DEEP DIVE — When the Biden administration rolled out its semiconductor export restrictions in October 2022, the logic seemed airtight: cut off Beijing’s access to advanced AI chips, and you’d put the brakes on China’s tech ambitions. Three years on, that bet looks a lot shakier than anyone in Washington expected.
Instead of paralyzing China’s AI sector, these controls have promoted domestic self-reliance. With no choice but to develop indigenous workarounds and architectural innovations, Chinese businesses are rapidly decoupling AI progress from sheer hardware volume. U.S. policies have undoubtedly bought time, but they have also ushered in a parallel innovation ecosystem totally independent of Western influence.
Now the Trump administration is trying a different tack. By allowing conditional chip exports to approved Chinese customers, complete with revenue-sharing arrangements, the White House appears to be hedging its bets. The strategy banks on maintaining some leverage through controlled access rather than outright denial. Yet critics caution this could backfire spectacularly: American companies might be financing the very Chinese AI capabilities designed to render them obsolete.
Both strategies, it seems, lead to uncomfortable places.
The DeepSeek Shock and Algorithmic Efficiency
A year ago, DeepSeek’s new R1 model sent tremors through Silicon Valley and Washington. The Chinese startup demonstrated AI reasoning capabilities closely mirroring those of OpenAI’s ChatGPT o1, without access to cutting-edge chips restricted by U.S. export controls.
“Algorithmic efficiency is a useful technique that enables developers to achieve reasonable performance compared to frontier models through inference-time optimization—essentially adapting the achievements of others for your own purposes,” Martijn Rasser, Vice President of Tech Leadership at the Special Competitive Studies Project, tells The Cipher Brief. “DeepSeek’s engineering is genuinely impressive, and this ‘fast follower’ strategy is very attractive for price-sensitive, commoditized AI applications.”
By optimizing inter-chip memory bandwidth on less sophisticated H800 chips, DeepSeek achieved competitive performance through algorithmic efficiency rather than brute force. According to the MIT Technology Review, Chinese startups are responding to export restrictions by prioritizing efficiency, resource pooling, and collaboration, thereby gaining an edge over competitors.
The breakthrough, however, should not be overinterpreted, Rasser warns.
“Algorithmic efficiency and raw compute aren’t substitutes at the frontier — you need both,” he says. “Training next-generation foundation models, achieving breakthroughs in areas like protein folding or materials science, and developing the most advanced autonomous systems all require massive compute that efficiency gains alone cannot replicate.”
Subscriber+Members get exclusive access to expert-driven briefings on the top national security issues we face today. Gain access to save your virtual seat now.
Other analysts see the dynamic playing out along different timelines.
“AI progress is shaped by two concurrent trends: pushing the capability frontier requires ever-greater computing power, yet over time, reaching any given capability level requires less and less computing power,” James Sanders, Research Associate at the Center for a New American Security, tells The Cipher Brief.
While American AI companies continue advancing along the compute-intensive frontier, Chinese firms constrained by chip shortages focus instead on the lower-cost, lower-capability portion of the market.
“If the U.S. continues to maintain its computational dominance, China will likely continue to play a role as a fast follower, rather than leader, in AI capabilities,” Sanders explains.
Smuggling Networks and Systematic Circumvention
Aside from inventiveness under constraints, Chinese companies have also shown remarkable versatility by smuggling goods through shell companies and elaborate smuggling operations. The procurement of chips for Huawei’s Ascend 910 AI processor is one of the most dramatic examples.
According to Congressional testimony, Huawei used shell companies to deceive Taiwan Semiconductor Manufacturing Company into manufacturing approximately 2 million computer chiplets.
In 2024, a smuggling ring operating under the company name “Luxuriate Your Life” reportedly purchased $390 million worth of servers from Dell and Supermicro containing banned Nvidia graphics processing units, then smuggled them into Malaysia for ultimate delivery to Chinese customers. Hundreds of controlled chips have subsequently been sold in Shenzhen markets as vendors trade restricted AI technology.
The dramatic scale of smuggling operations thus raises fundamental questions about enforceability.
“No export control regime is 100% leakproof,” Rasser acknowledges. “But the goal should be making evasion costly, risky, and limited in scale rather than expecting perfect enforcement.”
He contends that the most effective approach is drawing a clear line: semiconductors exceeding certain performance thresholds should not be sold to end users in countries of concern at all, regardless of stated end use.
Sanders takes a more optimistic view.
“America has the ability to reduce AI chip smuggling to negligible levels if it wanted to. It only requires the political will to do so,” he says.
Technology that verifies a chip’s location could reduce smuggling to negligible levels, Sanders argues, noting that Nvidia has reportedly built such a capability and several congressional bills have proposed mandating location verification for exports.
“Even with the level of smuggling we see today, export controls on advanced AI chips are still effective,” Sanders insists. “They impose a large differential cost on China’s AI development and make it much harder for these chips to be obtained and used by adversaries.”
The Self-Sufficiency Acceleration
Perhaps most significantly for long-term strategic competition, export controls have supercharged China’s decades-long quest for semiconductor self-sufficiency. At a February 2025 meeting between Chinese Communist Party Chairman Xi Jinping and technology executives, Huawei founder Ren Zhengfei reported that his previous concerns about domestic advanced semiconductor production had eased due to recent breakthroughs.
A CSIS analysis underscores that Ren stated he is leading a network of more than 2,000 Chinese companies working collectively to ensure that China achieves self-sufficiency of more than 70 percent across the entire semiconductor value chain by 2028.
Rasser is skeptical of the ambitious timeline.
“This timeline is not credible for leading-edge semiconductors,” Rasser observes. “If he means trailing-edge components and packaging, 2028 might be achievable. But if he means the advanced logic chips that power AI training and high-performance computing, it isn’t.”
Chinese firms still can't build the ultra-precise manufacturing equipment needed to produce chips at the most advanced nodes, he points out. But that hasn't stopped Beijing from doubling down. Overall chip investment dipped slightly in early 2025, yet spending on semiconductor equipment jumped more than 53 percent, a clear signal that China is dead serious about building out a complete domestic supply chain.
The Chinese government has channeled well over $150 billion into semiconductor development from 2014 through 2030, according to Semiconductor Industry Association analysis — an investment equivalent to the U.S. CHIPS Act virtually every year since 2014.
The Bifurcation Dilemma
The unintended consequence of U.S. export controls may be accelerating the bifurcation of global technology ecosystems into incompatible, competing supply chains.
Consequently, we are faced with a strategic paradox: Does a China that relies on Western chips but is constrained by American controls pose a lower long-term risk than one that operates in a technologically autonomous fashion beyond American eyesight?
“A technologically autonomous China certainly poses a greater risk of strategic surprise,” Rasser concedes. “But the alternative framing — that we could maintain Chinese dependency on Western technology indefinitely—isn’t realistic. It ignores history, Chinese strategy, decades of mercantilist industrial policy, and the inherent difficulty of controlling technology diffusion over long time horizons.”
The better frame, he suggests, is that export controls buy time.
“The question is what we do with that time,” Rasser continues. “Years of constrained Chinese progress are years during which the United States and allies can extend their own technological lead, build more resilient supply chains, strengthen alliances with key semiconductor nations, and shape the trajectory of AI governance.”
Sanders echoes this assessment.
“China is determined to develop a fully domestic AI chip industry, and will likely eventually be able to do so,” he notes. “The real question is whether we equip China to compete with American AI leadership today.”
The fundamental constraint on Chinese domestic AI chip production, Sanders points out, isn’t the lack of logic chips but access to advanced manufacturing equipment.
“Strengthening and fully enforcing current export restrictions on semiconductor manufacturing equipment is the most important step in preventing China from developing an AI chip industry that could outcompete America’s,” he says.
Yet the competitive landscape extends beyond manufacturing capacity alone.
The Supply Chain War
Leland Miller, co-founder and CEO of China Beige Book, who serves on the U.S.-China Economic and Security Review Commission, however, sees the entire framework shifting.
“The trade war is dead,” he tells The Cipher Brief. “This is a supply chain war. Beijing wants to weaponize supply chains for leverage. Tariffs are background noise now.”
Miller notes that China has advantages in data, talent, energy, and scale. Their weakness remains computing — advanced chips and lithography.
“That’s why H100s matter so much,” he says, referring to Nvidia’s high-end processors. On open-source AI platforms like Hugging Face, Chinese dominance is already evident. “Seventeen of the top 20 models are Chinese,” Mike Kuiken, Vice Chair of the U.S.-China Economic and Security Review Commission, tells The Cipher Brief. “Software plus enough hardware can close gaps. That’s why this moment is dangerous.”
The Trump administration’s December 2025 decision to allow conditional exports of Nvidia H200 chips to approved Chinese customers in exchange for a 25 percent revenue stake represents recognition that absolute denial may be neither achievable nor desirable. Yet Beijing has reportedly discouraged state-linked firms from adopting these chips, while Chinese companies like Huawei and Alibaba continue to advance domestic AI alternatives.
The controls have bought time — China’s AI development has slowed, and the U.S. still holds the edge in cutting-edge tech. The more complex question is whether Washington can consolidate that lead before Beijing circumvents the restrictions entirely, potentially causing an even bigger crisis.
“The goal isn’t perpetual dependency, it’s ensuring that when China does achieve greater autonomy, the strategic balance still favors us,” Rasser concludes.
The Cipher Brief is committed to publishing a range of perspectives on national security issues submitted by deeply experienced national security professionals. Opinions expressed are those of the author and do not represent the views or opinions of The Cipher Brief.
Have a perspective to share based on your experience in the national security field? Send it to Editor@thecipherbrief.com for publication consideration.
Read more expert-driven national security insights, perspective and analysis in The Cipher Brief because National Security is Everyone’s Business
