Founder index · Nvidia · 1963–, CEO since founding in 1993
Jensen Huang
The lesson that travels
“The best time to build a platform is a decade before the market for it exists, and the only way to afford that is a profitable business you are willing to tax.”
Huang was born in Taiwan, sent to a boarding school in Kentucky at nine that turned out to be a reform school, washed dishes at Denny's as a teenager, and was designing chips at LSI Logic and AMD before he was thirty. Nvidia was founded over breakfast at a Denny's in San Jose in 1993 with Chris Malachowsky and Curtis Priem, two Sun Microsystems engineers, on the bet that 3D graphics for games would need dedicated silicon. Roughly thirty other companies made the same bet. Nvidia's first product failed — it rendered curves when the industry standardised on triangles — and Huang laid off half the company and staked what remained on a single chip, tested in software emulation because there was no money for prototypes. He has been telling employees the company is thirty days from going out of business ever since, including in years when it was worth more than any company on earth.
His defining decision came in 2006, when Nvidia was a successful graphics company with about $3 billion of revenue and no reason to change. He committed to making every GPU programmable for general-purpose computing — CUDA — which added cost to every consumer chip for a feature gamers did not want, and served a market of scientists and researchers that could not yet pay for it. The stock fell, the analysts objected, and he funded it through the financial crisis and a costly defective-chip episode. When deep learning arrived in 2012, it arrived on Nvidia hardware because there was nowhere else for it to run. He then spent the next decade building for a customer that did not exist, personally delivering the first DGX supercomputer to OpenAI in 2016.
The contested parts are fewer than for most founders of his size, and should still be stated. In 2022 Nvidia paid $5.5 million to settle SEC charges that it had failed to disclose how much of its gaming revenue in 2018 came from cryptocurrency mining — a demand that then collapsed and left the company holding inventory. His management style is total: around sixty direct reports, no one-to-one meetings, public dressing-downs that he defends as 'refining' people, and a company that has no visible succession plan for a founder now in his sixties. He is, more than any chief executive of a company this large, the single point of failure. And Nvidia's largest export market has been closed and reopened by two governments since 2022, a dispute in which he has negotiated personally and in which the company's interest and the interest of its home country's policy have not always aligned.
§01 — The hard calls
- 1997
With months of cash left after the NV1 failure, skip physical prototyping and bring the RIVA 128 to market on software emulation alone.
The chip shipped, sold a million units in four months, and saved the company. It established the pattern of betting everything on the next architecture that Nvidia still runs on.
- 2006
Make every GPU programmable with CUDA, at a cost in die area and margin, for a market of researchers that could not yet pay for it.
Looked like a mistake for six years. Data centre revenue in FY2025 was about $115 billion, essentially all of it running on CUDA. The most valuable moat in technology was funded by gamers who never used it.
- 2016
Hand-deliver the first DGX-1 to OpenAI, a one-year-old non-profit, and build an AI systems business years before AI had a business.
ChatGPT ran on Nvidia hardware in 2022. The H100 became the most demanded product in the industry and Nvidia's market value passed $1 trillion in 2023, $4 trillion in July 2025 and $5 trillion that October.
- 2020
Buy Mellanox for $6.9 billion, over Intel's competing bid, to own the networking between chips.
Turned a chip company into a systems company. Networking now generates well over $10 billion a year and is the reason a rack of Nvidia GPUs behaves as one computer while a rack of rivals' chips does not.
- 2025
Accept a 15% revenue-share to the US government in exchange for licences to sell the H20 into China.
An arrangement without precedent in export control. China's government then discouraged purchases anyway, and Nvidia guided as though China revenue were zero — after a $4.5 billion write-down. The market it had spent two decades building was closed by both sides at once.
One of these is set up as a tutorial — you make the call before you find out what Jensen did.
Face the decision§02 — Around this founder
Others in Technology