Yih-Shyan "Wally" Liaw, cofounder of Super Micro Computer, was arrested last week and charged with conspiring to smuggle approximately $2.5 billion worth of Nvidia GPUs to China. The chips in question were H200 and B200 processors, both subject to U.S. export controls that ban their sale to Chinese entities. Liaw resigned from the Supermicro board the day after his arrest.
The alleged scheme routed servers containing the banned GPUs through an unnamed Southeast Asian company, which became Supermicro's eleventh-largest customer with $99.7 million in purchases during fiscal year 2024. According to the charging documents, compliance audits at the intermediary's facilities were staged with "dummy" servers, the compliance officer was kept off-site during inspections, and servers were repacked in unmarked boxes before being forwarded to Chinese buyers.
Supermicro's stock dropped 33% on the news. Short sellers made $860 million.
The company has been here before. Between 2001 and 2003, Supermicro exported servers, motherboards, and computer chassis to Iran through a distributor in Dubai without the required export licenses. The violations resulted in a $150,000 fine from the Department of Justice, a $125,400 penalty from the Commerce Department, and a $179,327 settlement with OFAC. The Iran scheme used the same structural playbook: a pass-through entity in a third country handling the physical routing while Supermicro's name stayed off the final destination paperwork.
Two decades later, the scale is different. Two and a half billion dollars in banned GPU shipments dwarfs the Iran case by orders of magnitude. The co-conspirators named in the current indictment include Ruei-Tsang "Steven" Chang, Supermicro's Taiwan general manager, who is currently a fugitive, and Ting-Wei "Willy" Sun, described as a fixer, who was also arrested.
Supermicro is one of the largest GPU server manufacturers in the world. The company builds the racks, the enclosures, and the thermal management hardware that ships alongside Nvidia's chips. When a Supermicro AI server arrives at a data center, it typically comes with integrated liquid cooling manifolds, cold plates, and CDU connections designed for the specific GPU configuration inside.
A $2.5 billion smuggling operation means thousands of high-density GPU servers ended up in facilities that were never part of the legitimate supply chain. Those servers need cooling. The H200 runs at 700 watts per GPU. The B200 pushes past 1,000 watts. A fully populated rack of either chip generates thermal loads that require liquid cooling infrastructure, whether that infrastructure was procured through proper channels or improvised from whatever components the end operator could source.
This matters for cooling vendors in two ways. First, the demand signal from export-controlled markets is larger than official figures suggest. If $2.5 billion in GPU hardware moved through a single intermediary in a single scheme, the total volume of diverted chips across all channels is almost certainly higher. That represents real cooling demand being served by someone, just not through the supply chains that Western cooling manufacturers typically track.
Second, the enforcement response will tighten procurement scrutiny across the entire server supply chain. Cooling components that ship as part of integrated GPU server packages will face the same end-use verification requirements as the processors themselves. CDU manufacturers, cold plate suppliers, and thermal interface material vendors selling into the server OEM channel should expect more questions about where their products end up.
The details of the alleged scheme reveal how thin the compliance layer actually is in high-volume server distribution. Dummy servers staged for audits. A compliance officer deliberately kept away from the inspection site. Servers repacked in unmarked boxes. These are not sophisticated evasion techniques. They are the kind of basic operational deceptions that work when nobody is looking hard enough.
For cooling vendors embedded in the server OEM supply chain, the risk is guilt by proximity. If a cooling company's CDUs or cold plates are found inside diverted servers at a Chinese data center, the question of whether the cooling vendor knew about the diversion becomes a legal matter, not a technical one. Export control enforcement has been expanding its scope beyond the controlled items themselves to include "items specially designed" for use with controlled technology. Liquid cooling systems designed for H200 and B200 servers could fall under that umbrella.
Supermicro's prior export violations and the scale of the current allegations will likely accelerate efforts by the Bureau of Industry and Security to extend end-use monitoring deeper into the component supply chain. Cooling hardware that ships integrated with controlled GPU servers may need its own documentation trail.
Supermicro holds roughly 10% of the global AI server market. If the company faces sanctions, debarment, or a sustained loss of customer confidence, that share gets redistributed to Dell, HPE, Lenovo, and a growing number of ODMs. Each of those competitors has its own preferred cooling vendors and integration partnerships. A Supermicro contraction reshuffles the cooling supply chain alongside the server supply chain.
The broader message is simpler. The GPU export control regime is creating a shadow market for AI compute hardware, and that shadow market generates cooling demand that exists outside normal procurement channels. The $2.5 billion that moved through Supermicro's back door bought chips that now sit in racks that need thermal management. The cooling for those racks is being built somewhere, by someone, outside the view of the companies and analysts who track this industry.
That is a market reality worth understanding, even if it does not show up in any forecast.