In the context of accelerating global digital transformation, artificial intelligence (AI) is becoming the core engine to promote a new round of industrial revolution. As an infrastructure that supports the development of AI technology, the importance of AI chips is becoming increasingly prominent. China, with its huge digital economic system, policy support and maturing local technology ecosystem, is rapidly growing into one of the most promising markets for AI chips in the world. Nvidia founder and CEO Jensen Huang's recent speech in an interview with CNBC pushed this topic to the forefront. He warned that U.S. companies would face significant economic and strategic losses if they were absent from the Chinese market. China Exportsemi will systematically analyze the deep logic behind Huang's remarks from the perspectives of industry, policy, technology and globalization.
Ⅰ The value of China's AI chip market should not be underestimated
In the interview, Huang noted that China's AI chip market is expected to reach $50 billion in the next few years. This prediction did not come out of nowhere. According to a number of reports by market research institutions IDC and TrendForce, China's AI accelerator market (including GPU, ASIC, FPGA, etc.) has a compound annual growth rate of more than 20%, and the overall AI chip market is expected to reach nearly US$50 billion by 2027, covering data centers, edge computing, smart terminals and other fields.
Nvidia's fiscal year 2023 (ending January 2024) annual report shows that about one-fifth of its revenue comes from the Chinese market. Although the share has declined due to the U.S. Department of Commerce's export controls, China remains one of Nvidia's most important overseas markets. However, flagship AI chips such as H100 and A100 were included in the export restriction list because they involved the "threshold of supercomputing power", and Nvidia had to launch the H800 and A800 with limited performance to deal with it. At the end of 2023, the company made a provision for $5.5 billion in related inventory impairment losses due to the difficulty of H20 chips to meet the latest regulatory standards. This figure accounts for nearly 40% of its single-quarter profit, which directly reflects the substantial impact of the policy on the business.
Picture: Nvidia Jensen Huang: American companies will lose a lot in the absence of China's AI chip market
Ⅱ The linkage effect of employment and taxation under the influence of the industrial chain
Huang further emphasized that access to the Chinese market is not only about revenue, but also about jobs and tax revenue in the United States. He pointed out that the AI chip business has a strong backward linkage effect on the U.S. domestic economy, that is, the expansion of overseas markets will directly drive the growth of domestic high-paying jobs, especially in chip design, EDA tool development, packaging and testing, etc.
Nvidia, for example, spends more than 25% of its annual revenue on R&D and employs a large number of chip engineers and AI researchers in Silicon Valley, Austin, Portland, and other places. If the Chinese market is restricted, the company's overall revenue will decrease, which will inevitably affect the R&D budget and recruitment plan, which in turn will affect the talent ecosystem and technology accumulation. In terms of taxation, the foreign profits of technology companies are taxed under the OECD/G20 Inclusive Framework, and the US government is promoting a 15% minimum corporate tax to re-tax the global earnings of multinational companies. Reduced profits in the Chinese market also mean a shrinking federal tax base.
Ⅲ The delicate balance between national security and global leadership
Huang's remarks also touched on a more sensitive question — whether national security and technology blockades can really help the United States maintain its leadership in science and technology. "If we can't participate in the Chinese market, we will lose our global technology leadership," he noted.
From the perspective of the collaborative logic of the global industrial chain, AI chips are not only end products, but also part of the infrastructure and platform ecology. Once U.S. companies are excluded from technology, market, and ecosystem construction, it will damage their dominance over next-generation computing paradigms (such as AI+5G, intelligent edge, and common foundation models) in the long run.
In addition, restricting exports to China will accelerate the advancement of China's "independent and controllable" strategy. For example, Huawei uses Ascend chips and the MindSpore framework to build a closed-loop AI ecosystem. Cambrian, Suiyuan, Biqian and other companies are rapidly iterating in the field of training and inference chips. In the "bottleneck" links such as EDA and packaging and testing, China has also gradually replaced it through mergers and acquisitions and independent research and development. Once the local technology chain matures, U.S. companies may have lost their strategic entry point even if the policy is relaxed in the future.
Ⅳ The rise of China's AI chip industry: from policy-driven to technological breakthroughs
The rise of China's AI chips is not accidental, but the result of the resonance of multiple factors such as policy, market, talent, and capital.
1. Policy dividends: China's 14th Five-Year Plan clearly lists AI and integrated circuits as strategic emerging industries. The Ministry of Industry and Information Technology, the Ministry of Science and Technology, the National Development and Reform Commission and other departments have issued special funds, industry guidance funds and local government supporting incentives. For example, Beijing's "Capital of Computing Power" plans to support the construction of large-scale AI infrastructure and promote the widespread deployment of AI chips in security, transportation, industrial Internet and other fields.
2. Market traction: China has the world's most active digital user group and the ability to implement scenarios. The wide application of AI in e-commerce, content recommendation, social media, smart cities and other fields provides a real and complex testing ground for local chip companies. For example, ByteDance, Meituan, and Alibaba have extremely high demand for AI computing power when training models, promoting local chip performance optimization and software and hardware collaboration.
3. Enterprise breakthroughs: The Ascend 910B is benchmarked against the A100, the Biqian BR100 has greatly improved AI inference performance, and the Suiyuan second-generation training chip has been put into trial operation in multiple AI cloud vendors. Although there is still a gap with NVIDIA and AMD at the architecture level, it has the ability to replace it in some specific vertical scenarios.
However, the challenges are just as daunting. At present, domestic AI chips still need to be improved in terms of compiler optimization, ecological compatibility, and power consumption control, and enterprises need to continue to invest in building a perfect ecosystem.
Ⅴ NVIDIA's China Strategy: Balancing Regulatory and Business Logic
In the face of the continuous tightening of export controls by the United States, NVIDIA has adopted a strategy of "down-regulation design + local customization". According to a report by Bloomberg in April 2024, Nvidia is informing Chinese customers of the "Blackwell architecture downgrade" chip plan, trying to retain the basic market of the Chinese business without violating BIS (United States Bureau of Industry and Security) regulations.
These "custom chips" will weaken some of the high-performance features to circumvent the threshold of limitations. For example, by reducing bandwidth, FP64 precision computing power, or cluster performance, the chip meets the export license criteria. However, this downgrade also reduces the attractiveness of the product to high-end customers, and it faces more fierce domestic substitution in key markets such as cloud computing and large-scale model training.
In addition, NVIDIA is also working more with neutral countries (such as Singapore and Malaysia) to try to bypass geopolitical risks and establish new technology and supply chain nodes. Behind this "multipolar" strategy is a long-term response to the long-term decoupling of science and technology between China and the United States.
Ⅵ Future direction: the integration and game of the global AI chip industry
Looking forward to the future, the global AI chip industry pattern will be in the dual tension of integration and game:
- Technology integration: The AI chip ecosystem is highly dependent on cross-border cooperation, from EDA tools to packaging and testing, to cloud platforms and training frameworks, each link cannot be completed by a single country independently. Whether it is the CUDA ecosystem in the United States or the optimization of edge computing scenarios in China, they are inseparable from learning and confrontation with each other.
- In terms of policy games: How China and the United States find synergy in competition will determine the long-term healthy development of the AI chip industry. If export control policies continue to be tightened, the global AI hardware system may be fragmented, which will ultimately weaken the development efficiency of global AI technology.
Conclusion: Revaluing the value of "participation".
Huang's statement is not only a practical consideration for business leaders, but also a reflection on the logic of global technology development. Under the general trend of technological globalization, short-term behaviors such as excessive decoupling and restricting cooperation may lead to the loss of long-term competitiveness. For U.S. companies, China is not only a market, but also a driving force for their own iteration. For the global AI industry, openness, mutual trust, and cooperation remain the only way to the future.
