How the Bureau of Industry and Security Uses the Foreign Direct Product Rule to Restrict Global AI Compute

The Bureau of Industry and Security (BIS) holds the authority to sever any foreign entity from the global semiconductor supply chain. Through the Foreign Direct Product Rule (FDPR), the agency can block the sale of artificial intelligence chips manufactured entirely outside the United States, provided the factory uses American software or tools. The BIS reviews these thresholds annually, with the next major revision of the Commerce Control List dictating which hardware clears the bar and which nations are denied access.[1][4]

The mechanism relies on a legal concept of extraterritoriality. If a foundry in Taiwan uses electronic design automation (EDA) software from California-based Synopsys, or lithography equipment containing US-patented components, the resulting silicon is subject to US export law. This allows the Commerce Department to regulate trade between two foreign nations without either party stepping foot on American soil.[2][4]

To define what constitutes a restricted AI chip, the BIS relies on a metric called Total Processing Performance (TPP). Calculated by multiplying the number of operations a chip can perform per cycle by the bit-length of those operations, TPP provides a hardware-agnostic baseline. In October 2023, the agency set the hard ceiling at 4,800 TPP.[1][2]

Raw performance is only half the equation. The 2023 interim final rule introduced a second metric: Performance Density (PD). This measures the TPP divided by the physical die area of the chip in square millimeters. The BIS established a PD threshold of 5.92, designed to prevent companies from bypassing the rules by packing multiple slightly weaker processors onto a single large die.[1]

The shift from the original 2022 rules to the 2023 framework fundamentally altered the market. Under the 2022 guidelines, the BIS used a chip-to-chip interconnect bandwidth threshold of 600 gigabytes per second to identify frontier hardware. By replacing that with the PD metric, the agency closed a loophole that allowed manufacturers to export slightly detuned versions of their flagship products.[1][2]

When normalising the 2022 bandwidth limits against the 2023 density thresholds, the effective permissible compute ceiling for mid-tier inference chips dropped by 85%. Hardware that easily cleared the 600 gigabytes per second interconnect limit in 2022 now triggers the 5.92 PD tripwire, capturing a massive swath of consumer-grade accelerators that were previously exempt.[3]

"The United States has essentially weaponized the semiconductor supply chain chokepoints," notes Gregory Allen, director of the Wadhwani Center for AI and Advanced Technologies at the Center for Strategic and International Studies (CSIS). "The FDPR ensures that no cutting-edge silicon can be fabricated without U.S. consent, regardless of where the fab is physically located."[2]

Enforcement remains a logistical challenge. The BIS operates with a budget of roughly $191 million for fiscal year 2024 and employs fewer than 600 enforcement agents globally. Tracking the end-use of millions of processors shipped through complex networks of shell companies in neutral third countries stretches the agency's resources to the breaking point.

The Semiconductor Industry Association (SIA) has repeatedly warned about the long-term consequences of aggressive FDPR application. The trade group argues that cutting off access to the Chinese market—which accounted for nearly 30% of global semiconductor revenue in 2023—deprives US firms of the capital necessary to fund next-generation research and development.

To circumvent the hardware restrictions, targeted entities are increasingly turning to Infrastructure as a Service (IaaS) providers. Instead of importing physical chips, foreign labs rent compute time on servers located in the Middle East or Europe. The Commerce Department proposed rules in early 2024 to mandate "know your customer" requirements for US cloud providers, but regulating foreign cloud subsidiaries remains legally ambiguous.[4]

Another vulnerability in the framework involves chiplet architecture. By manufacturing smaller, unregulated silicon dies and packaging them together post-export, foreign entities can theoretically assemble high-performance systems that bypass the 5.92 PD limit. The BIS has acknowledged this gap, noting in its 2023 filing that advanced packaging technologies require ongoing monitoring.[1]

The effectiveness of the FDPR ultimately hinges on the monopoly power of a few key suppliers in the EDA and semiconductor equipment markets. If targeted nations successfully develop indigenous lithography tools capable of sub-14-nanometer fabrication, the US software link is severed, and the FDPR loses its jurisdictional hook.[2][4]

The next verifiable checkpoint arrives in late 2026, when the BIS is scheduled to release its annual update to the Commerce Control List. For the current enforcement regime to hold, the agency must either lower the 4,800 TPP threshold to capture the next generation of efficient inference chips, or successfully implement the proposed cloud computing reporting mandates to close the IaaS loophole.[1][3]