Export Control

Normally, a company may sell to whomever it wants. For certain goods, that doesn’t apply. The state then requires a permit before a good may leave the country — or it bans the sale to certain buyers altogether. That is exactly what export control is. Classically, it concerned weapons and explosives. Today, high-performance computing chips are also on the lists, because they are the foundation of modern artificial intelligence.

Computing chips as a political lever

Modern AI systems don’t arise from especially clever programmers alone. They require enormous computing power. Training a large language model costs months of computing time on tens of thousands of specialized chips. Whoever cannot obtain these chips will struggle to keep up in cutting-edge research.

That is why the chips have become a tool of foreign policy. Since 2022, the United States has repeatedly banned the delivery of the fastest AI chips to China. The justification given is national security: the same chips that train chatbots can also power military systems and surveillance technology.

For investors, this is no fringe issue. Before the restrictions, chip company Nvidia generated a significant portion of its data center revenue in China. Every new rule shifts billions of dollars and causes stock prices to swing. The Dutch machine builder ASML, which manufactures the equipment used to produce chips, is also subject to such rules.

Thresholds, lists, and licenses

An export control needs a technical definition. Otherwise, no one would know which product is affected. For AI chips, authorities work with measurable thresholds, such as computing performance per second and the speed at which a chip exchanges data with other chips. If a product exceeds these, it requires a license.

On top of that come lists of recipients. If a company is on such a blacklist, nothing may be delivered to it at all without a license. A particular feature of US law reaches far beyond its own borders: even foreign products fall under these rules if American technology is used in their manufacture. Because this is the case for almost all chip factories, the regulations have worldwide effect.

In practice, it’s a race. Manufacturers initially designed weakened chips that fell just below the thresholds. The authorities then lowered the limits. At the same time, workarounds emerged: resale through third countries, smuggling, or renting computing power in foreign data centers, which physically does not constitute an export. Closing such loopholes is at the heart of the ongoing debate.

What makes it into the news

When Nvidia or AMD report quarterly earnings, export control is almost always a topic. Headlines like “special chip for China” or “license denied” are standard fare in the business press. Write-downs also come up: chips that have already been produced but can no longer be delivered must be booked as a loss by a company.

Indirectly, the policy also affects users. Chinese developers had to make do with less computing power and therefore optimized their models especially aggressively. Some experts consider this an unintended side effect of the restrictions: scarcity forces efficiency.

Two misconceptions are common. Export control is not a tariff — it’s not about money, but about permission. And it is not a sanction against an entire country, but a rule for specific goods and specific recipients. The EU and Germany also have their own export lists, for instance for surveillance software and equipment with potential military applications.