This dishwasher boy built a $5 empire

from the table of a Denny's diner.

He pitched the business plan over cheap

coffee in 1992, and 34 years later, his

company makes $20

every single hour.

Not from the safe bet, but from a gamble

nobody believed in that crushed entire

industries and took over the AI space,

the gaming industry, and even all of

your app algorithms.

And just as he started winning, the

government moved to [music] bring him

down.

Jensen Huang was born in Taiwan in 1963.

When he was nine, [music] his parents

sent him and his brother to America for

a better education.

The plan was simple. Live with relatives

in Kentucky, study hard, build a future.

This doesn't look like a school. It

looks more like somewhere people get

sent after they've done something very,

very wrong. That's because

>> [music]

>> it kind of is. Welcome to Oneida Baptist

Institute. You'll learn English,

discipline, and how not to complain

while cleaning bathrooms.

This is not how most tech CEO origin

stories start, but it didn't break him.

It taught him something no lecture hall

could.

When the situation is terrible, you

adapt [music]

or you disappear.

He eventually made it to Oregon State,

then Stanford, spent years at chip

companies,

AMD, then LSI Logic, studying

semiconductors with a level of obsession

that was, frankly, a little alarming.

By 1992, Jensen had an idea.

Graphics are going to change computing.

Games are getting bigger. 3D is getting

harder. CPUs can't carry this forever.

Someone is going to build a chip for

this new world.

I think it should be us.

Jensen, I have a mortgage, a stable job,

and a family that currently believes I

make responsible decisions.

Good.

Then this will be exciting. So, the plan

is, quit our jobs, start from zero, and

build a chip for a market that barely

exists?

Correct. I hate that this sounds insane

and also kind of logical.

They called the company Nvidia.

The idea was ambitious. Build one

powerful [music] graphics chip that

could handle everything. A clean

universal chip for the future of gaming

hardware.

For 2 years they worked. They spent

every dollar they had. And in 1995,

Nvidia launched its first product, the

NV1.

Jensen, we have a problem. Microsoft

just [music] published the Direct X

spec, full triangle based rendering

pipeline. Holy crap, the NV1 runs

quadratic mapping?

Yes.

We are the only ones going in the

opposite direction. Games are about to

be built from tiny flat triangles?

Our chip is built for curved surfaces.

We built the wrong chip with tremendous

confidence. 2 years, every dollar. A

chip built for a world that no longer

existed. The NV1 [music] flopped.

But Nvidia still had a lifeline. Sega

had contracted them to build the

graphics processor for the company's

next generation console,

the Dreamcast.

Real money, real chance to start over.

There was only one problem.

Sega's chip used the same dead-end

technology. [music] You're doing the

Sega math again, aren't you?

We'll spend the next year building

something we already know is dead. And

if we cancel, we lose the only customer

still paying us.

Uh, Pretty much.

I miss normal employment.

If we're going to die, let's die

building the right chip.

>> [music]

>> So Jensen flew to Japan.

The chip is a dead end. You should

cancel our contract and find someone

else.

We shouldn't finish the contract. It

would be a waste of your money.

>> [music]

>> There's one more thing. I still need the

money, the $5 million remaining on our

contract. Please put it into Nvidia as

an investment instead. Otherwise, we'll

vaporize overnight.

>> [music]

>> I have nothing to offer you. This money

will most likely be lost, but I'm asking

anyway.

I'll need a few days.

Jensen [music] bowed, walked out, and

got on a plane. Nothing to do now but

sit, and wait, and watch the Pacific

stretch out beneath him.

Jensen, you might want to see this.

He opens [music] it, eyes wide. It's a

massive 10% offer on the value of

BetterHelp.

Just kidding.

Let's be real for a second. When the

outcome is completely out of your hands,

the anxiety Did I make the right call?

That noise won't switch off. It doesn't

go away on its own.

The hardest part for me was noticing the

pattern.

I was pushing through instead of

actually dealing with anything.

My therapist asked me one question.

Are you solving something, or are you

rehearsing a disaster?

I didn't have an answer. Just having a

name for what I was doing,

that single reframe broke the loop.

BetterHelp, the paid [music] partner of

today's video, makes starting therapy

easier.

Take a short quiz, get matched with a

credential therapist, and meet by phone,

[music] video, or text. Whatever works

for your schedule.

If it's not the right fit, switching

costs nothing extra.

And it's working.

Over 6 million people have gotten help

[music] through BetterHelp, rated 4.8

out of five on the App Store.

Click the link in the description,

[music]

or go to betterhelp.com/crowncapital

for 10% off your first month of therapy.

Every great ship [music] needs

maintenance. So do you.

Hello.

The money is coming.

The reason the Sega CEO said yes had

nothing to do with business logic. He

just liked Jensen.

That $5 million bought Nvidia 6 months.

And now they had one last shot to build

the right chip, or

>> [music]

>> go bankrupt.

They called it the Riva 128,

triangle-based, [music]

DirectX compatible, built from scratch.

Weeks of late nights, bad coffee, and

absolutely zero plan B. They had exactly

1 month of payroll [music] left in the

bank to find out if it worked.

The chip launched in 1997, and this time

Nvidia did not miss. Game developers

could actually use it. Customers

actually bought it. From the brink of

bankruptcy to a million units sold in 4

months.

Most CEOs would have taken a long

vacation and slow down a bit.

He walked back into the office and wrote

one word on the whiteboard.

More.

So, by 1999, Nvidia launched the GeForce

256 and gave the world a term it didn't

really have yet, GPU.

Graphics processing unit.

It was a genuine leap. A chip that

handled complex 3D calculations that

previously required an entirely separate

processor. For gamers,

>> [music]

>> this meant smoother worlds, better

lighting, and monsters that looked

slightly less like wet cardboard. Gamers

loved it. Developers wept [music] with

joy.

Revenue exploded. And by the early

2000s, Nvidia had become the king of PC

graphics. Every serious gamer, every

visual effects studio, every game

developer on the planet. The obvious

move was simple. Keep making better

gaming chips, sell them to gamers,

become rich, buy a very large leather

jacket museum.

But Jensen started looking at something

other people missed.

A CPU is like one very smart employee

doing tasks one by one extremely fast.

Meanwhile, a GPU is like 10,000 interns

doing [music] tiny calculations at once.

Are the interns smart?

No.

But there are a lot of them, and if your

problem can be broken into thousands of

small calculations, physics, chemistry,

weather, biology, artificial

intelligence,

suddenly, the chip built for video games

[music] starts looking like something

else, a cheap supercomputer.

Jensen looked at this and thought, "What

if the same chip that rendered

explosions in video games could help

simulate molecules, predict weather,

model fluids, and maybe one day train

machines to think?"

So, in 2006, Nvidia launched CUDA,

Compute Unified Device Architecture, a

platform that let scientists,

researchers, and engineers use Nvidia

GPUs [music] for general-purpose

computing.

And then, he committed roughly $500

million to build it.

Jensen,

the gaming business is printing money.

Walk me through who CUDA is actually

for.

Researchers, scientists, people running

fluid dynamics simulations, climate

models, drug discovery?

So, not our customers.

Not yet.

I'm starting to hate that phrase. This

is our CUDA revenue after 2 years. Do

you see a number?

Not today.

Jensen, this is not a business. We are

subsidizing a hobby for PhD students.

PhD students are where the future

starts. For the first time, they can use

a gaming chip like a supercomputer.

We don't know what they'll build with

it. But when they build [music]

something important, it will run on us.

If they build nothing,

we still have the best gaming cards in

the world and a $500 million science

project. We've survived worse.

>> [music]

>> Six years passed. The gaming business

kept printing money. CUDA kept printing

almost nothing.

And Jensen kept waiting for the problem

big enough to prove him right.

Then, in Toronto, someone found it.

It's 2012. Every year, the world's top

AI researchers compete in a challenge

[music] called ImageNet. Tens of

millions of images, and your software

has to identify what's in them.

You see the Toronto submission? Some PhD

student [music] who running a a network

on gaming cards. We've been hand-coding

classifiers for a decade. Gaming cards

aren't going to crack this.

Ladies and gentlemen, thank you for

waiting.

The ImageNet [music] 2012 results are

final. Second place, University of Tokyo

with a 26.2% error rate. And this year's

winner,

University of Toronto, AlexNet, 15.3%.

I'm sorry, 15?

That's not a win. That's a murder scene.

For years, researchers have been trying

to teach computers how to see by

manually designing the features they

should look for.

Edges, textures, shapes, patterns.

AlexNet took a different path. Give the

neural network enough data, enough

layers, and enough computing power,

and let it learn the features itself.

Training a neural network requires

millions of small calculations happening

at the same time.

Exactly the kind of work that Nvidia had

been quietly perfecting for 6 years.

The $500 million bet

>> [music]

>> had finally found its customer.

Suddenly, every major AI lab started

paying attention. Stanford, MIT, Google,

Meta, Microsoft.

They weren't just buying graphics cards

[music] anymore.

They were buying computing power for

intelligence itself.

And Nvidia had the hardware, the

software, and the scale to deliver it.

Great news for Nvidia, until in 2022,

[music]

Washington made a phone call.

Before Washington called, Jensen had

already tried to fix Nvidia's biggest

weakness. Because Nvidia designed the

chips,

but it didn't make them.

That job belonged mostly to TSMC in

Taiwan.

And he didn't own the underlying

architecture, the fundamental

instruction set that tells processors

how to think.

That belonged to a British company

called ARM.

A company whose technology sat inside

virtually every smartphone on Earth.

Apple, Samsung, Qualcomm, your pocket.

Their DNA, not Jensen's.

So, for all of Nvidia's power, [music]

Jensen was still renting the building.

He owned the furniture.

So, in 2020,

>> [music]

>> Jensen made his move. Nvidia offered $40

to buy ARM.

Own the GPU, own the software, own the

architecture. Simple.

Unless you were literally everyone else

in tech.

Nvidia cannot own ARM. Agreed. Agreed.

Also agreed.

Wow, I hate how united we are. If Jensen

owns ARM, then every chip we build runs

through our most dangerous competitor.

Phones,

>> [music]

>> servers, AI chips, everything. So, what

do we do?

We complain to every regulator with an

email address.

And they did. The FTC sued. The European

Commission investigated. The UK blocked

it on national security grounds.

18 months later, Nvidia walked away.

Then Washington called. Mr. Huang,

effective immediately, advanced AI

chips, the H100,

>> [music]

>> can no longer be exported to China.

How much revenue are we talking about?

That sounds like a you problem.

Just like that, some of Nvidia's biggest

customers were cut off.

Alibaba, Baidu, ByteDance, China's AI

companies,

gone.

So, Nvidia did what Nvidia always did.

It engineered around the problem.

They built a weaker version, the H800,

designed to fit inside the rules.

Legal enough to ship. Powerful enough

[music] to sell.

For about 5 minutes, because by late

2023, Washington tightened the rules

again.

The H800 was restricted to

>> [music]

>> The message was clear. Nvidia was now a

strategic weapon in America's technology

war with China.

Jensen wasn't just a CEO anymore. He was

a piece on someone else's chessboard,

moved by people he couldn't negotiate

with.

But while politics got messier, demand

got insane. OpenAI, Google, [music]

Anthropic, xAI, Microsoft. Everyone

building large AI models needed the same

thing, more Nvidia chips, more data

centers, more power, more cooling, more

everything.

Blackwell, Nvidia's next [music]

generation chip, was being ordered

before factories could even finish

making it.

By October 2025, Nvidia [music] crossed

$5 trillion in market value.

The most valuable company on Earth,

ahead of Google, ahead of Apple.

Not bad for a company that once had 30

days of cash left and one very wrong

chip.

None of it was supposed to work.

Every single [music] bet looked insane

from the outside, and every single one

paid off.

As what he himself [music] said,

"My will to survive exceeds everybody

else's will to kill me."