What has happened
On 12 November 2026, Nvidia announced that Jensen Huang is stepping down as president, CEO and board member, with immediate effect. He co-founded the company in 1993 and led it for 33 years.
Nvidia designs the chips used in a large share of the world's artificial intelligence data centres. Colette Kress, its chief financial officer, will serve as interim CEO while the board looks for a successor, and Harvey C. Jones will chair the board. The company says it expects operations, customer commitments and its product roadmap to continue without interruption. It reports results tomorrow, 17 November.
A person is not a company
The usual reaction is to ask, "What will Nvidia be without him?" Daniel Kahneman, a Nobel laureate in economics, explains why that question comes so naturally: we prefer simple stories in which one person's talent and intentions explain everything, and we judge a manager by how their company is doing. It is known as the halo effect.
In Thinking, Fast and Slow, Kahneman cites the researcher Philip Rosenzweig, who concludes that stories of business success and failure exaggerate how much a manager's style matters to the results.
That does not mean Huang did not matter. It means that what holds Nvidia up today is not charisma, but something that outlasts any CEO.
What keeps customers is the software
In 2006 Nvidia launched CUDA, a program that lets people use its chips for tasks that have nothing to do with graphics. According to Chip War, Chris Miller's book, the company itself estimated in 2017 that it had invested at least $10 billion in that tool. It is free, but it only works on Nvidia chips.
That is the mechanism. Anyone who programs with CUDA builds up years of work that cannot be moved to another manufacturer without redoing it. No contract forces them to stay: leaving is simply expensive.
On a much smaller scale, the same can happen to any business: a management program that only exports your data in its own format, a website built on a platform you cannot take your content out of, a tool whose price rises once you can no longer leave it. The dependence goes unnoticed until the day you want to switch.
A few suppliers hold up the whole chain
Chip War shows another dependence: Nvidia designs its chips but does not make them. They are made mainly by TSMC, in Taiwan. And some of its biggest customers, the cloud giants, already design chips of their own.
That fragility runs through the whole industry. As Miller describes it (with data from 2022), almost every chip is designed with software from at least one of three US-based companies, the most advanced ones are made only by TSMC and Samsung (apart from the chips Intel makes for itself), and the machines needed to make them come from a single company, ASML of the Netherlands.
When so few control so many pieces, a problem at one of them, or a political decision, is felt by all the others. If your business depends on a single supplier for something essential, that fragility is yours too, only on a different scale.
What we still do not know
This week an analysis by the researcher Will DePue has also been circulating, according to which six artificial intelligence models from Chinese companies, released since April, carry a "fingerprint" that matches Nvidia's Blackwell chips. Exports of those chips to China are restricted.
It deserves a careful reading. The author himself makes clear that his analysis does not show where the chips were, how they were obtained or whether any rule was broken, and that the fingerprint reflects the hardware the training started on, not necessarily everything that was used. He estimates about 12,000 GPUs (the chips used to train AI) in total, within a range of 7,000 to 22,000, and writes that he is not confident in those figures. Other experts dispute the reading: some see a leak in the controls; others see a small scale that would suggest the controls are working.
The honest position is that nothing firm can be concluded today. And that is another lesson for your company: when a striking piece of news arrives, separate what has been shown from what is only suspected before you make a decision.
Five questions before relying on a supplier
- Who owns the code or the content when the work is finished?
- Can I take my data away in a format someone else can read, and what does it cost to leave?
- What happens if the supplier raises prices, changes the terms or disappears? Is there a plan B?
- How many essential pieces depend on a single company or a single person?
- Is the price fixed before the work starts?
In our projects, the code is yours when development ends, we give a closed quote before starting and we do not lock you in.
Vladyslav Dolnyi, CEO and founder of Gemelos Code
Sources: Nvidia's announcement of 12 November 2026, Will DePue's analysis of the same day, Chip War (Chris Miller) and Thinking, Fast and Slow (Daniel Kahneman).
See custom software