U.S. Closes Remote Computing Loophole with New AI Chip Export Controls

The Trump administration is tightening export rules to prevent Chinese firms from accessing high-end AI compute through overseas data centers, targeting the 'remote-access' workaround.

VentureGrill Desk
2 min read
U.S. Closes Remote Computing Loophole with New AI Chip Export Controls

The U.S. Commerce Department is rolling out an interim rule designed to block Chinese companies from bypassing existing export controls through remote access to international AI infrastructure. This expansion of trade restrictions requires companies to obtain licenses for the export, reexport, or transfer of advanced AI chips and servers to data centers globally if those facilities provide compute to entities in China or other embargoed nations. The policy specifically targets the D:5 group of countries, which includes Russia, Iran, and North Korea, signaling a more aggressive stance on protecting domestic 'Super Intelligence' semiconductors from being used to train rival advanced models.

This regulatory tightening arrives during a period of significant upheaval for the semiconductor industry, following the resignation of Nvidia co-founder Jensen Huang. His departure is linked to a scandal involving Megaspeed, a customer suspected of acting as a proxy to funnel thousands of Nvidia chips to Chinese businesses. Federal authorities had reportedly been considering placing Megaspeed on an entity list prior to the rule change. Recent reports from Reuters suggest that Chinese AI firms have been aggressively securing overseas computing capacity in anticipation of these tighter U.S. restrictions, making this move a direct response to market-driven workarounds.

The new technical threshold for these controls is set at 12 million total processing performance (TPP), a metric used by the government to quantify computing power. To put this in perspective for venture investors and infrastructure providers, 12 million TPP is roughly equivalent to the maximum capacity of 750 Nvidia H100 or H200 chips. Data center operators and exporters must now monitor customer ownership structures more rigorously to ensure they are not facilitating access for brokers or entities from the restricted D:5 list. Existing operators falling under this threshold have a 90-day window to secure the necessary licenses once the rule takes effect.

In addition to targeting remote access, the Commerce Department is adding new licensing requirements for chip transfers to Malaysia, Singapore, Indonesia, Thailand, and Brazil. These nations have been identified as posing a higher-than-normal risk of diversion for restricted hardware. However, the administration has carved out an exception for chips destined for U.S.-based companies operating within those regions. Notably, the rule excludes commercial recommendation systems used in social media and e-commerce, focusing the government's scrutiny on the high-level compute required for military and intelligence-related model training.

This interim rule also marks a definitive shift in policy by rescinding the previous administration's AI Diffusion Rule from early 2025. While the previous framework focused on country-level limits and model transfers, the current administration argues that the old approach placed unnecessary burdens on U.S. companies and their allies. By pivoting to a compute-level restriction based on TPP, the government aims to prevent the training of advanced models that could threaten national security while maintaining a competitive edge for American-headquartered firms operating abroad.

Sources

  1. 01 US tightens AI chip export controls to deter Chinese companies from accessing compute remotely — TechCrunch