Six hundred thousand processors sit idle in warehouses in Hong Kong, a stone’s throw from mainland China. They are Nvidia H200 chips: Washington has authorised their sale to selected Chinese customers, while Beijing is still withholding approval for large-scale imports so as not to tie its own companies to American technology. The news, reported by Reuters on 5 November, captures the game now being played: two governments competing to control the same dependency. A week later, on 12 November, Nvidia announced that Jensen Huang was stepping down as head of the company he co-founded in 1993. Chief financial officer Colette Kress has taken over on an interim basis.
The least calculable factor
The resignation follows revelations about Huang’s private life, beginning with his relationship with Alice Huang, a former executive at Megaspeed, a Singapore-based Nvidia customer that Washington suspects of helping China obtain restricted chips. According to the American press, US intelligence believes she was recruited by Beijing’s security services. China dismisses the allegations as a smear, Singapore says its investigations into her and into Megaspeed did not concern espionage, and Nvidia’s board has launched an independent inquiry. Responsibility has yet to be established. The paradox is plain: the most calculated infrastructure of the century has cracked on the least calculable factor of all, trust between people.
Sell or contain
Nvidia’s processors power the training of artificial intelligence models, with both civilian and military uses. In the last fiscal year revenue from Chinese customers fell from $25 billion to $19.7 billion, and in the quarter to July sales of data centre processors in China became marginal.
The Nvidia case · Chips and geopolitics
China’s weight in Nvidia’s revenue
Revenue from customers headquartered in China, including Hong Kong. Fiscal years ending in late January
Hover over the bars to see the share of total revenue
Revenue is attributed by customer headquarters. In the quarter ending 26 July 2026, shipments to China of Hopper-family data centre processors – the family of the H200 – amounted to less than 1% of data centre revenue. In its outlook for the following quarter, Nvidia assumed no data centre compute revenue from China.
Source: Nvidia, 2026 annual report, quarterly report and CFO commentary, August 2026. Compiled by SIR.
Meanwhile, in China, Huawei is closing in on Nvidia. Herein lies Washington’s dilemma: restricting sales slows China’s race but pushes Beijing towards home-grown alternatives, while allowing them under supervision supports American companies but hands tools to a rival. Huang was the face of the second approach, an outspoken critic of export controls. Now the pendulum is swinging back. House Speaker Mike Johnson has spoken of a growing consensus in Congress on curbing chip exports to China, while President Donald Trump went as far as naming a successor on social media. The question is who decides: the company, the White House or Congress. And how much room a private company should have in choices that bear on a state’s security.
The Nvidia case · Chips and geopolitics
Who sells AI chips in China
Shares of China’s market for data centre AI accelerators, first half of 2026
Hover over the bars for details
Nvidia delivered 678,000 accelerators in China, 44% fewer than a year earlier. Huawei rose to 572,000. The share of other suppliers is calculated as the remainder.
Source: IDC, data reported by Reuters (October 2026). Compiled by SIR.
The border moves into the cloud
A processor can sit in a data centre in South-East Asia and be used remotely by a Chinese company. According to Reuters, Chinese firms are now trying to secure computing capacity abroad, fearing that the scandal will give momentum to a US law on remote access to restricted technologies. Power no longer lies in owning an object, but in the right to use it, wherever it may be. And it rests on an exposed supply chain: Nvidia itself acknowledges that its supply chain is concentrated mainly in Asia, with Taiwan’s foundries in the front line. America’s lead in artificial intelligence runs through an island that Beijing regards as its own.
The price of concentration
When infrastructure this decisive is concentrated in a handful of companies, an executive’s personal crisis becomes a matter of national security, of markets, of diplomacy. Concentration, which looked like strength, turns out to be a vulnerability. For Europe, 12 November also marked the close of the European Commission’s call to build up to seven AI Gigafactories, the large computing hubs intended to strengthen the Union’s technological sovereignty. To secure the processors it needs, however, Brussels has signed letters of intent with three American manufacturers, Nvidia among them. Digital sovereignty is measured by the freedom to choose whom to depend on.
Six hundred thousand chips stranded in Hong Kong show that technology is no longer just an industry: it has become a frontier. Around those processors, part of the future distribution of economic, technological and military power is being negotiated. Huang’s fall leaves open a question that goes beyond Nvidia: who answers for a power that no one elected and on which everyone depends? When computing becomes power, its stewardship cannot be left to the resilience of a single man.