Washington just realized it accidentally spent a year feeding China's most ambitious artificial intelligence projects.
In a sudden, unusual weekend move, the US Department of Commerce stepped in to close a massive regulatory gap. This loophole allowed overseas subsidiaries of Chinese companies to legally buy the world's most powerful AI hardware. We are not talking about older, stripped-down legacy silicon here. The gap gave Chinese-owned units operating in places like Malaysia open access to top-tier hardware, including Nvidia's advanced Blackwell and Rubin architectures, alongside AMD's MI350x processors. Recently making news in this space: The Silent Leap to Autonomous Swarm Warfare in China.
If you want to know how this happened, you have to look at a classic case of bureaucratic whiplash.
The loophole opened wide in May 2025. That was when the Trump administration decided it would not enforce the "AI Diffusion Rule"—a strict, country-tiered global framework finalized by the Biden administration in its final days. By shelving that rule, Washington inadvertently dropped the guardrails. For nearly a year, if a company was headquartered in China but had an office in Southeast Asia, it faced no major hurdles buying restricted American chips. Additional information regarding the matter are explored by Gizmodo.
The damage is already done. While the Bureau of Industry and Security (BIS) issued fresh guidance to enforce license requirements for any entity headquartered in China regardless of physical location, the horse has left the barn. Industry insiders estimate that hundreds of thousands of frontier AI chips slipped through the crack while the door was left open.
The Logistics of the Overseas Bypass
How did a mistake of this scale happen right under the nose of US regulators? It comes down to corporate structure and physical geography.
When the US restricted direct chip shipments to mainland China, it drew a hard border around the country. But it didn't account for the globalized footprint of major tech firms. A Chinese AI company could easily set up a cloud infrastructure or data center subsidiary in a country like Malaysia or Thailand. Because these locations sat in an ambiguous regulatory tier, US chipmakers could legally fulfill orders to these foreign-registered businesses.
"The floodgates have quietly opened," warned an internal Washington briefing paper dated May 29, just days before the emergency enforcement action.
The physical hardware never had to cross into mainland China to deliver value. A developer in Beijing doesn't need to physically touch a server rack in Kuala Lumpur to train a large language model. They just need remote cloud access. By building out massive data centers overseas through these subsidiaries, Chinese firms effectively bypassed the domestic hardware blockade altogether.
Aggressive Smuggling Routes via Trusted Allies
Even as Washington scrambles to patch its official policy, physical diversion networks are operating in the background. The official loophole via corporate subsidiaries is only half the story.
Just days before the Commerce Department's weekend announcement, prosecutors in Taiwan detained three individuals for allegedly smuggling advanced Nvidia AI chips directly to China. The route they used reveals exactly why hardware embargoes are so difficult to police. The trio didn't ship the hardware directly across the Taiwan Strait. Instead, they falsified export documents for Super Micro Computer servers containing high-end Nvidia chips, routing the hardware through Japan first.
Japan is a bedrock US ally and a core piece of the American defense strategy in the Asia-Pacific. It is the last place regulators look for illicit tech transfers. From Japan, the hardware moved to Hong Kong, acting as the final transit point before hitting the mainland. This wasn't a crude operation; it was a highly sophisticated supply-chain fraud that exploited the high volume of trade between trusted economic partners.
What the New Rules Actually Change
The updated guidance from the BIS attempts to fix the structural corporate loophole by tying the restriction to ownership rather than location. If a corporate entity is ultimately headquartered in China, it now requires an explicit export license to receive advanced semiconductors, no matter where the subsidiary operates globally.
This directly impacts several front-line chip families:
- Nvidia Blackwell and Rubin: The absolute pinnacle of generative AI training hardware.
- AMD MI350x: AMD’s premier competitor in the high-density data center market.
- Intel Gaudi Lines: Specialized accelerators designed for enterprise deep learning.
Yet, the new guidance includes a massive catch that limits its immediate effectiveness. While it stops future shipments of these chips without a hard-to-get license, it does not require foreign data centers to stop using the chips they already bought. It also doesn't stop Western companies from servicing the high-end computing hardware already installed in those facilities.
The Shift to Domestic Reliance
While Washington plays regulatory whack-a-mole, Beijing is changing its playbook. Reliance on smuggled or loophole-acquired American silicon is a short-term fix. The long-term strategy for Chinese tech firms is total domestic independence.
We are already seeing the fruits of this push. When the US briefly restricted Nvidia's lower-spec, China-compliant H20 chip before back-tracking, Chinese buyers realized that US regulatory policy was too unpredictable to build a business on. Companies like Huawei have stepped into the vacuum, rapidly scaling up production of their own domestic AI chips.
The rise of domestic alternatives changes the competitive dynamics. Chinese AI startups, including companies like DeepSeek, are openly designing their newest models to run natively on domestic hardware. By forcing Chinese tech giants off American ecosystems, US export controls have inadvertently accelerated the creation of a parallel, completely independent semiconductor supply chain that Washington has zero power to regulate or monitor.
Next Steps for Compliance and Infrastructure Teams
If you run a global hardware procurement team, manage data center infrastructure, or handle international tech compliance, this sudden policy shift requires immediate action. Do not wait for an audit to fix your pipeline.
First, you need to conduct an immediate audit of your downstream customers and partners. The BIS guidance makes it clear that knowing the physical destination of your hardware is no longer enough. You must implement strict Know-Your-Customer (KYC) protocols that look past the local incorporation papers to identify the ultimate parent organization. If your buyer is an entity headquartered in China—even if they operate entirely out of Europe, the Middle East, or Southeast Asia—you must freeze the shipment until you secure an export license.
Second, re-evaluate your cloud leasing and remote compute agreements. While the new rule focuses primarily on physical chip shipments, Congressional pressure is mounting via the proposed Remote Access Security Act (RASA). This upcoming legislation aims to penalize firms providing remote cloud access to restricted chips. Start building compliance frameworks now to track who is renting your GPU capacity overseas, or risk getting caught flat-footed when the cloud loophole is officially closed next.
