What Does Jensen Huang Really Want?

What Does Jensen Huang Really Want?

The NVIDIA Story, Season Two, Part Two
22 November 2026

On 11 November 2026, NVIDIA’s board accepted founder Jensen Huang’s resignation as president, chief executive and a member of the board. On 17 November, he said publicly that he would be willing to return if the board believed it would be in the company’s interests. The board made clear that it had no plans to reinstate him as chief executive. [1][2]

That leaves a question that is hard to avoid. What does Jensen Huang really want?

Why would someone who has already amassed such wealth, and who spent more than thirty years leading a company he helped found, want his old job back after a crisis of this magnitude in both his personal life and the company? Money alone seems an unlikely explanation. He has far more than he could ever need to live on. Wealth and commercial rivalry can still matter to a very rich person, of course. The question is how much they tell us about this particular choice. Other considerations may well carry more weight.

In ‘Governance Reform and Accountability’, the first essay in this season, I looked at how NVIDIA was adjusting its governance after Huang’s departure, searching for a successor and responding to investigations as they unfolded. Here I want to turn to Huang himself. Accounts of his private life and the allegations surrounding it have appeared in the press. What a formal investigation has established is a separate matter. In particular, suspicions about foreign intelligence activity remain unproven. What interests me here is why someone might feel so strong an attachment to the enterprise he built, and how his ambitions for the technology might relate to his own experience and commercial interests.

Huang was born in Taiwan in 1963, lived in Thailand as a child and came to the United States with his older brother when he was about nine. The brothers spent time at a boarding school in Kentucky, in surroundings very different from anything they had known. Huang has recalled how young he was compared with the students he lived with, and how one of his duties was to clean the dormitory toilets. He has also spoken warmly of that period. His recollections hardly support the conclusion that the experience left him only with trauma. Still less do they justify making an unverified experience of discrimination the psychological explanation for everything he did afterwards. [3]

In ‘Why I Keep Following the Jensen Huang Story’, I discussed questions of identity, achievement, discrimination and cultural difference among people of Chinese heritage living overseas. For Huang, who grew up in the United States from a young age and later achieved extraordinary success there, national identity may be rooted in a complicated set of personal experiences. He can feel a genuine attachment to America, be grateful for the opportunities he found there and still disagree with particular US government policies.

A person’s cultural background can influence their judgement, though it is difficult to measure how much. Huang’s objections to some chip export restrictions give us little reason to question his loyalty to the United States. Nor does the importance he places on the Chinese market establish a particular political allegiance to China. For him, the relationship between country, markets, technology and his own life’s work is likely to be more complicated than a choice between two sides.

His relationship with NVIDIA deserves a closer look.

For Huang, running NVIDIA was never simply a management job he had been hired to do. He co-founded the company with two other engineers in 1993 and went on to shape its technology and business strategy for decades. Over more than thirty years, he lived through product failures, shifts in the market, industry competition and several occasions when the company’s survival was in doubt. He also helped turn the graphics processing unit, or GPU, into an important computing tool for the age of artificial intelligence. NVIDIA has been part of almost his entire working life.

For a founder in that position, a company can provide work and wealth while also giving shape to his sense of who he is, the professional judgements he makes, the people he knows and the goals he sets for himself. One way to understand the attachment is to consider how closely a person’s identity can come to be bound up with a role. Someone who has spent decades testing his abilities and finding a sense of his own worth through a particular endeavour may need time to feel separate from it after his formal role has ended.

The setting in which he worked was unusual too. In recent years NVIDIA has found itself at the centre of global AI infrastructure, capital markets and technological competition between the United States and China. A decision on Huang’s desk could touch on research priorities, chip supplies, a customer’s investment plans, international policy and NVIDIA’s reputation all at once. Success gave him greater influence and put the next goal higher still. In such a competitive and closely watched environment, other people’s expectations and his own ambitions could keep pushing him in the same direction.

From the outside, it is hard to know whether that environment affected his judgement or his personal conduct during the events in question. An observer might interpret an affair as a way of finding relief from pressure. There is no reliable evidence for that account, so it cannot serve as a psychological diagnosis or an explanation of his conduct. We can still judge particular actions against ethical standards and company rules. Understanding why someone might have made a choice does not relieve him of responsibility for it.

There is another possibility about Huang’s ambitions that I think deserves consideration. For some time now, he may have cared at least as much about the place the computing platform he helped create would ultimately occupy in the world as about how well NVIDIA’s next generation of chips would sell.

That brings us to CUDA.

NVIDIA originally developed CUDA so that programmers could use GPUs for general-purpose parallel computing. The name refers to a software platform that encompasses programming interfaces, compilers, drivers, a runtime, computing libraries and development tools. Its scope goes well beyond a single programming language, and the platform itself is distinct from the chip. Programmers can use CUDA to send work that lends itself to parallel processing to a GPU, including matrix calculations, scientific computing and AI model training. Over the years, a whole ecosystem of software, frameworks, practical knowledge and skilled people has grown up around it. [4]

One technical point is easy to overlook. In principle, the CUDA ecosystem can be separated from the GPUs that NVIDIA makes.

Software needs an interface through which to use hardware. NVIDIA publishes many of CUDA’s programming interfaces and documents its intermediate instruction set, PTX. A compiler then turns intermediate code into instructions that a particular GPU can execute. NVIDIA has disclosed only part of the detail of its machine instructions, firmware and internal chip design. For another chipmaker, the hard work would include implementing the necessary computing functions, drivers and software compatibility layer, while achieving competitive performance and energy efficiency. It could do all that with circuit designs of its own. [4]

Apple’s transition from Intel processors to its own Apple silicon offers a familiar example. The Mac’s processor architecture changed fundamentally, yet macOS and a large body of applications carried on. Developers could recompile their source code for the new architecture, and Rosetta 2 allowed some older software to run through translation during the transition. Users kept much of the software ecosystem they depended on even as the hardware underneath it changed. [5]

Apple controlled the operating system, the development tools and the hardware transition itself. NVIDIA controls the direction of CUDA, and rival chipmakers have far less control over the terms of a transition. The analogy has limits, but the underlying principle holds. With suitable interfaces and implementations, software can outlive the processor architecture it was first written for.

AMD already provides HIP and HIPIFY to help developers port some CUDA source code to AMD GPUs. Developers who have the source and the tools to build it can move software to another computing platform gradually, release by release. AI-assisted programming may reduce some of the effort involved. Some CUDA libraries still have no direct equivalent, though, and code that recompiles successfully may run more slowly or less reliably. Hardware, compilers and software libraries all require sustained optimisation. [6]

It would be too strong, then, to say that the CUDA ecosystem makes NVIDIA chips irreplaceable. Switching carries real costs, but those costs can in principle be overcome. Equally, the possibility of porting software tells us little about how hard it is to design and manufacture a competitive chip, deliver sufficient memory bandwidth, build high-speed interconnects and make a large system work as a whole.

Suppose Chinese companies eventually build sufficiently powerful GPUs and Chinese AI developers still make widespread use of CUDA-compatible software. NVIDIA might lose some of its hardware market, while the computing platform Huang championed continued to shape how the industry worked. His place in the history of that platform need not rise and fall with NVIDIA’s share of chip sales.

Whether NVIDIA would earn a corresponding share of the profits is another matter. A competitor that independently implemented the relevant interfaces without needing a licence or products from NVIDIA might even weaken its commercial control. Yet for someone who sees building a computing platform as his life’s work, its widespread adoption could matter in its own right, quite apart from near-term revenue.

There may be an element of technological idealism in this, perhaps even a sense of himself as someone who could personally shape the industry’s course. I think that possibility helps explain Huang’s longstanding interest in Chinese developers and the Chinese market. He may worry about losing Chinese chip buyers. A further concern may be that China develops a complete ecosystem of its own, from hardware and programming tools to a community of developers who no longer need to build around CUDA.

There is some basis for this reading in his public remarks. In a May 2025 interview at the Milken Institute, Huang said he wanted AI around the world to be built on American technology standards. If US companies left a market, he argued, other firms, including Huawei, could fill the gap. He also spoke explicitly of the enormous commercial value of the Chinese market. In his own account, the fate of technical standards and the commercial interests of his company were closely connected. [7]

That still leaves open how much his own place in history matters to him. Huang may want to protect NVIDIA’s profits, win over developers, extend the international reach of American technology and vindicate the technical direction he has pursued for decades, all at the same time. His objections to the effects of export restrictions also need to be considered separately from compliance with export law. Public disagreement with a policy gives us no grounds to conclude that he intended to circumvent it.

One economic consideration may matter even more in understanding his behaviour.

In recent years NVIDIA has sold chips to AI companies and also helped some of its customers build computing infrastructure by taking equity stakes, providing financing and working with financial institutions. In September 2025, NVIDIA and OpenAI announced a letter of intent. As the planned infrastructure was deployed, NVIDIA intended to invest up to US$100 billion in stages. That figure described an investment intention at the time of the announcement, and cannot be counted as money already paid. Separately, NVIDIA’s annual report for its 2026 financial year disclosed US$17.5 billion in investments in private companies and infrastructure funds, including AI companies that buy NVIDIA products. [8]

When a supplier invests in a customer and that customer in turn purchases the supplier’s products, the arrangement is known as circular financing. A November 2026 report from the US Congressional Research Service examined such arrangements in the AI industry, including their benefits and risks. [9]

Circular financing alone gives us no grounds to call a transaction fictitious or to dismiss demand for AI as unreal. A technology that requires enormous investment upfront may need outside finance so that companies can build infrastructure while their customer base and revenue grow. But if the companies using that computing capacity fail over time to earn enough from their own markets, and depend on ever larger investments to keep buying, suppliers, customers and investors may find themselves in a cycle that relies on continued expansion.

NVIDIA’s accounts record revenue from chip sales. The question is what those chips will ultimately produce. Will the services they enable, the productivity gains for businesses and the other economic returns justify so much spending on computing capacity over time? Rising chip sales tell us that customers are buying. They tell us little, on their own, about whether all those buyers have found a sustainable way to make money.

If these financing arrangements have already grown large enough, they may place a new constraint on Huang and NVIDIA. Opening up markets and helping customers keep investing can drive growth, but may also be necessary to sustain an industry that still depends on more than revenue from its end users. We have no evidence that this pressure prompted Huang to seek a return, much less that he has reached a point where he cannot stop pushing for growth. The financing does, however, bring a part of the industry’s economic reality into view that an account based on personality alone would miss.

A person may want the technology he helped create to endure. A company has to retain its customers, sustain sales and keep the confidence of capital markets. When a founder has lived with both pressures for decades, how much of his drive to keep expanding the market is still his own ambition, and how much has become a responsibility he has little choice but to carry? Huang himself may find that question even harder to answer than those of us looking on.

Sources

[1] NVIDIA. Announcement of 12 November 2026 and Form 8-K filed with the US Securities and Exchange Commission.
https://www.sec.gov/Archives/edgar/data/1045810/000104581026000082/nvda-20261111.htm

[2] TechCrunch, 17 November 2026. Huang’s offer to return and the board’s response.

Nvidia says it doesn’t plan to reappoint Jensen Huang as CEO

[3] Computer History Museum, Jensen Huang oral history. The New Yorker, 2023 profile of Huang.

https://www.newyorker.com/magazine/2023/12/04/how-jensen-huangs-nvidia-is-powering-the-ai-revolution

[4] NVIDIA CUDA Programming Guide. The CUDA platform, Driver API and PTX.
https://docs.nvidia.com/cuda/cuda-programming-guide/01-introduction/cuda-platform.html

[5] Apple Developer. Universal binaries for Apple silicon and the Rosetta translation environment.
https://developer.apple.com/documentation/apple-silicon/building-a-universal-macos-binary
https://developer.apple.com/documentation/Apple-Silicon/about-the-rosetta-translation-environment

[6] AMD. Official HIP and HIPIFY porting documentation.
https://rocm.docs.amd.com/projects/HIPIFY/en/latest/index.html
https://rocmdocs.amd.com/projects/HIP/en/develop/how-to/hip_porting_guide.html

[7] Milken Institute. Transcript of the interview with Jensen Huang at the 2025 Global Conference.

Click to access new-innovation-economy-conversation-nvidia-ceo-jensen-huang_Transcript_GC25.pdf

[8] NVIDIA and OpenAI. Partnership announcement, 22 September 2025. NVIDIA annual report for its 2026 financial year.
https://investor.nvidia.com/news/press-release-details/2025/OpenAI-and-NVIDIA-Announce-Strategic-Partnership-to-Deploy-10-Gigawatts-of-NVIDIA-Systems/default.aspx
https://www.sec.gov/Archives/edgar/data/1045810/000104581026000021/nvda-20260125.htm

[9] US Congressional Research Service. The Role of Circular Finance in Funding AI Investment, 16 November 2026.
https://www.everycrsreport.com/files/2026-11-16_IF13349_dd2c67e1f7f7b4b4d2194716fed54500cede8f9b.html