A recent smuggling case against a major AI company highlights the limitations of existing US export controls—and the ways they could be improved.

Somewhere in Southeast Asia last year, a woman blasted a server rack with a hair dryer. She wasn’t drying it. She was loosening the adhesive on a sticker with its serial number, so it could be peeled away—and pressed onto a different machine bound never to reach its declared destination.

The real servers, packed with NVIDIA’s most advanced B200 graphics cards, had already been repackaged into unmarked boxes and shipped to China. The dummy was waiting for inspection by an inattentive auditor.

This scene, captured in surveillance footage cited in a federal indictment unsealed last Friday, is the most precise image we have yet of how America’s semiconductor export controls are working in practice. To put it simply: they aren’t.

The Super Micro Case Shows the Limits of US End-Use Enforcement

The indictment of three co-conspirators—including the co-founder of Super Micro, a $12 billion AI server company—should disturb anyone who takes the US-China technology competition seriously. Export controls on advanced AI chips have been the bedrock of American strategy since President Trump’s first term, when the Commerce Department placed Huawei on the Entity List. The Biden administration subsequently expanded this with sweeping restrictions on high-end semiconductors. For more than a decade, a bipartisan consensus has correctly held that China should not have unfettered access to the computational power required to advance its intelligence collection and military capabilities.

The Commerce Department office responsible for enforcing export controls, the Bureau of Industry and Security, has fewer than 500 employees and only nine Export Control Officers posted at embassies worldwide—covering hundreds of billions of dollars in controlled trade. Its end-use verification regime assumes exporters accurately report where their products go, that foreign recipients will produce items for inspection, and that host governments will cooperate in scheduling compliance checks. End-use checks must be coordinated with host governments, meaning that the entity under scrutiny typically knows a visit is coming and can take measures to prepare. In the Super Micro case, conspirators fabricated end-use documentation, staged dummy servers in a rented warehouse, and arranged for a “friendly” auditor who never ended up visiting the site.

The “Chip Security Act” Could Hinder Chip Smuggling

This week, the House Foreign Affairs Committee will mark up legislation that could offer a reliable fix to some of these problems: the Chip Security Act, a bipartisan bill introduced by Senator Tom Cotton (R-MO) and Representative Bill Huizenga (R-MI) and cosponsored by Representatives Bill Foster (D-IL), Raja Krishnamoorthi (D-IL), John Moolenaar (R-MI), and Ted Lieu (D-CA). The bill requires that advanced AI chips exported from the United States be outfitted with location verification mechanisms before they leave the country.

The technology required for location verification has existed for years. A common approach involves having the chip send a “ping” to geographically dispersed servers, triangulating its position from the time delay of each response—the same way that police can locate personal phones from cell towers. This is precise enough for country-level accuracy and nontrivial to spoof. NVIDIA has acknowledged that its GPUs already have the capacity to transmit the necessary telemetry data, and geolocation can be achieved with off-the-shelf software.

Server geolocation is not surveillance. Nor is it particularly invasive. When Americans order goods online, Amazon, FedEx, and UPS can tell where their packages are at any moment, or whether a return shipment has been opened. The question before Congress is whether the most powerful AI hardware on earth should have a less sophisticated chain-of-custody record than a pair of sneakers.

Opposition to the Chip Security Act

The Semiconductor Industry Association has come out against the bill, arguing that mandating untested on-chip security mechanisms risks undermining global trust in American semiconductors. Other critics raise overlapping objections that are worth engaging seriously.

The most significant objection pertains to its cost. Some in industry insist that implementing geolocation at scale would be expensive and burdensome. But the mechanisms under discussion are software updates that use existing chip telemetry, not a fundamental redesign of hardware. If the United States is serious about forgoing billions of dollars in chip sales to China in the name of national security, it can certainly bear the marginal cost of verifying those restrictions are actually enforced.

BIS is already fabulously under-resourced for the mission it has been handed. But no number of agents will be capable of auditing every warehouse in Southeast Asia, even if they were given timely access. It is far simpler to build a system that automatically flags when a chip that was supposed to be in Singapore instead sends a location ping from Shenzhen.

Other experts worry about buyer preferences, and industry executives have speculated that location verification could drive allied buyers toward Chinese alternatives. But global demand for AI chips far outstrips supply. China is investing tens of billions of dollars into domestic semiconductor programs precisely because it cannot get enough American-made chips. No legitimate customer is in a position to refuse American GPUs on the grounds that their shipment may be geolocated. The list of potential buyers who would raise such a complaint is short — and composed mostly of intermediaries in the business of chip diversion.

In fact, the Chip Security Act is fundamentally pro-diffusion, and fully consistent with the Trump administration’s strategy of exporting the US AI stack. If the United States has an effective way of verifying that chips shipped to Malaysia or the UAE have not been rerouted to China, the case for loosening license requirements to those markets becomes even stronger. Unlike measures introduced and voted down in 2025, this bill does not restrict allied access to American technology—it creates the verification infrastructure that makes expanded access defensible.

This is the point that critics of the Chip Security Act miss most consistently. The alternative to geolocation is not a world of free and frictionless trade, but one of tightening restrictions and periodic scandals that corrode trust in the benefits of free enterprise. The single most effective way to preserve an open, pro-diffusion AI export policy is to make it enforceable.

The Super Micro indictment was not a unique failure, but the logical product of an export control system that runs on bribeable compliance auditors and stickers that can be peeled off with a hair dryer in a rented warehouse. Congress can keep pretending this system is adequate to police the most consequential technology race of the 21st century—or it can embrace technical modernization to ensure that the next $2.5 billion in AI servers doesn’t end up in an unmarked box bound for Shenzhen.

About the Author: Ryan Fedasiuk

Ryan Fedasiuk is a fellow at the American Enterprise Institute, where he focuses on US-China relations, technology, and national power. Concurrently, Mr. Fedasiuk is an adjunct assistant professor at Georgetown University. He was previously an advisor for US-China Bilateral Affairs at the State Department, where he served as the US government’s main point of contact with the Chinese Embassy in Washington.

The post America’s Export Controls Shouldn’t Run on the Honor System appeared first on The National Interest.