Peter Dasilva / © European Union, 2026 · CC BY 4.0 (cropped)

Back to Chapter 03 How Huang Actually Operates

The Operating
System

How he actually operates: 22 plays across survival, candor, platform, markets, cadence and ecosystem. Each carries an honest caveat, because much of Huang's edge rests on a gaming cash machine, AI timing and a stock that made leaving expensive. Filter by domain, or dive straight in.

Stay Thirty Days From Going Out of Business

Huang has run Nvidia for more than three decades as if it could die next month, and says the fear of failing drives him harder than the wish to win.

Lower Your Expectations, Raise Your Resilience

Huang argues that people who expect things to go well break when they don't, and that character is built by setbacks, not talent.

Confront the Mistake, Ask for Help

When Nvidia's first chip bet on the wrong technology, Huang told his biggest customer the truth, and asked them to pay anyway.

Shake It Off and Skate to the Next Opportunity

Treat setbacks as part of the route: absorb them fast and move to where the next opening is.

Know the Real Price Before You Start

Asked whether he would found Nvidia again, Huang says no: the most honest thing a founder of a multi-trillion-dollar company has said about what it costs.

Kill the Information Monopoly

Power at Nvidia isn't supposed to come from knowing things others don't, so information goes to everyone at once.

Improve People Instead of Replacing Them

Huang says he would rather push someone hard than give up on them: a demanding culture paired with unusually low turnover.

The Install Base Is the Company

Nvidia's real asset is the hundreds of millions of compatible GPUs developers can already write for, more than any single chip.

Own the Whole Stack, Not Just the Chip

To sell data centers, Nvidia had to make more than processors: software, systems and networking became one product.

Never Break Compatibility

Every Nvidia accelerator runs the same software, and Huang treats that as the one rule that is never up for negotiation.

Build Zero-Billion-Dollar Markets

Huang aims Nvidia at markets with no customers yet, because a market with no revenue also has no incumbents.

Sell Factories, Not Computers

Huang reframed the data center as an "AI factory" whose output is revenue, and Nvidia's sales followed the reframing.

Price on the Value Delivered

Huang argues the cost of each unit of AI output is falling fast, which is how he justifies margins that went from 35% to 75%.

Work Backwards From First Principles

Start from the end result and the physics of the problem, not from last year's plan or a competitor's roadmap.

Push Your Chips In When You Know It Will Work

The RIVA 128 bet looked reckless, but Huang de-risked it first by buying an emulator to prove the chip in software before committing.

Ship a New Architecture Every Year

Nvidia moved to an annual rhythm for its data-center platforms, so rivals are always chasing last year's chip.

Co-design the Whole System

When the problem no longer fits on one chip, design chips, systems, software and suppliers together.

Build a Network, Not a Moat

Huang rejects the moat metaphor, saying the point is to make everyone else more successful on your platform.

Win the Developers First

Customers follow developers, so Nvidia spent two decades making itself the first thing developers write for.

Back the Researchers

Nvidia puts compute in the hands of scientists, where tomorrow's demand is invented, and the UK has been a recurring beneficiary.

Invest in Your Own Customers

Nvidia now takes stakes in the clouds and labs that buy its chips, which grows the ecosystem and raises the "circular financing" question.

Make Countries Your Customers

Sovereign AI turns governments into buyers of national AI infrastructure, and the UK is one of them.