Export Controls for Artificial Intelligence

Artificial intelligence (AI) has rapidly emerged as a transformative technology with significant implications for economic growth, scientific advancement, and national security. The U.S. Department of Commerce's Bureau of Industry and Security (BIS), along with its Emerging Technology Technical Advisory Committee (ETTAC), has been evaluating the implications of AI export controls. A March 10 ETTAC public meeting highlighted some of the challenges, strategies, and policy directions to manage the proliferation of AI-related technologies while defending the West’s competitive edge.

Export Controls in AI Development

Export controls serve as a key regulatory tool to limit the transfer of strategic technologies to foreign adversaries. AI, which depends on advanced semiconductors, vast datasets, and high-performance computing infrastructure, presents unique challenges in enforcement. Traditional export controls, designed for tangible goods, must adapt to AI’s intangible nature, where software models, cloud-based inference, and research publications contribute significantly to capabilities.

AI Capabilities and Risks

Participants acknowledged the need for improved benchmarks to assess AI risks. AI models' ability to accelerate scientific discoveries, such as DNA sequence modeling and predictive chip fabrication create security risks. Companies like OpenAI and Anthropic have implemented safeguards to limit access to AI-driven knowledge in sensitive domains such as biotechnology and cyber warfare. However, committee members noted that small-scale fine-tuning can bypass these controls, making proliferation more challenging to prevent.

The Imminence of Artificial General Intelligence (AGI)

Forecasting models now predict that artificial general intelligence (AGI), defined as human-level intelligence capable of performing most laptop-based jobs, may be achieved as early as 2026. According to meticulous.ai, a leading forecasting platform, AGI development is accelerating at a pace that challenges existing regulatory frameworks. This projection underscores the urgent need for export controls that anticipate and mitigate national security risks associated with AGI-capable models.

Challenges in AI Export Controls Enforcement

Despite U.S. efforts to restrict access to advanced AI chips, evidence suggests that export controls have not effectively curtailed supply chain leakages. Data presented at the meeting indicated that NVIDIA’s revenue from the Sinosphere—including Taiwan, Malaysia, and Singapore—remains robust, matching U.S. domestic sales. These findings suggest that semiconductor diversion is occurring, undermining policy objectives.

Further complicating enforcement is the lack of effective end-user verification. Companies operating in adversarial regions have exploited regulatory loopholes, redirecting AI hardware and software through third-party intermediaries. Industry earnings calls have highlighted cases where executives decline to comment on diversion activities, raising concerns about compliance transparency. To address these gaps, policymakers must explore new verification mechanisms, including location-based tracking and digital safeguards embedded in AI models.

Conditional Export Controls and Tier II Diversion Verification

One of the most discussed recommendations at the ETTAC meeting was the concept of conditional export controls. This would involve imposing restrictions not only on direct exports but also on secondary markets, requiring AI chip vendors to verify user identity and location before granting access to high-performance computing resources. This approach would include:

• Country-Based Control Mechanisms – Implementing stricter verification procedures for high-risk countries suspected of facilitating unauthorized AI development.
• Wafer-Level Security Features – Embedding tamper-proof hardware mechanisms within AI chips to enable location verification and usage restrictions.
• Tier II Diversion Verification – Enhancing oversight of intermediary markets, including Singapore and Malaysia, where AI chips are often diverted before reaching restricted entities in China.

AI Dominance and Market Manipulation

China’s strategic investments in AI infrastructure present a direct challenge to U.S. efforts to maintain technological superiority. Initiatives like China’s $138 billion state-backed AI data center program and rapid adoption of the NVIDIA H20 chip underscore its commitment to achieving self-sufficiency in AI development. While the U.S. has imposed stringent chip export restrictions, Chinese firms continue to innovate by optimizing computational efficiency and leveraging domestic semiconductor production.

Sam Hammond, Chief Economist at the Foundation for American Innovation noted that inference chips—used for deploying AI models—are significantly easier to develop than training chips. He warns that China may attempt to flood the global market with inexpensive inference chips, leveraging economies of scale to establish hegemony in AI deployment. Such a strategy would not only diminish the impact of U.S. export controls but also enable China to shape global AI infrastructure to its advantage.

Four Pillars of AI Dominance

According to Hammond, AI supremacy hinges on four critical elements:

1. Algorithms – Advanced machine learning architectures that optimize AI performance.
2. Data – Access to vast, high-quality datasets for training and fine-tuning models.
3. Energy – Scalable and affordable power sources to sustain AI compute demands.
4. Chips – High-performance semiconductors for training and inference workloads.

While the U.S. currently leads in algorithm development and semiconductor design, China's superior access to energy and its ability to accumulate large-scale data sets presents a growing challenge.

Energy Constraints and AI Scalability

One of the most pressing concerns raised during the meeting was the growing energy demand of AI infrastructure. Leading AI firms in the U.S. are planning gigawatt-scale data centers to support AI workloads, yet grid modernization efforts remain sluggish. While China continues to expand its power generation capacity, adding 429 gigawatts in the last year alone, U.S. energy constraints may pose a long-term bottleneck for AI growth.

Jevons' Paradox

Jevons' Paradox states that as technological efficiency increases, demand for resources often rises instead of declining. This phenomenon is evident in AI computing. As AI models become more efficient, reducing the cost per computation, overall demand for high-performance computing surges. The introduction of AI models like DeepSeek, which reduce inference costs, has led to a sharp increase in GPU utilization, further straining available hardware and energy supplies. This paradox highlights the need for strategic planning in AI infrastructure to prevent supply shortages and ensure sustained U.S. competitiveness.  

Dario Amodei's take

In an influential January article Anthropic CEO Dario Amodei argues that recent advancements by DeepSeek, a Chinese AI company, do not weaken the case for U.S. export controls on AI chips but instead reinforce their necessity. 

"Export controls serve a vital purpose: keeping democratic nations at the forefront of AI development. To be clear, they’re not a way to duck the competition between the US and China. In the end, AI companies in the US and other democracies must have better models than those in China if we want to prevail. But we shouldn't hand the Chinese Communist Party technological advantages when we don't have to.

"Making AI that is smarter than almost all humans at almost all things will require millions of chips, tens of billions of dollars (at least), and is most likely to happen in 2026-2027. This means that in 2026-2027 we could end up in one of two starkly different worlds. In the US, multiple companies will definitely have the required millions of chips (at the cost of tens of billions of dollars). The question is whether China will also be able to get millions of chips.

"Well-enforced export controls are the only thing that can prevent China from getting millions of chips, and are therefore the most important determinant of whether we end up in a unipolar or bipolar world.

"

Given my focus on export controls and US national security, I want to be clear on one thing. I don't see DeepSeek themselves as adversaries and the point isn't to target them in particular. In interviews they've done, they seem like smart, curious researchers who just want to make useful technology.

"But they're beholden to an authoritarian government that has committed human rights violations, has behaved aggressively on the world stage, and will be far more unfettered in these actions if they're able to match the US in AI. Export controls are one of our most powerful tools for preventing this, and the idea that the technology getting more powerful, having more bang for the buck, is a reason to lift our export controls makes no sense at all."

On DeepSeek and Export Controls