On November 13, 2026, two stories about Nvidia and China landed on the same day.

The first was political. Jensen Huang had resigned as Nvidia's CEO the day before, after an internal investigation confirmed an extramarital relationship with a woman identified in media reports as Alice Huang, who was alleged to be a Chinese intelligence operative. U.S. lawmakers sought answers from the former CEO. Beijing's Foreign Ministry dismissed the espionage allegation as "groundless."

The second story was physical. Reuters published a review of Chinese corporate disclosures, investment plans, and procurement records. It found more than $3 billion in Chinese AI computing projects and equipment agreements designed around Nvidia's H100, H200, and H800 processors — chips that U.S. export controls restrict from sale to China. One project identified in the review, in Shaanxi province, specified a 3 billion yuan computing center built on H100 and H200 hardware, awarded to a subsidiary of the state-owned Sinomach group.

The first story is about a person. The second is about the hardware that person's company built. The political drama will dominate the headlines for weeks. The hardware will remain in China for years.

The CEO Left. The Chips Stayed.

The sequence matters for understanding what the scandal does and does not change.

Jensen Huang's departure removes the most visible architect of Nvidia's China strategy. He was the executive who negotiated the H200 arrangement with the Trump administration — the deal that allowed sales to approved Chinese customers in exchange for a 25 percent cut of revenue routed through the United States. He was the public face of Nvidia's argument that restricting chip sales to China would only accelerate domestic alternatives.

His resignation does not remove the H100, H200, and H800 processors that are already inside Chinese data centers. It does not remove the $3 billion in projects that Reuters identified, many of which are already under construction or in procurement. It does not remove the demand that created those projects.

A CEO resignation changes who negotiates the next licensing arrangement. The physical stock of chips that has already been installed, and the pipelines that are designed around them, remain where they are.

The $3 Billion Pipeline

The Reuters review is the most detailed public accounting of how restricted Nvidia hardware continues to reach Chinese buyers despite export controls.

The projects are not marginal. They are built by state-owned engineering groups, research institutions, and private developers. The Sinomach subsidiary contract in Shaanxi is a state-owned enterprise project specifying H100 and H200 processors. Other projects identified in the review span multiple provinces and include both new construction and equipment procurement agreements.

The review does not resolve how the chips reached China. Some may have been purchased before the restrictions took effect. Some may have arrived through transshipment routes. Some may have been sourced from the gray market that has developed around restricted accelerators. The Reuters review documents the demand and the procurement, not the logistics of every shipment.

But the scale of the pipeline tells you something about the effectiveness of the export control regime. If more than $3 billion in projects are being designed around chips that are supposed to be unavailable, the controls are not preventing the projects. They are changing the price, the routing, and the paperwork.

The Blackwell Fingerprints

The same week, a separate research effort added a more uncomfortable data point.

A team led by former OpenAI researcher Will DePue published findings that several leading Chinese open-weight models — including DeepSeek V4, Moonshot's Kimi K3, MiniMax models, and Ant Group models — bear numerical "fingerprints" consistent with training on Nvidia's Blackwell GPUs, which are banned from export to China or Chinese-owned firms.

The methodology is based on analyzing public model weights for computational signatures that correlate with specific hardware. TechCrunch's reporting noted that the fingerprints do not prove a rule was broken. They show that the training compute used to produce these models is consistent with hardware that the U.S. government has prohibited from reaching China.

The finding complicates the narrative that export controls have successfully constrained China's frontier training capacity. If the models that define China's open-weight frontier were trained on the most advanced restricted hardware, then the controls have not stopped the training. They have redirected the procurement.

The DePue team's analysis used direct bit-for-bit verification for DeepSeek-V4-Flash on a rented B300 system, and correlation analysis for Kimi K3 that reached a coefficient of 0.999995 across roughly 800 weight matrices, according to GateNode's technical review. The group also conducted black-box statistical tests on MiniMax models, DeepSeek-V4-Pro, and two Ant Group models, all of which the researchers found consistent with Blackwell-class hardware rather than prior-generation H100 systems.

Those findings are forensic correlations, not proof of diversion. But they narrow the range of explanations. If the models were trained on hardware the U.S. has restricted, the question becomes how the hardware got there — not whether it did.

What the Scandal Actually Changes

The political consequences of the Huang affair are real. U.S. lawmakers will seek testimony. Nvidia's board will face questions about its internal controls. The company's China strategy will be reviewed by a new CEO who did not design it.

But the structural facts that the scandal has exposed are not new. They were documented before Huang's resignation.

The $3 billion in restricted-chip projects was reported on November 13, the same day the resignation became public. The Blackwell fingerprints were published the same day. The two stories are not causally linked. They are two measurements of the same underlying reality: the demand for advanced Nvidia hardware in China did not disappear when the export controls were imposed, and it has not disappeared now that the CEO who negotiated the controls is gone.

A new CEO may take a different approach to China — more cautious on licensing, more restrictive on end-user verification, or more willing to walk away entirely. Those are policy choices.

The chips that are already in China are not a policy choice. They are inventory.

The Demand That Doesn't Resign

The Reuters and DePue findings point to a demand curve that is not responsive to political events.

Chinese AI developers need compute. The domestic alternatives — Huawei's Ascend, Cambricon, Hygon — are improving, but they are not yet substitutes at the frontier. The models that Chinese labs are training and deploying require the memory bandwidth, the interconnect, and the software ecosystem that Nvidia provides. The demand for that capability does not vanish because a CEO resigns.

The export control regime has made that demand more expensive, more complicated, and more dependent on gray-market channels. It has not eliminated it. The $3 billion pipeline is the measure of how much demand persists despite the controls. The Blackwell fingerprints are the measure of how much training has already happened on hardware the controls were meant to block.

The scandal may change who runs Nvidia. It will not change what the chips in China are being used for.

What Comes Next

The political process will produce testimony, documents, and headlines. The new CEO will produce a strategy review. The Chinese Foreign Ministry will continue to deny the espionage allegation. The news cycle will move on.

The chips will not move.

The H100, H200, and H800 processors that Reuters identified in Chinese projects will continue to train models, serve inference, and generate revenue for the companies that own them. The Blackwell GPUs that DePue's research detected will continue to leave their fingerprints in the next generation of Chinese open-weight models. The $3 billion pipeline will be joined by new projects, designed around whatever hardware is available at whatever price the market will bear.

Jensen Huang is gone. The chips are not. The scandal will not bring them back.

Sources: CNN (November 12–13, 2026); The Guardian (November 12–13, 2026); Reuters via Financial Express (November 13, 2026); NDTV (November 13, 2026); Economic Times Infra (November 13, 2026); CNBC TV18 (November 13, 2026); Sedaily English (November 14, 2026); TechCrunch (November 13, 2026); Dataconomy (November 13, 2026); GateNode on DePue's bit-for-bit verification and Kimi K3 correlation (November 13, 2026); Bernama (November 13, 2026); EFE (November 13, 2026); Antara News (November 9–10, 2026).

Disclaimer

The information provided in this article is for general informational and educational purposes only. It does not constitute legal, financial, or professional advice. The author and publisher are not responsible for any actions taken based on the content of this article. Readers should consult qualified professionals for advice specific to their situation. All trademarks and references to third-party products, services, or organizations are the property of their respective owners. The performance data and benchmarks discussed are based on specific research studies and may not generalize to all use cases or environments. As of the publication date, the AI landscape continues to evolve rapidly, and readers should verify current information independently.

Limitations

Reuters' $3 billion figure aggregates procurement notices, investment plans, and construction contracts; it is not a measure of Nvidia chip purchases alone.

Whether specific chips entered China legally, through transshipment, or via offshore data centers is unresolved; the weight-forensics findings are correlations, not proof of diversion.

Nvidia did not respond to requests for comment, and Chinese customs data on advanced chip imports is not publicly disclosed at this granularity.

The espionage allegation is attributed to U.S. officials and denied by Beijing; it is contested and unproven.

Reporting standards and disclosure requirements differ between Chinese and U.S. sources, so figures are not always directly comparable.

Market size and forecast estimates come from research firms whose methodologies vary; treat them as directional.

The sectors discussed are changing quickly, and readers should verify current figures independently.


Sources

  1. CNN (November 12–13, 2026)
  2. The Guardian (November 12–13, 2026)
  3. Reuters via Financial Express (November 13, 2026)
  4. NDTV (November 13, 2026)
  5. Economic Times Infra (November 13, 2026)
  6. CNBC TV18 (November 13, 2026)
  7. Sedaily English (November 14, 2026)
  8. TechCrunch (November 13, 2026)
  9. Dataconomy (November 13, 2026)
  10. GateNode on DePue's bit-for-bit verification and Kimi K3 correlation (November 13, 2026)
  11. Bernama (November 13, 2026)
  12. EFE (November 13, 2026)
  13. Antara News (November 9–10, 2026).

The information provided in this article is for general informational and educational purposes only. It does not constitute legal, financial, or professional advice. The author and publisher are not responsible for any actions taken based on the content of this article. Readers should consult qualified professionals for advice specific to their situation. All trademarks and references to third-party products, services, or organizations are the property of their respective owners. The performance data and benchmarks discussed are based on specific research studies and may not generalize to all use cases or environments. As of the publication date, the AI landscape continues to evolve rapidly, and readers should verify current information independently.

Limitations: Reuters' $3 billion figure aggregates procurement notices, investment plans, and construction contracts; it is not a measure of Nvidia chip purchases alone.; Whether specific chips entered China legally, through transshipment, or via offshore data centers is unresolved; the weight-forensics findings are correlations, not proof of diversion.; Nvidia did not respond to requests for comment, and Chinese customs data on advanced chip imports is not publicly disclosed at this granularity.; The espionage allegation is attributed to U.S. officials and denied by Beijing; it is contested and unproven.; Reporting standards and disclosure requirements differ between Chinese and U.S. sources, so figures are not always directly comparable.; Market size and forecast estimates come from research firms whose methodologies vary; treat them as directional.; The sectors discussed are changing quickly, and readers should verify current figures independently.