Amid intensifying technological warfare, the crisis of advanced processor shortages emerges as the greatest obstacle facing China’s ambitions to catch up with American leadership in generative artificial intelligence.

• Introduction to the leaks from the Chinese company founder
• Details of the massive gap in computing resources and chips
• U.S. escalation and evasion accusations via third-party countries
• Political ramifications and tighter export controls
• The future of the innovation environment and the bet on local alternatives
• Frequently asked questions

Introduction to the leaks from the Chinese company founder

In an unprecedented audio and informational leak revealing the harsh reality and hidden struggles facing China’s technology sector in the global dominance race, Liang Wenfeng, founder of prominent AI company DeepSeek, made critical remarks clearly highlighting the massive computing capacity deficit plaguing his country. These sensitive remarks, recorded during a lengthy closed-door meeting with potential investors lasting nearly four hours, were published via the Tencent Tech news platform in late July. These candid statements provided strong ammunition for officials in the U.S. administration who are constantly pushing for stricter and more resolute enforcement of advanced semiconductor export controls to China, especially as these leaks come at a sensitive time coinciding with new and serious U.S. allegations accusing Chinese tech companies of devising roundabout ways to bypass and circumvent imposed trade restrictions.

Details of the massive gap in computing resources and chips

During his transparent and candid conversation with potential financial backers, Liang made it abundantly clear that computing power and microchip availability—rather than a scarcity of engineering talent or a lack of big data required for training—are the sole decisive variables separating his ambitious company DeepSeek from its fierce counterparts and rivals in the U.S. market. He stated in unambiguous terms: all the differences and gaps we clearly see today, including the quality and availability of specialized talent, the superior operational capacity of large language models, and the diversity of practical applications, can be entirely attributed to the vast disparities in available computing resources. Liang provided precise figures to give vivid examples, noting that training a leading and advanced artificial intelligence model requires operating approximately 200,000 Huawei 950 processing cards, whereas his company, despite its size and impact, currently owns and operates only about 20,000 advanced H-class equivalent graphics processing units. Liang described the company’s current deal to acquire 16,000 Huawei 950 cards as merely a modest step, stressing that its computing power barely equals that of just 4,000 cards from Nvidia’s highly advanced B-series chips, commenting that this number is never a massive quantity sufficient for fierce competition. He concluded his remarks on this axis by saying: the major gap between us and the United States lies in material resources; on one hand, we are unable to purchase enough efficient chips from the domestic market, and on the other hand, the scale of our capital investment is much lower than what is being pumped into American laboratories.

U.S. escalation and evasion accusations via third-party countries

In an interesting coincidence, these revealing statements appeared in the same week the White House moved aggressively to expose what it described as systematic evasion and circumvention of existing export controls. On July 22, Michael Kratsios, director of the White House Office of Science and Technology Policy, leveled public and direct accusations via the X platform against Chinese AI startup Moonshot AI. He accused the company of indirectly acquiring high-performance servers equipped with Nvidia GB300 chips and routing computational workloads and big data through a cloud infrastructure based in Thailand in order to train and develop its advanced language model Kimi K3 away from direct oversight. Kratsios strongly argued that these complex procedures represent a blatant and convoluted violation of both strict U.S. federal export regulations and the explicit terms of service imposed by American AI-developing companies to protect their intellectual property. Furthermore, Business Insider reported that Kratsios also claimed the Chinese company performed unauthorized reverse engineering of the Claude model developed by American firm Anthropic during the phases of building its own Kimi K3 model, opening a new door for technical disputes.

Political ramifications and tighter export controls

These serious allegations prompted the U.S. Department of Commerce’s Bureau of Industry and Security to launch a broad, comprehensive, and urgent federal investigation to examine and scrutinize whether Chinese AI companies have indeed managed to bypass strict restrictions and skirt the technological embargo by routing heavy computational workloads through overseas subsidiaries or by hiring cloud service providers located in neutral third countries not subject to strict oversight. This notable chronological convergence between Liang’s admissions of infrastructure weakness and the Chinese evasion accusations leveled against Moonshot gives hawks and hardliners in Washington—who advocate imposing harsher restrictions on Beijing—a stronger negotiating position and a solid argument. Liang’s clear and direct words indicate that U.S. sanctions and export restrictions are working with high effectiveness and achieving their desired objectives of restricting and stunting the growth and development of AI in China by denying it easy and accessible access to advanced computing capabilities; however, the Moonshot case clearly highlights the regulatory and legal loopholes that still exist and that the Chinese cleverly exploit to access banned technology.

The future of the innovation environment and the bet on local alternatives

Anticipating such loopholes and expanding the strategy of technological strangulation, the U.S. Department of Commerce had already taken proactive legislative steps last May to expand the scope of complex export license requirements to include all foreign-based companies, branches, and entities primarily owned by Chinese parent companies in order to thwart evasion attempts. For his part, despite the gloominess of the current landscape, DeepSeek founder Liang expressed a tone of cautious optimism about the future, noting his high confidence in the ability of the local technological ecosystem led by Chinese giant Huawei to bridge this massive gap and gradually close it within just one year. This ambitious and bold bet by Chinese industry represents a living test and direct practical experiment that will prove in the coming days whether U.S. restrictions are truly capable of continuing to hinder China’s progress, or if local innovation will find its way to overcome these forcibly imposed challenges to break the Western monopoly and establish an independent national infrastructure.

Frequently asked questions

Question: What is the biggest challenge facing AI development in China according to the DeepSeek founder?

Answer: The company founder emphasized that the acute shortage of computing resources and the lack of advanced chips are the sole decisive factors delaying Chinese companies from their American competitors.

Question: How many chips are required to train leading models compared to what the company currently owns?

Answer: Training a leading model requires 200,000 Huawei 950 processing cards, while DeepSeek currently owns only the equivalent of just 20,000 advanced processing units.

Question: Why did the White House recently accuse Chinese company Moonshot AI of violating sanctions?

Answer: The Chinese company was accused of using servers equipped with advanced Nvidia chips via a cloud architecture in Thailand to train its models, which is considered an explicit circumvention of U.S. export restrictions.

Question: What step did Washington take to close the loopholes of evasion through neutral countries?

Answer: The U.S. Department of Commerce expanded export license requirements in May to include all foreign branches of Chinese companies to prevent them from purchasing chips via third parties.