Jensen Huang

Why we read him as an Architect

5 moments from his public record, and what each one shows of the Architect pattern.

• In April 1993, Jensen Huang met two engineering friends, Chris Malachowsky and Curtis Priem, at a Denny's in San Jose, and the three of them founded Nvidia there, aiming at a market that barely existed yet: chips that could put real 3D graphics on an ordinary PC. Huang later said the restaurant "had all the coffee you could drink and no one could chase you out." He designed the company's shape before he had a product built or a market proven. [3]
• In 1999, Nvidia shipped the GeForce 256 and marketed it as the world's first graphics processing unit, a term Nvidia had not used before that product. The chip was not the first to take graphics work off a computer's main processor. Nvidia was the company that named the category, and the rest of the industry organised around the term it chose. [9]
• In 2006, Nvidia unveiled CUDA, a way to program its graphics chips for general computing, not just games, and kept funding it through roughly a decade in which the company's market value sat flat between about $8.5 billion and $10 billion, spending close to $12 billion on research and development between 2006 and 2017 while revenue stayed in the single billions. Engineer Bryan Catanzaro said it took "about 10 years before Wall Street really started to believe this investment was worth anything." Huang built the platform layer years before he had proof anyone outside Nvidia needed it. [4]
• In early 2013, Bryan Catanzaro built cuDNN, a library that let Nvidia's chips train neural networks, after Nvidia's own software team had criticised the project. Catanzaro took it straight to Huang. Huang cleared his weekend to study deep learning, and when the two next met he had written three words on his whiteboard: "Once In A Lifetime Opportunity." He authorised Catanzaro to pull in whoever he needed from across the company. Huang did not build the AI pivot himself. He recognised someone else's working prototype of a new pattern and rebuilt the company's priorities around it fast. [5]
• Huang runs Nvidia with around 60 people reporting to him directly and no scheduled one-on-one meetings. "I don't have one-on-ones with them because it's impossible," he has said. "We present a problem, and all of us attack it." He calls the method extreme co-design: a flat structure built so that no layer of managers holds information the rest of the company cannot see. [6]

What made him so effective

Where his type’s weakness shows up

Every type has a known weakness. For the Architect it is this: can treat a finished design as finished work, when nothing has yet met a real user or a real load. Here is where that shows up in his record.

• Nvidia unveiled its Blackwell chip family in March 2024 and initially expected to ship it by the second quarter. A design flaw discovered afterward was cutting manufacturing yield, and Nvidia had to work with its manufacturing partner TSMC to fix it before large-scale production could resume. Huang later said plainly: "We had a design flaw in Blackwell. It was functional, but the design flaw caused the yield to be low. It was 100% Nvidia's fault." [7]
• Building a Blackwell computer required seven different types of chips, each designed from scratch and ramped into production at the same time. The flaw became public in August 2024, and Huang did not confirm it was fixed until 23 October 2024, months during which Amazon, Google, Meta and Microsoft, customers together worth more than 40% of Nvidia's revenue, waited on delayed shipments. [8]
• This is the architect's blind spot working exactly as the type predicts: a finished design, unveiled and dated in public, that had not yet met the real load of high-volume manufacturing. Blackwell worked on paper and in early samples. It took contact with the production line to show the yield problem that cost Nvidia months. [8]

What to copy

Three things he is on record actually doing, that you can start doing this week.

1. 1

   Fund the platform layer underneath your product years before you have proof anyone outside your company needs it. Nvidia kept paying for CUDA through close to a decade in which its market value barely moved, before AI research gave the bet a market. [4]

2. 2

   When someone below you keeps showing you a working prototype of a pattern you have not backed yet, clear real time to study it yourself before you decide. Huang cleared a weekend for Catanzaro's cuDNN library rather than delegating the decision. [5]

3. 3

   Flatten your own reporting line and say the important things to the whole group at once, not to individuals in private, if you want everyone building against the same information. [6]

What not to copy

Sources

Every numbered claim on this page links to one of these.

1. [1] Our History: Innovations Over the Years
2. [2] Jensen Huang - CHM
3. [3] NVIDIA Founder and CEO Jensen Huang Returns to Denny's Where NVIDIA Launched a Trillion-Dollar Vision
4. [4] Going all-in with Nvidia: How Jensen Huang's high-stakes bets paid off
5. [5] "Once In A Lifetime": The Bryan Catanzaro Story of Nvidia
6. [6] Nvidia CEO Jensen Huang Never Has One-on-One Meetings With His 60 Direct Reports, Here's Why
7. [7] Nvidia's design flaw with Blackwell AI chips now fixed, CEO says
8. [8] Nvidia CEO Jensen Huang: Blackwell AI Chip Design Flaw Fixed
9. [9] Nvidia's first GPU: GeForce 256