The Huang spy story, and who decides which AI you can run
Jensen Huang has resigned from Nvidia and the spy allegation is unproven. A scandal about one executive is not evidence against open models, and it should not decide which models the rest of us can use.
A scandal about one person is not evidence about open models, and it should not become the reason to limit which models the rest of us can run. I would change my mind if someone showed that a specific Chinese model leaks data or carries a backdoor, but nothing published about the Huang affair does that.
On 12 November 2026 Nvidia said Jensen Huang had resigned as chief executive and director, effective immediately, with chief financial officer Colette Kress as interim CEO1. Days earlier Huang had acknowledged the affair and apologised in an open letter. The woman, Alice Huang, is a former executive of Megaspeed International, a Singapore data centre company, and the two are not related. The New York Times reported that US intelligence had determined she was gathering information for Beijing2. That is an allegation. US Attorney General Todd Blanche said on 12 November that the Justice Department had reached no conclusions about who she is or is not3, and Beijing rejects the claims4.
In my earlier LinkedIn post I said the timing looked odd. I would put it more carefully now. The Times first reported on her closeness to Huang on 9 October 20252. The FBI has been investigating for at least a year, according to CNN3. The Times also reported that Trump was told about it and took no action, and Trump said on 9 November that he did not hear about it before his China trip5. A short seller, Culper Research, had flagged the relationship in May5. So the story was not new. Who prompted the latest round of coverage, I don't know. I found no named source and I won't guess.
Now the chips. Nvidia designs them and TSMC manufactures them. Even if design details leaked, manufacturing is what China lacks at the leading edge, and Nvidia's board said it had no reason to believe national security or company secrets were compromised4. Smuggling is a separate question. The Times reported that Megaspeed bought more than $4.8 billion of Nvidia chips over two years, and news reports said the Commerce Department was investigating whether it helped sidestep export rules. Nvidia said its compliance team had found nothing illegal2. I treat all of that as allegation. What is documented is that Beijing has held chips back itself. In January it reportedly told some firms to halt orders for Nvidia's H2006, and Reuters reported on 5 November that about 600,000 chips were waiting in Hong Kong while Beijing withheld approval4. China is also deciding which chips it takes.
Power sits on the other side of the equation. The usual summary is that the US has the chips and is short of power, while China has the power and is short of chips7. BloombergNEF estimates China will add more than six times as much generation capacity as the US over the next five years7. A Bernstein analyst, quoted in December 2025, said China's lack of leading-edge chip capacity was the tighter constraint in the near term8. Both can be true. Neither bottleneck is permanent.
Then there are the models. Chinese models took a plurality, 41%, of Hugging Face downloads over the year to February 20269. By August, Qwen had 2,045 million downloads in 2026 and 151,448 derivative models on the hub10. One caution: models under 1 billion parameters account for 83% of all-time downloads10, so downloads say more about small models than about the frontier. Kimi K3 was reported to rank third on the Artificial Analysis index, behind two US closed models11. I run these models myself. They are cheap and very usable.
This is what I keep saying about harness. When the best open model sits that close to the closed ones, the model is not what separates one agent from another. The harness does: the tools, the context and the workflow around the model. That makes the freedom to pick, inspect and adjust the model more valuable, not less.
Huang's argument is also commercial. He told Axios on 22 July that cheaper open models bring more people to AI and raise demand for chips12, and he sells chips. I take that as given. It does not decide whether open weights are good policy. The other side has real points. The administration was reported to be weighing limits on Chinese open-weight models after Moonshot's Kimi K3, amid allegations that it was trained on outputs from a US model13. The industry letter Huang shared on 24 July asks for "targeted legal and commercial frameworks rather than sweeping restrictions"14, which treats misconduct and the models as separate questions. I find that fair. No ban has been announced, and the Times reported that the administration backed off for now after pushback from Silicon Valley15.
My worry is the sequence. A scandal about one executive lands while the policy fight over open models is unsettled, and a list of "government approved" models becomes easier to argue for. I am not claiming the two are connected, because the timeline above does not show that. But a list like that decides who gets to inspect and change the tools people think with, as well as what counts as secure. Australia has few model providers of its own. We should build the infrastructure to host open models here, so that no single country's policy decides what we can use.
Nothing here is financial, legal or investment advice. Every figure is published by its named source and carries the date it was published; figures may since have changed.
References
-
NVIDIA, NVIDIA Announces CEO Transition (12 November 2026, filed with the SEC) ↑
-
Associated Press via PBS NewsHour, Nvidia and Huang face lawsuit alleging investors were misled about spy affair (10 November 2026) ↑
-
CNN, US intel isn't sure if businesswoman spied on Nvidia for China, doesn't mean she wasn't a risk, sources say (12 November 2026) ↑
-
Open Magazine, EXPLAINED | Trump Backs Jensen Huang. Why Are Nvidia's Chips Stuck in Hong Kong? (November 2026, citing Reuters) ↑
-
Newsweek, What did Trump admin know about Nvidia spy claims? Unanswered questions (November 2026) ↑
-
Reuters, China asks tech firms to halt orders for Nvidia's H200 chips, Information reports (7 January 2026) ↑
-
Al Jazeera, China's secret weapon in AI race with US? Lots of cheap energy (28 May 2026) ↑
-
Implicator.ai, The AI race America is losing happens underground (11 December 2025) ↑
-
Hugging Face, State of Open Source on Hugging Face: Spring 2026 (17 March 2026) ↑
-
Hugging Face, State of Open Models: Summer 2026 Observations (14 August 2026) ↑
-
TweakTown, NVIDIA CEO Jensen Huang joins X with post backing open-weight AI (July 2026) ↑
-
Axios, Exclusive: Nvidia's Jensen Huang defends Chinese AI amid Kimi panic (22 July 2026) ↑
-
eWeek, Nvidia, Microsoft Back Open-Weight AI Models (27 July 2026) ↑
-
NVIDIA and other signatories, Open Weights and American AI Leadership (24 July 2026) ↑
-
The Decoder, Silicon Valley's rift over open source pushes back contemplated White House bans on Chinese AI (2026) ↑