4 Nvidia H200 Licensees Flagged: Empty Offices [2026]

The U.S. Commerce Department approved dozens of Chinese companies to handle Nvidia’s H200 artificial-intelligence chips, and several of them look very little like the businesses described in their paperwork, The New York Times reported on Nov. 7. One approved firm, New Matrix, listed a Beijing address that turned out to be an unfinished concrete shell. A second, Maginfra, listed an office that reporters also found deserted.

The Nvidia H200 China approvals matter because the chip sits at the high end of what Washington lets Chinese buyers touch. Other approved companies, according to the reporting, have ties to firms that serve the Chinese military or sit on a U.S. trade blacklist. Nothing in the sources we reviewed accuses Nvidia of breaking a rule. The open question is whether the government’s screening checked who really stands behind each applicant.

We covered the New Matrix address in our earlier report on the empty Beijing office and the Commerce approval. This analysis widens the lens to all four named licensees, the corporate trails that connect them to blacklisted names, and the line between what is confirmed and what remains analyst opinion.

What the Reports Say About the Nvidia H200 China Approvals

The core finding is simple. Commerce cleared certain Chinese companies to handle H200 chips or assemble them into AI servers. The Times examined those companies and found that some had no visible operating footprint at the addresses on file. The Business Times in Singapore ran the story under a headline that captures the two problems in one line: empty offices and ties to blacklisted firms.

Four companies carry most of the weight in the coverage. New Matrix and Maginfra drew attention for their empty addresses. Zhejiang Yindun Cloud Technology and Shanghai Jiatang drew attention for the people and companies behind them. The Washington Examiner, The Hill and Reuters have also carried the story, which pushed it from a single investigative piece into a Washington talking point within a day.

Precision matters here. The reporting describes companies approved to handle or assemble chips, which is not the same as end users running models on them. An assembler can build servers for another customer, so the identity of that customer is the piece that determines the real risk. The available coverage does not name those customers, and it does not confirm any purchase quantity, dollar value, or shipment date.

New Matrix: A Beijing Address That Turned Out to Be a Concrete Shell

New Matrix is the headline case. The company won approval to assemble Nvidia H200 chips, and the Beijing address it listed was found vacant. The Times described the site as an unfinished concrete shell. A shell cannot house a clean room, a test bench, or the staff needed to build AI servers, so the address alone raises a basic question about how the application was reviewed.

The age of the company adds to the concern. A secondary summary of the Times reporting says New Matrix was 17 months old when examined. That figure does not appear in the primary excerpt we could review, so treat it as reported but not independently confirmed. A young company is not suspect on age alone, since plenty of legitimate AI infrastructure firms started in 2024 and 2025.

The ownership trail is where the story gets sharper. Corporate records tie New Matrix’s owners to Wise Road, a Chinese private-equity fund. The U.S. government placed Wise Road on a trade blacklist in 2024. Some analysts told reporters that New Matrix looks like a front or facade for firms linked to Wise Road. That description is an assessment, not an official finding, and no U.S. agency has said so on the record in the sources we reviewed.

Maginfra and the Inspur Question

Maginfra was approved under the same program, and its listed office was also reported as deserted. Two empty addresses across two approvals suggest a pattern worth examining, though two cases do not prove a systemic failure. What matters is whether Commerce ever visited, called, or asked for photos of the sites before signing off.

The link to Inspur drives the second half of the Maginfra story. Inspur is a Chinese server company reported as blacklisted by the United States for work with the Chinese military. Maginfra reportedly shares personnel and patents with Inspur. Analysts quoted in the coverage believe Maginfra is an offshoot of the larger server maker.

Here the line between fact and inference runs thin. Shared staff and shared patents are documented overlaps. An offshoot relationship, with Inspur controlling or directing Maginfra, is a conclusion the available reports do not establish. The distinction decides whether the approval violated the spirit of the blacklist or merely brushed against it.

Zhejiang Yindun Cloud and the Military Contractor Link

Zhejiang Yindun Cloud Technology presents a different profile. The reports describe it as an approved company tied to Range Intelligent Computing Technology Group, which has sold cloud-computing services to Chinese military contractors. No empty office features in this part of the story. The concern is the neighbor and the customer list.

For AI infrastructure, that distinction is significant. A cloud operator that rents compute to defense contractors can pass H200 capacity downstream without ever moving a chip. Hardware export rules were built around shipments, but cloud access blurs the border between a buyer and a user. A licensee with a legitimate office and a real data center can still raise the same policy problem as a shell company.

Note what the reporting does not say. It does not accuse Zhejiang Yindun of supplying the military, and it does not claim any H200 has reached a defense customer. It documents a connection to a company that sells cloud services to such customers. That is a due-diligence flag, and the burden of explaining it now falls on the agency that approved the firm.

Shanghai Jiatang: Owners With Defense-Linked Ties

Shanghai Jiatang rounds out the group. The reports identify it as another licensee whose owners have ties to companies serving the Chinese military. The coverage gives the fewest specifics here, so the safest reading is narrow: an approved company whose ownership overlaps with defense-serving businesses, with no confirmed chip transfer attached.

Taken together, the four names show three different failure modes. New Matrix suggests a missing physical presence. Maginfra suggests hidden affiliation with a blacklisted manufacturer. Zhejiang Yindun and Shanghai Jiatang suggest customer and owner links that a basic ownership search would surface. A screening process strong on one of those checks could still miss the others.

Licensee Scorecard: Who Reports Say Is Behind Each Firm

The table below condenses the reporting into one view. It lists only what the source excerpts support and flags the status of each claim.

Company	Role in the program	What reporters found	Claim status
New Matrix	Approved to assemble Nvidia H200 chips	Listed Beijing address vacant, described as an unfinished concrete shell	Reported by The New York Times
Maginfra	Approved under the same program	Listed office deserted, shares personnel and patents with Inspur	Office reported, Inspur offshoot unconfirmed
Zhejiang Yindun Cloud Technology	Approved Chinese company	Ties to Range Intelligent Computing Technology Group	Tie reported, no chip transfer alleged
Shanghai Jiatang	Identified as a licensee	Owners tied to companies serving the Chinese military	Tie reported, no chip transfer alleged
All approved firms	Cleared to handle or assemble H200 chips	Dozens approved by Commerce	Reported

Wise Road, Inspur and the Entity List Problem

Two blacklisted names anchor the controversy: Wise Road and Inspur. The U.S. trade blacklist exists to stop restricted parties from receiving sensitive technology, directly or through intermediaries. An intermediary with the same owners, the same engineers, or the same patents defeats that purpose. That is why analysts focus on ownership trails rather than on company names.

Wise Road went on the list in 2024. The Chinese corporate records cited by the Times connect it to New Matrix’s owners, which means the connection predates the H200 approval by a wide margin. A screener who searched Chinese business registries could have found it. Whether anyone ran that search is the central unanswered question.

Blacklisted entity	Reported link	Linked approved firm	Evidence type
Wise Road	Chinese private-equity fund, on U.S. trade blacklist since 2024	New Matrix	Corporate records tie it to New Matrix’s owners
Inspur	Chinese server company reported blacklisted for military work	Maginfra	Shared personnel and patents
Range Intelligent Computing Technology Group	Sells cloud services to Chinese military contractors	Zhejiang Yindun Cloud Technology	Reported tie between the two
Unnamed military-serving companies	Customers or affiliates of owners	Shanghai Jiatang	Ownership ties

Confirmed Versus Unconfirmed: Where the Reporting Stops

Readers will see strong language in the headlines, so a clean ledger helps. The facts below rest on named outlets. The inferences rest on analysts’ beliefs. Mixing them would overstate the case against any single company and weaken the legitimate accountability questions.

Claim	Status	Basis
Commerce approved dozens of Chinese firms to handle Nvidia H200 chips	Reported	The New York Times and outlets carrying its findings
New Matrix’s listed Beijing address was vacant	Reported	The New York Times
Wise Road was blacklisted in 2024	Reported	Corporate and government records cited in coverage
New Matrix is a front for Wise Road-linked firms	Unconfirmed	Analysts’ assessment, no official finding
Maginfra is an offshoot of Inspur	Unconfirmed	Shared staff and patents plus analyst belief
Any purchase quantity, dollar value, or shipment date	Unconfirmed	None provided in available excerpts
Any named U.S. official quote on the approvals	Unconfirmed	None provided in available excerpts

No expert quotes are included in this piece because our quote-verification pass returned nothing we could attribute with a source link. We prefer a shorter article to an invented one.

Why the Screening Gap Matters for AI Compute

The H200 is Nvidia’s high-end accelerator for large-scale AI training and inference. Access to that class of silicon sets the pace for how fast a country can train frontier models. Washington’s chip policy rests on a bet that controlled access slows military-relevant AI work without cutting off ordinary commerce. That bet only holds if approved intermediaries are who they claim to be.

Consider the three pathways the reports imply. A shell company could receive hardware and pass it along. An affiliate of a blacklisted manufacturer could assemble servers that end up serving the same customers the blacklist targets. A cloud operator could rent H200 capacity to defense contractors. Each pathway needs a different control, and no single paperwork check closes all three.

For Nvidia, the market impact is mostly about policy risk. China sales under license are a revenue line the company defends, and a scandal over licensee quality hands critics a reason to narrow or end that channel. We have no confirmed share-price figure tied to this specific story, so we make no claim about one.

For Chinese buyers, the picture splits. Legitimate firms that hold approvals now face extra scrutiny that could slow deliveries. For cloud and server assemblers outside the program, the story is a reminder that their ownership structures will get read closely by compliance teams at suppliers. The effect on rival chip suppliers is harder to call, and we leave it as an open question rather than guess.

How the Approval Model Compares With Other Control Approaches

The reporting invites a comparison between three ways to police AI chips. This is our analytical framing, not a ranking drawn from published data. Each approach catches a different kind of risk.

Paper Licensing of Named Companies

This is the model on display. Commerce approves a company, and that company handles the chips. It scales well and creates a clear legal record. It also depends entirely on the quality of the application review, which is the weak point the Times exposed. An empty address is the kind of problem a single site visit or satellite image check would reveal.

Ownership-Chain Verification

A tougher model traces owners, directors, shared patents, and prior business names before approval. This would have surfaced the Wise Road records and the Inspur overlaps the press found. It costs more analyst time and requires fluency in Chinese corporate registries. It also needs a standard for how much overlap is too much, because legitimate firms often share suppliers and talent.

Compute-Level and Use-Based Controls

The third model tracks where the chips run and who uses the capacity. It addresses the Zhejiang Yindun scenario, where a clean company hosts or sells compute to a sensitive customer. It is the hardest to enforce because it requires visibility into cloud tenants. Our earlier piece on the one-page accord and its four oversight layers shows how thin the oversight framework around AI remains.

Historical Context: From Chip Curbs to the H200 Reversal

Washington began restricting advanced AI chip sales to China in 2022, and the rules have shifted repeatedly since. The 2024 blacklisting of Wise Road came in that period.

The H200 episode follows a policy reversal described in the reports, under which some H200 shipments to China became possible after earlier restrictions meant to limit Beijing’s access to advanced computing hardware. That reversal created the licensing program, and the licensing program created the screening burden. The Times story is the first major look at how the new gate works in practice.

The timing also overlaps with other Nvidia headlines. The Verge has reported that Nvidia is investigating Jensen Huang’s relationship with a Chinese spy, a separate matter we examined in our report on the China spy claim and its three unknowns. Nothing in the available reporting links that probe to the H200 approvals, and we do not suggest a link.

Washington Fallout: Congress, Nvidia and the Questions Ahead

Congress was already circling Nvidia before the Times story landed. House committee leaders have separately sought a briefing from Huang, a request we detailed in our piece on the Nov. 18 House briefing sought by three chairmen. The empty-office findings hand those lawmakers a concrete example to raise in the room.

Expect the questions to fall into a few buckets. Which agency staff reviewed each application, and what evidence did they weigh? Did the approvals require site verification? Does Commerce keep a list of end customers behind each assembler? Who signs off when a licensee turns out to share people with a blacklisted company? None of those answers appear in the public reporting so far.

Nvidia’s position is delicate. The company sells to approved parties under rules it did not write, and it can argue that Commerce decides who qualifies. Critics will counter that a chipmaker with deep China experience can run its own checks on partners. Both arguments have merit, and neither answers the factual question of what each licensee actually does.

Five Predictions for the Weeks Ahead

These are our judgments based on how similar episodes have unfolded, not confirmed events. Treat them as informed forecasts that could miss.

Document requests from Congress. Lawmakers will likely ask Commerce for the full licensee list, the vetting files, and any site-visit records. A written demand is the cheapest first move and the most probable.
New verification conditions. Commerce may add address checks and ownership disclosures to future approvals, since an empty-building finding is an easy failure to prevent on paper.
Pressure on existing approvals. At least one of the named firms could see its approval reviewed or paused. This is plausible but not certain, because the agency would have to publicly admit a screening miss.
Tighter partner diligence at chipmakers and server vendors. Suppliers that sell into China will probably expand their own ownership screening to avoid being the next headline.
A bargaining chip in US-China AI talks. The episode will feed arguments about how to measure and police AI progress, a theme from our coverage of the US-China talks and the shared yardstick proposal.

What Developers and AI Teams Should Watch

Engineers who build on global GPU supply chains should track three signals. First, any change to H200 availability or lead times for Chinese buyers, which would show up in cloud pricing and server delivery schedules. Second, new compliance language in vendor contracts that asks who owns the buyer and how the chips will be used. Third, any official statement from Commerce on how it will review existing approvals.

Teams that resell or host GPU capacity should also review their own customer onboarding. The Zhejiang Yindun angle shows that a legitimate data center can still become part of a sanctions story through its tenants. A simple know-your-customer step, covering ownership, address verification, and intended use, is cheaper than explaining a gap later. For broader context on the models and chips driving this demand, browse our AI and machine learning coverage.

The Bottom Line on the Nvidia H200 China Approvals

The takeaway: the reported facts are narrow but serious. Dozens of Chinese companies received approval to handle H200 chips, two of the named firms had empty or deserted listed offices, and others have ties to companies that serve the Chinese military or sit on a U.S. blacklist. The stronger claims, including front companies and offshoots, remain analyst opinion for now.

What happens next depends on Commerce. If the agency can show it verified these companies by other means, the story fades into a debate about definitions. If it cannot, the Nvidia H200 China program becomes a case study in how a licensing gate can look strict on paper and porous in practice. Smaller outlets such as AI Weekly and Santage have summarized the Times findings, and we will update this analysis as the agency, Nvidia, or the companies respond.

Frequently Asked Questions

What did the U.S. approve for Chinese companies?

The Commerce Department approved dozens of Chinese companies to handle Nvidia H200 AI chips or assemble them into AI servers, according to The New York Times. The approvals cover handling and assembly, and the coverage does not confirm any quantity or value.

Which companies are named in the reporting?

The coverage names New Matrix, Maginfra, Zhejiang Yindun Cloud Technology, and Shanghai Jiatang. It also names Wise Road, Inspur, and Range Intelligent Computing Technology Group as connected entities.

Is New Matrix confirmed as a front company?

No. Analysts describe New Matrix as a front or facade for Wise Road-linked firms, but the available reports present that as an assessment. Corporate records do tie its owners to Wise Road, and its listed Beijing address was found vacant.

Is Maginfra an Inspur offshoot?

That is unconfirmed. Reports say Maginfra shares personnel and patents with Inspur, and analysts believe the two are related. No source we reviewed establishes a confirmed corporate relationship.

Did any H200 chips reach the Chinese military?

The reporting does not say so. It documents ties between approved firms and companies that serve military contractors, but it does not allege a chip transfer to a defense customer.

Has Nvidia been accused of wrongdoing over the approvals?

Not in the sources we reviewed. The approvals came from the Commerce Department. Nvidia faces separate scrutiny elsewhere, including the matter The Verge reported involving Jensen Huang, which is not connected to these approvals in the available coverage.

What happens next?

No official response is confirmed. Congressional questions to Commerce and Nvidia are the likeliest next step, and any change to the approvals would come from the agency.