Chinese State Financing Linked to Nvidia Blackwell Chips in Restricted Data Centers
Semi-Tech Leasing, controlled by Shenzhen and Beijing municipal entities, financed 700+ servers including at least 32 equipped with export-restricted Nvidia B300 Blackwell chips, then retroactively redacted hardware details from China's public credit registry, per Bloomberg.

A Chinese municipal-backed leasing firm quietly funded the acquisition of export-banned Nvidia AI chips, then scrubbed the evidence from its own public filings.
Key takeaways
- Semi-Tech Leasing Group, majority-controlled by Shenzhen and Beijing municipal governments and originally seeded by China's national semiconductor fund, financed more than 700 servers for a Chinese tech firm, including at least 32 equipped with export-restricted Nvidia B300 Blackwell chips, per People's Bank of China credit registry filings reviewed by Bloomberg.
- After Bloomberg reviewed the filings, Semi-Tech refiled the registry documents with hardware descriptions, supplier names, and server model numbers stripped out, a retroactive redaction suggesting the parties knew the transactions were legally problematic.
- U.S. export controls target hardware movement, not the state-backed financing layer funding the acquisitions. The gap is structural, and neither Nvidia's investigation nor Asus's boilerplate compliance statement closes it.
Semi-Tech Leasing Group, a Chinese financing company majority-controlled by municipal government entities in Shenzhen and Beijing and originally seeded by China's national semiconductor investment fund (the "Big Fund"), financed the purchase of more than 700 servers for Glory View Technology, per PBOC credit registry filings first reported by Bloomberg on October 1, 2026. At least one contract in those filings specifically identified 32 Asus-built servers equipped with Nvidia's B300 Blackwell chips, a chip class that requires a specific U.S. government license to export to China. No such license has been confirmed.
The financing began in August 2025. The first lease, per Bloomberg, covered 401 million yuan (approximately $60 million) for 171 servers, structured as a sale-and-leaseback arrangement repayable over 63 months. Combined borrowings from Semi-Tech exceeded 3 billion yuan (approximately $420-450 million at current exchange rates, per various reports citing Bloomberg). The servers, including the Blackwell-equipped units, were recorded in the filings as installed at a China Mobile data center park in Zhongwei, Ningxia.
The Redaction Is the Story
The filings prove financing contracts that name specific hardware. They do not, by themselves, confirm physical delivery or establish a legal violation. That distinction matters for anyone tempted to overstate what the documents show.
What the documents show unambiguously is that someone decided the details needed to disappear. Per Bloomberg's reporting, after its review of the PBOC credit registry filings, Semi-Tech refiled those documents with hardware descriptions, supplier names, and model numbers stripped out. The original filings, which explicitly named the chips, had been public. The refiling occurred before Bloomberg published on October 1, according to Bloomberg's account.
That sequence is not a compliance accident. State-adjacent entities do not quietly purge their own public registry records by mistake. The redaction is confirmation that the parties recognized legal exposure and moved to eliminate the paper trail inside China's own financial infrastructure.
Nvidia told Bloomberg it is investigating the matter with its equipment-manufacturing partners. Asus, identified as the server manufacturer in the original filings, said it remains committed to complying with international export control regulations. Both statements were paraphrased by Bloomberg; no verbatim quotes were available.
Export Controls Have a Financing Blind Spot
The Bureau of Industry and Security's enforcement framework focuses on physical export and end-use verification. It targets the hardware movement. It does not, in any operational sense, target the capital structures that fund the acquisitions upstream.
Semi-Tech is a leasing vehicle majority-controlled by Chinese municipal governments, originally capitalized by a state semiconductor fund, now deploying more than 11 billion yuan (approximately $1.6 billion, per Bloomberg) into compute infrastructure. The mechanism here is fiat credit: sale-and-leaseback arrangements, municipal government guarantees, and a financing entity whose ownership structure runs directly to city governments in Shenzhen and Beijing. That is sovereign capital being deployed through a structure that looks, on paper, like a commercial leasing company.
Washington's compute policy will remain theater unless enforcement extends to the capital layer. Sanctioning entities that provide financing for restricted chip acquisitions, not just entities that physically move chips, is the logical extension. Without it, the export control regime is a toll booth on one lane of a six-lane highway.
The geographic detail sharpens the picture. Zhongwei, Ningxia is the same region that hosted large-scale Bitcoin mining operations before China's 2021 ban, drawn by cheap power and proximity to renewable generation. China cleared that grid capacity and built out state-backed AI data center infrastructure in its place. That infrastructure is now reportedly being stocked with export-banned Nvidia hardware financed by municipal government credit vehicles.
What to Watch
The operative question is whether BIS or Treasury moves against the financing chain, not just the hardware chain. If Semi-Tech, Glory View, or the intermediary seller faces enforcement action that demonstrably deters future deals, the thesis that financial-layer controls are absent weakens. If enforcement stays focused on the physical export chain while leasing vehicles continue funding restricted acquisitions, the control regime's strategic value is limited. Watch for any BIS Entity List additions in the leasing and semiconductor finance sector, and watch whether the Taiwan indictment track, Taiwan's Keelung District Prosecutors' Office indicted nine people in August 2026 over alleged B300 server smuggling to China, expands to include financing intermediaries.
Update, November 4, 2026
Nvidia has moved beyond hardware-level export controls and is now pressing its own distribution network. The Financial Times reported Tuesday that Nvidia has instructed its global data center partners to ensure at least 51% of their cloud computing services are sold to non-Chinese companies, a restriction that did not previously exist.
Nvidia communicated these requirements verbally through its sales force and regional teams rather than in written documents. The directive applies to partners worldwide, not just those operating in Asia.
The RTX Pro 5500 situation crystallizes how thin the remaining legal channel into China has become. While B300s are banned from export to China outright, the RTX Pro 5500 workstation card is cleared for shipping to the country, and Alibaba, Tencent, ByteDance, and other Chinese AI and cloud firms ordered roughly six million units in total. They will receive only a few tens of thousands over the next four months.
Bloomberg reported Nvidia will fill just about 1% of the RTX Pro 5500 orders placed over the past four months, with the total order book estimated at roughly $70 billion at more than $11,000 per unit.
As Nvidia and TSMC allocate capacity toward next-generation products like Rubin, RTX Pro 5500 supply is expected to remain tight.
The downstream effect is the story the Semi-Tech financing structure anticipated. The move creates additional obstacles for Chinese technology companies pursuing AI development, as many have relied on leasing advanced compute capacity through overseas cloud arrangements for model training.
Nvidia has already more than halved the number of Asian customers authorized to buy its AI chips. Chinese hyperscalers squeezed out of compliant supply chains have direct incentive to pursue exactly the kind of state-backed workarounds the Semi-Tech filings documented. The financing layer and the distribution squeeze are two ends of the same pressure campaign.
Update, November 5, 2026
A new Reuters exclusive adds a concrete inventory figure to the squeeze already documented here. About 600,000 Nvidia H200 chips are stored in Hong Kong awaiting shipment to mainland China, with Beijing yet to approve large-volume imports, though several major Chinese internet companies have each obtained about 10,000 H200 chips in recent months.
The scale of the backlog had not been previously reported, according to Reuters.
The bottleneck is now running in both directions. The situation is unusual because the U.S. side has loosened its stance: the Bureau of Industry and Security now reviews H200 export applications on a case-by-case basis, subject to specific criteria, including a requirement that exporters certify sales to China will not reduce semiconductor capacity available to U.S. customers, a condition that could cap shipments at 50% of equivalent U.S. end-use volumes.
Beijing has imposed its own separate constraints on H200 imports, reflecting concerns about dependence on U.S. technology.
The price signal tells you everything about what the inventory freeze means on the ground. Small quantities have reached China through intermediaries and indirect routes, and by end of October an eight-card H200 server was being offered in mainland China for more than 6 million yuan, or about $900,000, roughly three times its price a year earlier.
Servers fitted with Nvidia's newer Blackwell processors, also covered by U.S. export controls, were offered for upwards of 20 million yuan apiece. That is the black-market premium for compute that state-backed financing structures like Semi-Tech's exist to access through other means.
Update, November 7, 2026
The New York Times added a named corporate vehicle to the evasion picture this week. U.S. officials suspect that Megaspeed, a Singapore-based data center company, and its affiliates helped Chinese companies gain access to advanced computing capacity through data centers in Southeast Asia. Megaspeed's Malaysian affiliate purchased at least $4.8 billion in AI equipment containing Nvidia chips between June 2024 and February 2026, per trade records cited by the Times.
The company developed data centers in Southeast Asia and rented computing capacity to Chinese technology giant Alibaba. That is the third-country routing model in practice: buy chips legally in Malaysia, deploy them in a Southeast Asian data center, sell access to Chinese firms that cannot receive the hardware directly.
The intelligence dimension is what escalated this into a congressional matter. U.S. intelligence agencies concluded that Alice Huang, an executive at Megaspeed suspected of helping China obtain restricted Nvidia chips, was also gathering intelligence for Beijing, and that surveillance revealed she had developed a close relationship with Jensen Huang, raising concerns about potential access to sensitive Nvidia or Trump administration policy information.
President Trump was briefed on the intelligence findings roughly a year ago but chose not to act, according to the Times, citing concerns about disrupting Nvidia's chip sales and complicating trade relations with China.
Congress is now moving. In a letter dated November 4, Brian Mast, who chairs the House Foreign Affairs Committee, John Moolenaar, who leads the China Select Committee, and Intelligence Committee Chair Eric Crawford asked Jensen Huang to provide a closed-door briefing on November 18 about the Times report and Nvidia's "security protocols against foreign adversary espionage."
Nvidia maintains that regional cloud customers receive ordinary business information, not sensitive trade secrets or national-security information, and says earlier chip-diversion allegations against Megaspeed were investigated and debunked. The Megaspeed structure is a retail-level version of the same logic the Semi-Tech financing documents: when the front door is locked, state-adjacent capital finds a window, and the hardware moves anyway.
Update, November 14, 2026
The evasion story moved from financing contracts to published model weights this week. Will DePue, a former OpenAI researcher who now runs the nonprofit Unemployment Labs, published a forensic analysis on November 12 showing that six models, specifically DeepSeek-V4-Flash, DeepSeek-V4-Pro, Kimi K3, Kimi K3.1-Pro, MiniMax M3/M3.1-Flash, and Ant Group's Ling-3.1-flash, contain weight patterns matching Blackwell B200/B300 and GB200/GB300 chips. The method is straightforward: DePue's report counts 132 processing cores on Nvidia's older Hopper chips, 148 on the Blackwell B200 and B300, and 152 on the GB200 and GB300.
The same companies' older models match Nvidia's previous generation, Hopper, while the newest ones match Blackwell.
The finding is technical, not circumstantial. There is no leak here and no inside source. Just math, run on the models the labs chose to publish.
The report includes the initial seed values that linked DeepSeek-V3 to Nvidia's Hopper generation and DeepSeek-V4-Flash to Blackwell, so that other researchers can attempt to reproduce the results. The important caveat: strictly speaking, the fingerprint shows which chip set a model's starting numbers, and the report says the evidence "supports, without proving, the inference that Blackwell also performed the training," though it is hard to see why any lab would generate initialization weights on one machine and train on another.
The access mechanism remains an open question, and that is where the Megaspeed structure documented here last week becomes directly relevant. The report notes that Blackwell systems can be rented from data centers outside China, and U.S. rules generally allow Chinese firms to rent chips abroad, even though selling the chips to them is banned.
The Commerce Department's export enforcement arm has reportedly opened a review of how Chinese AI firms reach Nvidia chips overseas. Whether the compute came through third-country rental arrangements or through harder evasion routes, the result is the same: the Unemployment Labs report provides the first publicly documented technical evidence of Blackwell hardware access by Chinese AI developers, where previously U.S. officials and press reports had alleged the access but no technical proof existed.
Update, November 15, 2026
Treasury Secretary Scott Bessent escalated the federal response to Nvidia's China compliance failures well beyond the Commerce and congressional pressure already documented here. Bessent on Sunday questioned whether Nvidia's earnings might have been inflated by inappropriate sales to China and urged the company to hire an outsider as CEO. Speaking on Fox News's Sunday Morning Futures, he asked "what was the culture at Nvidia, and were they pressing these seemingly illicit sales to goose their revenues and their earnings," and said "we need answers in the next 10 days from the board and from the outside counsel."
The backdrop is a leadership vacuum that opened three days before Bessent spoke. Nvidia's board announced November 12 that Jensen Huang has resigned as President, Chief Executive Officer, and board member, effective immediately.
Colette Kress was appointed interim CEO while continuing to serve as CFO, with the board initiating a search for a permanent successor and saying it will consider internal and external candidates. Bessent's answer to that last point was direct: "This makes me think that, as the board conducts a search for a new leader, that they need to go outside the company."
Bessent also said it was "obvious" that some companies doing business with Nvidia were shell companies for Chinese interests, and accused the chipmaker of a "culture of noncompliance." That framing matters for this story specifically. The shell-company network documented in the Megaspeed reporting from November 7 was not a rogue operation the company failed to notice; Bessent is now asserting it reflects how Nvidia conducted business. Nvidia's board has already sounded out Arm chief executive Rene Haas, who spent seven years at Nvidia as vice president and general manager of its computing products business before joining Arm in 2013.
Whether the board or its counsel will answer within 10 days is not established, and Bessent did not say what follows if they do not. Nvidia did not immediately respond to requests for comment on Bessent's remarks.
Sources
- Bloomberg Technology, October 1, 2026 (first reported by Bloomberg; PBOC credit registry filings and Chinese corporate registry documents are the underlying primary records, reviewed by Bloomberg and not independently linkable as standalone public URLs)
- People's Bank of China Credit Registry (underlying primary record; accessible via Bloomberg's review)
Frequently Asked Questions
Are the Nvidia B300 Blackwell chips actually inside China, or only documented in financing contracts?
The PBOC credit registry filings document financing contracts that name the hardware and record the servers as installed at a China Mobile facility in Zhongwei, Ningxia. They do not, by themselves, confirm physical delivery or constitute proof of a legal violation. Bloomberg's reporting does not definitively confirm physical delivery either. The filings establish what was contracted and financed. Whether the chips are physically in a rack in Ningxia is an open question the public record does not fully resolve.
Why don't U.S. export controls cover the leasing and financing layer?
BIS export control enforcement focuses on physical export transactions and end-use verification. It was designed to catch hardware moving across borders, not to monitor the capital structures that fund acquisitions upstream. A Chinese municipal government deploying credit through a leasing vehicle to fund a sale-and-leaseback arrangement for servers is structurally invisible to a framework built around shipment records and export licenses. Extending controls to the financing layer would require either new regulatory authority or a significant expansion of OFAC sanctions targeting entities that provide capital for restricted chip acquisitions, not just entities that touch the hardware directly.
What does Ningxia have to do with Bitcoin mining?
Zhongwei, Ningxia was a significant hub for Bitcoin mining before China's 2021 ban, chosen for cheap electricity and proximity to hydro and wind generation. After the ban cleared that load off the grid, the region became a target for state-backed AI data center development. The same grid characteristics that made Ningxia attractive for proof-of-work mining now make it attractive for GPU compute clusters. The irony is direct: China displaced Bitcoin miners to free up grid capacity, then filled that capacity with AI infrastructure now reportedly stocked with U.S. chips that Washington explicitly prohibited.


