Whither U.S. AI and Chip Policies Toward China Go?

TMTPOST--With U.S. President Joe Biden’s dropout of his re-election drive and former President Donald Trump revving up his campaign to return to the White House, many are wondering how the U.S. administration's AI and chip policies will evolve.

On August 1, the U.S. government is reportedly considering imposing restrictions on China's use of AI large models and AI storage chips, including HBM, by the end of August.

A spokesperson for the U.S. Department of Commerce said that they are assessing the evolving threat environment and updating export controls that are necessary to protect U.S. national security and maintain the technology ecosystem.

Republican presidential nominee Trump believes that chip manufacturing should be moved to the U.S. His allies are drafting a sweeping AI executive order that would launch a series of “Manhattan Projects” to develop military technology and immediately review “unnecessary and burdensome regulations”—signaling how a potential second Trump administration may pursue AI policies favorable to Silicon Valley investors and companies.

After over seven-year-long tech war between China and the U.S., China seems to be in a better position to tackle the adversity. China is building an independent and controllable industrial and supply chain system and narrowing the AI tech gap with the U.S. despite the U.S. restrictions on AI chip exports.

In the AI field, while the U.S. expresses a desire for dialogue with China, it plans to curb the development of China's AI technology, exposing its hypocrisy. The U.S. actions will not stop China's technological progress but will motivate Chinese enterprises towards self-reliance and resilience, China’s foreign ministry spokesman Lin Jian said on July 31.

Industry experts told TMTPost that China is speeding up in building a domestic AI and chip system. Zhang Yi, the founder and chief analyst of iMedia Research, believes that China's construction of computing resource infrastructure will help alleviate concerns about the lack of advanced chips.

China's large market size, public sector demand for AI and its applications, and the pressing need for efficiency improvements in many entities favor the further development of AI technology in China, Zhang added.

"Making AI Work for the American People" can be seen on the Biden administration's AI website, ai.gov. This philosophy, coupled with the escalating U.S.-China tech and trade friction, is driving the global tech field towards "tech decoupling," a term coined by Taiwan-based chipmaking giant TSMC founder Morris Chang.

Since the then-President Trump initiated the trade war with China in 2017, the U.S. government has escalated sanctions and restrictions on Chinese companies.

According to U.S. Secretary of State Antony Blinken, over 1,300 Chinese companies in AI, chips, quantum computing, and other frontier technologies have been added to the "Entity List."

Since the advent of ChatGPT and Nvidia GPUs, AI computing power has become all the more important in the new wave of generative AI. The U.S. has intensified efforts to prevent China from obtaining advanced AI chips from Nvidia and AMD, which has affected the global tech supply chain.

In November 2022, Nvidia released the H200 Tensor Core GPU, the world's most powerful AI chip. The H200 outperforms its predecessor by 60-90% and is 110 times more efficient than CPUs. However, Nvidia has stated that without export licenses, the H200 cannot be sold in China. Even the "China-specific" versions H20 and H800 are now restricted to be sold in China.

U.S. Secretary of Commerce Gina Raimondo emphasized that the U.S. cannot allow China to acquire cutting-edge chips, fearing China's advancement in training state-of-the-art AI models.

Nvidia's revenue from Chinese clients dropped from 19% of data center business in 2023 to mid-single digits in 2024 due to new export controls. Nvidia CEO Jensen Huang expressed hopes to maximize business with new China-specific chips despite these restrictions.

In addition to sanctions and containment, the U.S. employs subsidies and policy guidance in its AI and chip competition with China. The U.S. National AI Research Resource Task Force plans to invest over $2.6 billion in generative AI research over the next six years. The 2024 federal budget proposes over $251.1 billion for AI-related research and services, involving the Department of Defense, Department of Energy, and Department of Homeland Security. Combined with private and external funding, the U.S. AI investment is expected to exceed trillions of dollars.

Policy-wise, on July 31, the U.S. Commerce Department announced three final guidance documents from National Institute of Standards and Technology and AI safety guidelines from the AI Safety Institute. These aim to enhance AI system security, reliability, and trustworthiness. The Commerce Department's documents mentioned companies like Microsoft, OpenAI, and Apple participating in the Biden administration's AI development plan, emphasizing safe and responsible AI innovation.

Commerce Secretary Raimondo likened U.S. semiconductor development initiatives to a moonshot similar to NASA's Apollo program. Regardless of economic conditions, the U.S. aims to maintain a leading position in AI and chips, shaping global tech competition and industry structure over the long term.

Winston Ma, the author of The Digital War: How China’s Tech Power Shapes the Future of AI, Blockchain and Cyberspace, notes that U.S. export restrictions have driven China to improve AI efficiency.

Chris Miller, the author of Chip Wars, believes that the seven-year U.S.-China chip war is reshaping the technological supply chain. The intensified U.S.-China tension over technology — and especially semiconductors — has shifted electronics supply chains in slow but significant ways. Foxconn’s Wisconsin facility is far smaller than initially promised, but TSMC, Taiwan’s most valuable company and the world’s biggest producer of processor chips, will soon open a new facility in Arizona. Previously, almost all of TSMC’s recent investment had been in China. Now it is diversifying its fabrication footprint, building a new chip fab in Japan and exploring one in Singapore, too.

Miller emphasizes that for the U.S., chip manufacturing subsidies have only a short-term effect. In the longer term, American success requires maintaining a technological edge, which means prioritizing innovation and research.

Beijing Daily on Tuesday published an editorial, arguing that while the U.S. can temporarily tighten its grips on China's innovation through "stranglehold" tactics, it cannot keep China out of the high-tech competition circle indefinitely. Such tactics only strengthen China's resolve and ability to become technologically self-reliant, motivating China to break through blockades and "secure its own rice bowl."

The increasing U.S. restrictions on China's AI, chip, and other tech sectors are reshaping China's technological development landscape.

On July 31, Reuters reported that the U.S. plans to add about 120 Chinese entities to its restricted trade list this month, involving companies in wafer manufacturing, EDA (Electronic Design Automation) software, AI technology, and other fields. These entities would need a license from the U.S. Department of Commerce to supply to industry chain companies, but these licenses are likely to be denied.

Regarding U.S.-China chip development, Yin Zhiyao, chairman and general manager of Advanced Micro-Fabrication Equipment Inc. of China, said that while there is still a long way to go for China to catch up with the world's most advanced semiconductor levels, it is expected to achieve self-sufficiency in chip equipment by the third quarter of this year and reach international advanced technology levels within the next five to ten years.

Additionally, AI is at the center of U.S. restrictions. Although the U.S. currently leads in AI technology development, exemplified by ChatGPT, China is rapidly catching up.

Recently, several Chinese companies have launched open-source AI model technologies comparable to those of the U.S.' OpenAI, available to global consumers, businesses, and independent software developers.

In May, Baichuan AI, founded by Wang Xiaochuan, released the new Baichuan 4 series models and the AI assistant "Baixiao Ying." The Baichuan 4 model ranked first domestically in benchmark evaluations.

In June, Zhipu AI launched the new open-source large model GLM-4 series. In July, Kuaishou released the video generation model Kling in China, which is now open to global users, and Zhipu AI's Sora video model has also been open-sourced. In August, TikTok’s parent company ByteDance's Jimeng AI app was launched on Apple and Android app stores, with video and image generation capabilities rivaling OpenAI.

Qu Dongqi, a co-founder of Llama Chinese Community, said that Kuaishou's Kling video model technology is on par with OpenAI's Sora video model.

Just before the release of Kling, the YI model created by Kai-Fu Lee's Lingyi Wanwu ranked nearly equal to leading U.S. technologies in the latest LMSYS blind test competition recognized by OpenAI CEO Sam Altman. The latest trillion-parameter model Yi-Large ranked seventh globally and first among Chinese models, surpassing Llama-3-70B and Claude 3 Sonnet. In the text division, Yi-Large tied for first place worldwide with GPT-4.

"We have overturned the common belief that China lacks the talent or technology to compete with the U.S. in AI," Kai-Fu Lee said. "That perception is completely wrong."

Kai-Fu Lee told TMTPost App, "A year ago, Chinese large models felt far behind (the U.S.). But today, we can proudly say that we have caught up with the top U.S. models released a year ago, though we still need to keep working on that."

Lee also emphasized that the current gap in computing power is a significant issue. "Our computing power has always lagged far behind." However, he believes that better models can be created based on lower computing power costs.

McKinsey & Company Senior Partner Violet Chung believes that generative AI, based on traditional AI and advanced analytics, can further unleash economic benefits of $2 trillion annually in China. Investment experts are poised to make significant investments in the Chinese AI industry over the next decade, with annual amounts reaching trillions of dollars, driving strategic growth in China's economy.

Moonshot AI on Tuesday had just completed a new round of financing exceeding $300 million, with a post-investment valuation of $3.3 billion, attracting other investors including Tencent, Gaorong Capital, and existing shareholder Alibaba.

Previously, Baichuan AI announced the completion of a five billion yuan in round A financing, valuing it at 20 billion yuan. Investors included Beijing Artificial Intelligence Industry Investment Fund, Shanghai Artificial Intelligence Industry Investment Fund, Alibaba, Xiaomi, Tencent, China International Capital Corporation (CICC), and Shenzhen Capital Group.

In 2023, China's AI core industry scale reached 578.4 billion yuan, with a year-on-year growth rate of 13.9%, according to the Blue Book on AI and Rule of Law in the World (2024).

In terms of industry investment, there were 815 AI investment and financing events in China last year, an 18.2% decrease from 2022; however, the total financing amount reached 263.1 billion yuan, an increase of about 51% from 173.1 billion yuan in 2022, indicating a decrease in the number of investment events but an increase in total financing amount. 可以作爲abstract

As of the first quarter of 2024, there were nearly 30,000 AI enterprises worldwide, with 4,500 from China. Currently, there are 1,328 AI large models globally, with the U.S. accounting for 44% and China for 36%, according to the China Academy of Information and Communications Technology.

As the U.S. government continues to utilize export controls, tariffs, and long-arm jurisdiction to curb China's development in advanced manufacturing and high-tech fields, China is accelerating the development of a self-sufficient technological supply chain.

Unlike the U.S., Chinese companies are accelerating innovation-driven development and improving their autonomous innovation capabilities while actively sharing technological advancements with other countries for mutually beneficial development.

With China's AI computing power, foundation model, and application market sizes rapidly expanding, the AI market is expected to reach 5.2 trillion yuan and the industrial AI market to hit 9.4 trillion yuan by 2030, according to Peng Wensheng, the chief economist of CICC.