLMs to AI agents and physical AI, China — which has collected data at massive scale — could be better positioned in humanoid robots, autonomous driving software and other areas."

80% of U.S. Startups Use Chinese Models; China's GPU Self-Sufficiency to Hit 76% by 2030

Reuters estimated on Wednesday that approximately 80% of American AI startups currently use Chinese open-source models. DeepSeek's "R1," released last year, became the most downloaded AI model on the U.S. app store. Alibaba's "Qwen" has also surpassed Meta's "Llama" in cumulative downloads. USCC Vice Chairman Michael Quigken said "there is a widening cumulative gap between the U.S. and China in physical AI," adding that "we are watching how China deploys AI in biotechnology, quantum computing and advanced materials." Siemens CEO Roland Busch also praised Chinese open-source AI, citing cost advantages and convenient user customization, saying "there are no downsides to using Chinese open-source AI to train models specialized in industrial automation for German companies."

According to Chinese-language media outlets including Gongshang Shibao and Kuaike Technology, U.S. investment bank Morgan Stanley also estimated in a recent report that China's AI GPU self-sufficiency rate could more than double from 33% in 2024 to reach 76% by 2030. Chinese companies leading technology self-reliance against intensified U.S. export controls on advanced AI products include Huawei, Alibaba, Cambricon and Moore Threads. The report projected that China's AI chip market could grow from $6 billion (approximately 9 trillion won) in 2024 to $51 billion (approximately 76.1 trillion won) by 2030, representing a compound annual growth rate of 42%. It also forecast that China's monthly production capacity for sub-12-nanometer process products would increase from 8,000 wafers (based on 12-inch wafers) last year to 20,000 next year, 42,000 in 2028, and 50,000 in 2030. Related yields are expected to reach 50% by around 2030.

The report projected that capital expenditure in China's cloud computing industry would reach $130 billion (approximately 194.1 trillion won) by 2030, with roughly 51% going to AI GPU-related equipment. This would drive the AI GPU market to grow at an annual average of 23% from 2024 to 2030, reaching approximately $67 billion (approximately 100 trillion won).

The report further noted that China is gradually reducing its dependence on Taiwan's TSMC, the world's largest foundry, and increasing use of products from domestic chipmaker SMIC. While China remains reliant on foreign suppliers for high-bandwidth memory (HBM) and deep ultraviolet (DUV) lithography equipment, the report assessed that progress has been made in certain core technologies.

China's Sovereign Fund Paused Wall Street Talks Over Iran War; Investment Expected to Resume Amid Private Credit Crisis

Wall Street, where cash demands have risen to meet redemptions amid concerns over private credit defaults, is paying attention to another dimension of Chinese power: capital. Expectations are growing that an influx of Chinese "big money" into Wall Street, catalyzed by the U.S.-China summit, could help avert a crisis.

Bloomberg reported on Thursday that China Investment Corporation (CIC), which manages $1.57 trillion (approximately 2,350 trillion won) in assets, had discussed investment plans with Blackstone, the world's largest private equity firm, and TPG, a global alternative asset manager, in recent weeks. The report noted that CIC had pulled significant capital from the U.S. private equity market after Trump's return to power last year.