A $21 million diversion scheme, and the charge that almost wasn't available
On August 24, the Keelung District Prosecutors' Office indicted nine people over a scheme to divert advanced Nvidia AI servers to China, including a distribution manager at Nvidia's Taiwan office, two sales managers at Super Micro's Taiwan subsidiary, the head of a Super Micro distributor, and executives at a data-center operator, a customs broker, and a trading firm (Taipei Times). Prosecutors say the group used falsified end-user documents to claim that 130 Super Micro servers packed with Nvidia's B300 GPUs would be installed at a rented data center in Taiwan. Instead, 74 were routed to Chinese buyers through direct shipments and transshipment via Indonesia, Japan, and Hong Kong; the remaining 56, destined for a similar handoff, were intercepted by customs before leaving Taiwan (Focus Taiwan).
What makes the case notable isn't the smuggling — chip diversion schemes have become a predictable feature of the US-China tech rivalry — but the charge sheet. The nine defendants face forgery and breach of trust, not a violation of Taiwan's export control statute. That is not prosecutorial choice; it's the only tool available.
Why the obvious charge wasn't on the table
Taiwan's Foreign Trade Act does have a criminal provision for illegal tech exports: Article 27 carries up to five years' imprisonment and fines of up to NT$3 million for exporting strategic high-tech commodities to a restricted region without authorization (Laws & Regulations Database, Ministry of Justice). Shipping the same category of goods to Iran, North Korea, or Syria would trigger exactly that penalty. China does not carry the same designation. Legislator Chung Chia-pin, who has pushed to close this gap, traces it to an administrative order issued during the Ma Ying-jeou administration that narrowed China's control status from a broadly restricted region to one covering only a handful of categories — today, just 12 types of wafer-fabrication equipment out of more than 10,000 controlled items on Taiwan's books (Taipei Times). Advanced AI accelerators like the B300 were never added to that narrow list, so exporting them to a Chinese buyer who isn't on a company-specific blacklist (like Huawei or SMIC) simply isn't the crime that exporting them to Tehran would be.
That's why Keelung prosecutors reached for forgery and breach of trust instead — real charges, but ones built around the paperwork fraud used to disguise the destination, not the diversion itself. It is a narrower net: it requires proving falsified documents or breached corporate duty, rather than the export event itself, and the maximum exposure under Taiwan's securities-law aggravated breach-of-trust provisions or general forgery statutes doesn't automatically track Article 27's dedicated five-year ceiling for the underlying conduct.
The case for closing the gap — and the case for doing it carefully
The strongest argument for legislative action is straightforward: if the predicate offense conveniently avoids naming the actual harm, prosecutors are left improvising with adjacent statutes every time a scheme like this surfaces, and the deterrent signal is weaker than it should be for conduct US officials have explicitly flagged as damaging to the credibility of Taiwan-US trade cooperation. Washington has restricted export of the most advanced Nvidia chips to China since 2022, and every transshipment scheme that succeeds — however briefly, however partial — hands Beijing hardware it was never supposed to get and hands critics of Taiwan-US supply-chain coordination a talking point. Chung Chia-pin's proposed "mainland China semiconductor chip clause" amendment to the Foreign Trade Act, along with broader talks between Taipei and Washington about extending controls beyond entity-specific blacklists to cover Chinese customers generally, is a reasonable response to a genuinely identified hole in the statute (Taipei Times).
But the fix should target the diversion mechanism, not chip ownership generally. Taiwan's semiconductor sector, and Nvidia's Taiwan operations specifically, function because Taiwan is simultaneously the fabrication hub and a compliance-conscious commercial partner for US chip designers — over-broad criminalization that sweeps in legitimate commercial activity, or that gives prosecutors discretion to charge routine paperwork errors as felonies, would raise compliance costs for the thousands of legal, whitelisted transactions Nvidia and its partners already process under existing end-user verification rules (the same rules, notably, that this scheme had to forge its way around). The five-point evasion strategy prosecutors describe — false end-user filings, a Japan shell company, laundered documentation, and multi-hop transshipment — succeeded despite Nvidia and Super Micro's own on-site inspection requirements for orders above eight units, suggesting the corporate compliance layer mostly worked and was defeated by outright fraud, not by a policy loophole in the ordinary sense.
The better legislative target, then, is precise: restore criminal parity between diversion-to-China and diversion-to-other-restricted-destinations specifically for chips already covered by US export controls, rather than reclassifying China wholesale as a controlled region for the 10,000-plus SHTC line items that have nothing to do with AI accelerators. That would close the exact gap this case exposed without turning Taiwan's export bureaucracy into a chokepoint for its own chip industry.