The Fall of Jensen Huang: A Fratricidal War Inside the Empire
Bappa Sinha
JENSEN HUANG, who co-founded Nvidia in 1993 and built it into a company valued at over five trillion dollars at its October peak, resigned as its chief executive on November 12. The official story is a sordid one. On October 29, the New York Times, citing anonymous US officials, reported that Huang had a close relationship with Alice Huang (no relation), the former chief executive of Megaspeed, a Singapore data centre firm and Nvidia customer, whom American intelligence suspected of working for China. Jensen Huang admitted the affair. Reports of a second affair with a subordinate followed, and within two weeks the most powerful man in the artificial intelligence (AI) industry was gone.
It is worth stating at the outset that Huang is no saint. He is a billionaire who argued for selling chips to China because China is a huge market for Nvidia. The issue is not his personal conduct. The issue is who brought him down, and why.
A MEMO, A LEAK, A SCALP
The Washington Post has traced the affair to Marco Rubio, Secretary of State, National Security Adviser and one of Washington's most influential China hawks. Rubio took a CIA report and rewrote it in sharper language as a memo to Trump. The memo's findings were then leaked to the press. CNN reports that CIA officials are not yet convinced Alice Huang knowingly passed anything to Beijing, and Singapore says it has not investigated her for espionage. A former official told the Post that Trump's national security team "wanted to destroy the influence Jensen had over the president".
The tussle is now in the open. Treasury Secretary Scott Bessent has publicly blamed Commerce Secretary Howard Lutnick for failing to blacklist Megaspeed, and has demanded a "dramatic shakeup" at Nvidia with an outsider as chief executive. David Sacks, Trump's former AI czar who now co-chairs his science advisory council, has defended Huang, saying the spy narrative is collapsing and that the only confirmed crime is the leak of a classified memo. Meanwhile the Justice Department is running a criminal probe into Megaspeed, Republican committee chairs threaten to subpoena Huang, and three bills tightening chip controls, already in the Senate version of the annual defence bill, are being pushed into its final text.
TWO FACTIONS, ONE AIM
This is not a morality tale. It is a fight between two factions of the US ruling class over how to contain China.
The first faction wants total denial: no chips, no chipmaking equipment and, increasingly, restrictions on Chinese open source AI models. It includes the security establishment, a bipartisan bloc in Congress and the frontier labs, OpenAI and Anthropic, which sell access to their closed models. Anthropic's chief Dario Amodei has said exporting chips to China is "a bit like selling nuclear weapons to North Korea". The Financial Times notes that Huang's fall swings power back towards these labs. Huang was the chief financier of their rivals, the open models that anyone can download and run on their own hardware, and Nvidia had agreed to buy Hugging Face, the main open model repository, for $12.9 billion.
The second faction, led by Huang, argued that the US should sell China slightly older chips, keep it dependent on American systems, and plough the revenue into research to stay ahead. His motive was profit. He wanted AI models to become cheap commodities so that the monopoly profits stay with the chipmaker. Both factions want US supremacy. They differ only on the means, and on which monopoly collects the rent.
FROM VIDEO GAMES TO AI
Nvidia began by making graphics cards for video games. Drawing a game scene means calculating the colour of millions of pixels many times a second. Each calculation is simple, but millions must be done at once. An ordinary processor (CPU) is like a few brilliant mathematicians solving problems one after another. The graphics processing unit (GPU), a term Nvidia popularised in 1999, is like thousands of schoolchildren each doing a simple sum at the same moment.
In 2006 Nvidia released CUDA, software that let scientists use this army for any calculation, not just graphics. In 2012 a neural network called AlexNet, trained on Nvidia GPUs, beat every rival at recognising images. Neural networks turned out to be exactly this kind of work: billions of simple calculations done in parallel. Most major AI models since have been trained on Nvidia chips, and the company's revenue reached $215.9 billion last years. The GPU chips, together with CUDA, on which a generation of researchers has been trained, form a lock-in. That is the monopoly.
THE CENTRAL BANK OF AI
Huang used these monopoly profits to become what one Wall Street analyst called "the central bank of AI". Nvidia put $30 billion into OpenAI and committed $105 billion to backstop its data centre. It took a 7 per cent stake in CoreWeave, which rents out Nvidia chips, and planned a $10 billion stake in Anthropic's public offering. This is circular financing: Nvidia funds the companies that buy Nvidia chips, sales rise, valuations rise, and more money flows in. Three months ago Huang brought together six financial giants, including BlackRock and Goldman Sachs, in a plan, not yet finalised, to raise over $500 billion in debt against Nvidia hardware.
Many of these deals rested on Huang personally. With a criminal probe under way, congressional hearings ahead and the Treasury Secretary demanding a purge, Nvidia's ability to act as lender of last resort is in doubt. The weakest links, as analyst David Rosenberg notes, are the borrowers who depend on Nvidia's guarantees. The parallel with the telecom bubble of the late 1990s is clear. Equipment makers lent carriers the money to buy their own equipment, and when demand fell short the carriers went bankrupt. The hawks, in striking at Huang, may have stuck a pin in the bubble that props up the US stock market.
THE DOOR CHINA HAS ALREADY CLOSED
The irony is that Washington is fighting over a door that China has shut from its own side. Fortune reports that the biggest barrier to Nvidia chips entering China today is not Washington but Beijing, which has largely withheld import permission and pushed its firms towards Huawei's Ascend chips. Analyst Bernstein projects Nvidia's share of China's AI chip market falling from about two-thirds in 2024 to about 8 per cent this year, with Huawei at around 50 per cent. Years of technology denial have produced the opposite of what was intended: a protected home market for a Chinese chip industry.
Nor are chips the binding constraint. Electricity is. China generates more than twice as much electricity as the US and added 543 gigawatts of capacity in 2025 alone, most of it solar and wind. BloombergNEF expects it to add almost six times as much as the US over the next five years. The US grid, run by private utilities for profit, cannot connect data centres fast enough. Socialist planning builds ahead of demand. The market waits for it.
Chinese labs release many of their leading models openly. Anyone in India, Africa or Latin America can download, run and adapt them. That is why the hawks' next target is the open models themselves. AI is the most advanced productive force of our time. The question is whether it is enclosed as the private property of three or four American corporations or remains common knowledge that every country can build on. For the Global South, sovereignty lies in building on open models and domestic computing capacity, not in renting intelligence from Silicon Valley.
The fall of Jensen Huang is not a verdict on one man's conduct. It is a verdict on an imperial order that cannot stop China's advance and has begun to devour its own.


