Advanced AI requires advanced chips, and advanced chips depend on a remarkably concentrated global supply chain. This has made semiconductors a central arena of geopolitical competition, especially between the United States and China. Export controls — legal restrictions on selling certain technologies abroad — are the main policy tool.

Why chips are strategic

Training and running frontier AI models requires large numbers of cutting-edge accelerators. Producing them involves:

  • chip design (dominated by US companies such as NVIDIA and AMD);
  • manufacturing at the leading edge (concentrated at TSMC in Taiwan, with Samsung and Intel also producing advanced chips);
  • lithography equipment (the most advanced extreme ultraviolet machines come from ASML in the Netherlands);
  • high-bandwidth memory (from a few suppliers, mainly in South Korea);
  • advanced packaging, design software and specialized materials.

Each step has few suppliers, creating “chokepoints” that governments can use.

How the controls work

Beginning in October 2022, the US Department of Commerce’s Bureau of Industry and Security (BIS) introduced sweeping rules restricting exports to China of:

  • advanced AI chips above certain performance thresholds;
  • semiconductor manufacturing equipment for advanced nodes;
  • support from US persons for advanced chip production in China.

The rules have been updated several times, tightening thresholds and closing loopholes. Chipmakers responded with lower-performance versions designed to comply, some of which were later restricted too. Allied countries, including the Netherlands and Japan, introduced their own restrictions on chipmaking equipment.

Goals and debates

Supporters argue the controls slow the development of military and surveillance AI by strategic rivals and protect national security. Critics and skeptics raise several points:

  • Effectiveness: smuggling, cloud access and stockpiling can undermine controls.
  • Domestic acceleration: restrictions give targeted countries strong incentives to build their own chip industries.
  • Commercial cost: companies lose access to a large market, reducing revenue that funds research.
  • Efficiency innovation: constrained developers may find ways to train strong models with fewer or weaker chips.

Industrial policy

Alongside controls, governments invest heavily in domestic production: the US CHIPS and Science Act (2022), the European Chips Act (2023), and large programs in China, Japan, South Korea and elsewhere. Building leading-edge fabs takes years and tens of billions of dollars.

Beyond chips

Geopolitical competition in AI also involves talent, data, energy for data centers, model access and standards-setting. “Sovereign AI” initiatives, in which countries build national AI infrastructure and models, are part of the same picture — see sovereign AI explained.

Effects on companies and markets

Export controls affect chip designers, manufacturers and their customers. Chipmakers have designed modified products to meet thresholds for restricted markets, only to see some of those products restricted later as rules changed. Companies in affected countries have accelerated domestic chip development and stockpiled equipment. Cloud providers must check where customers are and what they use computing for. For investors, the rules add uncertainty to revenue forecasts for chip companies, as access to large markets can change with policy.

For context, read why NVIDIA dominates AI, sovereign AI explained and why big tech builds its own AI chips, which explain the technology and strategy behind the policy debate.

Bottom line

Export controls are reshaping where advanced AI chips are sold, how they are designed and how countries invest in domestic alternatives. Expect the rules to keep changing.

What to watch next

  • Updates to chip performance thresholds and country-specific rules.
  • Progress of domestic chip manufacturing in China and elsewhere.
  • Enforcement against chip smuggling and remote access loopholes.
  • New fabs coming online under industrial-policy programs.

Key takeaways

  • Advanced AI depends on a concentrated supply chain with few suppliers at each step.
  • Since October 2022, US export controls restrict advanced AI chips and chipmaking tools to China.
  • Allies such as the Netherlands and Japan restrict key equipment as well.
  • Debates focus on effectiveness, domestic acceleration and commercial costs.
  • Industrial policies such as the CHIPS Act aim to expand domestic production.

Frequently asked questions

Why can’t countries just make their own advanced chips?

Leading-edge manufacturing requires extremely specialized equipment, materials and expertise developed over decades, plus massive investment.

Do export controls affect ordinary consumers?

Mostly indirectly. They target data-center accelerators and manufacturing equipment rather than everyday devices.

What is a chokepoint in the chip supply chain?

A step controlled by very few suppliers, such as extreme ultraviolet lithography, where restricting access can significantly limit others’ capabilities.