PLA Universities Bought Nvidia A100s via Super Micro Shell Companies

Beihang University — one of China's "Seven Sons of National Defense," on the U.S. Entity List — completed a purchase of a Super Micro server with four Nvidia A100 GPUs in March 2026. Harbin Institute of Technology, another Entity List university, bought a system with eight A100s in July 2025. Both purchases happened after the October 2022 export controls that were specifically designed to stop this from happening.

How the Shell Company Routing Worked

Super Micro's co-founder Yi-Shyan "Wally" Liaw was arrested in March 2026 and charged with conspiracy to violate export control laws. Federal prosecutors allege he helped route restricted hardware through a network of shell companies — firms with no real operations that existed solely to pass paperwork and obscure the end destination of the chips.

The routing chain worked roughly like this: Super Micro sold to a nominally independent distributor, who sold to a reseller in a third country (typically Singapore, Malaysia, or the UAE), who then shipped to the Chinese university. At each step, the end-user certificate declared a legitimate non-military recipient. By the time the chips reached Beihang's server room, the trail was three transactions deep.

This is not a new technique. The same shell-company routing has been documented for Russian military procurement since 2022. What is notable here is that it worked for hardware explicitly named in U.S. export controls — the A100 was added to the restricted list in August 2022 specifically because of its AI training capabilities.

What the Seven Sons of National Defense Actually Are

Beihang (Beijing University of Aeronautics and Astronautics) and Harbin Institute of Technology are two of seven Chinese universities with deep, formally documented ties to the People's Liberation Army. The U.S. government placed all seven on the Entity List between 2020 and 2021, meaning U.S. companies need a license — effectively denied by default — to export technology to them.

The seven are: Beihang University, Harbin Institute of Technology, Nanjing University of Aeronautics and Astronautics, Nanjing University of Science and Technology, Northwestern Polytechnical University, Beijing Institute of Technology, and Harbin Engineering University.

These are not fringe institutions. They have combined enrollment in the hundreds of thousands and produce a significant fraction of China's aerospace, missile, and AI talent. Research published by their faculty regularly appears in open journals while simultaneously feeding classified PLA programs — a pattern the U.S. intelligence community calls "civil-military fusion."

The Scale of the Problem

Federal prosecutors put the total value of illegally exported hardware at $2.5 billion across the broader Super Micro investigation. That figure covers multiple defendants and multiple years, not just the university purchases. But the A100 numbers matter more than the dollar total.

Each A100 GPU delivers roughly 312 TFLOPS of FP16 performance — good enough for serious language model training and inference. A cluster of eight A100s, as Harbin acquired, can fine-tune a 7B-parameter model in hours. Four A100s, as Beihang received, are enough to run inference on a 70B-parameter model. These are not research toys. They are production AI training hardware.

The U.S. introduced the H800 and A800 — deliberately crippled versions of the H100 and A100 — for the China market to allow some sales while limiting the most sensitive applications. The universities did not buy the crippled versions. They bought the real thing, through a chain of companies designed to hide where the chips were going.

Super Micro's Ongoing Entity List Problems

Super Micro (SMCI) has a complicated history with U.S. regulators. The company was founded in 1993 by Charles Liang and Wally Liaw, both Taiwanese-American engineers. It grew into one of the world's largest server manufacturers, competing directly with Dell and HPE for hyperscaler contracts.

In 2023, Super Micro delayed its annual financial filing and disclosed an internal investigation into accounting practices. The SEC launched a formal inquiry. The stock dropped more than 70% from its 2024 peak. The company eventually restated earnings and avoided delisting, but the accounting episode raised questions about internal controls that now look relevant in the export control context.

Liaw's arrest suggests federal prosecutors have been building this case for some time. Export control violations carry penalties of up to 20 years in prison. The charges against Liaw are conspiracy charges, which means prosecutors believe multiple people inside or connected to Super Micro knew what was happening.

What Developers Need to Know About the Next Round of Controls

Every time a smuggling case like this becomes public, it produces a predictable regulatory response. Expect the following in the next 6-12 months:

Enhanced end-user verification requirements for GPU exports. The current system relies heavily on self-certification — the buyer signs a document saying they will not re-export. The Liaw case is evidence that self-certification is inadequate for hardware with this risk profile. Expect mandatory third-party audits for sales above certain thresholds.

Expanded Entity List additions. Any distributor or reseller identified in the Liaw prosecution will likely be added to the Entity List. This may include legitimate-seeming companies in Singapore, Malaysia, and the UAE that were used as transit points.

Tighter restrictions on A800/H800 equivalents. The existence of crippled-but-legal versions of restricted chips has always been a policy compromise. If the A100 — a chip that was supposed to be fully restricted — can be smuggled at scale, expect pressure to restrict the A800/H800 equivalents more aggressively or eliminate them entirely.

Cloud compute scrutiny. Chinese firms have used U.S.-based cloud providers (AWS, Azure, GCP) to access compute that they cannot import. This case will amplify calls for mandatory know-your-customer requirements on cloud GPU rentals. Several bills in Congress are already moving in this direction.

The Developer Infrastructure Angle

If you run AI workloads or build on top of GPU infrastructure, the export control tightening matters to you even if you are not in China. When regulators restrict which chips can be sold where, they also shape which chips get manufactured in volume. Nvidia has made significant revenue from the crippled A800/H800 China market. If that market closes entirely, the investment calculus for next-generation chip design changes.

For developers building RAG pipelines or fine-tuning workflows on Nvidia hardware, watch the BIS (Bureau of Industry and Security) rule changes coming this quarter. The Liaw arrest suggests enforcement is accelerating, which typically precedes a new rule update.

Key Takeaways

• Beihang University (Entity List) bought a 4x Nvidia A100 server through Super Micro shell companies in March 2026 despite full export restrictions
• Harbin Institute of Technology (Entity List) bought an 8x A100 system in July 2025 using the same routing
• Wally Liaw, Super Micro co-founder, arrested on conspiracy to violate export control laws
• $2.5 billion estimated total value of illegally exported hardware in the broader investigation
• The shell-company routing chain used Singapore, Malaysia, and UAE intermediaries to obscure end destination
• A100 GPUs — four units are enough to run inference on a 70B model, making these genuinely strategic hardware
• Expect new BIS rules on end-user verification and possible restrictions on A800/H800 equivalents within 6-12 months