Nvidia China chip sales are back under an uncomfortable spotlight after reporting by The New York Times found that one approved customer, New Matrix, appeared to lead to an empty office in China. That is not a minor paperwork embarrassment. When the product is an Nvidia H200, one of the world’s most sought-after AI accelerators, the question is whether US export controls are actually controlling anything.
- Nvidia China chip sales reportedly included an approved buyer whose listed office appeared vacant and unknown to local property managers.
- The Nvidia China chip sales process is under pressure because some approved firms reportedly have links to previously restricted Chinese entities.
- The case exposes how export controls can fail when officials must assess opaque corporate structures across borders.
- Nvidia’s H200 is a valuable training chip, making buyer screening central to US national-security policy.
Table of Contents
Nvidia China chip sales meet an empty office
According to the Times, New Matrix was cleared to receive Nvidia’s H200 chips despite scant public evidence that it operates as a real technology business. Reporters visiting the company’s listed address found what the paper described as a hollow gray space, with unfinished wiring where lights should have been. The building directory reportedly did not list New Matrix, and staff at the industrial park’s management office said they had no record of it.
A dormant office, on its own, doesn’t prove a crime. Companies move, use registered-agent addresses, launch quietly, and occasionally have terrible facilities administration. But that charitable reading gets hard to sustain when a firm is buying high-end AI silicon that Washington has spent years treating as a strategic asset. Nvidia China chip sales should involve more than a name on an approval list and a plausible-looking corporate registration.
The Times also reported that New Matrix was formed roughly 17 months ago and that its owners had connections to a Chinese private-equity group blacklisted by the US government in 2024. That alleged connection is the part officials cannot casually wave away. Export rules are supposed to prevent advanced computing equipment from reaching military-linked users or organizations that could support surveillance, weapons development, or other restricted activities. Shell companies are the oldest trick in the sanctions-evasion playbook; chips simply happen to be the current prize.
The H200 is not an ordinary graphics card
Here is the important distinction: Nvidia’s H200 is a data-center GPU built for the sort of AI work that consumes enormous quantities of memory bandwidth: training large language models, serving them to millions of users, and running scientific or industrial workloads at scale. It is not something a tiny, barely visible company buys for a couple of workstations.
That does not mean every H200 shipment automatically has military value. AI hardware is used across medicine, finance, manufacturing, search, robotics, and consumer apps. The problem is that the same computing clusters can be repurposed quickly, rented out, or broken into smaller allocations through cloud providers and resellers. Once an advanced GPU is inside a country, following its ultimate use can resemble tracking a specific drop of water after it hits a swimming pool.
Nvidia China chip sales have therefore become a litmus test for the broader US approach to technology competition. The Biden administration tightened restrictions on exports of advanced AI chips and manufacturing tools to China, arguing that leadership in compute has direct national-security implications. The official rules, administered by the Commerce Department’s Bureau of Industry and Security, remain available through its export administration portal, but rules are only as durable as the checks behind the licenses.
A policy built on screening has to screen
The Trump administration reportedly relaxed the earlier limits after Nvidia agreed to give the US government 15 percent of all sales. That arrangement is unusual enough to invite skepticism. A revenue share may give Washington a financial return, but it does not answer the central question raised by the New Matrix episode: who exactly receives the hardware, and who ultimately controls it?
Frankly, mixing a national-security licensing system with a percentage-of-sales arrangement looks messy. Export policy should not be a toll booth where the government collects a fee and assumes the truck is carrying what the manifest says. If the chips are sensitive enough to require restrictions, the due diligence has to be exhaustive; if they are not, the restrictions are political theater. There is very little sensible middle ground.
Nvidia has a legitimate commercial interest in serving markets where it is allowed to sell. Cutting American firms out of China can strengthen domestic Chinese rivals, particularly Huawei, while reducing US influence over the global AI stack. That is a serious argument, not corporate boilerplate. Remember the telecom equipment fights of the last decade? Restricting a market leader can create room for a domestic alternative to mature faster.
But Nvidia cannot outsource responsibility to Washington simply because it has a license in hand. Nvidia China chip sales of this scale should demand a customer profile that makes sense: identifiable operators, credible data-center capacity, disclosed ownership, an intelligible use case, and continuing checks after delivery. A vacant suite with dangling electrical wires is not an encouraging signal.
The loophole is the real story
The reported New Matrix case matters less because of one obscure company than because it may reveal the limits of a system designed around formal entities. Chinese corporate ownership can be difficult to trace even for experienced investigators, with investment vehicles, affiliates, local partners, and shifting registrations obscuring who is funding or directing a business. Add cross-border enforcement and a frantic AI hardware market, and the incentive to accept thin documentation becomes obvious.
That is why Nvidia China chip sales need scrutiny after the license is issued, not only before. Regulators could require verified operating addresses, beneficial-ownership disclosures, installation records, periodic audits, and proof that chips remain at declared facilities. None of that is foolproof. It would, however, raise the cost of using a paper company as a pass-through.
Here is the awkward possibility: the US may be trying to preserve American chip revenue while slowing China’s AI progress, and those goals increasingly collide. China is too large a market to ignore, Nvidia’s products are too strategically significant to treat like ordinary exports, and enforcement agencies have finite staff. Something gives.
My read is that this episode should force a public accounting from both Commerce and Nvidia: what standards got New Matrix approved, what verification occurred, and whether the chips have actually been delivered to the declared end user. Without convincing answers, Nvidia China chip sales risk becoming a policy slogan rather than a defensible security framework. And once a rack of H200s disappears into an opaque supply chain, getting it back is about as realistic as putting toothpaste back in the tube.
