The United States has spent years tightening export controls on advanced semiconductors to slow China's artificial intelligence progress, but a structural flaw in this approach may be achieving the opposite effect. Rather than disrupting China's drive to build AI capabilities, export restrictions are actually reinforcing the government incentive systems that motivate Chinese firms to pursue advanced technology development in the first place.

How Are Export Controls Backfiring on U.S. Strategy?

The core problem lies in what policy experts call a mismatch between the tools Washington uses and the systems it's trying to disrupt. Export controls work on the supply side, restricting what technology can flow into China. But they cannot reach what analysts call the "demand-side incentive architecture" that determines what Chinese firms actually build and deploy.

While U.S. officials debated how to impose costs through sanctions and entity list designations, Chinese municipalities were actively distributing rewards. Beijing's 2026 compute voucher program was in active enrollment, with local governments rolling out subsidies to attract AI developers. These aren't new tactics; China has used market-access gates, differential subsidies, and exit barriers in industries from solar panels to electric vehicles to semiconductors since the early 2000s.

What's different in AI is how these mechanisms interact with U.S. technology denial. The resulting alignment doesn't just govern product specifications; it shapes the content standards, decision logic, and information architecture of AI systems deployed at scale across China.

What Is China's Registration Gate and Why Does It Matter?

The most consequential piece of Chinese AI regulatory architecture that English-language export control analysis has largely overlooked is the generative AI registration system, known as "da moxing bei'an." Under China's 2023 Interim Measures for the Management of Generative AI Services, any firm offering generative AI services to the Chinese public must complete formal registration with the Cyberspace Administration of China.

This isn't voluntary. It's a regulatory prerequisite for market access. By the end of 2025, 748 generative AI services had completed formal registration, and 435 generative AI applications or features drawing on registered models had completed a separate filing process.

The registration requirements are extensive. Applications require security self-assessment reports exceeding 100 pages that cover training data provenance, content safety benchmarks, and emergency response protocols. Models that fail or skip registration face removal from app stores, fines, or operational suspension under China's revised Cybersecurity Law, which took effect January 1 and incorporated AI governance provisions into statutory law for the first time.

Registration functions as a survival condition, not a privilege tier. Unregistered models cannot legally operate in the Chinese market. This means every firm participating in compute voucher programs, every firm competing for government procurement, and every firm building applications on domestic AI infrastructure has already passed through the registration gate and demonstrated compliance with state content and security standards.

How Do Subsidy Programs Create Dependency on Domestic Technology?

Inside China's registration gate, a second mechanism shapes firm behavior: differential subsidies that make compute infrastructure choices financially predetermined. The subsidy programs are extensive and growing rapidly across the country.

• Shenzhen's Program: Distributed its first batch of "training power vouchers" in March 2025, totaling nearly 200 million yuan across approximately 40 firms, with individual awards reaching up to 10 million yuan, and committed 4.5 billion yuan in combined funding for digital economy and technology initiatives with AI as a designated priority
• Beijing's Yizhuang Zone: Issues 100 million yuan in annual compute vouchers, reimbursing 30 percent of firms' compute service costs at an individual cap of 20 million yuan, with a crucial difference: 40 percent reimbursement when firms use domestic AI chips versus only 30 percent for nondomestic alternatives
• Hangzhou's Initiative: Allocates 250 million yuan per year with the same logic: 30 percent reimbursement for firms using domestic compute infrastructure and 20 percent for others

The differential subsidies don't merely lower costs; they create dependency structures. A firm that trains on subsidized domestic compute becomes operationally dependent on continued access to those subsidies. Over a year of model training, a 10-percentage-point subsidy premium on domestic compute translates into millions of yuan. The price signal accomplishes what no directive needs to.

In January 2026, eight central government ministries led by the Ministry of Industry and Information Technology issued a joint guideline on "AI Plus Manufacturing," which became one of the first national-level policy documents to reference "compute vouchers" as a policy instrument. The subsidy field is being institutionalized from municipal budgets to national policy architecture.

What Real-World Evidence Shows Export Control Evasion?

The theoretical concerns about export control effectiveness have concrete manifestations. A Bloomberg investigation revealed that executives tied to server manufacturer Supermicro allegedly used a Thailand-based government-related entity to route restricted Nvidia AI GPUs to China, including to tech giant Alibaba.

The intermediary, described in court documents as "Company-1," was allegedly Obon, a Bangkok-based company connected to Thailand's sovereign AI initiatives. Some of the systems sold through Obon may have gone to Alibaba, though the Chinese technology giant denied any involvement and stated it had never deployed prohibited Nvidia hardware in its data centers.

The servers reportedly included Nvidia H200-based systems, which fall under U.S. export restrictions intended to prevent China from obtaining top-tier AI compute hardware without government approval. The indictment accused Supermicro co-founder Yih-Shyan "Wally" Liaw, Supermicro Taiwan sales manager Ruei-Tsang "Steven" Chang, and third-party broker Ting-Wei "Willy" Sun of organizing a large-scale diversion network that rerouted Nvidia Hopper-based AI servers.

Prosecutors alleged that the group used a Southeast Asian front company, falsified documentation, and maintained inventories of dummy servers to conceal actual shipments. The alleged perpetrators even used hair dryers to transfer serial-number labels from legitimate servers onto empty chassis to deceive inspectors and install the restricted hardware afterward. U.S. authorities estimate that the operation generated roughly 2.5 billion dollars in sales beginning in 2024.

How Is China Responding to Export Controls With Domestic R&D?

Rather than being crippled by U.S. export controls, China's chipmakers have turned restrictions into motivation for accelerated research and development. China's leading chipmakers are funneling unprecedented sums into R&D as Beijing accelerates efforts to reduce reliance on foreign technology.

Moore Threads, a Beijing-based chip company, spent half of its entire revenue on research and development in the first quarter of 2026. Shanghai-based MetaX spent 45 percent over the same period. To put that in context, consider a restaurant that takes in 1,000 dollars on a given day and immediately puts 500 of it back into developing new recipes and kitchen equipment. It represents an extraordinary level of reinvestment and signals how urgently these companies feel the need to catch up.

Compare that to American counterparts. U.S. chipmakers such as AMD and Intel have typically spent between 20 and 30 percent of their revenue on research and development in recent years. China's newer chip companies are operating at roughly double that rate, a sign of focused strategic intent rather than desperation.

At the national level, the commitment is equally clear. China's total domestic research and development spending reached the equivalent of roughly 569 billion dollars in 2025, with R&D intensity rising to 2.8 percent of GDP. Adjusted for the lower cost of conducting research in China, that figure stretches even further. From 2019 to 2023, China's R&D investment grew at an annual rate of 8.9 percent, compared with just 4.7 percent in the United States.

The results are already visible in market share. China-based firms now account for 33 percent of global production capacity for foundational chips, up from just 19 percent in 2015. These are the chips that power cars, appliances, factories, and infrastructure across the modern world. Analysts project that Chinese chipmakers could account for nearly half of all new capacity in this segment over the next few years.

What Does This Mean for U.S. Export Control Strategy?

The evidence suggests a fundamental strategic problem with relying primarily on export controls. The policy tool designed to constrain capability acquisition simultaneously reinforces the system that motivates it. When Washington denies access to advanced chips, it deepens the resource dependence that motivates Chinese firms to align with their government's priorities in the first place.

The case raises broader concerns about the effectiveness of export-control enforcement across Southeast Asia and beyond. As long as a market like China exists with strong incentives to acquire restricted technology, the supply will always find a way.

The structural challenge is that export controls address only one side of the equation. They restrict supply but cannot reshape the demand-side incentive architecture that drives Chinese firms to pursue advanced capabilities. Meanwhile, China's registration system, subsidy programs, and national R&D investments create a comprehensive ecosystem that makes firms dependent on government support and aligned with state priorities.