Do export controls slow down a restricted country's AI progress?

Real Near-Term Friction, Less Certain Long-Term Effect

The evidence generally supports the idea that export controls create genuine near-term obstacles for a restricted country trying to access the most advanced AI chips, which can meaningfully slow its ability to train the largest, most computationally demanding AI models at the same pace as countries with unrestricted access. Acquiring cutting-edge chips through normal commercial channels becomes more difficult, more expensive, or simply unavailable, which has real consequences for organizations that depend on that hardware for large-scale AI training efforts.

However, the picture becomes considerably more complicated when you look beyond the immediate, near-term impact. Restricted countries don’t simply stop trying to develop advanced AI capabilities; they typically respond in ways that can partially offset, or in some cases potentially overcome, the disadvantage created by restricted chip access.

How Restricted Countries Typically Respond

One common response is accelerated investment in domestic semiconductor manufacturing capability, aimed at reducing long-term dependence on foreign chip suppliers entirely. This is a difficult, capital-intensive, and time-consuming undertaking, since advanced chip manufacturing involves deeply complex technical processes and specialized equipment that isn’t easily replicated quickly. But sustained investment and government support can meaningfully accelerate this kind of domestic capability development over time, potentially narrowing the gap that export controls were designed to maintain.

Another common response involves adapting AI research and development approaches to work more efficiently within existing hardware constraints. Researchers facing restricted access to the most powerful chips may focus more intensively on techniques that require less raw computational power, such as more efficient model architectures or training methods, potentially achieving meaningful AI capability with less reliance on the most advanced hardware than would otherwise be the case.

Why the Net Effect Remains Genuinely Uncertain

Given these dynamics, whether export controls ultimately produce a lasting slowdown in a restricted country’s overall AI progress, or instead simply shift the pathway that progress takes, without necessarily reducing its eventual pace, is a genuinely open and actively debated question among analysts. Some evidence supports meaningful near-term impact, while other evidence suggests restricted countries are finding ways to adapt that could reduce the long-term effectiveness of these controls. This uncertainty is part of why export control policy remains a continually revisited and adjusted area of government action rather than a fixed, one-time decision.

Bottom Line

Export controls do appear to create real near-term obstacles for a restricted country’s access to the most advanced AI chips, which can slow certain kinds of AI progress. However, restricted countries often respond with accelerated domestic chip investment and more hardware-efficient AI research approaches, making the long-term net effect on their overall AI trajectory considerably less certain than the near-term impact alone might suggest.

Go deeper

• What Is the Difference Between Near-Term AI Risks and Long-Term Existential Risks?