The Hair Dryer Test
On Thursday, federal agents arrested Yih-Shyan “Wally” Liaw, co-founder of Super Micro Computer, at his Silicon Valley home. The charge: conspiring to smuggle $2.5 billion worth of AI servers to China in violation of U.S. export controls.
On Friday morning, the White House released its “National AI Legislative Framework” — a four-page document outlining how Congress should govern artificial intelligence.
The juxtaposition is almost too perfect.
The Scheme
The alleged smuggling operation reads like a heist movie written by someone who’s never seen a heist movie.
Servers built in the United States were shipped to Taiwan, where Super Micro has manufacturing facilities. From Taiwan, they were routed to unnamed countries in Southeast Asia. There, workers used hair dryers to heat and peel serial number labels off the real servers, transferring them onto non-functional “dummy” replicas. The real servers — packed with Nvidia AI chips banned from export to China — were placed in unmarked boxes and shipped onward.
When compliance inspectors showed up, they found thousands of properly labeled, properly located servers sitting exactly where they were supposed to be. The actual servers were already in China.
More than half a billion dollars worth of servers were diverted between April and mid-May 2025 alone. The total over the full scheme: at least $2.5 billion.
Three days before his arrest, Liaw was standing next to Super Micro CEO Charles Liang at Nvidia’s GTC conference in San Jose, shaking hands with Jensen Huang. The photograph was posted on X by Super Micro’s official account.
The Framework
The White House document, by contrast, exists in a cleaner universe. It outlines six objectives:
- Protecting children and empowering parents
- Safeguarding and strengthening American communities
- Respecting intellectual property
- Preventing censorship and protecting free speech
- Enabling innovation and ensuring American AI dominance
- Educating Americans and developing an AI-ready workforce
The framework explicitly calls on Congress to preempt state AI legislation — creating one federal standard instead of what it calls a “patchwork” of state rules. It proposes limiting developer liability for harms caused by AI systems, particularly shielding companies from responsibility for how third parties use their models.
It mentions energy. It mentions children. It mentions censorship. It mentions workforce training.
It does not mention export controls. It does not mention chip smuggling. It does not mention the supply chain that makes every other provision in the document possible.
The Gap
This is not a criticism of either document in isolation. Export controls fall under Commerce and Defense, not the legislative framework being proposed here. The White House framework addresses a different problem.
But the gap between the two tells you something about how we think about AI governance.
The framework operates at the level of abstraction. It imagines AI as a policy object — something that needs rules about children’s safety, intellectual property, free speech. Important things. Real things.
The Super Micro case operates at the level of physics. It reveals AI as a physical object — servers with serial numbers, chips that draw 1,200 watts each, boxes that need to be loaded onto ships. And people willing to use hair dryers to circumvent the rules that govern those physical objects.
Policy frameworks assume enforcement. The hair dryer test reveals what enforcement actually looks like.
The Precedent
Super Micro is no stranger to trouble. In 2020, the company was temporarily delisted from Nasdaq over accounting irregularities. Last year, short-seller Hindenburg Research accused it of “accounting manipulation, sibling self-dealing, and sanctions evasion.” Its auditor, Ernst & Young, resigned.
Now, its co-founder — a man who has been with the company since its founding in 1993, who sat on its board of directors — has been arrested for what prosecutors call a “brazen” scheme to circumvent export controls.
Nvidia, whose chips were allegedly smuggled, issued a careful statement: “Strict compliance with export laws is a top priority.” The company did not say whether it was aware of the diversion.
Reuters had reported back in 2024 that China was acquiring Nvidia chips through Super Micro servers. The chips kept flowing for at least another year.
What This Actually Means
The policy conversation about AI governance is overwhelmingly about software. Models. Alignment. Bias. Censorship. Important conversations.
But the Super Micro case is a reminder that the binding constraint on AI is still hardware. Export controls work — or fail — at the physical layer. The most consequential AI policy of the last three years wasn’t any framework or executive order. It was the October 2022 chip export restrictions to China, followed by the October 2023 expansion that closed the A800/H800 loopholes.
Those restrictions created the economic incentive for exactly the scheme that Liaw is now accused of running. When you make a chip worth $40,000 unavailable through legitimate channels, you create a $40,000 incentive to find illegitimate ones.
This is not an argument against export controls. It’s an observation about what happens when policy meets physics. Export controls don’t make the demand disappear. They move it underground. And underground, the enforcement mechanism isn’t a four-page framework from the White House. It’s surveillance cameras catching people with hair dryers.
The Real AI Governance
The White House framework will get debated in Congress. Think tanks will publish responses. Lobbyists will argue over liability shields. States will fight preemption. This will take years.
Meanwhile, the actual governance of AI — the decisions about who gets the chips, who builds the servers, who runs the data centers — is happening in shipping containers and industrial parks, governed by export licenses, customs inspections, and apparently, the adhesive strength of serial number labels.
One framework imagines how to govern AI. The other reveals how AI is actually governed.
The hair dryer test: if your entire AI governance regime can be defeated by a $15 appliance from Walmart, you might want to rethink your assumptions about where the real power lies.