Jensen Huang is no longer Nvidia's chief executive. That sentence would have sounded impossible a year ago. The circumstances of his departure make it worse: according to documents reviewed by Bloomberg on November 13, 2026, banned Nvidia chips, routed through Singapore, ended up in a Shanghai data center linked to a company that Alice Huang founded. Alice Huang is the woman at the center of Jensen Huang's admitted affair. The two threads, corporate scandal and national security, have collided into the biggest crisis in Nvidia's 32-year history, and they are doing so at the worst possible moment, when AI chip demand has never been higher and when the gap between US policy and Chinese compute capability has never been more consequential.
What Bloomberg's Documents Actually Show
In January 2025, a subsidiary of Shanghai Longteng, a Chinese company founded by Alice Huang in 2015, pledged 193 servers as collateral for a loan, according to Chinese loan filings reviewed by Bloomberg. The loan filings disclose the specific hardware: 22 Dell servers equipped with Nvidia H800 chips and 52 Super Micro systems carrying Nvidia H100 and H200 chips. All three chip models were restricted from export to China. The H800 was banned from Chinese sales in 2023 by the Bureau of Industry and Security. The H100 and H200 required a US export license that the transaction does not appear to have obtained.
The servers were physically located at a data center in Shanghai linked to Megaspeed International, a Singapore-based entity that the US Justice Department had been investigating for alleged chip smuggling prior to Bloomberg's report. Alice Huang sold her shares in Shanghai Longteng in 2017, and Bloomberg has not established that she personally directed the purchase of the servers. But her founding role at the Chinese company, her alleged ties to Megaspeed, and her relationship with Jensen Huang have drawn the attention of US federal prosecutors regardless of that technicality.
Some of the servers traced back to Aperia Cloud Services, also known as Aperia International, a Singapore-headquartered Nvidia Cloud Partner. Aperia has denied wrongdoing and declined to comment further. Dell halted all sales and support to Aperia International in February 2025 after chip smuggling concerns surfaced internally at Dell's compliance team. Nvidia removed Aperia from its public partner list after the company's CEO was arrested. Nvidia's own SEC 10-Q filing for the first quarter of fiscal 2027 acknowledges the company is "effectively foreclosed from competing in China's data center computing market," an on-record admission that the authorized channel into China has closed, though it does not explain how the chips already there arrived.
The Route the Chips Took: Singapore, Malaysia, and the Transshipment Loop
The 193 servers in the Shanghai loan filing represent a documented instance of a larger, well-established smuggling architecture. Chips leave authorized US distributors destined for entities in Singapore, Malaysia, or the United Arab Emirates. They are incorporated into servers, reshipped, and end up in Chinese data centers with tampered or missing Nvidia serial number records. TechCrunch, citing Epoch AI analysis, estimated that approximately 2 million advanced chips entered China through these schemes between 2024 and 2025 alone. That figure comes from Epoch AI's analysis of import records, shipping manifests, and secondary market data, and the organization notes it represents a floor, not a ceiling.
The Malaysia route is now documented in Epoch AI's data: 275,000 Nvidia AI accelerators were imported to Malaysia across those two years, volumes that far exceed any plausible local demand from Malaysia's data center market at that time. The H200 chip occupies a gray zone in this trade. It is less powerful than the H100 and was initially permitted for export to some markets outside China. But exporting it to China without a US license is still prohibited, and there is no indication that the Shanghai Longteng loan collateral servers carried the required documentation. Meanwhile, approximately 600,000 H200 chips are reportedly stockpiled in Hong Kong awaiting onward movement into China, their shipment currently on hold for reasons that have not been publicly explained by any government or company involved.
The scale of the transshipment architecture reflects how quickly the market adapted to US export rules. When restrictions went into effect, prices for restricted chips on secondary markets in China spiked to two and three times their US list price, creating an enormous financial incentive for anyone positioned between an authorized Nvidia distributor and a Chinese buyer. Megaspeed, investigators now believe, was one of the best-positioned intermediaries, with offices in Singapore giving it access to Nvidia's authorized supply chain and corporate relationships that provided cover for what was actually moving through its logistics network. The arrested Aperia CEO is the most direct link now in investigators' hands.
Jensen Huang's Exit and What It Changes for Nvidia
Jensen Huang admitted to a romantic affair with Alice Huang and departed Nvidia this week, as CNBC reported on November 12. The timing is not coincidental. Nvidia's potential corporate liability for export control violations grows materially if federal investigators can establish that company executives had knowledge of, or actively facilitated, the diversion of restricted chips. In May 2025, Jensen Huang publicly vouched for Megaspeed, stating that Nvidia officials had visited its data centers and found no evidence of chip diversion. That statement now carries real legal consequence as a direct contradiction of the Bloomberg documents, and it is exactly the kind of on-record denial that prosecutors look for when building an obstruction or false statement case.
In May 2026, Culper Research published a report arguing that Alice Huang's companies were backed by Chinese entities barred from purchasing US chips, and that Nvidia had a "China problem" that management was not disclosing to investors. Nvidia disputed those findings at the time and its stock price held. The Bloomberg documents, showing banned chips in a data center physically connected to Megaspeed's network, now give federal prosecutors concrete facts to work from rather than a short-seller's allegations. The question is no longer whether the chips got there. The question is who knew and when, and whether anyone at Nvidia's senior level made statements that federal investigators can prove were false.
Commerce Secretary Howard Lutnick is also reportedly under pressure. Sources cited by TechCrunch say Lutnick could face calls to resign after reports that his investigators failed to identify suspicious Nvidia chip buyers and maintained unusually close relationships with Nvidia officials during the period when the smuggling was occurring. That adds a second political dimension to what was already a corporate crisis, complicates the enforcement response going forward, and raises the question of whether the investigation will be allowed to reach Nvidia's senior management or be contained at the distributor level where individual actors can be prosecuted without implicating the company that made the chips.
When Banned Silicon Trains Frontier Models
The chip smuggling problem would matter less if the chips were sitting idle in warehouses. They are not. Independent researcher Will DePue, writing for Unemployment Labs and cited by TechCrunch, analyzed open-weight models from DeepSeek, Moonshot AI, MiniMax, and Ant Group and found fingerprints consistent with Blackwell chip training in Moonshot's Kimi K3 and DeepSeek's V4. Blackwell chips are banned from export to China under any license. White House technology adviser Michael Kratsios had already stated, citing US intelligence sources earlier in 2026, that Moonshot used export-banned Nvidia chips to train its models. The FBI investigation into Jensen Huang's conduct is now running parallel to the chip-tracing exercise, connecting the corporate and national security threads in a single investigative file.
What does it mean when two of China's most capable frontier models were trained on hardware that US export controls were specifically designed to block? At minimum, it means the gap between what US policy assumed China could compute and what China actually computed is larger than the official record shows. DeepSeek V4 and Kimi K3 have been released as open weights, meaning the capabilities embedded in that compute are now globally accessible regardless of what happens next to the chip supply chain. The policy damage is not reversible by tightening future export rules. The inference capacity is already out there, already in the hands of developers worldwide who downloaded the weights without asking where the training hardware came from.
The Bear Case for US Export Controls
Critics argue, however, that the focus on individual smuggling networks misses a structural reality: the global semiconductor supply chain was not designed for the granular custody tracking that US export policy now demands. Nvidia sells through authorized distributors who sell to resellers who sell to cloud partners who sell to end users. Each step adds opacity. The 193 servers that appeared in a Shanghai loan filing did not arrive in a suitcase. They passed through multiple commercially legitimate-looking entities in a chain too long for standard compliance audits to trace reliably, and the loan filing that exposed them was not discovered by US enforcement agencies but by Bloomberg journalists reviewing Chinese court documents.
Skeptics of the current enforcement regime also note that the chip models documented in Shanghai, the H800, H100, and H200, were frontier-class in 2023 and 2024 but have since been superseded by Nvidia's Blackwell architecture. The Blackwell fingerprints DePue found in Chinese models suggest that the enforcement gaps that allowed H-series chips to reach China remain open for Blackwell-class hardware. If 2 million chips reached China under a regime nominally being enforced, the number that will reach China under a regime that investigators now suspect was undermined from within is an open question. The policy may have slowed China's AI compute accumulation by months, not years. And China's domestic chip industry, which had been gaining ground at SMIC and Huawei's semiconductor division before the latest restrictions, now has a clear mandate to close that gap entirely rather than depend on smuggling routes that are increasingly visible to US intelligence.
What Investigators Need to Prove and When
Federal prosecutors face a layered evidentiary task. The loan filings establish that banned chips were physically present in a Shanghai data center. What they do not yet establish is a chain of custody that implicates specific individuals who can be prosecuted, extradited, or sanctioned. Alice Huang's role in Shanghai Longteng ended in 2017, before the H800 chip existed. Any prosecution would need to prove that the chips were knowingly diverted, that specific actors facilitated that diversion, and that those actors fall within US legal jurisdiction or the reach of allied enforcement partners in Singapore or the European Union. Per reporting by Reason, the investigation into the chip diversion network stretches back several years and involves multiple agencies, suggesting prosecutors have been building the evidentiary record longer than the Bloomberg story implies.
The Megaspeed investigation, led by the Justice Department, is now the central thread. Megaspeed operated in Singapore, which gives US prosecutors leverage through extradition agreements and international financial cooperation. Nvidia's earnings call in the coming weeks will force new management to answer questions about the company's compliance programs, its prior endorsements of partners now under federal investigation, and the potential cost of any civil or criminal settlement with the Commerce Department or the DOJ. For a company generating more than $44 billion in quarterly revenue, a financial settlement may be manageable in isolation. A prolonged investigation that ties up Nvidia's ability to close new data center partnerships, combined with a CEO transition at a moment when AI chip demand has never been higher and China is recommitting to its domestic semiconductor industry, is a risk of a different magnitude.
Two million advanced chips reached China while Nvidia vouched for the partners that moved them. That gap between what was permitted and what happened is now simultaneously a federal investigation, a CEO departure, and a foreign-policy reckoning compressed into one week.
Key Takeaways
- 193 servers with banned Nvidia H800, H100, and H200 chips were pledged as loan collateral in Shanghai in January 2025, confirming hardware from three restricted chip models reached a Megaspeed-linked data center, per Bloomberg documents reviewed November 13.
- Jensen Huang departed Nvidia this week after admitting to an affair with Alice Huang, whose company Shanghai Longteng is linked to the servers at the center of the Bloomberg investigation, raising questions about what senior Nvidia leadership knew about Megaspeed.
- Approximately 2 million advanced chips entered China through smuggling schemes in 2024 and 2025, per Epoch AI analysis, with 275,000 Nvidia AI accelerators transiting Malaysia alone as part of a transshipment architecture that adapted within months of each new US export restriction.
- DeepSeek V4 and Moonshot Kimi K3 show fingerprints consistent with Blackwell chip training, per independent analysis, suggesting export-banned compute is already behind China's most capable open-weight models, now globally distributed and irreversible.
- 600,000 H200 chips are reportedly stockpiled in Hong Kong awaiting movement into China, while the Justice Department-led Megaspeed investigation widens and Commerce Secretary Howard Lutnick faces pressure over enforcement failures during his tenure.
Questions Worth Asking
- If Blackwell fingerprints already appear in Chinese open-weight models now globally available for download, what can tightening future export controls realistically achieve when the most dangerous capabilities are already publicly distributed?
- How much of Nvidia's compliance failure came from deliberate choices by senior executives versus the structural impossibility of policing a supply chain that passes through dozens of intermediary entities across multiple jurisdictions?
- If Commerce Secretary Lutnick faces accountability for enforcement gaps that allowed 2 million chips to reach China on his watch, does that signal a wholesale restructuring of US chip export oversight, and which countries and competitors benefit most from continued enforcement confusion?
