U.S. Closes Export Loophole Enabling Chinese Subsidiaries to Acquire Advanced AI Chips

Announcement

The U.S. Bureau of Industry and Security (BIS) released updated export‑control guidance on May 31, 2026 that mandates a license for any advanced AI chip shipped to a company whose ultimate parent is located in China’s Country Group D:5, regardless of where the subsidiary itself is incorporated. The clarification directly addresses a gap that let Chinese‑owned firms purchase Nvidia GB200, AMD MI350 x, and similar accelerators through offshore entities in Malaysia, Singapore and other “friendly” jurisdictions.

Image credit: Getty / iStock

Technical specs and the loophole

Chip family	Nominal performance (FP32)	Power envelope	Export classification
Nvidia GB200 (H100 successor)	1.2 TFLOP/SM, ~60 TFLOP total	400 W	EAR99 → License Required
AMD MI350 x (MI300 x successor)	1.1 TFLOP/SM, ~55 TFLOP total	350 W	EAR99 → License Required
TSMC N5‑based AI ASICs (e.g., custom AI‑optimized SoCs)	0.8‑1.0 TFLOP/mm²	150‑250 W	EAR99 → License Required

The original BIS rule set a Country Group D:5 threshold that applied only when the destination was physically inside that group. Chinese firms sidestepped the rule by establishing subsidiaries abroad; the subsidiaries technically qualified for the “non‑Group D:5” exemption, even though the ultimate controlling shareholder remained Chinese. Export documentation listed the foreign address, so the shipments cleared without a license.

Industry sources estimate hundreds of thousands of GPUs and accelerators moved through this channel, a figure that aligns with the 366 % surge in AI‑chip imports reported by Malaysian customs in early 2026. The scale is significant because a single H100‑class GPU can train a 175‑billion‑parameter model in roughly 30 hours; multiplying that by 200 000 units translates to a potential 6‑million‑GPU‑hour boost for Chinese AI developers.

Market implications

1. Supply‑chain risk re‑calibration – Companies that previously counted on a “clean” export pipeline now face additional compliance steps. License applications for each shipment can add 2‑4 weeks of lead time, prompting OEMs to increase safety stock. Forecasts from IDC suggest a 3‑5 % upward revision in AI‑chip inventory levels for U.S. distributors through Q4 2026.
2. Pricing pressure on Nvidia and AMD – The tighter controls reduce the effective addressable market in China by an estimated 12‑15 % for the GB200 and MI350 x families. Historical data shows a 7 % price premium when supply tightens; analysts expect a $150‑$200 per‑GPU uplift for the next generation of H100‑class parts.
3. TSMC’s exposure – Although TSMC has pledged to prioritize U.S. customers, the new guidance does not explicitly tie its enhanced‑due‑diligence rule to the BIS license requirement. If Chinese firms continue to route orders through foreign subsidiaries, TSMC could see additional scrutiny on its N5 and N4 capacity allocations, potentially accelerating the shift of some AI‑chip fab bookings to U.S.‑based foundries such as GlobalFoundries.
4. Strategic shifts for Chinese AI players – With the offshore route closed, firms may revert to more covert methods (e.g., bulk shipments of storage drives, remote‑rental of overseas servers). The increased enforcement risk could push them toward domestic chip development programs, accelerating China’s own AI‑accelerator roadmap but also raising the likelihood of a fragmented global AI‑hardware ecosystem.
5. Compliance ecosystem growth – Export‑control software vendors are already reporting a 40 % surge in demand for automated screening tools that can trace ultimate ownership across corporate hierarchies. Companies that fail to integrate such solutions may face penalties exceeding $10 million per violation under the Export Administration Regulations.

Outlook

The BIS clarification marks a decisive step in tightening the AI‑chip export regime, but it also introduces new friction points for the global supply chain. In the short term, expect higher inventory buffers, price adjustments, and a rise in compliance‑technology spend. Over the longer horizon, the measure could accelerate the diversification of AI‑hardware sourcing away from a single dominant market, reshaping the competitive dynamics for Nvidia, AMD, and TSMC alike.

For a deeper dive into the regulatory text, see the official BIS notice here.