Washington’s AI strategy reflects genuine strategic concerns: that the United States is locked in a decisive race with China for artificial general intelligence (AGI), that whoever crosses the finish line first wins a permanent strategic advantage, and that maximum acceleration, paired with maximum denial of chips and technology to Beijing, is the only rational play. While the competition with China is real and the risks are genuine, key elements of this narrative deserve closer examination. And building a strategy on an incomplete understanding risks producing outcomes far worse than the threat it purports to address.

I write this not to minimize real security concerns, nor to suggest naïveté about Chinese strategic ambitions. After 35 years helping U.S. MNCs build technology businesses in China and leading AI, semiconductor and cybersecurity efforts across both ecosystems, I have seen how quickly assumptions about technological dominance can unravel. The argument here is simply this: we need to reexamine the race we think we’re running, the game we think we’re playing, and the strategy we’ve chosen. Because right now, the evidence suggests our current strategy may be incomplete or overconcentrated on a single dimension of competition.

Questioning the Decisive Strategic Advantage Thesis

The dominant narrative in Washington’s AI policy circles goes something like this: whoever achieves AGI first will gain a decisive strategic advantage (DSA), a durable, possibly permanent lead that reshapes the global balance of power. This belief drives hundreds of billions in capex commitments, aggressive export controls, and a general willingness to treat AI development as a wartime mobilization.

The DSA thesis depends on several assumptions, each of which deserves scrutiny. First, that there is a clear finish line. There is not. AI development is a continuous, multi-dimensional process with no single threshold that confers omnipotence. AGI, however defined, would not be a static superweapon but a rapidly evolving ecosystem that competitors can and will replicate. History reminds us: America’s nuclear monopoly lasted just four years.

Second, the thesis assumes that model capability gaps can be maintained through compute denial. Recent market data point in a different direction. The U.S.-China frontier model gap, once estimated at 12-14 months, has narrowed to roughly two to three months, during a period when the U.S. has steadily ramped up export controls on even mid-tier data center GPUs. On the April Monthy 2026 OpenRouter leaderboard, four of the top six most-used models globally were Chinese. MiniMax, Kimi, DeepSeek, Xiaomi and Zhipu models are all surging in popularity, driven by their cost advantages and the thirst for highly capable open-source models by AI agents. In video generation, Kuaishou’s Kling 3.0 and ByteDance’s Seedance 2.0 significantly outperform OpenAI’s Sora and Google’s Veo on independent benchmarks and real-world use cases as discussed in my recent CCA essay.

Third, the DSA narrative assumes a winner-take-all dynamic. In practice, AI models are rapidly commoditizing. Cursor usage data from 2025 shows the top model changing nearly every month, with no single lab holding sustained dominance. The recent batch of China models launched around Lunar New Year all are within a few percent of leading U.S. frontier models (at the time of release) but available at approximately 1/10^th the API price. There are now even small models from Qwen which perform at about the level of frontier models from a year ago that can run natively on an iPhone. When the commodity is cheap and abundant, strategic advantage shifts from who builds the best model to who deploys AI most effectively across their economy.

And this is precisely where the current strategy is most in tension with its own goals. While Washington and Silicon Valley pours massive resources into chasing the elusive AGI finish line, Chinese companies are deploying “good enough” AI at massive scale. ByteDance’s Doubao chatbot exceeded 100 million daily active users. Alibaba’s Qwen models have surpassed 700 million downloads globally, spawning 180,000+ derivative models. Chinese open-source models are becoming the de facto platform for sovereign AI efforts across the Global South’s 150 Belt and Road Initiative partner countries. Leading open-source models are usually much smaller than closed-source frontier models, so they can be downloaded and installed on smaller on-premise systems, and the smaller models will run smoothly on personal laptops and even modern phones. This is a significant diffusion advantage by offering greater data sovereignty and privacy, while also enabling lower deployment costs domestically and globally.

Today, most Chinese models are designed to run on Nvidia chips, but the Chinese are already developing alternatives to Nvidia’s GPUs, and a number of the new Chinese models (e.g., Deepseek and Zhipu) are already being optimized for them with some already trained on them. As Chinese domestic chips mature, the long term plan is to offer full stack solutions to the international market. Denying US chips to Chinese ecosystem now will only accelerate the drive for domestic alternatives, speeding local innovation and reducing reliance on US technology.

Beyond the Arms Race Framework

The U.S. approach treats the AI competition as an arms race: zero-sum, secretive, defined by denial and containment. But what’s actually unfolding looks far more like a platform race and an innovation race, where value expands, multiple winners coexist, and the key metric is diffusion, how broadly and effectively AI is adopted across industries--not who achieves the highest benchmark score this week.

Consider the four types of AI competition. In a simple race (first to a benchmark), there’s a clear finish line and a single winner, but in practice, benchmark leads flip monthly and confer no durable advantage. In an arms race, secrecy and denial dominate, but the AI technology stack is too diffuse and fast-moving to contain. In an innovation race, the game is multi-turn with many winners—think robotics, agentic AI, and clean energy integration. In a platform race, it’s an infinite game where ecosystem adoption determines outcomes—think open-source AI, cloud APIs, global developer communities.

The United States is investing heavily in an arms race and simple race dynamics. China, increasingly, is competing in the innovation and platform quadrants (which offer more long-term value). The results speak for themselves: China leads in 66 of 74 critical technologies tracked by the Australian Strategic Policy Institute, dominates open-source AI model offerings, accounts for 54% of global industrial robot installations, produces about half of the world’s AI researchers, and builds more new electricity capacity annually than the rest of the world combined. These are the foundations of AI deployment at scale, and chip denial alone can’t offset them.

There are claims that most Chinese AI advances are due to stealing technology or distilling models from the United States. Given that now more than 40% of the world’s AI papers are published by researchers in China, and about half of the world’s AI researchers are born and educated in China, it’s hard to justify this claim. For some background on the practice of distilling models, it’s commonly recognized that most/all labs practice this internally and on each other (even domestically) to optimize model accuracy and efficiency. Since the frontier U.S. models are closed-source, Chinese labs are more easily tracked if and when they do it. Whereas most Chinese models are open-source, so the models can be run on the users’ local machines or hosted by each country’s domestic cloud services, thus Chinese labs cannot track their models’ detailed usage. Recently Cursor got into PR issues when it claimed to have trained its own advanced AI coding model (Composor 2), when it turned out that they just fine-tuned the Chinese Kimi K2.5 model. Additionally, Chinese labs usually publish their new technical findings as detailed papers with each major release of models, thus the need for U.S. labs to distill for learning breakthroughs from new Chinese models are reduced.

Rethinking the Strategic Framework

Washington’s strategic framework implicitly models the U.S.–China AI dynamic as a Prisoner’s Dilemma, a game where defection (competing aggressively, withholding cooperation) is the dominant strategy. But this framing has two fatal flaws.

First, the Prisoner’s Dilemma assumes a one-shot game with no communication. We live in a multi-turn world where actions are observable, reputations accumulate, and in such a case, the mathematically optimal strategy is tit-for-tat, starting with cooperation. We can see China’s actions and words. They can see ours. We are not in separate interrogation rooms.

Second, and more fundamentally, we are playing the wrong game entirely. The better model is the Stag Hunt. In this classic game, two hunters enter a forest and must decide: cooperate to catch a stag that feeds both for a month, or hunt hares separately that provide a week’s sustenance each. The stag requires both hunters working together. If one goes for the stag while the other chases hares, the stag hunter goes home empty-handed.

The United States is currently going after the stag, the projected artificial superintelligence (ASI) that promises permanent dominance. China, meanwhile, hunts the hares: practical AI diffusion across manufacturing, healthcare, education, and sovereign AI platforms for the developing world. The Stag Hunt has two Nash equilibria: both cooperate (stag) or both hunt alone (hares). But the worst outcome is precisely where the United States finds itself now: pursuing the stag alone while the other player collects reliable, real-world gains from hares. The more accurate and nuanced game theory representation of the world today is likely closer to a multi-turn Stag Hunt for the commercial market portion and a multi-turn Prisoner’s Dilemma for the national security portion. In both cases, increasing communication and finding common ground would create a better outcome for both sides. A more effective approach is to adopt a multi-turn Stag Hunt strategy where we first prioritize the hare equilibrium (diffusion focus) to address immediate risks and garner AI’s economic benefits, then later jointly pursue the long-term stag equilibrium in collaboration with China when we have a better sense of how to manage the shared safety risks of a truly transformative AI (ASI). Given the clear interdependence of both sides across multiple dimensions, this would deliver the optimal outcome, without the excessive capex costs or potential societal and safety risks of our current path.

The consequences of this mismatch are not theoretical. U.S. tech firms have committed over $800 billion annually in AI capex for 2025–2027, while job openings in the United States have declined sharply since 2022 even as the S&P 500 soared, a divergence that recent Stanford research attributes to AI automation of early-career positions in software development and customer service. World Bank data shows the U.S. workforce faces over 60% AI exposure, the highest globally, compared to about 40% in China. We are building the most powerful automation technology in history while having no plan for the workers it will displace.

The loudest proponents of the DSA narrative are AI labs and their investors who have strong commercial interests in lighter regulation and increased government funding. The refrain from Silicon Valley labs that “we can’t be regulated because if we slow down, China won’t” is not based on facts. China’s AI industry is one of the most heavily regulated in the world and their infrastructure buildout is just a fraction of the size of the United States. Putting in some sensible guardrails and regulations for AI development and deployment around safety, privacy, and security would not slow down the United States, but rather help ensure a less chaotic diffusion into broader society. In fact, China is slowing down their own AI labs but restricting their ability to buy Nvidia GPUs. Even after U.S. export control restrictions allowed for H200 GPUs to be sold to China, the Chinese government has required approvals for their companies to buy them. Recent remarks from Howard Lutnick suggest that to date, zero H200 GPUs have been sold to China.

The Military Dimension: From Benchmarks to Operational Readiness

Proponents (government and industry) of the current strategy often invoke national security: “If China gets AGI first, they’ll weaponize it against us or would prevent us from winning.” This is a legitimate security concern worthy of serious attention. However, the military advantage argument warrants closer examination. What the military needs is not the latest frontier model. It needs models that are fit to task, certified, tested, and integrated into operational systems. Here, strategic advantage depends not just on the invention of frontier models but on the institutional capacity to operationalize them, integrating them into doctrine, training personnel, and adapting operational concepts to leverage their capabilities.

The U.S. military’s vendor and model certification process can take over 12 months for national security level accreditation (DoD IL6). Recent reports suggest that the recent Iran military operations were planned using the 20-month-old Claude 3.5 Sonnet model, not because newer models didn’t exist but because the certification pipeline couldn’t keep pace, in addition to the need to remove standard safety guardrails from the base consumer model. Meanwhile, Chinese government agencies have far fewer testing and evaluation (T&E) delays, and their AI models already go through review by the government before public release. The gap between invention and deployment matters more than the gap between benchmarks.

In military settings, the tangible advantage may lie less in who trains the most capable model than in who can certify, integrate, and deploy usable systems fastest. On that metric, the current U.S. system is a structural disadvantage. But even if we achieve ASI first and can somehow expedite the certification process, unless we are willing to deploy it untested in a pre-emptive all-out strike against China, the temporal advantage could evaporate in a matter of weeks or months. It’s unclear that this result would justify the trillions in spend. And even if such an attack succeeds in taking down China’s infrastructure, the inevitable response will be swift and devastating for both sides. It’s the exact reason why we have not seen the use of nuclear weapons in real conflict for over 80 years.

In recent weeks, the new Anthropic Mythos model has attracted a lot of attention for its cybersecurity prowess, and some point to this as an example that the United States is “winning” and the current strategy is correct. Unlike large amounts of fissile materials, securing software and know-how is almost impossible. There are already reports that the Mythos model is now accessible to unauthorized personnel, not to mention the recent unintended leak of the full source code of Anthropics’ Claude Code solution. Recent new benchmarks are also showing that the OpenAI GPT 5.5 model is actually as performant in cybersecurity capabilities as Mythos and that model is publicly available to all users, thus showing that any perceived cyber advantages are highly temporary. We need to accept that any software-based advantage is temporary and adjust our strategy to the reality that DSA may just be a mirage. In fact, the bigger and more certain threat is the misuse of these advanced models by bad actors. Hackers and terrorists would not hesitate to make use of these models for nefarious purposes, whereas state actors still fear reprisals and the risk of escalation. This is a shared risk for both the United States and China, and a place where we can find common ground for cooperation.

What We’re Not Seeing: A Competitive Ecosystem, Not a Monolith

Washington tends to view Chinese AI as a state-directed monolith. The reality is a fiercely competitive commercial ecosystem where Chinese labs compete with each other (just to survive) far more than they coordinate to conspire against the United States. ByteDance’s Doubao is a closed-source consumer product fighting for domestic market share. Z.AI generates over 60% of its revenue from enterprise on-premise deployments. MiniMax earns roughly 70% of revenue from international API sales. Moonshot AI’s Kimi competes directly with all of them in consumer chatbots and AI agent inference. As an example, neither the Chinese nor U.S. government had any significant idea about DeepSeek’s innovations prior to its rise to prominence in January 2025.

The open-source strategy that has made Chinese models globally ubiquitous is an emergent outcome of commercial competition, not a directed national strategy. Alibaba open-sourced Qwen to drive cloud adoption; DeepSeek open-sourced to attract research talent; MiniMax did so to build developer ecosystems. Even within these companies, there are significant internal tensions. Alibaba is actively debating open-source diffusion versus revenue protection, and Tencent faces similar strategic discussions on agent-focused models. Framing this diverse, commercially motivated ecosystem as a unified strategic threat leads to policy responses that are simultaneously too broad (restricting all Chinese AI) and too narrow (focused on chips while ignoring the deployment gap). That said, while these firms operate with substantial commercial independence, they remain embedded in a system where the Chinese state retains the ability to shape or direct their alignment when it perceives critical national interests to be at stake. The recent push by the Chinese government for some of its leading labs to optimize models for local chip offerings before release is an example of this.

Additionally, it’s worth mentioning that U.S. export control policies also exist around advanced tools and equipment targeting chip and semiconductor manufacturing players (e.g., Huawei and SMIC) with it’s core aim to constrain compute supply to Chinese AI labs. These players have a clear association with the state, and such restrictions have shown near-term effect in limiting their capabilities, but has also clearly accelerated the development of domestic alternatives, albeit still one to two generations behind the most advanced Western offerings. As the two ecosystems bifurcate, there will be increasingly less dependency on the United States and less leverage for future negotiations.

A Different Path Forward

None of this means we should be complacent about security. Genuine threats exist: AI-enabled CBRNE (Cyber, Bio, Radioactive, Nuclear, Explosive) risks, autonomous weapons proliferation, and the misuse of AI by bad actors. But these threats are best addressed through targeted, narrowly scoped controls and shared intelligence, not through a blanket technology-denial strategy that was designed for a world that no longer exists.

The “small yard, high fence” doctrine, originally conceived as a targeted national security tool, has expanded into a comprehensive technology approach that now touches chips, lithography, EDA software, HBM, chemicals, and advanced AI models, as discussed in a recent CCA article. This expansion has paradoxically accelerated China’s drive toward full supply chain independence, fueled innovations born of constraint (like DeepSeek’s efficiency breakthroughs), and alienated allied semiconductor industry firms who see shrinking addressable markets. That said, export controls serve legitimate national security purposes, and policymakers may be operating with intelligence assessments and threat evaluations not available in the public domain.

Adjusting course does not require abandoning competition. It requires reframing it. Three shifts would immediately improve America’s strategic position:

First, make efforts to narrow export controls to genuine military and CBRNE threats, not all advanced technology. The current regime tries to prevent China from developing any frontier capability, which is both unenforceable and counterproductive. In fact, given Chinese labs’ propensity for open-sourcing their models, most advances they produce will naturally flow into U.S. models rapidly, turning them into subsidiary development arms for U.S. players. It must still be noted that in the unlikely event of direct conflict, having some export controls may still buy time for domestic adaptation, alliance coordination, and military hardening even if such controls cannot achieve the wished-for permanent dominance.

Second, shift from a Cold War arms-race mental model to a Space Race innovation model. The Space Race generated GPS, Teflon, the internet’s precursor, bringing massive spillover benefits for the whole economy. An AI innovation race focused on healthcare, climate, education, and productivity could do the same, with far more participants winning. It would also encourage the U.S. labs to release comparable level open-source models to China so our tech stack can compete on a level playing field for global diffusion. Additionally enacting some reasonable safety and socially responsible regulations on U.S. labs (as China has already done) will lessen the growing societal fear over AI that’s been brewing domestically which is certain to hinder the nation’s ability to deploy this technology effectively into the economy and reduce our global competitiveness.

Third, open bilateral and multilateral channels on AI safety. The Stag Hunt only works when hunters can communicate before entering the forest. Both China’s “Global AI Governance Action Plan” and America’s “AI Action Plan” were released in July 2025, and neither mentions the other country. We cannot manage the multifaceted risks of the most powerful technology in human history while refusing to talk to the other nation building it. There’s clear common ground on AI safety; both countries want to avoid catastrophic risks from runaway AI; and both have a shared interest in preventing AI-enabled bad actor proliferation.

Conclusion: There Is No Finish Line, but Many Medals to Win

A race is not a race if the competitors are headed to different destinations. The United States is sprinting toward AGI and ASI as if crossing that threshold confers permanent dominance. China is building industrial AI infrastructure that will shape how much of the world deploys this technology for decades. Both paths have value, and neither inherently zero-sum, if proper coordination occurs.

The biggest risks we face from AI are misuse by bad actors, mass displacement of workers, concentration of economic power, and escalation of military conflict. These are shared risks that require shared solutions. Every dollar spent on denial should be matched by greater investment in national preparedness for AI diffusion and adaptation. Every diplomatic channel closed over chips is a safety conversation not happening. A deeper discussion on the underlying technology issues and a detailed framework for addressing both economic and geopolitical issues around the AI race can be found in this Stanford paper, Beyond Rivalry.

It is time to lift the veil of assumptions that have constrained our thinking. The playing field is not the one we were told exists. But the real one offers enormous opportunities if we have the clarity to see it and the courage to adjust course. There is no finish line in AI, but along the way, there will be many medals awarded. We should make sure America is competing for the ones that actually matter.

Leave a comment

China’s evolving export control system is reshaping industry decisions and production in real time. Licensing regulations, extraterritorial concerns, end-use ambiguity, and disclosure requirements that can shift with little notice are creating significant uncertainty for companies trying to plan production, manage inventory, and meet customer demand. This uncertainty is increasing risks to global supply chains and accelerating the push to diversify away from China, even at higher cost.

In this commentary, Cameron Johnson, Senior Partner at Tidalwave Solutions, shares his inside-out perspective on China’s export control system. With decades of experience managing supply chains across industries in China, Cameron presents first-hand observations on the challenges firms currently face under China’s rare earth export controls. CCA Honorary Senior Fellow Paul Triolo weighs in with reflections on how U.S. policymakers might approach China’s rare earth export control regime, emphasizing the need for clearer policy alignment in Washington, sustained dialogue with Beijing on licensing transparency, and pragmatic tradeoffs that could help reduce uncertainty and stabilize strategically critical supply chains.

Navigating China’s Rare Earth Export Controls: Challenges for Industry and Paths Toward Greater Clarity

Cameron Johnson, Senior Partner, Tidalwave Solutions

In recent conversations with supply chain operators and multinational firms, I have heard a consistent concern: China’s current rare earth export control regime is creating operational uncertainty that is increasingly difficult for companies to manage.

Risks of Trade Secret Disclosure

Some companies that have applied for licenses have been asked to provide sensitive information, such as product specifications, photographs of production lines, customer identities, chemical compositions, and processing methods. Disclosing such information poses the risk of reverse engineering or information leakage. Companies with unique products or advanced materials face the possibility that their “recipe” — how the product is made — could be replicated once disclosed, along with their customers being identified.

Declaring precise chemical ratios or processing methods could reveal intellectual property that took years of R&D to develop. These processes can be recreated, circumventing the costs, risks, and time required for development. As a result, firms are extremely cautious and hesitant to provide information beyond end-use verification, even when they want to comply, because a single disclosure can erode years of competitive advantage and R&D.

In many industries, a customer’s identity is considered confidential and a strategic asset, often protected by contracts prohibiting disclosure. Firms may be forced to breach contractual obligations, therefore choosing between protecting critical business relationships and complying with export control regulations. It could also trigger situations in which they have to inform customers, resulting in lost business.

For example, companies across industries, such as automotive and aerospace, have contracts that prohibit disclosing technical details, supply chain structures, or customer identities to third parties, including governments, without prior authorization. Jet engine producers face strict confidentiality requirements because their supply chains include specialized coatings, alloys, and precision parts. Automotive companies also have strict requirements because component sourcing reveals costs, future model plans, and designs.

Uncertainty about Licensing and Regulatory Requirements

Beijing has now made case-by-case and individual export control review a core process, even for companies with long-standing compliant exports and established end use. For example, shipments that are almost identical could face different analyses, documentation requirements, and timelines. For companies operating on strict production schedules, planning becomes exceedingly difficult.

Security assessments also add additional layers of scrutiny, complicating the process, especially for dual-use classified components and advanced materials. An application is reviewed first by MOFCOM and then sent to other ministries for additional review, each of which might request additional information. These reviews are not transparent to applicants, who often do not know why their applications were escalated or whom to contact to resolve them. This lack of transparency adds uncertainty and can sometimes significantly extend approval times. It also contributes to perceptions of regulatory inconsistency or discrimination against certain applicants.

Uncertainty grips companies as license review timelines can shift, leaving them to wonder whether a license will be approved in weeks or months. The 45-working-day review period for license approval helped ease this challenge, but uncertainty persists. Regulations offer no guidance on detailed status updates, reasons for delay, the maximum review period, or other aspects of the process. Companies are left to navigate this uncertainty, making it difficult to plan production, manage inventory, or confirm delivery schedules. This unpredictability and the resulting operational challenges are disrupting global production flows, customer relationships, and industry foresight, forcing some businesses to turn away customers or prompting customers to find new suppliers, leading to the end of long-term relationships.

Some regulations do not clearly define the extent of their extraterritorial reach, and some applicants have been affected by political dynamics, with rare earth restrictions at times arbitrarily imposed on companies as part of China’s response to tensions with the United States. This leaves companies facing broader political undercurrents beyond their control, which can influence decisions in ways that are nearly impossible to predict. This level of uncertainty, faced by businesses around the world, makes routine business decisions much harder and is pushing companies to question their long-term reliance on Chinese-origin rare earths and products.

Conclusion

I believe predictable, transparent regulations give businesses a clear understanding of how they are implemented and what can trigger changes to licensing. Companies can adapt to any condition and plan production with greater confidence when they understand potential delays, restrictions, and the purpose of controls. This understanding allows businesses to organize themselves to meet customer needs while complying with regulations. Knowing the “rules of the road” is key to every business and stabilizes expectations in a fast-moving environment.

How U.S. Policymakers Might Approach China’s Rare Earth Export Control Regime

Paul Triolo, Honorary Senior Fellow on Technology, Center for China Analysis

From a policy standpoint, it is important to recognize that rare earth and magnet supply chains have moved well beyond the realm of normal commercial or industrial policy questions. They now sit squarely within a broader geopolitical and national security context. China’s policies in this space are closely intertwined with how Beijing views the actions of the United States, Europe, Japan, and other partners as they move to diversify supply chains, strengthen export controls, and reduce strategic dependencies. Because of this dynamic, it is difficult to approach the issue solely through the lens of what China might do to provide clarity or stability. Outcomes will increasingly be shaped by reciprocal policy choices on all sides.

For U.S. policymakers, the first priority should be achieving greater alignment within Washington on objectives and expectations. There is still a degree of adjustment underway as policymakers and industry stakeholders continue to assess the implications of China’s evolving export control framework around rare earth elements and permanent magnets. Different parts of the U.S. government approach the issue based on different institutional mandates. The United States Trade Representative and the Treasury Department, for example, both have important roles to play, but their priorities are not always fully aligned in terms of what the United States might seek from Beijing and what forms of flexibility might realistically be offered in return. Achieving greater internal clarity will be important before the United States formulates a durable external approach.

At the same time, policymakers should recognize that the April 4, 2025, and October 9, 2025 rules have already altered how markets and industry view the reliability of these supply chains. Whether these measures are actively enforced or simply remain a credible policy lever, they have shifted expectations across the sector. Many large U.S. manufacturers that rely on these materials are still mapping their exposure. In some cases, companies are only now realizing the extent of their dependence on Chinese rare earth processing and magnet manufacturing, and how limited near-term alternatives remain.

History offers a useful reminder of how slowly these markets adjust. During the 2010–2011 rare earth disruption, the export restrictions themselves lasted only a relatively short period. Yet even after supply resumed, it took nearly two years for markets to stabilize and for pricing and trade relationships to normalize. The current situation is considerably more complex. Supply chains are larger, more integrated, and far more strategically contested. Even under favorable circumstances, adjustment is likely to take years rather than months.

These realities point toward a dual-track policy approach. The United States should continue to invest in diversifying supply chains, expanding alternative processing capacity, and strengthening resilience across critical materials ecosystems with allies and partners. At the same time, maintaining channels of communication with Beijing on export control practices, licensing transparency, and technical implementation will remain important. Reducing uncertainty where possible can help stabilize global markets while the longer-term work of supply chain diversification proceeds.

In addition, the administration will want to consider whether there are tradeoffs that can be made that will reduce supply chain uncertainties and build even a small level of trust with Beijing, to avoid a sustained, high risk of “snapback” if the United States takes actions Beijing perceives as violating the spirit of the Busan agreement. For example, as I have laid out in my recent article, relaxing export controls in areas that have proven to be counterproductive, such as on memory and related semiconductor tooling, in exchange for guarantees from Beijing on broader and more expeditious licensing and transparency around the licensing process (e.g., sectoral white lists) should be considered as part of the bilateral trade negotiations. Such a deliverable would help to stabilize markets in the near term and allow time for companies to adjust and derisk supply chains.

MORE CCA FLAGSHIP PRODUCTS

Decoding Chinese Politics (DCP):

https://asiasociety.org/policy-institute/decoding-chinese-politics

Taiwan Policy Database (TPD): https://asiasociety.org/policy-institute/taiwan-policy-database

Global Public Opinion on China (GPOC): https://asiasociety.org/policy-institute/global-public-opinion-china

Cure4Cancer(C4C): https://asiasociety.org/policy-institute/cure4cancer

By Paul Triolo, Honorary Senior Fellow on Technology, Center for China Analysis, and partner, senior vice president for China, and technology policy lead at DGA-Albright Stonebridge Group (DGA-ASG)

Editor’s note:

In this essay, Paul Triolo examines the evolution of U.S. technology export controls since 2022 and questions whether the “small yard, high fence” framework has expanded beyond its original intent. What was initially framed as a targeted effort to safeguard a narrow set of national security–critical technologies, he argues, has gradually widened to encompass large segments of the semiconductor and advanced computing ecosystem.

Triolo assesses both the strategic rationale behind the controls and their effects. He notes the growing complexity of implementation, the challenges of defining what qualifies as “advanced” in fast-moving industries, and the economic and supply chain consequences for U.S., allied, and Chinese firms alike. He also highlights Beijing’s countermeasures, particularly in critical minerals and rare earths, as part of a broader dynamic of mutual leverage that is reshaping global technology networks.

The essay urges a recalibration: clearer definitions, more adaptive thresholds aligned with industry realities, and stronger coordination mechanisms. Triolo’s argument is that export controls remain an important policy tool—but one that must be continually refined to balance national security objectives with economic resilience and long-term technological competitiveness.

These questions remain contested. Advocates of tighter controls argue that advanced compute and enabling technologies could shape future military balances and that limiting access is prudent even at clear and mounting economic costs. Others contend that overly expansive restrictions risk accelerating Chinese self-reliance, fragmenting global innovation networks, and weakening the competitiveness of U.S. firms without clearly defined or measurable security gains. We invite rigorous debate on these trade-offs and welcome diverse perspectives on how best to align technology policy with enduring national security interests – seeking to balance national and economic security interests, rather than treating them as coterminous.

Key takeaways

• The “small yard, high fence” idea has effectively widened: treating advanced compute, chips, SME, AI, biotech, and green tech as national-security imperatives pulls whole supply chains into the regime, with broad spillovers for firms and partners.

• Beijing’s retaliation—especially rare earths and critical minerals licensing—has turned export controls into a two-way leverage contest, with potentially long-lived consequences for both commercial supply chains and the defense industrial base.

• Questions remain about how to weigh security benefits against economic costs, especially as Chinese AI models and semiconductor capabilities continue to advance despite existing controls. Restrictions have also accelerated Chinese domestic investment, workarounds, and ecosystem-building, while potentially eroding U.S. firms’ market share, revenue, and (by extension) R&D capacity.

• Defining what qualifies as “advanced” technology has proven challenging in fast-moving industries; static technical thresholds can quickly fall out of step with commercial and technological realities, prompting repeated rule adjustments.

• Policy debate is shifting toward a sliding-scale approach: the H20/H200 controversy highlights three paths, ban all, keep current thresholds, or adopt a clearer performance-based sliding scale with guardrails that preserves U.S. market presence and visibility while protecting the true frontier.

• Export controls should be a tool, not the strategy: Potential reforms include narrowing and clarifying the “yard,” building a standing technical advisory process, updating thresholds regularly, improving interagency/allied coordination, and focus monitoring on genuinely sensitive technologies. The long-term emphasis should shift toward sustaining U.S. comparative advantages (software, system integration, research, SME) and pairing competition with cooperation (e.g., standards and AI governance).

U.S. policy with respect to the export of certain advanced technologies has been supposedly governed by what I have called the “Sullivan Doctrine” – also known as the “small yard, high fence” framework. This policy, outlined by former National Security Advisor Jake Sullivan in a series of speeches in the fall of 2022, includes some critical formulations. First, the U.S. government has asserted that it seeks to maintain an absolute or as large as possible, versus sliding scale, lead over China in key technology areas. Second, government officials have further indicated that the United States would pursue a “small yard, high fence” approach, limiting the technologies defined as critical to U.S. national security, and erecting high barriers around these. And finally, in a somewhat contradictory formulation, they have further claimed that advanced compute technologies, including semiconductors, semiconductor manufacturing equipment, AI, quantum computing, biotechnology, and green technology, etc., were all major national security concern to the U.S. government. With that formulation, the “small yard” became quite wide, and the definition of what constitutes an “imperative” for U.S. national security became massive, covering essentially entire supply chains for semiconductors, pharmaceuticals, wind turbines, EVs, and batteries.

The effect of this formulation has been huge, both on global supply chains, U.S. and Chinese companies, trusted relationships, and bilateral relations. Beijing’s response to sweeping U.S. export controls around “advanced compute” for example, caused Beijing to respond with broad controls on critical mineral and rare earths, used across many industrial sectors, and critical to defense industrial production for critical inputs to military modernization and national security. Since Beijing imposed a licensing regime on seven rare earth elements and magnets on April 4, 2025, the impact of the broad definition of national security inherent in the “small yard, high fence” doctrine has come home to roost, with the potential blowback heavily impacting commercial and defense industrial supply chains and production, forcing the U.S. government to spend what will ultimately be hundreds of billions of dollars and many years to try and duplicate some portion of critical mineral and rare earths supply chains.

In addition, the national security definition that included semiconductor manufacturing equipment (SME), has imposed major collateral damage on U.S. technology leaders. SME, a critical input for manufacturing all types of semiconductors, including the most advanced, has been dominated historically by U.S. companies—manufacturing advanced SME is one of the few areas where U.S. companies do advanced manufacturing and provide hardware inputs to this critical industry. Most of the actual manufacturing of semiconductors is done in Asia, logic in Taiwan, South Korea, and now Japan, and memory in South Korea, Japan, and China (via SK companies). It is now becoming increasingly evident that U.S. export controls launched in October 2022 will cripple U.S. companies in the critical SME space. One of the few independent studies of this done by the New York Fed in April 2024 concluded that “export controls impose significant costs on domestic firms producing the very technologies these policies intend to protect.” An excellent CSIS study in November 2024 concluded that “current U.S. export controls risk jeopardizing the economic and national security of the United States by hindering U.S. companies’ market share and accelerating China’s relative technological gains.” I also documented some of this in papers I wrote in late 2024 and 2026 on how China’s semiconductor industry was responding to U.S. controls. Despite this, former U.S. government officials, think tank supporters, and other commentators are still suggesting that the United States impose more controls on SME exports to China. A particularly telling example of this was the recent testimony before the House Foreign Affairs Committee by a former Biden administration official.

Given the continuing fallout from U.S. technology policy -- both domestically and across global supply chains -- the massive collateral damage it has caused, and the prospect of further damage to U.S. innovation and innovation value chains in general, as both the United States and China, and other regions such as the EU pursue efforts to duplicate complex supply chains to cope in response to the “small yard, high fence” framework, is it time for a rethink? By this I mean a rethink of the concept of technology and national security, and what should constitute a real “small yard, high fence” approach. It also means a rethink of whether the seemingly limitless expansion of the national security justifications— recently illustrated by the U.S. Department of Defense asserting that large consumer facing Chinese private sector firms such as Alibaba, Baidu, BYD, and Tencent are associated with China’s military—has reached a point of clearly diminishing returns. In February the DoD pulled down an updated list of Chinese “military companies” after pressure from the White House, a sign that high level support for this type of continuation of previous policies remains uncertain. Even now, proponents of this expansionary vision of national security and technology cannot really articulate credibly what the national security gains are that offset the considerable costs of the technology control framework, because no cost benefit analysis was done for the October 2022 controls. For example, proponents of the controls continue to assert that they have slowed down the ability of Chinese companies to develop advanced artificial intelligence capabilities, even as we wake up daily to the release of more advanced AI models coming from leading Chinese AI labs, the use of these advanced models by firms and organizations in the United States and other key regions outside China, and the continued progress of China’s semiconductor industry, which is becoming ever more capable of providing the advanced AI hardware that will enable further advances in this arena. What has been slowed and what has been the gain? There are no clear answers here coming from former or senior U.S. government officials.

Export Controls: What They’re Supposed to Do and What They Are Actually Doing

An analysis of documents coming from the Commerce Department’s Bureau of Industry and Security (BIS) since the initial addition of Chinese telecommunications firms to the Entity List in the 2016-2019 timeframe, the massive controls in October 2022 and then October 2023 and December 2024 on advanced computing and semiconductor manufacturing equipment, and continuing through the addition of over a thousand Chinese firms since 2022 suggests that the goals of export controls include:

• Denying adversaries access to technologies that could enhance their military, surveillance, or cyber capabilities. This justification is typically operationalized via pointing to China’s “military-civil fusion” initiative, a fairly vague and diffuse policy that holds that China’s military should leverage advances in civilian technology, something that the militaries of all countries do––numerous Chinese sources suggest the initiative is based on the approach of the United States and other developed countries to defense industrial production. So far, there is no evidence that any military in the world is using an advanced generative AI model that would provide it with some type of a new capability decisive in a future conflict. In fact, all of the leading generative AI companies are civilian private sector firms and the vast majority of AI applications are being used in China’s civilian sector. It is in the United States that leading AI companies have openly partnered with the U.S. military and intelligence services. Despite efforts by U.S. think tanks to link DeepSeek to the Chinese military, there is no evidence that the company is doing any direct support for the Chinese government––the company is small, with just 139 engineers and researchers at the time of its R-1 model release, compared with OpenAI’s over 1000 research staff at that time.

• Preserving U.S. advantages in advanced computing, semiconductors, and AI—despite the fact that export controls, according to former senior BIS official Kevin Wolf, were not designed to maintain U.S. technology dominance. The deeper argument that U.S. officials make in the case of AI and advanced semiconductors supporting the development of AI is that there is a race to achieve artificial super intelligence (ASI), which will confer a “decisive strategic (military) advantage” to the country that gets there first. This is an argument less about the potential advantages the deployment of advanced AI will have for conventional military conflicts, and more about a theoretical future advantage—in five to ten years—that political leaders in the “winning country” will choose to wield to bring the losing country to its knees, and even force regime change. Some leading U.S. AI labs, such as Anthropic and its CEO Dario Amodei, have justified support for U.S. export controls in part by asserting that “democratic AI” must triumph over “autocratic AI”, and that this advantage can be used to force regime change, with the clear target being China.

• Shaping the global technology ecosystem in favor of trusted supply chains and shared norms, here referring to something akin to “democratic values.” U.S. officials, prior to the heavy use of export controls targeting Chinese firms, claimed that the country of origin of technology inputs should not be a consideration in how countries managed imports. Over the past decade, in the age of offensive cyber operations, this argument has been turned completely on its head, with U.S. officials arguing that the mere fact that something was manufactured or touched any part of the Chinese technology ecosystem is grounds for considering that product suspect, making it potentially exploitable by Chinese cyber operators wherever it is. This is despite the fact that there are many ways to reduce risks and ensure the security of hardware, and the fact that the vast majority of cyber intrusions occur in the cloud via software exploits rather than hardware from a particular country.

So, these are the justifications for sweeping export controls. But how have the controls––which, as a reminder, were not designed to address any of the above issues––been implemented? Here are some of the most important ways.

• Restrictions on high-performance AI hardware and semiconductor manufacturing tools. Through multiple releases of new rules from BIS, the Commerce Department has steadily ratcheted up controls on advanced AI hardware and SME, while adding hundreds of Chinese companies, some far upstream in the supply chain, to the Entity List.

• Expansion of controls to third countries through mechanisms such as the Foreign Direct Product Rule (FDPR), end-use and domestic persons controls, and the Affiliates Rule. The FDPR rule has dragged Taiwan into the U.S.-China technology competition space; the end-use and domestic persons controls dragged in the Netherlands, Japan, and South Korea; and the Affiliates Rule, now suspended, again created major issues for the Netherlands in the Nexperia case.

• Licensing for specific firms or products, such as allowing some licenses for Huawei, domestic foundry leader SMIC, and domestic DRAM leader CXMT, along with changing licensing requirements that have allowed the shipment to China of some advanced AI hardware, including NVIDIA’s A800 and H20, and potentially soon the H200, though conditions imposed on the export of H200 class GPUs as of late February had so far meant that no exports had been approved.

But what about the actual benefits versus the real costs of all this? Here is a reality check that is sorely needed. For all the justifications, arguments about—“military-civil fusion,” supposed benefits to China’s military modernization, and claims that the policy will slow China’s development of advanced AI capabilities—there has been a major failure to account for the many orders of magnitude of collateral damage caused by U.S. technology policies centered on export controls beginning in 2019.

• Outcomes since 2022 include the significant modernization and upgrading of China’s semiconductor industry. Here, constrained access to top-tier U.S. technology, both GPUs and SME, has resulted in rapid Chinese domestic substitution, the development of gray-market workarounds, and quick adaptation through coordinated national investment in the domestic semiconductor ecosystem that has in turn increased the competitiveness of leading Chinese semiconductor toolmakers outside of China, where they now compete with leading U.S. technology firms.

• The cumulative impact of the U.S. controls has forced Beijing to finally get serious about implementing a comprehensive export control regime around critical minerals and rare earths. In the making since 2022, the global licensing regime MOFCOM rolled out on April 4, 2025, and expanded on October 9 of the same year before temporarily suspending, has fundamentally changed the nature of the bilateral relationship. At a minimum, it will constrain U.S. action targeting China for a decade, despite the false hope of some administration officials that the United States can significantly reduce dependence in one year. It has also in turn forced the U.S. government to reconsider its approach to critical minerals, scramble to find alternatives, and come up with a long-term strategy that will face considerable challenges. The recent Pax Silica, Project Vault, FORGE, and Critical Minerals Ministerial initiatives are all part of this effort.

• In addition, Beijing is now unsurprisingly putting in place a mirror image of U.S. controls, including end- use controls and extraterritorial measures. MOFCOM is also unlikely to ever approve a license for companies supplying rare earth magnets and materials for military end-use in the United States. The full impact of this reaction from Beijing to U.S. export controls has yet to be felt, but it will be considerable. All U.S. missile guidance systems contain rare earth magnets, and despite U.S. government investments over the past few months, a solution to the cutoff of these key inputs remain years away.

So what is the net assessment of the controls and whether they are working in a “small yard, high fence” manner?

• Mixed outcome on incentives: the controls have demonstrably accelerated Chinese development of domestic alternatives while undercutting U.S. firms’ market share and leverage, with direct consequences for research and development budgets of leading U.S. technology companies in a critical sector.

• Confusion around enforcement of controls that are too broad has created challenges. Controls are enforced more at the extreme frontier but are leaky and insufficiently executed or monitored in practice, allowing for the creation of workarounds and legal uncertainties for companies up and down the supply chain about what is required or allowed.

• Perpetuation of a climate of uncertainty for companies across the supply chain. Uncertainty and confusion following the Affiliates Rule pause, Chinese retaliation, and the Nexperia case fallout––all related to the imposition or threat of imposition of U.S. export controls––have created mixed signals and major supply chain disruptions for major companies in the auto, medical devices, semiconductor industry and other areas across the United States, EU, and Asia.

• Increasing fragmentation of global technology networks, as other countries navigate an ever more complex set of rules and regulations. This creates a perceived need to divert limited resources into efforts to duplicate certain supply chains, reducing overall innovation and commercial viability for some companies, while stretching governments’ ability to intervene effectively to try and alter complex supply chain dependencies.

• Reduced visibility for U.S. firms and technologies into China’s evolving semiconductor and AI landscape. The export controls, particularly domestic persons controls that forced U.S. firms to pull personnel from Chinese fabs, have drastically reduced the visibility of key industry players on what is happening within China’s semiconductor manufacturing and toolmaking sectors.

One of the challenges has been the effort, also part of the “small yard, high fence” framework, to maintain an absolute lead over China in key technology domains. Part of this effort has been an attempt to define what “advanced” technology means. But with technological frontiers constantly advancing, and government processes continuing to move slowly, the impact of this approach has been primarily to create further damage across supply chains and to U.S. companies’ ability to operate in markets for “less-than-‘advanced’” technologies. With no clear definitions of “advanced” and no effort to change definitions over time consistent with industry understanding, the “small yard” ends up expanding massively, resulting in additional collateral damage, and no clear U.S. national security gains from each expansion.

The problem is that the definition of what constitutes “advanced or cutting-edge” technology was never clearly established under the framework, and the technology has continued to evolve––in some cases rapidly––quickly overcoming attempts to draw hard lines. For example, the October 2022 controls set the bar for “advanced” node production at 16/14 nm for semiconductor logic. Even at that moment, no one in industry considered 16/14 nm “advanced or cutting edge”—at this point 7 and 5 nm were already where leading industry players were producing “advanced” semiconductors. The same held true for memory end-use controls at 128 layers for NAND and 18 nm pitch for DRAM.

• Historically, in the nascent stages of export control policy, restrictions were based on straightforward technical parameters. These measurable indicators reflected an era when technological capability could be defined by hardware specifications alone, and most parameters did not change significantly from year to year.

• However, applying this approach to dynamic, global, and fast-developing industries such as semiconductors has clearly struggled to capture the true nature of technological capability and competition today.

• The boundary between “restricted” and “non-restricted” technology has therefore remained fluid, shifting in response to the broader strategic competition with China. A good example is AI training versus inference. U.S. controls in October 2022 were aimed at the most advanced GPUs for training frontier AI models. But by October of 2023, U.S. officials realized that the GPUs NVIDIA had modified to meet the performance levels of the October 2022 controls were no longer adequate, and they changed the goalposts. Later, after NVIDIA made another modification to meet the October 2023 performance parameters, U.S. officials again realized that the modified H20 GPU was ideal for AI inference, which was becoming a more important part of the overall operating picture for advanced AI models, and in April 2025 the Trump administration issued “is informed” letters to NVIDIA, AMD, and Intel that exports of previously approved GPUs would require licenses, with presumption of denial—all of this within only two years.

• This technological evolution has moved the export control narrative beyond its original intended purpose of regulating specific technologies for unclear long-term strategic goals toward serving also as a broader instrument of geopolitical and industrial strategy. Under the Trump administration, policymakers with strong technology backgrounds, such as White House AI Czar David Sacks, have questioned the logic of maintaining an absolute lead, rightly pointing to the downsides of such a policy, namely massively incentivizing the Chinese industry to replace dependence on U.S. hardware and software, and creating competition for the U.S. AI stack outside of China. Sacks and others, including, of course, NVIDIA CEO Jensen Huang, have argued that the United States should allow sales of less-advanced GPUs to China, to keep Chinese AI developers “addicted” to portions of the U.S. AI stack. Huang has highlighted the fact that nearly half the world’s AI developers are in China.

• In addition, by including memory in the October 2022 controls, the United States has both disadvantaged U.S. tool makers in a major way, and paved the way for Chinese domestic tool makers to improve and compete outside China for both memory and logic. Here, as a result of controls, every fab in China that is restricted from getting U.S. and other foreign tools becomes a technology incubator for Chinese domestic toolmakers. It turns out the memory is easier than logic, does not require the most advanced lithography, and the memory fabs in China are also now not only already using domestic tools for memory production, but also serving as incubators for advances that can be applied on the logic side also—a trifecta.

• Finally, the inclusion of memory has resulted in major negative consequences for production capacity just as a memory supercycle has driven demand for both NAND and DRAM to historic highs. This AI infrastructure buildout-driven cycle is set to last for years, and by crippling the ability of Korean memory makers to expand much needed capacity in China, and slowing the progress of innovation and capacity expansion at YMTC and CXMT, U.S. export control policies have taken offline major capacity to meet the needs of AI development globally, while also impacting other major consumers of memory such as PC, auto, and electronics companies. This self-inflicted wound is likely to have significant economic impacts in 2026.

Given the change of attitude in Washington, it would appear that in pushing back against China critics and former Biden administration arguments that the United States should stick to the absolute rule and deny Chinese firms all GPUs, a clearer and more adaptive definition of what constitutes “advanced” hardware is needed to minimize the damage to U.S. firms and articulate clear strategic objectives. We are at a point where the major limitations of the “small yard, high fence” framework are now glaringly apparent, and a new formulation is required. U.S. industry is well versed in complying with U.S. export control laws, particularly in the weapons of mass destruction (WMD) era, when definitions were clear and policy goals even clearer. The framework, however, used broadly against “advanced compute”, with poorly defined technology and strategic goals, has proven to have major liabilities for U.S. firms, global supply chains, and bilateral relations. For the latter in particular, the policy has resulted in the generation of a major counterreaction from Beijing that could cripple the U.S. defense industrial base––a possible outcome no one in Washington foresaw in the fall of 2022, but which should have been anticipated.

So what are the challenges in defining how to think about “advanced” or “cutting-edge” technologies -- what risks they currently and potentially pose, and what the actual boundaries of export controls should be—given the complexity of global supply and value/innovation chains, and the rapidly changing nature of many technologies that served as key inputs to the AI stack more broadly?

• AI capability depends increasingly on system-level performance: how GPUs interact across clusters, how efficiently they use memory and software, how these efficiencies translate into real-world economic applications, and how quickly they diffuse globally.

• The manufacturing and supply-chain dynamics of this industry create natural time delays: what counts as “advanced” or “cutting-edge” when restrictions are drafted may already be outdated by the time those technologies meaningfully reach the Chinese market. This lag complicates enforcement and undermines the long-term relevance of fixed control categories.

• China’s AI strategy challenges the logic of restricting the “cutting-edge” of “advanced GPUs.” China’s AI blueprint, as outlined in the AI+ Initiative Implementation Plan and other documents, with a 2030 goal of 90% integration of AI into the economy, focuses on deploying AI for economic growth. This approach prioritizes deployment to address real “pain points” over the rapid expansion of compute capacity using the most advanced GPUs. Chinese firms can achieve “good enough” outcomes by using larger quantities of less advanced GPUs that fall below U.S. export thresholds.

• The criticality of AI diffusion for China’s economic development also makes U.S. controls––aimed at keeping Chinese firms frozen at a level the United States predetermined as “advanced” but no longer clearly so––appear to be mostly, if not entirely, designed just to slow China’s economic development rather than U.S. national security. This fuels the hardliners within the Chinese system, those calling for decoupling, and complicates diplomatic negotiations that now include similar arguments being made by the United States on China’s use of critical minerals controls.

• When a senior Chinese trade official asserted in late October 2025 that “we cannot have rare earths coming back to us in missile guidance systems,” U.S. officials became uncomfortable. But Beijing is arguably attempting to narrow the use of its controls to clear military end-users, while U.S. controls target firms far back in the semiconductor supply chain via an expanding––and hardly small–– “yard,” without clear national security justification.

The H20, B30, and H200 Debate: Call for Going Back to Sliding Scale?

The debate within the Trump administration, which includes representatives with deep technical and industry background such as Sacks, who are sympathetic to arguments around the need to keep allowing Chinese firms access to U.S. GPU technology, is crystallized by the H200 debate. After President Trump approved licensing of the H20 GPU, based on the older Hopper architecture––which had been legal to sell to Chinese customers since October 2023––Beijing reacted badly to the “addiction” rhetoric coming from U.S. officials, and likely issued internal guidance to Chinese firms to not buy the H20. There was brief discussion about a so-called “B30”, allegedly built on NVIDIA’s Blackwell architecture, but Nvidia never actually designed this downgraded chip for China sales. The Blackwell systems are currently the most advanced available and will be deployed through 2026 by U.S. companies and other western AI developers, with the next generation Rubin architecture ready later in 2026. If Nvidia had gone through with the “B30”, this would have represented a third attempt by NVIDIA to produce a GPU that qualifies for export to China.

Twice before, NVIDIA designers have been forced to redesign––at great cost to the company––a modified version of the company’s latest architecture, to comply with U.S. export thresholds while remaining saleable in China. However, although the performance parameters defining GPUs as advanced or cutting-edge were changed in October 2023 and again sometime in 2025, the updated criteria remain unknown because BIS has not published a document explaining what the parameters would be for any new GPUs designed to comply with changes in the controls resulting from President Trump’s decision in May 2025 to allow H20 sales to China. Then in December 2025, President Trump decided to end the practice of setting parameters that forced Nvidia to downgrade an existing GPU for export to China, when he approved sales of the Hopper-architecture H200 class GPUs to China. The issue has gained political visibility after reports that the “B30” had been discussed during the trade negotiations throughout the summer of 2025, turning it into a focal point of debate in both Washington and industry circles.

Competing arguments around the issue of H200 sales to China now illustrate the broader dilemma.

• Proponents argue that allowing limited H200 sales and a product with similar performance from AMD would maintain the presence of U.S. firms in the Chinese market, which supports their research and development budgets and sustains U.S. jobs. They also argue this would keep Chinese companies using the U.S. AI stack. They also contend that continued engagement preserves U.S. insight into China’s evolving AI ecosystem and provides a calibrated alternative to blanket bans that can and have provoked Chinese retaliation.

• Opponents counter that even these downgraded GPUs could meaningfully enhance the ability of Chinese AI developers to do both training and inference. They warn that permitting such exports risks creating a slippery slope that blurs the line between national security and commercial interests. This despite the fact that the GPUs would be licensed and the only allowed customers would be leading Chinese private sector firms developing AI models, for overwhelmingly if not exclusively civilian applications, including Alibaba, Tencent, Baidu, Bytedance, Moonshot, Zhipu, and DeepSeek.

• This argument gained renewed traction in Washington in early December as Huang and Sacks were able to persuade President Trump to allow sales of the H200 GPU, given that it is already a full generation—more than 18 months—behind the cutting edge. Whether Beijing will welcome the move and not seek to discourage or bar Chinese AI developers from importing the H200 remained unclear as of late February. The move is important as here Trump has overriden the concerns of former Biden officials, including former Biden White House AI Czar Ben Buchanan, who has argued against sales of even the deprecated H200, the H20. But even sales of the H200 would still not “hand China a major AI win that erodes the U.S. lead in AI”, as many have argued.

• In late January, the Commerce Department and BIS issued a rule around the conditions for export of the H200 class GPUs to head off the sales of advanced GPUs to China that went against the intention of the President’s December statement on the issue. The conditions included some that Beijing would not likely let Chinese companies comply with, such as revealing customers, while also likely limiting the ability of H200 class GPU purchasers to use them in data centers outside of China.

Here, going forward, U.S. policy choices can be organized into three main scenarios:

• Scenario 1: Do not allow any exports of advanced AI hardware or SME to China.

• Scenario 2: Allow exports of both AI hardware and SME under existing thresholds.

• Scenario 3: Establish a middle ground for AI hardware and SME with a clear sliding scale of performance that comports with industry realities, is acceptable to the industry, and is based on deep understanding of technology trajectories across an evolving AI stack, along with additional and flexible guardrails and transparency mechanisms.

For each scenario, U.S. policymakers should assess:

• The effects on China’s AI development trajectory.

• The effects on NVIDIA, AMD, Intel, Broadcom, Qualcomm, Applied Materials, Lam Research, KLA, and the broader U.S. technology sector, and on ASML, TEL, Samsung, SK Hynix, and other companies in allied countries.

• Broader implications for U.S.–China technological competition and strategic stability.

The Current Landscape of Chinese GPUs

In order to adopt a new policy approach which narrows the scope of controls and clearly defines what “advanced” AI hardware means, it is necessary to understand where Chinese players across the hardware part of the stack stand. China’s domestic semiconductor sector has seen rapid progress over the past 30 months. Leading players such as Huawei and new and capable players such as Moore Threads, MetaX, Enflame, and Biren, are developing increasingly competitive products tailored to domestic AI firms’ needs. These companies are supported by strong local demand, a large engineering workforce, and government incentives.

China’s progress is characterized by growing domestic ecosystem integration and a strong emphasis on open-source and open-weight development models. Innovation has been driven by necessity, leading to creative approaches that compensate for hardware limitations. Examples include the use of mixture-of-experts (MoE) architectures and low-precision computing methods to stretch available computing power. Chinese firms are also investing heavily in compiler and framework optimization to improve the performance of locally designed chips and software stacks.

Chinese industrial policy planners continue to adapt and provide significant support for the country’s semiconductor ecosystem.

• Market-based Financing: Active pipeline of GPU firms pursuing new funding rounds and public listings. Capital markets have become more supportive, with more lenient listing rules and increased access to domestic equity financing.

• 15th Five-Year Plan: the new FYP, which is expected to finalize in March 2026, emphasizes technological self-reliance, offering strong policy backing for semiconductor development. Government support spans infrastructure investment, R&D support, energy cost reductions, etc. These measures aim to accelerate domestic capability building and ensure sustained progress in building out a supportive ecosystem for AI.

Policy Choices Around Export Controls Need to Be Redone

Given these developments, and the fact that all Chinese semiconductor industry players are now operating beyond the end-use controls that defined “advanced” in October 2022, there is a growing sense within the industry that there should be a redefinition of “advanced” and “cutting-edge” for both GPUs and SME. This would involve significant re-tooling of U.S. export controls targeting China. The path here though has already been shown by the decision in December by President Trump to allow licensing for the H200 class GPUs. This is clearly not a cutting-edge or advanced GPU, it is based on the now outdated Hopper architecture. The “cutting-edge” of GPU design in 2026 will be the Blackwell architecture, which will give way later in 2026 to the Rubin architecture, in 2027 to the Rubin Ultra, and then in 2028 to the Feynman version. Redefining cutting edge GPU architecture as the Rubin or Feynman level could go a long way to providing clarity on U.S. policy with respect to advanced compute, and mark a significant shift away from the failed “small yard, high fence” doctrine.

At the same time, given that H20 (and H200) are products of a 5 nm process at TSMC, using HBM3 (and H200 using HBM3e), the definition of “advanced” semiconductor nodes––and hence SME associated with these nodes––should be completely revamped. Setting a new advanced node definition, for example, at 3 or 2 nm, which is where the industry is, would make sense here. Like GPU architecture, future GAAFET and CFET architectures are already in development, highlighting the rapidly changing outward movement of the “cutting-edge,” which of course comes with some lag and overlap with previous generations, but this is the way the industry works, and slavishly holding to the 16/14 nm definition of “advanced” semiconductors nodes in 2025 makes little sense, if it ever did. A simple acknowledgment by the Commerce Department and U.S. officials of this reality would be a welcome step, given the degree to which Chinese firms are already operating well beyond those “limits” and the degree to which the industry has changed since October 2022. Continuing to deny the reality of rapid technological advancements is not useful for making progress either on national security concerns or avoiding further blowback to U.S. industry in terms of R&D losses and Chinese retaliation.

There are a number of pragmatic ways to update the current outdated export control framework to reflect requirements of new technologies and the global industry realities.

• Establish a standing industry technical advisory panel to consult with experts and industry participants to review the state of the industry and the full implications of any rule changes. This would help align regulatory measures with the realities of global innovation.

• Institutionalize regular review and updates of export control thresholds in light of technological advances, the changing commercial landscape, and new and emerging technologies. Assess the economic impact of current rules to ensure they remain technically current and strategically relevant.

• Enhance interagency coordination to ensure consistent communication and application of rules. This is especially important as China possesses multiple counter-leverage tools, such as rare earth exports, and its centrality in the global supply chain remains a reality that cannot be ignored. Agencies should consult rigorously to minimize market disruptions and prevent unintended damage to the U.S. economy and defense industrial base.

• Strengthen and narrow monitoring and execution. Modernize tracking systems to improve visibility, enforcement efficiency, and data-driven oversight of export control compliance. By narrowing the window of monitoring––for example, not devoting resources to the smuggling of outdated GPUs to China, such as the A100 and A800 (and now the H200)––the actual enforcement or rules targeting only a narrow and sliding subset of advanced hardware would reduce the burden of export control enforcement.

Looking Ahead: Competition Beyond Technology Controls...

Over the course of 2025, U.S.-China relations and the centrality of export controls to the trade negotiations and overall state of the relationship highlighted the need for a major rethink of how these tools are used. In the case of the United States, tools designed for containing the proliferation of weapons of mass destruction (WMD) during the Cold War have proven inadequate and disruptive to global supply chains in the age of AI. For China, using controls on critical minerals for national security goals is new, and Beijing must be careful not to overshoot on the controls and disrupt commercial supply chains that require long-term, stable, sustainable access to critical inputs.

The United States and China will remain systemic competitors in advanced technology, but competition need not preclude cooperation in areas of shared global concern. Significantly narrowing the scope of controls through a redefinition of what is “advanced,” along with clarity about the ultimate goals of controls and acknowledgement that both sides have legitimate national security goals in using export controls, would go a long way toward avoiding further disruptions to supply chains, innovation, and business operations for industry, while maintaining the focus on real risks and national security priorities.

The United States should continue to compete vigorously in strategic technologies while seeking avenues for cooperation in fields where mutual stability and global welfare are at stake. These areas include data standards and responsible AI governance. Establishing communication channels for cooperation in such areas can reduce the risk of miscalculation and help maintain transparency in a deeply interdependent global system. Here, the willingness of Beijing to participate in new forms of collaboration––bilateral dialogues on AI governance, the scope of national security controls, and other key digital economy issues such as data governance––could be enhanced by a U.S. willingness to narrow the definition of national security related to technology, to avoid giving Beijing the impression that U.S. efforts in this domain are really a pretext for slowing China’s economic development.

The NVIDIA H20/B30/H200 controversy underscores a defining tension in U.S. technology policy. Export controls cannot serve as permanent barriers; at best, they are temporary instruments that must be situated within a broader long-term strategy of renewal. Technology associated with advanced compute and semiconductor manufacturing equipment is applied science: trying to lock it into a permanent box and assuming that Chinese companies will not find another way to apply it or will not find system-level solutions to overcome the underlying limitations on raw computing power appears short-sighted and fails to recognize the core drivers of innovation.

On memory, the best approach would be to lift all the controls, which have only served to disrupt global supply chains––including major Samsung and SK Hynix operations in China––while damaging U.S. companies’ ability to conduct R&D, and jump starting China’s toolmaking industry in a major way. In a classic example of how the memory controls could backfire in a big way, U.S. officials in January put in place another temporary replacement for validated end-user (VEU) status for the big South Korea memory makers in China, as these firms unsurprisingly want to expand production of DRAM and NAND in response to a massive increase in demand driven by the buildout of AI data centers. But to do this, U.S. officials must grapple with the growing contradiction that, for Korean facilities in China, they will have to approve licenses for tools to allow these firms to expand––tools that will go way beyond the end-use controls for DRAM and NAND applied to Chinese firms. But the market follows demand, and damaging the ability of global leaders such as Samsung and SK Hynix to meet AI-driven demand would ultimately work against U.S. interests.

For companies, removing all controls on memory and SME for memory—including reconsidering the Entity List designation for YMTC—would be the rational response to the realities of the memory sector. YMTC is suing in court for the release of documents at BIS related to the deliberations about putting the firm on the list. These documents remain unclear and do not appear to have been related to any clear national security concerns—the most plausible reason, noted in the Entity List document, that YMTC could “divert” NAND memory for consumer devices at Huawei, such as smartphones—does not hold up to close scrutiny, it is almost tautological: putting a company on the Entity List so it cannot supply another Entity List company with products for consumer devices that have never been identified as a threat to national security. Most of the semiconductor industry has never supported the memory end-use controls, and most in the industry do not support putting companies like YMTC and CXMT on the Entity List for dubious reasons that mostly damage U.S. companies. Lifting the memory controls and Entity List designations for memory companies would not be a goodwill gesture or a “win” for China, but an acknowledgement that the implications were not considered, that they serve no national security use, and that they have mostly damaged U.S. companies.

A better approach overall than reliance on controls would be to focus on sustaining leadership in the domains where the United States retains a comparative advantage, including software ecosystems, system integration, and foundational research, while ensuring that any outbound restrictions are flexible enough to evolve with technological and market realities. At the same time, a better approach would be to recognize that the United States also maintains a comparative advantage in SME and the ecosystems that support U.S. companies. Ensuring that U.S. SME leaders have access to such a major market as China is imperative to keeping these companies viable globally, and the sooner this reality is recognized, the better, given the already damaging impact of U.S. controls on this sector and the trendlines this portends.

The issue of strengthening allied coordination is complex and also requires a rethink. While the effectiveness of U.S. controls depends on global alignment, the goals of the controls and the ability to keep them narrow and focused on real national security concerns is critical to gaining that alignment. Broad, unclear, and inflexible controls are unlikely to gain support from countries such as Japan and the Netherlands, where SME companies are national champions and major employers. This has already been the case for more than three years, as governments in Tokyo, The Hague, and Seoul in particular continue to chafe under U.S. controls imposed unilaterally, followed by strong arm tactics to gain some alignment. Unilateral restrictions applied too broadly and without clear justification have proven to be less effective, and the result has been that U.S. allies or partners continue to supply alternative pathways for Chinese access to some restricted technologies––neither Japan nor South Korea, for example, have ever agreed that memory, NAND and DRAM, should be the subject of end-use controls, let alone static end-use controls in an industry without significant legacy production.

The United States should work toward a more realistic and structured coordination mechanism with Japan, the Netherlands, South Korea, Taiwan, and other countries that control only a limited number of critical segments of the semiconductor supply chain, such as lithography and advanced packaging. Cooperative efforts should aim to develop consistent control thresholds and shared enforcement standards while ensuring that measures remain economically and politically feasible for allies. U.S. policies should also be sensitive to the potential for U.S. arm-twisting of allies to impose broad controls without clear justification, which raises the risks of Chinese retaliation against those countries. This is another reason for narrowing the controls and ensuring the definition of “advanced” does not just continue to uncontrollably expand beyond limits that companies in allied countries believe are reflected in the evolving technological and commercial reality.

...And Moving Away from Zero-Sum Thinking Toward Flexible Cooperation

With a narrower definition of national security and technology, the United States should also pursue the expansion and strengthening of institutional capacity to help sustain a more narrow, dynamic, and flexible export control system. Allowing U.S. companies to compete in the dynamic and expanding China market, under new and much narrower limitations, would be a corollary to this and feed into a second policy priority. Even narrowed export control restrictions should not become a substitute for a real long-term strategy of innovation. In the wake of the Biden-era CHIPS Act, the Trump administration should also consider expanding and sustaining incentives for innovation across the U.S. semiconductor ecosystem.

At the same time, as the AI stack becomes the real focus of U.S. government domestic technology policy, the administration should build on the U.S. AI Action Plan call for supporting the development of open and secure AI frameworks that allow U.S. firms to shape global standards. But this should not be viewed as a zero-sum game. Chinese firms, developing open-source/weight models using less advanced U.S. AI hardware on a sliding scale are becoming and will become the preferred suppliers to some developing countries. This is not in any way a bad thing or a strategic threat.

By allowing exports of U.S. AI hardware, Chinese developers will continue to be tied into the U.S. AI software ecosystem, and those Chinese models, appropriate for the financial resources of small- and medium-sized companies in developing markets, can still run on U.S. hardware, or Chinese hardware that is less expensive and appropriate for some applications and workloads. Abandoning the zero-sum thinking of some proponents of endless export controls would be conducive to acknowledging real-world situations that are already happening, and avoid creating areas of contention and competition between Washington and Beijing where collaboration would benefit all parties.

This more flexible and less zero-sum approach would also improve the outlook for new efforts in the U.S. to fund programs to attract, retain and work with global talent through research grants, immigration reforms, and academic–industry partnerships. These programs should not be U.S.-centric or involve only U.S. technology ecosystems. To be successful in underserved markets they should include support for the deployment of elements of the Chinese AI stack as appropriate and not be viewed as a zero-sum game where only the U.S. AI stack should “dominate”. This outcome in a world of competitive Chinese AI models and AI hardware is unlikely to happen in any case, and simply turns every region into a zero-sum battleground where no country wants to play the game. To this end the United States could encourage regional innovation clusters in developing countries that link semiconductor production with next-generation applications such as quantum computing, biotechnology, and energy systems, while including both U.S. and Chinese firms in the mix, to avoid forcing countries to choose sides and allowing appropriate solutions that draw on the strengths of companies in both countries.

Overcoming the clear opposition to a deeper rethink of U.S. controls will be challenging. But if the U.S.-China relationship is to be put back on real solid footing, and we are to avoid a constant back-and-forth on export controls around semiconductors, SME, rare earths, magnets, and critical minerals, that could expand to other technologies and sectors in the event of even a small breakdown in the bilateral relationship, a profound rethinking in particular of U.S. export controls will be needed. Here, the Trump administration will be better served by listening to proven industry experts as well as to the increasing number of analyses that highlight the shortcomings of previous export control frameworks that have failed to present clear costs and benefits, expanded well beyond their intended application domain, and created clear and mounting levels of collateral damage. Now is the time to rethink—before Beijing has time to resource its own export control apparatus on rare earths and critical minerals, and likely beyond—while the trade truce provides a modicum of breathing space to step back and reconsider a better approach than the “small yard, high fence” doctrine - one that actually takes into account the path and realities of commercial and technological development in the Age of AI.