DeepSeek Shockwave: Huawei, Export Controls, and the Future of the China-US AI Race

Industry Context: Why DeepSeek Caused Global Shock

On January 20, 2025, DeepSeek released its R1 model, coinciding with President Trump's inauguration. This timing seemed to herald a new phase in US-China AI competition. Just one week later, DeepSeek's iOS app surpassed ChatGPT in downloads on the U.S. App Store, becoming the top free app. Nvidia's stock plummeted in a single day, with market value evaporating over $600 billion—though it later recovered somewhat, the market had realized: Chinese AI research labs are no longer chasers, but participants in frontier innovation.

DeepSeek did not appear out of nowhere. Its parent company, High-Flyer Capital Management, is one of China's top quantitative hedge funds. As early as 2022, it had accumulated more than 10,000 Nvidia A100 chips and invested 1.2 billion RMB to build data centers. This extreme pursuit of computing infrastructure stems from high-frequency trading's sensitivity to millisecond-level latency. DeepSeek founder Liang Wenfeng once revealed in an interview that High-Flyer started with a single GPU in 2015, owned 1,000 by 2019, and by the time DeepSeek was founded, it already had ample computing reserves.

More importantly, DeepSeek is not simply copying American technology. MIT Technology Review noted that DeepSeek has indeed achieved a "real technological breakthrough." Although some innovations already had precedents at American AI companies, its optimizations in model efficiency, inference costs, and other areas have forced the global AI industry to re-evaluate the innovative capabilities of Chinese companies.

Market Impact: From Nvidia's Stock Price to Global AI Investment Logic

DeepSeek's shockwave first appeared in the capital markets. Nvidia lost over $600 billion in market value in a single day, setting a historic record. Although the stock price later rebounded, investors' "unbounded optimism" about AI computing demand began to be rationally corrected. DeepSeek proved that more efficient algorithms can achieve near-frontier performance under constrained computing power, directly challenging the assumption that "computing power is a moat."

For tech giants, DeepSeek's low-cost training methods may change the cost structure of AI commercialization. If companies can achieve the same capabilities with fewer GPUs, the growth curve of cloud providers' computing power demand will flatten, while competition at the AI application layer will intensify.

For Chinese tech companies, DeepSeek's rise has boosted investor confidence. According to the Financial Times, foreign investors' interest in Chinese tech stocks has increased significantly. DeepSeek itself is also beginning to consider accepting external venture capital for the first time, marking its transformation from a "research laboratory" into a "commercial entity."

Competitive Landscape: The Huawei and SMIC Challenge CSIS report specifically notes that DeepSeek's success partly reflects loopholes in U.S. export controls, but the more critical challenge comes from the combined efforts of Huawei and SMIC. Huawei founder Ren Zhengfei said at a symposium with Xi Jinping on February 17, 2025, that earlier concerns about insufficient domestic advanced semiconductor capacity had been "eased," and revealed that Huawei is leading more than 2,000 Chinese companies in an effort to achieve self-reliance and control over more than 70% of the semiconductor supply chain by 2028.

If this goal is achieved, it will fundamentally weaken the effectiveness of U.S. export controls. Currently, the combination of Nvidia and TSMC remains the golden standard for global AI chips, but Huawei's Ascend chips and SMIC's advanced process nodes are gradually narrowing the gap. The CSIS report argues that export controls can only "slow down" and "disrupt" China's development, not completely stop it.

DeepSeek's rise has also reshaped China's AI competitive landscape. Previously, China's AI research was dominated by large companies such as Baidu and Alibaba, but DeepSeek, as a subsidiary of High-Flyer, quickly rose to the forefront of China's frontier AI labs thanks to the capital and technical accumulation from quantitative trading. This sends a signal to the market: AI innovation is no longer limited to tech giants; institutions with deep understanding of computing power and engineering capabilities can also break through.

Enterprise Implications: Recalibrating Enterprise AI Strategy

For global enterprise decision-makers, the lessons from DeepSeek are at least fourfold:

1. Computing efficiency first: DeepSeek's MLA (Multi-head Latent Attention) and MoE (Mixture of Experts) architecture demonstrates a viable path to reducing inference costs. When designing AI applications, enterprises should focus more on model efficiency rather than simply accumulating GPUs.

2. Rebalancing open source and closed source: DeepSeek chose to open-source its model weights, intensifying competition between open-source and closed-source models. When evaluating AI vendors, enterprises need to weigh the customization freedom of open-source models against the balance of commercial support.

3. Diversifying supply chain risk: Against the backdrop of U.S.-China tech decoupling, the risk of relying on a single chip supplier has risen significantly. Enterprises should assess the feasibility of alternative computing solutions (such as Huawei Ascend, AMD, and self-developed chips).

4. Front-loading regulatory compliance: The United States may further tighten export controls, while China is pushing for self-reliance and controllability. Multinational enterprises need to simultaneously meet the data and chip compliance requirements of both countries, and this will become the norm in AI strategy design.12 months: DeepSeek will continue to release new models, and its commercialization path (such as API services, enterprise partnerships) will gradually become clear. US export controls may escalate, focusing on cracking down on chip smuggling channels. The ecosystem maturity of Huawei's Ascend chips will become a key observation point for China's AI supply chain.

24 months: The performance gap between Chinese and US AI models may narrow to the "same level." If Huawei + SMIC achieve stable mass production of 7nm-class chips, China's AI computing power bottleneck will be greatly alleviated, and the global AI chip market will form a "dual-supplier" landscape. Enterprise AI applications will shift from "experimentation" to "large-scale implementation," with ROI becoming the primary metric.

3 years: The AI industry may split into two independent ecosystems—one centered on the US, relying on NVIDIA + TSMC; the other centered on China, revolving around Huawei + SMIC + DeepSeek. Global companies will have to choose between the two ecosystems, or maintain both systems simultaneously, which implies higher technology costs and strategic complexity.

The CSIS report believes that if the US can effectively prevent large-scale chip smuggling and curb Huawei + SMIC's alternative solutions, it can still maintain its lead in the AI race. But success is not guaranteed, and China's self-sufficiency process is progressing much faster than expected. For industry observers, DeepSeek is just the beginning; the real competition will unfold at the infrastructure level.