WEBVTT

00:00:00.430 --> 00:00:02.866 align:center line:81%
(upbeat instrumental music)

00:00:12.376 --> 00:00:15.178 align:center line:79%
Welcome back
to the stage, Patrick Collison.

00:00:15.379 --> 00:00:18.348 align:center line:81%
(upbeat instrumental music)

00:00:28.825 --> 00:00:30.394 align:center line:81%
All right, good afternoon, folks.

00:00:30.527 --> 00:00:32.896 align:center line:81%
I hope you've enjoyed the sessions

00:00:33.030 --> 00:00:36.066 align:center line:79%
between now
and when we last saw you this morning.

00:00:37.134 --> 00:00:40.037 align:center line:79%
For this afternoon's keynote
or fireside chat,

00:00:40.137 --> 00:00:43.974 align:center line:79%
I suppose, I'm about to introduce
somebody who needs little introduction.

00:00:44.341 --> 00:00:45.509 align:center line:81%
Although, a fun fact

00:00:45.842 --> 00:00:47.878 align:center line:79%
that you may not know
about Jensen Huang is

00:00:48.011 --> 00:00:51.548 align:center line:79%
that he's been a CEO
of NVIDIA for 31 years this month,

00:00:51.882 --> 00:00:54.318 align:center line:81%
making him the longest-serving CEO

00:00:54.451 --> 00:00:55.986 align:center line:81%
in the technology industry.

00:00:56.587 --> 00:00:58.388 align:center line:81%
And guess, therefore, logically that is...

00:00:58.488 --> 00:01:00.591 align:center line:81%
(Audience cheers and whistles.)

00:01:03.560 --> 00:01:07.831 align:center line:79%
So, John and I have only been doing it
for a mere 14 years, so you know

00:01:07.931 --> 00:01:10.434 align:center line:79%
even if we double that,
we'll still be second to him.

00:01:10.567 --> 00:01:15.472 align:center line:79%
And, Jensen,
we'll talk about this on stage,

00:01:15.906 --> 00:01:20.010 align:center line:79%
attended
the Oneida Baptist Institute in Kentucky.

00:01:20.911 --> 00:01:22.713 align:center line:81%
We'll definitely be asking about it.

00:01:23.880 --> 00:01:25.148 align:center line:81%
Oregon State.

00:01:25.282 --> 00:01:28.619 align:center line:81%
Worked as a waiter at Denny's, then...

00:01:28.719 --> 00:01:30.621 align:center line:81%
(Audience laughs.)

00:01:30.821 --> 00:01:32.022 align:center line:81%
Denny's close to here, actually.

00:01:32.823 --> 00:01:36.660 align:center line:79%
LSI Logic, and then AMD,
which is of course now run

00:01:36.793 --> 00:01:39.930 align:center line:81%
by his first cousin once removed.

00:01:40.263 --> 00:01:41.865 align:center line:81%
We'll definitely be asking about that.

00:01:43.033 --> 00:01:45.402 align:center line:81%
Before he founded NVIDIA in 1993—

00:01:45.802 --> 00:01:50.240 align:center line:81%
and NVIDIA's market cap was $8 billion

00:01:50.741 --> 00:01:53.143 align:center line:81%
when Stripe launched in 2011,

00:01:54.111 --> 00:01:56.947 align:center line:79%
and it is now, of course,
more than 200 times that.

00:01:57.314 --> 00:01:59.216 align:center line:81%
So, he's been busy since.

00:01:59.449 --> 00:02:01.585 align:center line:79%
Please,
welcome to the stage, Jensen Huang.

00:02:01.885 --> 00:02:04.187 align:center line:79%
(Audience applauds.)
(upbeat music)

00:02:17.134 --> 00:02:18.135 align:center line:81%
Hey, everybody.

00:02:20.170 --> 00:02:21.405 align:center line:81%
(Audience hoots.)

00:02:21.538 --> 00:02:23.473 align:center line:79%
-So you watched the keynote earlier?
-I did.

00:02:23.573 --> 00:02:25.208 align:center line:81%
I've never seen a duet before.

00:02:26.910 --> 00:02:30.914 align:center line:79%
I've never seen a duet before.
You were so synchronized.

00:02:31.048 --> 00:02:34.051 align:center line:79%
It seemed like the two
of you knew each other.

00:02:34.184 --> 00:02:35.585 align:center line:81%
It's incredible.

00:02:35.686 --> 00:02:37.320 align:center line:81%
Some acquaintance.

00:02:38.055 --> 00:02:40.857 align:center line:79%
But, okay,
you've been doing keynotes a long time.

00:02:40.957 --> 00:02:43.627 align:center line:81%
You are the keynotes GOAT, so...

00:02:45.028 --> 00:02:46.029 align:center line:81%
Stop it.

00:02:46.163 --> 00:02:49.700 align:center line:79%
So give us your... we don't have
even a signature outfit yet.

00:02:49.800 --> 00:02:50.934 align:center line:81%
We're just amateurs here.

00:02:52.235 --> 00:02:53.637 align:center line:81%
So give us...

00:02:53.770 --> 00:02:55.272 align:center line:81%
It's because you're still young.

00:02:55.839 --> 00:02:56.873 align:center line:81%
(Audience laughs.)

00:02:56.973 --> 00:02:59.810 align:center line:79%
Give us your keynote performance review.
What'd you think?

00:03:00.310 --> 00:03:01.344 align:center line:81%
I thought it was A+.

00:03:01.912 --> 00:03:03.213 align:center line:81%
I thought it was A+.

00:03:03.313 --> 00:03:05.115 align:center line:81%
I thought it was... really.

00:03:05.215 --> 00:03:06.216 align:center line:81%
(Audience cheers and applauds.)

00:03:07.384 --> 00:03:10.721 align:center line:81%
You explained perfectly

00:03:11.421 --> 00:03:13.924 align:center line:81%
the purpose of the company,

00:03:14.658 --> 00:03:16.927 align:center line:79%
what inspires you guys,
what keeps you guys up

00:03:17.060 --> 00:03:20.464 align:center line:79%
what makes you work so hard,
the ecosystem that you serve

00:03:20.797 --> 00:03:22.699 align:center line:81%
the incredible platform you've built

00:03:23.133 --> 00:03:26.169 align:center line:79%
the amazing contribution
you make to the world's economy.

00:03:26.536 --> 00:03:28.138 align:center line:81%
It's incredible. I thought it was great.

00:03:28.271 --> 00:03:30.207 align:center line:79%
And there was a whole
bunch of technology stuff,

00:03:30.340 --> 00:03:31.742 align:center line:81%
feature stuff, money stuff.

00:03:31.875 --> 00:03:33.577 align:center line:81%
I didn't understand any of that, but...

00:03:33.677 --> 00:03:35.145 align:center line:81%
(Audience laughs.)

00:03:35.712 --> 00:03:37.647 align:center line:81%
But something about a CYK or something.

00:03:37.781 --> 00:03:38.782 align:center line:81%
What was that?

00:03:39.816 --> 00:03:41.685 align:center line:79%
-KYC.
-KYC. Yeah.

00:03:41.818 --> 00:03:43.186 align:center line:81%
(Audience laughs.)

00:03:43.386 --> 00:03:45.455 align:center line:79%
-I thought it was...
-It's a big deal in our world.

00:03:45.555 --> 00:03:47.224 align:center line:81%
Is that right? Kentucky Fried Chicken?

00:03:48.592 --> 00:03:49.860 align:center line:81%
We take care of KYC

00:03:49.993 --> 00:03:52.429 align:center line:79%
so that you can associate us
with Kentucky Fried Chicken.

00:03:52.562 --> 00:03:53.563 align:center line:81%
Okay, got it.

00:03:54.030 --> 00:03:55.465 align:center line:81%
Did you...

00:03:55.832 --> 00:03:57.634 align:center line:81%
Software-defined financial services

00:03:57.868 --> 00:03:58.869 align:center line:81%
this idea, did that...

00:03:58.969 --> 00:04:00.871 align:center line:79%
-Incredible.
-Does that make sense to you?

00:04:01.004 --> 00:04:02.939 align:center line:79%
Well, first of all,
I think it's a giant idea.

00:04:03.774 --> 00:04:05.342 align:center line:81%
Do you know where it came from?

00:04:07.611 --> 00:04:08.779 align:center line:81%
You're going to tell me.

00:04:10.046 --> 00:04:12.883 align:center line:81%
So Jensen and I were catching up, maybe...

00:04:13.016 --> 00:04:17.320 align:center line:79%
But the part that I loved was
how you realized, in the very beginning

00:04:17.621 --> 00:04:23.093 align:center line:79%
that financial payments was 
about code, not finance.

00:04:23.193 --> 00:04:24.427 align:center line:81%
I thought that was incredible.

00:04:24.561 --> 00:04:26.863 align:center line:79%
And you explained that
the first time we met.

00:04:26.997 --> 00:04:29.766 align:center line:79%
So Jensen and I were catching up
18 months ago or so

00:04:29.900 --> 00:04:32.803 align:center line:79%
and I guess it was
a couple years since we'd last spoken.

00:04:32.936 --> 00:04:34.738 align:center line:79%
So he was kind of asking
for the update on Stripe.

00:04:34.871 --> 00:04:39.042 align:center line:81%
And I was explaining, and you said

00:04:39.309 --> 00:04:41.978 align:center line:79%
“Oh, so it's like
software-defined networking

00:04:42.112 --> 00:04:43.113 align:center line:81%
but for money.”

00:04:43.780 --> 00:04:46.650 align:center line:79%
That was still ricocheting
around in my mind.

00:04:46.783 --> 00:04:50.921 align:center line:79%
So that's where we got to this idea
for software-defined financial services.

00:04:51.021 --> 00:04:53.156 align:center line:79%
So I hope we don't have to pay
a licensing fee for that or something.

00:04:53.256 --> 00:04:54.958 align:center line:81%
I got zero equity for that good idea.

00:04:55.058 --> 00:04:57.394 align:center line:81%
All right, you guys are doing okay.

00:04:57.494 --> 00:04:59.029 align:center line:81%
I was thinking about this.

00:04:59.396 --> 00:05:01.965 align:center line:81%
Tesla's earnings were of course yesterday

00:05:02.098 --> 00:05:05.936 align:center line:79%
and Elon announced that
I think Tesla is going to have 85,000

00:05:06.570 --> 00:05:09.139 align:center line:81%
H100s by the end of this year.

00:05:09.673 --> 00:05:11.241 align:center line:81%
I was just reflecting on...

00:05:12.175 --> 00:05:15.779 align:center line:79%
it's quite a success to sort of build
a business where CEOs kind of compete

00:05:15.912 --> 00:05:17.781 align:center line:81%
with each other to announce

00:05:17.881 --> 00:05:20.016 align:center line:81%
who has spent more buying your product.

00:05:20.150 --> 00:05:22.686 align:center line:79%
So, I think you've done
something quite impressive.

00:05:22.819 --> 00:05:25.021 align:center line:79%
But anyway, I actually want
to start out talking a bit about...

00:05:25.121 --> 00:05:28.658 align:center line:79%
All of my CEO friends,
they all have the most.

00:05:28.758 --> 00:05:30.994 align:center line:81%
(Audience laughs.)

00:05:32.229 --> 00:05:34.831 align:center line:81%
I want to start out talking a little bit

00:05:34.965 --> 00:05:39.970 align:center line:79%
about a remark you made
at a Stanford event recently.

00:05:40.136 --> 00:05:42.038 align:center line:81%
I was thinking of GSB, I think.

00:05:42.172 --> 00:05:49.112 align:center line:79%
You said, “I wish upon you
ample doses of pain and suffering.”

00:05:50.447 --> 00:05:51.581 align:center line:81%
Elaborate.

00:05:51.748 --> 00:05:53.550 align:center line:81%
(Jensen laughs.) (Audience laughs.)

00:05:53.984 --> 00:05:55.852 align:center line:81%
Well, let's see.

00:05:57.220 --> 00:05:59.322 align:center line:81%
There is a misunderstanding.

00:06:01.458 --> 00:06:02.826 align:center line:81%
There's a phrase that said

00:06:02.926 --> 00:06:07.631 align:center line:79%
“You should choose
your career based on your passion.”

00:06:08.565 --> 00:06:13.303 align:center line:79%
And usually, people connect
passion with happiness.

00:06:14.771 --> 00:06:18.041 align:center line:79%
I think there is
something missing in that.

00:06:18.174 --> 00:06:20.944 align:center line:79%
Nothing there is wrong,
but there's something missing.

00:06:21.077 --> 00:06:23.780 align:center line:79%
And the reason for that is because,
if you want to do great things

00:06:23.914 --> 00:06:27.717 align:center line:79%
and I know this to be true
about you creating Stripe.

00:06:27.851 --> 00:06:32.489 align:center line:79%
By the way, this is one
of the world's finest CEOs.

00:06:32.856 --> 00:06:34.491 align:center line:81%
Young as he may be. 

00:06:34.591 --> 00:06:36.559 align:center line:81%
(Audience applauds.)

00:06:36.660 --> 00:06:39.863 align:center line:79%
You guys know I've met a lot of CEOs.

00:06:40.063 --> 00:06:41.831 align:center line:81%
I've heard about a lot of companies

00:06:41.965 --> 00:06:45.368 align:center line:79%
and this is genuinely one
of the world's great visionary companies.

00:06:45.502 --> 00:06:48.171 align:center line:81%
So anyways, I just wanted to say that.

00:06:48.305 --> 00:06:50.507 align:center line:79%
And it's the reason
why I'm here, I just love what...

00:06:50.640 --> 00:06:53.576 align:center line:79%
No more compliments allowed.
It makes us terribly uncomfortable.

00:06:53.710 --> 00:06:55.445 align:center line:81%
I know. I could tell. I could see him.

00:06:55.545 --> 00:06:56.646 align:center line:81%
He's starting to sweat.

00:06:56.780 --> 00:06:58.081 align:center line:81%
(Audience laughs.)

00:06:58.214 --> 00:07:00.617 align:center line:79%
And so, the thing is,
when you want to build

00:07:00.750 --> 00:07:02.452 align:center line:81%
something great

00:07:03.820 --> 00:07:05.388 align:center line:81%
it's not easy to do.

00:07:05.689 --> 00:07:07.557 align:center line:79%
And when you're doing
something that's not easy to do

00:07:07.691 --> 00:07:09.492 align:center line:81%
you're not always enjoying it.

00:07:10.493 --> 00:07:15.398 align:center line:79%
I don't love every day of my job.
I don't think every day brings me joy

00:07:15.699 --> 00:07:18.969 align:center line:79%
nor does joy have to be
the definition of a good day.

00:07:19.636 --> 00:07:21.237 align:center line:81%
And every day, I'm not happy.

00:07:21.371 --> 00:07:23.473 align:center line:79%
Every year
I'm not happy about the company,

00:07:23.573 --> 00:07:26.743 align:center line:79%
but I love the company
every single second.

00:07:27.944 --> 00:07:31.614 align:center line:79%
So I think
that what people misunderstand is

00:07:31.748 --> 00:07:36.019 align:center line:79%
somehow the best jobs are the ones
that bring you happiness all the time.

00:07:36.152 --> 00:07:37.687 align:center line:81%
I don't think that that's right.

00:07:38.088 --> 00:07:41.257 align:center line:81%
You have to suffer.

00:07:41.658 --> 00:07:45.128 align:center line:79%
You have to struggle.
You have to endeavor.

00:07:45.261 --> 00:07:48.365 align:center line:79%
You have to do those hard things
and work through it

00:07:48.932 --> 00:07:51.401 align:center line:79%
in order
to really appreciate what you've done.

00:07:52.235 --> 00:07:55.905 align:center line:79%
There are no such things that are great
that were easy to do.

00:07:56.539 --> 00:07:59.776 align:center line:81%
So by definition, I would say therefore

00:08:01.511 --> 00:08:05.782 align:center line:79%
I wish upon you greatness,
which by my way of saying it,

00:08:05.882 --> 00:08:08.551 align:center line:79%
I wish upon you
plenty of pain and suffering, and so.

00:08:09.419 --> 00:08:11.788 align:center line:81%
(Audience applauds.)

00:08:16.426 --> 00:08:19.596 align:center line:79%
Anything in your upbringing
that taught you that idea?

00:08:19.729 --> 00:08:22.332 align:center line:79%
Or is it just
somehow innate to your makeup?

00:08:29.406 --> 00:08:31.574 align:center line:79%
I didn't realize I had
to lay down for this but...

00:08:31.674 --> 00:08:32.675 align:center line:81%
(Audience laughs.)

00:08:35.912 --> 00:08:39.949 align:center line:79%
I'm about to tell you things
I've never told anyone, not even my family.

00:08:40.083 --> 00:08:41.184 align:center line:81%
(Audience laughs.)

00:08:42.519 --> 00:08:44.721 align:center line:81%
I was an immigrant.

00:08:45.055 --> 00:08:48.324 align:center line:81%
And when I came in 1973, I was 9.

00:08:48.458 --> 00:08:50.660 align:center line:81%
My older brother was almost 11.

00:08:51.194 --> 00:08:52.662 align:center line:81%
This was a foreign country

00:08:52.796 --> 00:08:57.734 align:center line:81%
and there was nothing easy about that.

00:08:59.569 --> 00:09:04.741 align:center line:79%
We also grew up in a...
really, really terrific parents,

00:09:04.841 --> 00:09:06.509 align:center line:81%
but we weren't wealthy.

00:09:06.643 --> 00:09:08.912 align:center line:79%
And so, they worked hard.
They work hard today.

00:09:09.579 --> 00:09:12.315 align:center line:81%
So they passed along a lot of life lessons

00:09:12.449 --> 00:09:13.917 align:center line:81%
by working hard.

00:09:14.050 --> 00:09:16.219 align:center line:81%
Now, I had all kinds of jobs.

00:09:18.154 --> 00:09:23.059 align:center line:79%
We went to a school
that included a lot of chores.

00:09:23.526 --> 00:09:27.430 align:center line:79%
-It was in Kentucky.
-Yeah. Kentucky, Oneida Baptist Institute.

00:09:28.131 --> 00:09:29.866 align:center line:81%
And I...

00:09:30.967 --> 00:09:32.869 align:center line:81%
I don't think it's the same as MIT

00:09:32.969 --> 00:09:37.107 align:center line:79%
that “I” is not the same.
It's the same word, but it's different.

00:09:37.240 --> 00:09:39.109 align:center line:81%
It's a different type of institute.

00:09:39.242 --> 00:09:42.545 align:center line:79%
But my institute required you
to go to school,

00:09:43.046 --> 00:09:45.348 align:center line:79%
and it was a dormitory,
and so there were a lot of chores.

00:09:45.648 --> 00:09:47.484 align:center line:81%
I was the youngest kid in school

00:09:47.817 --> 00:09:49.853 align:center line:79%
and so all
of the other kids got the hard work.

00:09:49.986 --> 00:09:52.489 align:center line:81%
They had to work in the tobacco farm.

00:09:52.622 --> 00:09:54.958 align:center line:79%
I got the easy job.
I was 9 years old.

00:09:55.091 --> 00:09:57.694 align:center line:79%
So after they left,
I had to clean all the bathrooms.

00:09:58.728 --> 00:10:01.531 align:center line:81%
I never felt that I got the easy job

00:10:01.965 --> 00:10:04.033 align:center line:81%
because what they left behind was...

00:10:05.101 --> 00:10:07.003 align:center line:81%
you can't unsee that kind of stuff.

00:10:07.137 --> 00:10:09.205 align:center line:81%
(Audience laughs.)

00:10:09.339 --> 00:10:12.642 align:center line:79%
But that was my job
and so I did it delightfully.

00:10:12.809 --> 00:10:16.846 align:center line:79%
Then I had plenty of other jobs,
and Denny's was one of them.

00:10:16.980 --> 00:10:21.451 align:center line:79%
I started out as a dishwasher
and became a busboy and became a waiter.

00:10:21.584 --> 00:10:23.820 align:center line:79%
And I loved every one of them.
I loved every one of them.

00:10:23.953 --> 00:10:27.323 align:center line:81%
Somehow, I've always found...

00:10:30.326 --> 00:10:31.961 align:center line:79%
I want to say joy,
but that's not quite right.

00:10:32.095 --> 00:10:36.766 align:center line:79%
I just, everything that I was doing,
I wanted to do the best I could.

00:10:37.000 --> 00:10:40.370 align:center line:79%
And maybe that was kind
of ingrained from the very beginning,

00:10:40.770 --> 00:10:44.374 align:center line:79%
but I was definitely
the best bathroom cleaner

00:10:44.474 --> 00:10:46.442 align:center line:81%
the world's ever seen, I'm sure of it, yeah.

00:10:47.544 --> 00:10:50.380 align:center line:81%
So if we fast-forward just a little bit

00:10:51.080 --> 00:10:52.815 align:center line:81%
to the NVIDIA of today.

00:10:53.816 --> 00:10:55.518 align:center line:81%
How large is your leadership team?

00:10:55.652 --> 00:10:58.288 align:center line:81%
How large is your leadership team?

00:10:59.656 --> 00:11:04.093 align:center line:79%
NVIDIA's leadership team is
60+, 60 people, yeah.

00:11:04.294 --> 00:11:05.995 align:center line:81%
And they all report to you?

00:11:06.095 --> 00:11:07.564 align:center line:81%
Yeah, they all report to me.

00:11:07.664 --> 00:11:10.066 align:center line:79%
Sixty direct reports.
-Sixty direct reports, yup.

00:11:10.200 --> 00:11:13.169 align:center line:79%
Which is not conventionally considered
a best practice.

00:11:13.836 --> 00:11:16.105 align:center line:81%
(Audience laughs.)

00:11:16.239 --> 00:11:17.907 align:center line:81%
I agree that may not be the best practice kind of...

00:11:18.041 --> 00:11:19.576 align:center line:81%
I'm certain that's the best practice.

00:11:19.709 --> 00:11:23.446 align:center line:79%
It's not conventional,
but I am certain it's the best practice.

00:11:24.180 --> 00:11:27.183 align:center line:81%
(Audience applauds.)

00:11:28.851 --> 00:11:31.087 align:center line:79%
By the end of this,
I'm going to convince all of you

00:11:31.187 --> 00:11:32.956 align:center line:81%
to have 60 people on your direct reports.

00:11:33.089 --> 00:11:34.357 align:center line:81%
(Audience laughs.)

00:11:34.490 --> 00:11:35.858 align:center line:81%
The floor is yours.

00:11:36.626 --> 00:11:41.931 align:center line:79%
First of all, the reason is
because the layer of hierarchy

00:11:42.065 --> 00:11:43.900 align:center line:81%
in your company really matters.

00:11:44.033 --> 00:11:46.002 align:center line:81%
Information really matters.

00:11:46.302 --> 00:11:49.405 align:center line:81%
I believe that your contribution

00:11:49.539 --> 00:11:53.209 align:center line:81%
to the work should not be based

00:11:53.343 --> 00:11:56.079 align:center line:81%
on the privileged access to information.

00:11:57.380 --> 00:11:58.748 align:center line:81%
I don't do one-on-ones

00:11:58.881 --> 00:12:02.252 align:center line:81%
and my staff is quite large.

00:12:03.119 --> 00:12:05.822 align:center line:79%
Almost everything that I say
I say to everybody all at the same time.

00:12:06.756 --> 00:12:08.524 align:center line:81%
The reason for that is

00:12:08.658 --> 00:12:11.794 align:center line:79%
because I don't really believe
there's any information that I operate on

00:12:12.128 --> 00:12:15.164 align:center line:79%
that somehow only one
or two people should hear about.

00:12:15.465 --> 00:12:17.834 align:center line:81%
“These are the challenges of the company”

00:12:17.934 --> 00:12:19.402 align:center line:79%
or “This is
the problem I'm trying to solve”

00:12:19.502 --> 00:12:20.937 align:center line:79%
or “This is the direction
we're trying to go into.

00:12:21.971 --> 00:12:24.674 align:center line:79%
These are the new endeavors.”
“This isn't working. That's working well.”

00:12:25.675 --> 00:12:29.245 align:center line:79%
And so all of this type of information,
everybody should be able to hear.

00:12:30.079 --> 00:12:33.516 align:center line:79%
I love that everybody's working off
of the same song sheet.

00:12:33.950 --> 00:12:37.253 align:center line:79%
I love that there is
no privileged access to information.

00:12:37.387 --> 00:12:40.890 align:center line:79%
I love that we're able
to all contribute to solving a problem.

00:12:41.291 --> 00:12:43.559 align:center line:81%
And when you have 60 people in a room

00:12:43.693 --> 00:12:47.697 align:center line:79%
and oftentimes, my staff meetings are
once every other week,

00:12:48.197 --> 00:12:51.467 align:center line:79%
it's all based on issues,
whatever issues we have.

00:12:51.701 --> 00:12:54.170 align:center line:79%
Everybody's there working
on it at the same time.

00:12:54.337 --> 00:12:57.740 align:center line:79%
Everybody heard
the reasoning of the problem.

00:12:57.874 --> 00:12:59.976 align:center line:79%
Everybody heard
the reasoning of the solution.

00:13:00.109 --> 00:13:02.111 align:center line:81%
Everybody heard everything.

00:13:02.245 --> 00:13:04.681 align:center line:81%
And so that empowers people.

00:13:05.214 --> 00:13:07.050 align:center line:79%
I believe that when you give
everybody equal access

00:13:07.183 --> 00:13:08.651 align:center line:81%
to information, it empowers people.

00:13:08.785 --> 00:13:10.653 align:center line:81%
And so, that's number one, empowering.

00:13:11.421 --> 00:13:12.689 align:center line:81%
Number two

00:13:14.857 --> 00:13:19.062 align:center line:81%
if the CEO's direct staff is 60 people,

00:13:19.162 --> 00:13:20.563 align:center line:81%
the number of layers you've removed

00:13:20.697 --> 00:13:22.699 align:center line:79%
in a company is probably
something like 7.

00:13:22.832 --> 00:13:24.534 align:center line:81%
Depending on how it is.

00:13:24.667 --> 00:13:26.869 align:center line:81%
Is it 60 at every layer? 

00:13:27.003 --> 00:13:31.074 align:center line:81%
As in, if I'm one of the fortunate 60,

00:13:31.541 --> 00:13:33.476 align:center line:81%
do I also have 60 direct reports?

00:13:33.576 --> 00:13:34.577 align:center line:79%
-No.
-Okay.

00:13:34.677 --> 00:13:37.046 align:center line:79%
I don't think that
that's scalable downward.

00:13:37.146 --> 00:13:40.416 align:center line:79%
And the reason for that is
because you need more

00:13:40.550 --> 00:13:45.121 align:center line:79%
and more supervision
depending on certain levels.

00:13:45.421 --> 00:13:48.624 align:center line:81%
And at the E-staff level,

00:13:48.725 --> 00:13:51.627 align:center line:79%
if you're so unfortunate
to be serving on NVIDIA's E-staff,

00:13:52.261 --> 00:13:56.065 align:center line:79%
it's very unlikely
you need a lot of managerial.

00:13:57.667 --> 00:14:00.303 align:center line:81%
So I rarely find myself having to...

00:14:01.304 --> 00:14:03.005 align:center line:81%
stand up for conventional wisdom.

00:14:03.139 --> 00:14:05.341 align:center line:81%
But if I were to steel man the other side

00:14:05.475 --> 00:14:08.478 align:center line:79%
I'd say, “Well, one-on-ones
are where you provide coaching,

00:14:08.611 --> 00:14:10.680 align:center line:79%
where you maybe talk
through goals together,

00:14:10.813 --> 00:14:13.716 align:center line:79%
personal goals,
career advancement, what have you.

00:14:14.083 --> 00:14:16.185 align:center line:81%
Where maybe you give feedback on something

00:14:16.319 --> 00:14:18.855 align:center line:79%
that you see somebody systematically
not doing so well and so forth.”

00:14:18.955 --> 00:14:21.791 align:center line:79%
And there is
all these things that one is, again

00:14:21.924 --> 00:14:23.693 align:center line:79%
conventionally supposed
to do in the one-on-one.

00:14:23.860 --> 00:14:26.329 align:center line:79%
Do you not do those things,
or do you do them in a different way?

00:14:26.896 --> 00:14:28.598 align:center line:81%
Really good question.

00:14:28.731 --> 00:14:29.732 align:center line:81%
I do it right there.

00:14:31.167 --> 00:14:32.735 align:center line:81%
Right there in...

00:14:32.869 --> 00:14:35.405 align:center line:79%
I give you feedback
right there in front of everybody.

00:14:36.672 --> 00:14:39.342 align:center line:81%
In fact, this is really a big deal.

00:14:39.776 --> 00:14:42.445 align:center line:81%
First of all, feedback is learning.

00:14:42.745 --> 00:14:43.846 align:center line:81%
Feedback is learning.

00:14:43.980 --> 00:14:46.816 align:center line:79%
For what reason are you the only person
who should learn this?

00:14:48.017 --> 00:14:49.986 align:center line:81%
Now, you created the conditions

00:14:51.521 --> 00:14:55.191 align:center line:81%
because of some mistake that you made

00:14:56.058 --> 00:15:00.263 align:center line:79%
or silliness that you
brought upon yourself.

00:15:00.596 --> 00:15:03.433 align:center line:81%
(Audience laughs.)

00:15:04.901 --> 00:15:07.737 align:center line:81%
We should all learn from that opportunity.

00:15:07.870 --> 00:15:10.540 align:center line:79%
So you created the conditions,
but we should all learn from it.

00:15:10.640 --> 00:15:13.743 align:center line:79%
Does it make sense?
And so, for me to explain to you

00:15:13.876 --> 00:15:16.746 align:center line:79%
why that doesn't make sense
or how I differ from it

00:15:16.879 --> 00:15:19.081 align:center line:81%
half the time, I'm not right.

00:15:19.215 --> 00:15:20.616 align:center line:81%
But, for me to reason through it

00:15:20.850 --> 00:15:21.851 align:center line:81%
in front of everybody

00:15:21.951 --> 00:15:23.820 align:center line:79%
helps everybody
learn how to reason through it.

00:15:24.487 --> 00:15:26.889 align:center line:81%
So the problem I have

00:15:27.023 --> 00:15:29.759 align:center line:79%
with one-on-ones
and taking feedback aside is

00:15:29.892 --> 00:15:32.995 align:center line:79%
you deprive a whole bunch
of people that same learning.

00:15:33.129 --> 00:15:34.797 align:center line:81%
Learning from mistakes,

00:15:34.931 --> 00:15:37.800 align:center line:79%
other people's mistakes,
is the best way to learn.

00:15:37.934 --> 00:15:39.602 align:center line:81%
Why learn from your own mistakes?

00:15:41.070 --> 00:15:43.239 align:center line:79%
You know, why learn
from your own embarrassment?

00:15:43.372 --> 00:15:45.408 align:center line:79%
You got to learn
from other people's embarrassment.

00:15:45.541 --> 00:15:48.244 align:center line:79%
That's why we have case studies,
and isn't that right?

00:15:48.578 --> 00:15:50.780 align:center line:79%
We're trying to read
from other people's disasters,

00:15:50.913 --> 00:15:52.348 align:center line:81%
other people's tragedies.

00:15:52.482 --> 00:15:54.283 align:center line:81%
Nothing makes us happier than that.

00:15:54.383 --> 00:15:55.885 align:center line:81%
(Audience laughs.)

00:15:56.586 --> 00:15:58.821 align:center line:79%
Have you succeeded
in getting other leaders

00:15:58.955 --> 00:16:00.723 align:center line:81%
at NVIDIA to adopt this practice?

00:16:00.857 --> 00:16:02.191 align:center line:81%
Or is that difficult?

00:16:02.291 --> 00:16:05.061 align:center line:79%
I give people the opportunity
to decide for themselves

00:16:05.194 --> 00:16:07.196 align:center line:81%
but I really discourage one-on-ones.

00:16:07.730 --> 00:16:09.031 align:center line:81%
I really discourage one-on-ones.

00:16:09.131 --> 00:16:13.236 align:center line:79%
Nothing is worse
than the idea that somebody says

00:16:13.369 --> 00:16:15.438 align:center line:81%
“Oh, Jensen wants us to do this.”

00:16:16.973 --> 00:16:19.475 align:center line:79%
Why does that have to be said to anybody?
Everybody should know.

00:16:20.243 --> 00:16:23.079 align:center line:81%
Or somebody said,

00:16:23.713 --> 00:16:25.381 align:center line:81%
“That E-staff said that.”

00:16:25.515 --> 00:16:27.383 align:center line:81%
Nothing drives me nuttier than that.

00:16:28.584 --> 00:16:31.687 align:center line:79%
You once told me
that you really didn't like firing people

00:16:31.821 --> 00:16:34.223 align:center line:81%
and very seldom did it.

00:16:34.357 --> 00:16:35.658 align:center line:81%
Can you elaborate on that?

00:16:37.126 --> 00:16:40.796 align:center line:79%
Well, I'd rather improve you
than give up on you.

00:16:41.397 --> 00:16:44.567 align:center line:79%
When you fire somebody,
you're kind of saying...

00:16:45.535 --> 00:16:47.436 align:center line:81%
Well, a lot of people say

00:16:47.537 --> 00:16:51.107 align:center line:79%
“Well, it wasn't your fault,”
or “I made the wrong choice.”

00:16:52.074 --> 00:16:53.543 align:center line:81%
There are very few jobs.

00:16:53.676 --> 00:16:56.579 align:center line:79%
Look, I used to clean bathrooms,
and now I'm the CEO of a company.

00:16:56.746 --> 00:16:58.281 align:center line:81%
I think you could learn it.

00:16:59.415 --> 00:17:01.083 align:center line:81%
I'm pretty certain you can learn this.

00:17:01.484 --> 00:17:04.153 align:center line:79%
There are a lot of things in life
that I believe you can learn

00:17:04.253 --> 00:17:06.489 align:center line:79%
and you just have
to be given the opportunity to learn it.

00:17:06.622 --> 00:17:09.392 align:center line:79%
I had the benefit of watching
a lot of smart people do a lot of things.

00:17:09.525 --> 00:17:12.461 align:center line:79%
I'm surrounded by 60 people
that are doing smart things all the time.

00:17:12.595 --> 00:17:14.397 align:center line:81%
They probably don't realize it

00:17:14.530 --> 00:17:16.966 align:center line:79%
but I'm learning constantly
from every single one of them.

00:17:17.366 --> 00:17:21.804 align:center line:79%
So I don't like giving up on people,
because I think they could improve.

00:17:22.071 --> 00:17:24.840 align:center line:79%
So there's...
It's kind of tongue in cheek,

00:17:24.941 --> 00:17:29.445 align:center line:79%
but people know that
I'd rather torture them into greatness.

00:17:29.912 --> 00:17:32.315 align:center line:79%
That was the phrase
that I was hoping to uncover.

00:17:32.415 --> 00:17:33.849 align:center line:81%
Yeah, I remember you mentioned that.

00:17:33.983 --> 00:17:35.818 align:center line:79%
Yeah.
So I'd rather torture you into greatness

00:17:35.918 --> 00:17:37.687 align:center line:81%
because I believe in you.

00:17:38.020 --> 00:17:43.092 align:center line:79%
And I think coaches
that really believe in their team

00:17:43.192 --> 00:17:44.694 align:center line:81%
torture them into greatness.

00:17:44.994 --> 00:17:47.496 align:center line:81%
And oftentimes they're so close.

00:17:47.997 --> 00:17:49.532 align:center line:81%
Don't give up. They're so close.

00:17:49.665 --> 00:17:51.467 align:center line:81%
Greatness, it comes all of a sudden.

00:17:51.567 --> 00:17:53.803 align:center line:79%
One day it's like, “I got it.”
Do you know what I'm saying?

00:17:54.370 --> 00:17:56.005 align:center line:79%
That feeling
that you didn't get it yesterday

00:17:56.138 --> 00:17:58.641 align:center line:79%
and all of a sudden one day
something clicked, and “Oh, I got it.”

00:17:59.442 --> 00:18:01.677 align:center line:79%
Could you imagine if
you gave up just that moment

00:18:01.811 --> 00:18:02.878 align:center line:81%
right before you got it?

00:18:03.312 --> 00:18:04.947 align:center line:81%
So I don't want you to give up on that.

00:18:05.081 --> 00:18:06.749 align:center line:81%
So let's just keep torturing you.

00:18:09.151 --> 00:18:10.319 align:center line:81%
How is...

00:18:12.021 --> 00:18:13.389 align:center line:81%
How's your work-life balance?

00:18:14.523 --> 00:18:16.258 align:center line:81%
(Audience laughs.)

00:18:21.731 --> 00:18:23.499 align:center line:81%
Well, it depends on who you ask.

00:18:24.834 --> 00:18:27.169 align:center line:79%
I think
my work-life balance is really great.

00:18:27.737 --> 00:18:29.171 align:center line:81%
It's really great.

00:18:29.972 --> 00:18:33.075 align:center line:81%
I work as much as I can.

00:18:33.743 --> 00:18:35.945 align:center line:81%
(Audience laughs.)

00:18:38.414 --> 00:18:39.949 align:center line:81%
I feel like he's judging me.

00:18:40.082 --> 00:18:41.951 align:center line:81%
(Audience laughs.)

00:18:42.585 --> 00:18:45.988 align:center line:79%
You know, I'm older than you.
I have more wisdom than you.

00:18:46.122 --> 00:18:47.156 align:center line:81%
So what I...

00:18:47.289 --> 00:18:48.524 align:center line:81%
(Audience laughs.)

00:18:48.658 --> 00:18:50.660 align:center line:79%
These are all the highlights
from our conversations

00:18:50.760 --> 00:18:53.195 align:center line:79%
that I think
more people should get to hear, so.

00:18:55.031 --> 00:18:58.267 align:center line:81%
Well, I work from the moment I wake up

00:18:58.401 --> 00:18:59.535 align:center line:81%
to the moment I go to bed.

00:18:59.669 --> 00:19:01.637 align:center line:81%
And I work seven days a week.

00:19:02.304 --> 00:19:04.240 align:center line:79%
When I'm not working,
I'm thinking about working.

00:19:05.341 --> 00:19:08.811 align:center line:81%
And when I'm working, I'm working, and so.

00:19:10.246 --> 00:19:13.816 align:center line:79%
I sit through movies,
but I don't remember them

00:19:13.916 --> 00:19:15.084 align:center line:81%
because I'm thinking about work.

00:19:15.317 --> 00:19:16.752 align:center line:81%
(Audience laughs.)

00:19:18.187 --> 00:19:19.188 align:center line:81%
You know?

00:19:19.555 --> 00:19:24.660 align:center line:79%
And so that's...
but my work is not as, you know...

00:19:25.594 --> 00:19:27.596 align:center line:81%
It's not “working” as in...

00:19:28.531 --> 00:19:31.267 align:center line:79%
There's this problem,
and you're trying to solve this problem.

00:19:31.767 --> 00:19:34.170 align:center line:79%
You're thinking
about what the company can be

00:19:34.303 --> 00:19:37.139 align:center line:79%
and are there things
that we could do even better.

00:19:37.273 --> 00:19:39.408 align:center line:79%
Or sometimes,
it's just trying to solve a problem.

00:19:40.543 --> 00:19:42.511 align:center line:81%
But sometimes you're imagining the future

00:19:42.645 --> 00:19:44.547 align:center line:81%
and boy, if we did this and that.

00:19:45.014 --> 00:19:46.649 align:center line:81%
And it's working.

00:19:46.782 --> 00:19:48.818 align:center line:81%
You're fantasizing, you're dreaming.

00:19:49.251 --> 00:19:51.087 align:center line:81%
I mean, that's incredible.

00:19:51.220 --> 00:19:53.222 align:center line:79%
Well, so yeah,
to concretize this a little bit

00:19:53.355 --> 00:19:54.824 align:center line:81%
and then, we will get to talk about AI,

00:19:54.957 --> 00:19:57.593 align:center line:81%
which I hear is a thing these days, but...

00:19:58.094 --> 00:20:01.297 align:center line:79%
-It's a thing.
-Yeah. Officially a thing, TM.

00:20:01.430 --> 00:20:04.467 align:center line:81%
But, to concretize this a bit,

00:20:04.600 --> 00:20:07.436 align:center line:79%
what does a day
in Jensen's life look like?

00:20:08.270 --> 00:20:10.339 align:center line:79%
-Like when do you wake up?
-Well, I used to wake up at 5.

00:20:10.439 --> 00:20:12.308 align:center line:79%
These days I wake up
at 6 because of my dogs.

00:20:14.143 --> 00:20:16.812 align:center line:79%
And the reason why,
6 is somehow we decided

00:20:16.946 --> 00:20:18.781 align:center line:79%
that 6 o'clock is
when they should wake up.

00:20:19.715 --> 00:20:22.017 align:center line:81%
And I don't know what it is.

00:20:22.151 --> 00:20:24.186 align:center line:81%
I don't mind waking anybody up,

00:20:24.920 --> 00:20:26.822 align:center line:79%
but I feel guilty
when I wake the puppies up.

00:20:27.423 --> 00:20:29.024 align:center line:81%
(Audience laughs.)

00:20:29.291 --> 00:20:31.727 align:center line:81%
It actually burdens me.

00:20:32.361 --> 00:20:34.964 align:center line:81%
So I don't want to move, it might...

00:20:35.097 --> 00:20:36.999 align:center line:81%
they pick up on any vibration

00:20:37.099 --> 00:20:39.301 align:center line:81%
in the house and it wakes them up.

00:20:40.069 --> 00:20:42.638 align:center line:81%
So we stay in bed

00:20:43.239 --> 00:20:47.076 align:center line:79%
and I just read in bed
until 6 o'clock and it's time.

00:20:47.176 --> 00:20:48.911 align:center line:81%
But you're thinking about GPUs?

00:20:49.211 --> 00:20:51.313 align:center line:81%
Oh, yes, yeah, yeah, sure.

00:20:51.781 --> 00:20:54.483 align:center line:79%
I'm obsessed about GPUs.
I mean, what can you do?

00:20:54.717 --> 00:20:57.553 align:center line:81%
I'm constantly... no, I'm just...

00:20:57.686 --> 00:21:00.356 align:center line:79%
Then the day is all,
I guess, group meetings

00:21:00.489 --> 00:21:02.024 align:center line:81%
because it can't be one on one meetings.

00:21:02.124 --> 00:21:04.026 align:center line:79%
Yeah.
I get my work done before I go to work.

00:21:04.693 --> 00:21:06.128 align:center line:81%
And then when I get to work...

00:21:07.163 --> 00:21:09.465 align:center line:81%
(Audience laughs.)

00:21:09.598 --> 00:21:11.433 align:center line:81%
How many meetings in a typical day?

00:21:12.701 --> 00:21:14.003 align:center line:81%
Pretty much all day long.

00:21:14.503 --> 00:21:17.072 align:center line:79%
So I select the meetings
that are really important to me.

00:21:17.206 --> 00:21:21.243 align:center line:81%
I try not to have regular meetings

00:21:21.877 --> 00:21:24.146 align:center line:81%
regular operational meetings

00:21:24.313 --> 00:21:25.848 align:center line:79%
because I've got
amazing people in the company

00:21:25.948 --> 00:21:28.317 align:center line:79%
who are doing
regular operational meetings.

00:21:28.450 --> 00:21:30.653 align:center line:81%
So we're pinch hitters.

00:21:31.086 --> 00:21:32.188 align:center line:81%
CEOs are pinch hitters.

00:21:32.321 --> 00:21:35.324 align:center line:79%
We should be working on the things
that nobody else can or nobody else is.

00:21:35.457 --> 00:21:37.893 align:center line:79%
So you're jumping
into projects that are stuck

00:21:38.027 --> 00:21:39.195 align:center line:81%
or offtrack.

00:21:39.328 --> 00:21:40.429 align:center line:79%
-That's right.
-Or new ideas.

00:21:40.529 --> 00:21:41.997 align:center line:81%
Wherever we can move the needle.

00:21:42.298 --> 00:21:44.233 align:center line:81%
No reporting. No reporting meetings.

00:21:44.366 --> 00:21:45.568 align:center line:81%
I hate reporting meetings.

00:21:45.701 --> 00:21:48.137 align:center line:79%
They don't want to report to me,
and just problem meetings.

00:21:48.270 --> 00:21:50.105 align:center line:81%
So problem meetings or idea meetings

00:21:50.239 --> 00:21:52.441 align:center line:79%
or brainstorming meetings
or creation meetings

00:21:52.541 --> 00:21:53.642 align:center line:81%
or whatever it is.

00:21:53.776 --> 00:21:55.344 align:center line:81%
Those are the meetings I go to.

00:21:55.477 --> 00:21:57.546 align:center line:81%
Usually I call them.

00:21:58.214 --> 00:22:01.417 align:center line:79%
I try really hard not to have
Outlook manage my life.

00:22:01.784 --> 00:22:04.320 align:center line:81%
We purposefully decide what kind of things

00:22:04.453 --> 00:22:06.388 align:center line:81%
that we want to do, we want to work on.

00:22:06.522 --> 00:22:10.993 align:center line:79%
So I try to live a life of purpose
and I manage my time accordingly.

00:22:12.094 --> 00:22:15.531 align:center line:81%
You used a phrase once

00:22:16.332 --> 00:22:18.000 align:center line:81%
“Zero-billion-dollar markets,”

00:22:18.133 --> 00:22:20.502 align:center line:79%
that zero-billion-dollar markets are
your favorite markets.

00:22:20.636 --> 00:22:22.238 align:center line:79%
-Yeah.
-What do you mean?

00:22:24.607 --> 00:22:29.745 align:center line:79%
If you take a step back, our purpose,
almost all of our purposes should be

00:22:30.479 --> 00:22:35.985 align:center line:79%
to go and do something
that has never been done before

00:22:37.119 --> 00:22:38.888 align:center line:81%
that is insanely hard to do—

00:22:39.521 --> 00:22:42.458 align:center line:79%
that if you achieve it,
it could make a real contribution.

00:22:43.826 --> 00:22:46.896 align:center line:79%
-I know your company does that.
-I try to do that.

00:22:47.596 --> 00:22:50.299 align:center line:79%
If that's the case,
it hasn't been done before

00:22:50.432 --> 00:22:52.234 align:center line:81%
it's incredibly hard to do.

00:22:52.368 --> 00:22:54.637 align:center line:79%
It's probably,
and it's never been done before.

00:22:55.271 --> 00:22:59.008 align:center line:79%
That market is
probably zero billion dollars in size

00:22:59.141 --> 00:23:00.943 align:center line:81%
because it has never been done before.

00:23:01.310 --> 00:23:05.981 align:center line:79%
I'd rather be a market maker,
market creator than a market taker.

00:23:08.183 --> 00:23:10.286 align:center line:81%
You know, to create something new

00:23:10.419 --> 00:23:12.988 align:center line:79%
that never existed before
versus thinking about share.

00:23:13.122 --> 00:23:16.425 align:center line:79%
I don't love thinking about share.
I don't like the concept of share.

00:23:17.293 --> 00:23:18.427 align:center line:81%
The reason for that is

00:23:18.560 --> 00:23:22.031 align:center line:79%
because if you think
about it in the big picture,

00:23:23.766 --> 00:23:25.868 align:center line:81%
Stripe existed out of thin air.

00:23:26.101 --> 00:23:29.438 align:center line:79%
You vaporized.
You created something out of vapor.

00:23:29.571 --> 00:23:31.640 align:center line:81%
It wasn't as if there was another...

00:23:32.875 --> 00:23:34.677 align:center line:81%
something else.

00:23:34.810 --> 00:23:38.647 align:center line:79%
So I'd like to think
that we can come up

00:23:38.781 --> 00:23:42.351 align:center line:79%
with something
that is zero billion dollars.

00:23:42.484 --> 00:23:44.186 align:center line:81%
A zero-billion-dollar market is

00:23:44.420 --> 00:23:46.622 align:center line:79%
a good way to cause
the company to think

00:23:46.755 --> 00:23:49.692 align:center line:79%
about how to go create
something for the first time.

00:23:50.192 --> 00:23:52.861 align:center line:79%
So our mission is to grow
the GDP of the internet.

00:23:52.995 --> 00:23:55.197 align:center line:81%
I mean, the GDP of the internet—

00:23:55.331 --> 00:23:59.768 align:center line:79%
a clause in that
usually gets most of the attention.

00:23:59.902 --> 00:24:03.472 align:center line:79%
But I think the most important part is
just the verb grow.

00:24:03.706 --> 00:24:04.840 align:center line:81%
Yeah.

00:24:05.140 --> 00:24:08.644 align:center line:79%
Because, to your point
we shouldn't be thinking about, well...

00:24:08.744 --> 00:24:11.347 align:center line:79%
which are the transactions
that are already happening

00:24:11.447 --> 00:24:13.148 align:center line:79%
or which are
the businesses that already exist.

00:24:13.282 --> 00:24:15.617 align:center line:79%
We should be thinking about which are
the transactions that don't exist.

00:24:15.751 --> 00:24:17.386 align:center line:79%
-That's right.
-Which are the businesses

00:24:17.486 --> 00:24:18.988 align:center line:79%
-that don't exist.
-Exactly.

00:24:19.088 --> 00:24:21.590 align:center line:79%
The GDP of the world is
around $100 trillion

00:24:21.690 --> 00:24:23.459 align:center line:81%
but it doesn't have to be $100 trillion.

00:24:23.592 --> 00:24:25.961 align:center line:79%
It could be
$200 trillion or $1,000 trillion.

00:24:26.095 --> 00:24:28.263 align:center line:79%
That's exactly right.
That's exactly right.

00:24:28.397 --> 00:24:30.833 align:center line:81%
Most of the value we're going to create

00:24:30.966 --> 00:24:32.835 align:center line:81%
over the next several decades

00:24:33.369 --> 00:24:38.240 align:center line:81%
are likely not limited by physical things.

00:24:38.841 --> 00:24:41.577 align:center line:81%
So this is a pretty extraordinary time.

00:24:41.710 --> 00:24:45.147 align:center line:79%
So with this concept
of zero-billion-dollar markets

00:24:45.280 --> 00:24:46.648 align:center line:81%
if I'm, again, at NVIDIA

00:24:46.782 --> 00:24:49.418 align:center line:79%
am I coming to you
with some proposal for some project

00:24:49.551 --> 00:24:52.888 align:center line:79%
and maybe there's several
billion dollars of CapEx involved

00:24:53.022 --> 00:24:56.625 align:center line:79%
or, you know,
it's a many-year pursuit or something

00:24:56.759 --> 00:25:00.295 align:center line:81%
and there are no customers for it today.

00:25:00.396 --> 00:25:02.531 align:center line:79%
There's no demand
that I can demonstrate for it.

00:25:02.664 --> 00:25:06.468 align:center line:79%
And you guys are just making
a gut call to say that

00:25:06.969 --> 00:25:10.005 align:center line:79%
“Yes, nobody's doing this today.
We think they could.

00:25:10.139 --> 00:25:12.508 align:center line:79%
We think they should.
And therefore, we're going to pursue it.”

00:25:13.008 --> 00:25:15.110 align:center line:79%
Really close.
Yeah, it's kind of like that.

00:25:15.778 --> 00:25:17.880 align:center line:81%
It's a gut call in the sense

00:25:18.013 --> 00:25:21.917 align:center line:79%
that your intuition says
something as a starting thesis

00:25:22.551 --> 00:25:24.520 align:center line:81%
but then you have to reason through it.

00:25:24.820 --> 00:25:27.656 align:center line:79%
And the reasoning of it is
much, much more important to me

00:25:27.790 --> 00:25:28.891 align:center line:81%
than a spreadsheet.

00:25:29.158 --> 00:25:30.159 align:center line:81%
I hate spreadsheets

00:25:30.292 --> 00:25:32.661 align:center line:79%
because you can make
spreadsheets do whatever you want.

00:25:32.795 --> 00:25:35.064 align:center line:79%
You can make any chart
you want out of a spreadsheet.

00:25:35.364 --> 00:25:36.999 align:center line:81%
You just got to type in some numbers.

00:25:37.433 --> 00:25:39.668 align:center line:79%
So I don't love
spreadsheets for that reason.

00:25:39.802 --> 00:25:41.603 align:center line:81%
I love words for that reason.

00:25:41.737 --> 00:25:44.506 align:center line:79%
Words are reasoning.
Tell me, how did you reason through this?

00:25:44.640 --> 00:25:47.443 align:center line:79%
What's our intuition?
Why do we believe that matters?

00:25:47.576 --> 00:25:48.877 align:center line:81%
Why do we think it's hard?

00:25:49.878 --> 00:25:52.714 align:center line:79%
I like hard things,
because it takes a long time to do.

00:25:53.082 --> 00:25:54.783 align:center line:81%
And if it takes a long time to do,

00:25:56.018 --> 00:26:00.022 align:center line:79%
a lot of people who are less committed
probably won't do it.

00:26:01.223 --> 00:26:03.926 align:center line:79%
If it's really, really hard to do,
it takes a long time to do

00:26:04.259 --> 00:26:06.462 align:center line:81%
it takes us a really resilient

00:26:06.595 --> 00:26:09.898 align:center line:79%
and a really dedicated,
really committed person to go after it.

00:26:10.365 --> 00:26:13.335 align:center line:81%
And if it also takes a long time to do

00:26:13.469 --> 00:26:15.237 align:center line:79%
you can kind of flounder around
for a couple years

00:26:15.370 --> 00:26:17.106 align:center line:81%
nobody notices.

00:26:18.474 --> 00:26:21.510 align:center line:79%
So I could be
incompetent for several years

00:26:21.643 --> 00:26:24.246 align:center line:81%
and everybody goes, well, who saw it?

00:26:25.214 --> 00:26:27.783 align:center line:81%
And where did CUDA come from?

00:26:29.384 --> 00:26:32.955 align:center line:81%
CUDA came originally from two ideas.

00:26:33.088 --> 00:26:34.156 align:center line:81%
One is called...

00:26:36.792 --> 00:26:39.928 align:center line:81%
I hate to get technical, but we created—

00:26:40.028 --> 00:26:42.131 align:center line:79%
we pioneered this idea called
accelerated computing.

00:26:42.564 --> 00:26:48.070 align:center line:81%
Accelerated computing is like an IO device,

00:26:48.537 --> 00:26:52.207 align:center line:79%
something that you sit on PCI Express,
if anybody's in the computer business,

00:26:52.341 --> 00:26:55.410 align:center line:81%
an IO device that allows

00:26:56.512 --> 00:26:59.515 align:center line:79%
the application to interact
with that IO device in such a way

00:26:59.648 --> 00:27:02.384 align:center line:81%
as to accelerate parts of the application.

00:27:03.185 --> 00:27:10.058 align:center line:79%
And UDA was an invention in 1993,
and this really profound invention allows

00:27:10.192 --> 00:27:14.630 align:center line:79%
the software programmer
to directly program

00:27:14.763 --> 00:27:18.767 align:center line:79%
an IO device, write
an application directly to the IO device

00:27:18.901 --> 00:27:20.669 align:center line:81%
because the IO device is virtualized

00:27:20.802 --> 00:27:22.237 align:center line:81%
and it's...

00:27:24.373 --> 00:27:26.975 align:center line:79%
architecturally compatible
across multiple generations.

00:27:28.810 --> 00:27:31.513 align:center line:79%
Anyways, we invented this idea
called accelerated computing

00:27:31.647 --> 00:27:34.483 align:center line:79%
and that was, we called it
Unified Driver Architecture

00:27:34.616 --> 00:27:35.617 align:center line:81%
for whatever reason.

00:27:36.718 --> 00:27:39.254 align:center line:81%
And then, several years later

00:27:39.388 --> 00:27:42.558 align:center line:79%
we thought we could make
our GPUs more programmable

00:27:42.691 --> 00:27:44.126 align:center line:81%
to high-level programming languages

00:27:44.259 --> 00:27:46.895 align:center line:81%
and we invented this idea called CG.

00:27:47.429 --> 00:27:48.864 align:center line:81%
C for graphics.

00:27:49.131 --> 00:27:51.333 align:center line:81%
C for graphics processors.

00:27:51.934 --> 00:27:57.439 align:center line:79%
That opened up
some really exciting opportunities.

00:27:57.539 --> 00:27:59.041 align:center line:81%
And we thought, you know what?

00:27:59.141 --> 00:28:01.109 align:center line:81%
This is going to work, but CG,

00:28:01.243 --> 00:28:03.345 align:center line:79%
the programming model,
wasn't exactly right.

00:28:03.478 --> 00:28:06.448 align:center line:81%
And so we extended, we invented CUDA,

00:28:06.582 --> 00:28:09.952 align:center line:79%
which is compute with...
So anyways, that's how.

00:28:10.452 --> 00:28:12.721 align:center line:79%
It's a horrible story, frankly.
(Audience laughs.)

00:28:12.821 --> 00:28:15.190 align:center line:79%
Anyways, we invented this idea
called accelerated computing.

00:28:15.324 --> 00:28:16.625 align:center line:81%
We pioneered this approach.

00:28:16.758 --> 00:28:20.295 align:center line:79%
I guess the real question is,
was it a smash hit overnight?

00:28:21.430 --> 00:28:22.965 align:center line:81%
No, it was a...

00:28:24.533 --> 00:28:26.735 align:center line:81%
It was an incredible disaster overnight.

00:28:28.403 --> 00:28:30.272 align:center line:81%
And it kind of went like this.

00:28:30.439 --> 00:28:32.674 align:center line:79%
So this is one of your zero-
billion-dollar markets you went after.

00:28:32.774 --> 00:28:33.875 align:center line:81%
Yeah.

00:28:33.976 --> 00:28:35.744 align:center line:79%
-And it was a disaster.
-Yeah.

00:28:35.944 --> 00:28:38.313 align:center line:79%
Because it was
a zero billion dollar we went after,

00:28:38.447 --> 00:28:41.650 align:center line:79%
but it cost so much to go
after that zero-billion-dollar market

00:28:42.184 --> 00:28:45.887 align:center line:79%
it actually crushed
the $1 billion market we were enjoying.

00:28:46.021 --> 00:28:48.390 align:center line:81%
So, and the reason for that is

00:28:48.523 --> 00:28:51.526 align:center line:81%
because CUDA

00:28:52.494 --> 00:28:54.997 align:center line:81%
added a ton of cost into our chips.

00:28:55.731 --> 00:28:58.767 align:center line:79%
But there were no applications.
And there are no applications.

00:28:58.900 --> 00:29:02.704 align:center line:79%
Customers don't value the product,
and they won't pay you a premium for it.

00:29:03.338 --> 00:29:04.973 align:center line:79%
And if people aren't willing
to pay you for it

00:29:05.107 --> 00:29:08.110 align:center line:79%
but your cost went up,
then your gross margins get crushed.

00:29:08.243 --> 00:29:09.244 align:center line:81%
And we got...

00:29:09.378 --> 00:29:12.781 align:center line:81%
Our market cap was low

00:29:13.215 --> 00:29:15.117 align:center line:81%
and it went down to really low.

00:29:15.250 --> 00:29:17.019 align:center line:79%
It was like,
I think our market cap went down

00:29:17.152 --> 00:29:19.187 align:center line:79%
to like a billion dollars
or something like that.

00:29:19.321 --> 00:29:21.523 align:center line:79%
-I wish I bought it, but anyways...
-Okay.

00:29:21.623 --> 00:29:24.893 align:center line:79%
So therefore,
you immediately canceled CUDA

00:29:24.993 --> 00:29:26.595 align:center line:81%
and went back to the old strategy.

00:29:26.928 --> 00:29:28.563 align:center line:81%
No, no, I believed in CUDA

00:29:29.197 --> 00:29:30.666 align:center line:81%
because you reasoned about it.

00:29:30.799 --> 00:29:33.268 align:center line:79%
You reasoned about it.
Look, we really believe

00:29:33.402 --> 00:29:36.505 align:center line:79%
that accelerated computing
was going to be able to solve

00:29:36.638 --> 00:29:39.474 align:center line:81%
problems that normal computers couldn't.

00:29:40.409 --> 00:29:42.177 align:center line:81%
And if we wanted to extend

00:29:42.311 --> 00:29:46.315 align:center line:79%
the architecture to be
much more general purpose

00:29:46.448 --> 00:29:49.418 align:center line:81%
we had to make that sacrifice.

00:29:49.851 --> 00:29:55.157 align:center line:79%
So I deeply believed
in the mission of our company.

00:29:55.290 --> 00:29:57.893 align:center line:81%
I deeply believed in its opportunities.

00:29:58.026 --> 00:29:59.194 align:center line:81%
And so were analysts...

00:29:59.328 --> 00:30:01.396 align:center line:79%
And I deeply, deeply believed
that people were wrong.

00:30:02.130 --> 00:30:04.700 align:center line:81%
They just didn't appreciate what we built.

00:30:04.833 --> 00:30:06.134 align:center line:81%
I deeply believed it.

00:30:06.268 --> 00:30:08.770 align:center line:79%
And so, weren't analysts
and the board and employee

00:30:08.870 --> 00:30:11.606 align:center line:79%
like, “You've torpedoed
this existing revenue stream.

00:30:11.740 --> 00:30:14.142 align:center line:81%
You've this hyped thing that...

00:30:14.976 --> 00:30:18.847 align:center line:79%
you're selling a lofty dream around
that nobody seems to actually want.

00:30:18.980 --> 00:30:20.816 align:center line:81%
The business is really suffering.”

00:30:22.784 --> 00:30:24.319 align:center line:81%
Talk us through that.

00:30:24.453 --> 00:30:25.454 align:center line:81%
You believed.

00:30:26.154 --> 00:30:29.124 align:center line:79%
You just go something like this.
“Oh gosh, they're so dumb.”

00:30:30.058 --> 00:30:31.626 align:center line:81%
(Audience laughs.)

00:30:32.561 --> 00:30:33.729 align:center line:81%
Something like that.

00:30:33.862 --> 00:30:36.732 align:center line:79%
You know, denial.
No, I'm just kidding.

00:30:37.232 --> 00:30:38.800 align:center line:81%
No, you go back to what you believe.

00:30:39.334 --> 00:30:40.635 align:center line:81%
And if you believe something...

00:30:40.769 --> 00:30:43.305 align:center line:79%
Did the board put
pressure on you during this?

00:30:45.841 --> 00:30:48.744 align:center line:79%
I'd start every conversation 
with what I deeply believed.

00:30:49.945 --> 00:30:52.347 align:center line:81%
And they believed it

00:30:52.447 --> 00:30:54.449 align:center line:81%
because they saw me deeply believe it.

00:30:54.583 --> 00:30:55.884 align:center line:81%
And I reasoned about it.

00:30:56.017 --> 00:30:59.488 align:center line:81%
It wasn't like it was a spreadsheet

00:30:59.621 --> 00:31:01.590 align:center line:79%
and therefore,
you've got to believe the spreadsheet.

00:31:01.723 --> 00:31:03.658 align:center line:79%
They had to believe
the reasoning, the words.

00:31:03.792 --> 00:31:05.660 align:center line:81%
How long did it take it to start working?

00:31:09.898 --> 00:31:11.533 align:center line:81%
Probably 10 years, yeah.

00:31:13.235 --> 00:31:15.137 align:center line:79%
-Yeah. It wasn't that long.
-Yeah.

00:31:16.471 --> 00:31:18.173 align:center line:81%
Ten years. It comes and goes.

00:31:19.174 --> 00:31:21.076 align:center line:79%
-Ten years.
-Less than a third of your tenure.

00:31:21.209 --> 00:31:23.912 align:center line:79%
Yeah, it comes and goes.
It was, I barely remembered it.

00:31:24.012 --> 00:31:26.081 align:center line:81%
The suffering, I barely remembered it.

00:31:26.314 --> 00:31:31.553 align:center line:79%
Could NVIDIA be
as successful in AI without CUDA?

00:31:31.820 --> 00:31:33.088 align:center line:81%
No. Impossible.

00:31:33.221 --> 00:31:37.159 align:center line:79%
It is potentially one
of the most important inventions

00:31:37.292 --> 00:31:38.727 align:center line:81%
in modern computing.

00:31:38.860 --> 00:31:41.696 align:center line:79%
We invented this idea
called accelerated computing.

00:31:41.830 --> 00:31:46.802 align:center line:79%
And the idea is so simple,
but deeply profound.

00:31:47.302 --> 00:31:50.038 align:center line:81%
It says the vast majority...

00:31:50.172 --> 00:31:53.308 align:center line:79%
a small percentage
of the code of programs

00:31:54.342 --> 00:31:58.780 align:center line:81%
occupies, consumes 99.999% of the runtime.

00:31:59.881 --> 00:32:02.851 align:center line:79%
This is true for a lot
of very important applications.

00:32:03.285 --> 00:32:06.688 align:center line:81%
That small little kernel

00:32:06.822 --> 00:32:09.024 align:center line:81%
or, you know, some, several kernels...

00:32:11.426 --> 00:32:12.761 align:center line:81%
can be accelerated.

00:32:13.128 --> 00:32:14.429 align:center line:81%
And they tend...

00:32:14.563 --> 00:32:17.265 align:center line:79%
It's not all just parallel processing.
It's not as simple as that.

00:32:17.532 --> 00:32:20.168 align:center line:79%
But the idea is
that we can take that kernel

00:32:20.302 --> 00:32:22.337 align:center line:79%
that piece of software,
that part of the software

00:32:22.437 --> 00:32:23.772 align:center line:79%
and accelerate
the living daylights out of it.

00:32:24.506 --> 00:32:27.375 align:center line:79%
And today,
when Moore's Law has run its course

00:32:27.476 --> 00:32:29.878 align:center line:81%
and CPU scaling is basically stopped.

00:32:30.011 --> 00:32:33.048 align:center line:81%
If we don't accelerate every software

00:32:33.415 --> 00:32:37.352 align:center line:79%
you're going to see
extraordinary computation inflation.

00:32:37.486 --> 00:32:39.421 align:center line:79%
Because the amount
of computation the world does is

00:32:39.521 --> 00:32:42.424 align:center line:79%
doubling every year still,
and yet, if CPUs

00:32:42.557 --> 00:32:44.593 align:center line:79%
and general-purpose computers
are not increasing

00:32:44.726 --> 00:32:46.528 align:center line:81%
in performance because it's stopped,

00:32:46.661 --> 00:32:47.929 align:center line:81%
then what's your alternative?

00:32:48.063 --> 00:32:50.832 align:center line:79%
Or your cost of computing is going 
to keep going up exponentially.

00:32:50.999 --> 00:32:53.568 align:center line:81%
So the time has come for us to do that.

00:32:53.702 --> 00:32:56.071 align:center line:81%
So everyone here runs a business, and...

00:32:56.605 --> 00:32:59.808 align:center line:79%
-Accelerate everything.
-And you heard it here first.

00:33:00.041 --> 00:33:02.377 align:center line:81%
And probably everyone has...

00:33:02.878 --> 00:33:04.513 align:center line:81%
some version of CUDA

00:33:05.614 --> 00:33:08.416 align:center line:79%
or a thing
that they think really makes sense

00:33:08.550 --> 00:33:11.920 align:center line:79%
for the sector or makes sense
for their technology or what have you

00:33:12.654 --> 00:33:15.323 align:center line:81%
but where the market doesn't see it yet.

00:33:15.457 --> 00:33:18.927 align:center line:79%
Do you think it's possible to extract
any kind of generalizable principles

00:33:19.060 --> 00:33:23.198 align:center line:79%
around when you should
really doggedly trust that vision

00:33:23.331 --> 00:33:26.968 align:center line:79%
and when perhaps it's worth
reconsidering in a fashion that, yeah

00:33:27.102 --> 00:33:29.137 align:center line:79%
we could extrapolate
from, in the case of CUDA

00:33:29.271 --> 00:33:32.173 align:center line:79%
and other CUDAs that have existed
over the course of NVIDIA's history?

00:33:32.374 --> 00:33:35.377 align:center line:81%
Yeah, the question is determination

00:33:35.510 --> 00:33:37.679 align:center line:81%
and commitment versus stubbornness.

00:33:38.246 --> 00:33:40.682 align:center line:81%
And that line is fuzzy.

00:33:43.718 --> 00:33:47.656 align:center line:79%
Look, I gut-checked
against my core beliefs every day.

00:33:47.789 --> 00:33:50.625 align:center line:79%
I still do.
And you gut-check against it.

00:33:51.960 --> 00:33:54.262 align:center line:81%
The first principles by which you reasoned

00:33:54.396 --> 00:33:56.097 align:center line:81%
about your strategies

00:33:56.231 --> 00:33:58.767 align:center line:79%
the first principles by which you reasoned
about your strategies

00:33:58.900 --> 00:34:01.136 align:center line:79%
those first principles are
easy to remember.

00:34:02.003 --> 00:34:04.339 align:center line:81%
It's not a long list.

00:34:05.240 --> 00:34:10.111 align:center line:79%
Now the question is, 
did those principles...

00:34:10.245 --> 00:34:12.514 align:center line:81%
did they change in some fundamental way?

00:34:12.647 --> 00:34:16.484 align:center line:79%
Are external conditions
such that they no longer matter

00:34:16.618 --> 00:34:17.986 align:center line:81%
as much as before?

00:34:18.119 --> 00:34:20.188 align:center line:81%
Did somebody else solve the problem

00:34:20.322 --> 00:34:22.757 align:center line:79%
and therefore,
that problem has now disappeared?

00:34:24.059 --> 00:34:27.095 align:center line:81%
Is it, there will never be any need?

00:34:28.363 --> 00:34:30.398 align:center line:81%
You gut-check it constantly

00:34:30.765 --> 00:34:34.736 align:center line:79%
to the extent that
that's number one, gut-check.

00:34:34.869 --> 00:34:37.572 align:center line:79%
You have to, first of all, you really have
to be careful to distill down

00:34:37.706 --> 00:34:40.208 align:center line:79%
the first principle
instead of “I want something.”

00:34:40.342 --> 00:34:41.443 align:center line:81%
That's stubbornness.

00:34:43.078 --> 00:34:44.913 align:center line:81%
You can't reason about it. I just want it.

00:34:45.380 --> 00:34:47.515 align:center line:81%
We're not 5-year-olds, right?

00:34:49.150 --> 00:34:50.485 align:center line:79%
So you got to reason
about it, number one.

00:34:50.619 --> 00:34:52.887 align:center line:81%
Number two, you have to be clever.

00:34:53.021 --> 00:34:54.189 align:center line:81%
The fact of the matter is

00:34:54.289 --> 00:34:56.024 align:center line:79%
there are a lot
of new companies being created here.

00:34:56.157 --> 00:34:59.461 align:center line:79%
It's amazing how many great companies
are in the audience

00:35:00.195 --> 00:35:02.397 align:center line:81%
and young companies in the audience.

00:35:02.530 --> 00:35:03.565 align:center line:81%
You have to be clever.

00:35:03.698 --> 00:35:08.069 align:center line:81%
So we found ways to monetize

00:35:08.670 --> 00:35:12.440 align:center line:81%
even in a small way, CUDA.

00:35:12.540 --> 00:35:16.678 align:center line:79%
And so, we found app,
we looked everywhere for applications.

00:35:16.811 --> 00:35:19.114 align:center line:79%
We found an application
with CT reconstruction.

00:35:19.247 --> 00:35:21.816 align:center line:79%
We found an application
with seismic processing.

00:35:21.950 --> 00:35:24.586 align:center line:79%
We found another application
with molecular dynamics.

00:35:25.120 --> 00:35:27.756 align:center line:79%
And so we're constantly looking
for applications.

00:35:27.889 --> 00:35:30.825 align:center line:81%
They didn't make it a home run

00:35:31.292 --> 00:35:34.562 align:center line:79%
but it sustained us
just enough, just enough

00:35:34.996 --> 00:35:37.932 align:center line:79%
and bought us time
for it to really happen.

00:35:38.767 --> 00:35:42.203 align:center line:81%
Okay, so let's talk about AI.

00:35:42.337 --> 00:35:44.773 align:center line:79%
Maybe just going to do
some math to ground things here.

00:35:44.906 --> 00:35:48.810 align:center line:79%
Let's just say
that the total compute capacity

00:35:48.943 --> 00:35:51.946 align:center line:81%
of all GPUs in the world today is X.

00:35:53.348 --> 00:35:54.916 align:center line:81%
What do you think...

00:35:56.284 --> 00:36:00.822 align:center line:79%
What multiple of X
will we be at in five years?

00:36:04.592 --> 00:36:08.129 align:center line:79%
First of all, you know that
I'm going to regret saying this.

00:36:09.631 --> 00:36:10.665 align:center line:81%
And this is...

00:36:11.199 --> 00:36:13.902 align:center line:81%
I'm a public company, you crazy person.

00:36:14.035 --> 00:36:16.137 align:center line:81%
(Audience laughs.)

00:36:16.271 --> 00:36:21.042 align:center line:79%
See, this is...
how nice is it to be private?

00:36:21.176 --> 00:36:22.744 align:center line:81%
(audience laughs)

00:36:24.212 --> 00:36:26.648 align:center line:81%
(Audience applauds.)

00:36:29.451 --> 00:36:31.052 align:center line:81%
Safe to say considerably more?

00:36:31.986 --> 00:36:34.222 align:center line:81%
Well, let's reason about it, shall we?

00:36:34.622 --> 00:36:37.659 align:center line:79%
Okay, so let's reason about it.
Let's reason our way through, okay?

00:36:37.792 --> 00:36:39.494 align:center line:81%
So first of all, it goes like this.

00:36:39.794 --> 00:36:41.062 align:center line:81%
The world has installed

00:36:41.196 --> 00:36:43.231 align:center line:79%
about a trillion dollars
worth of data centers.

00:36:43.898 --> 00:36:46.301 align:center line:79%
Those trillion dollars
worth of data centers used

00:36:46.434 --> 00:36:47.936 align:center line:81%
as general-purpose computing.

00:36:48.069 --> 00:36:51.072 align:center line:79%
General-purpose computing has run
its course.

00:36:51.206 --> 00:36:53.608 align:center line:81%
We cannot continue to process that way.

00:36:53.742 --> 00:36:55.810 align:center line:79%
And so, the world is going
to accelerate everything

00:36:55.944 --> 00:36:57.746 align:center line:81%
data processing, you name it, okay?

00:36:57.879 --> 00:36:59.547 align:center line:81%
So we're going to accelerate everything.

00:37:00.215 --> 00:37:02.517 align:center line:79%
When we accelerate everything,
every single data center,

00:37:02.650 --> 00:37:05.386 align:center line:79%
every single computer will be
an accelerated server.

00:37:05.920 --> 00:37:08.189 align:center line:79%
Well,
there's about a trillion dollars worth

00:37:08.323 --> 00:37:10.325 align:center line:81%
of computers if we don't grow at all

00:37:10.892 --> 00:37:13.661 align:center line:79%
over the next, call it four years
that we have to go replace.

00:37:14.162 --> 00:37:16.498 align:center line:79%
Four years, six years,
pick your number of years.

00:37:16.831 --> 00:37:18.700 align:center line:79%
But, if the computer
industry continues to grow

00:37:18.833 --> 00:37:20.201 align:center line:81%
at some 20% or so,

00:37:20.335 --> 00:37:22.837 align:center line:79%
we'll probably have to replace
over the course of next...

00:37:23.271 --> 00:37:26.841 align:center line:79%
pick your number of years,
about $2 trillion worth of computers

00:37:26.975 --> 00:37:28.376 align:center line:81%
with accelerated computing.

00:37:29.077 --> 00:37:31.446 align:center line:81%
So, just make that GPUs, okay?

00:37:32.113 --> 00:37:34.215 align:center line:79%
That's number one.
And this is the second part.

00:37:34.315 --> 00:37:38.553 align:center line:81%
This is the reason why all of you, Stripe

00:37:38.686 --> 00:37:41.823 align:center line:79%
you're onto something
just absolutely monumental.

00:37:42.257 --> 00:37:45.527 align:center line:81%
This idea called, and...

00:37:46.227 --> 00:37:48.696 align:center line:79%
you've heard me say
an industrial revolution.

00:37:48.830 --> 00:37:50.131 align:center line:81%
Lemme tell you why.

00:37:50.365 --> 00:37:52.600 align:center line:79%
We are producing something
for the very first time

00:37:52.734 --> 00:37:54.502 align:center line:81%
that has never been produced before.

00:37:55.270 --> 00:37:57.572 align:center line:79%
And we're producing it
in extremely high volume.

00:37:57.939 --> 00:38:01.142 align:center line:81%
And the production of this thing requires

00:38:01.276 --> 00:38:03.344 align:center line:79%
a new instrument
that never existed before.

00:38:03.478 --> 00:38:04.479 align:center line:81%
It's a GPU.

00:38:05.814 --> 00:38:07.015 align:center line:81%
The thing that we're producing

00:38:07.148 --> 00:38:09.651 align:center line:79%
for the very first time,
for the mathematicians

00:38:09.784 --> 00:38:12.053 align:center line:79%
and all the computer scientists
in the room, for all of you

00:38:12.187 --> 00:38:14.522 align:center line:81%
you know that we're producing tokens.

00:38:14.923 --> 00:38:18.760 align:center line:79%
We're producing
floating point numbers at high volume

00:38:18.893 --> 00:38:20.361 align:center line:81%
for the first time in history.

00:38:20.495 --> 00:38:22.263 align:center line:81%
The floating-point numbers have value.

00:38:22.397 --> 00:38:24.966 align:center line:79%
The reason why they have
value is because it's intelligence.

00:38:25.099 --> 00:38:26.467 align:center line:81%
It's artificial intelligence.

00:38:26.601 --> 00:38:28.469 align:center line:81%
You can take these floating-point numbers,

00:38:28.603 --> 00:38:31.906 align:center line:79%
you reformulate it
in such a way that it turns into English,

00:38:32.040 --> 00:38:36.744 align:center line:79%
French, proteins,
chemicals, graphics, images,

00:38:36.878 --> 00:38:41.382 align:center line:79%
videos, robotic articulation,
steering wheel articulation.

00:38:41.516 --> 00:38:44.752 align:center line:79%
We're producing tokens
at extraordinary scale.

00:38:45.286 --> 00:38:48.122 align:center line:79%
Now, we've discovered a way
through all of the work that we do

00:38:48.256 --> 00:38:50.458 align:center line:79%
with artificial intelligence
to produce tokens

00:38:50.592 --> 00:38:51.726 align:center line:81%
of almost any kind.

00:38:52.827 --> 00:38:56.397 align:center line:81%
So now, the world is going to produce

00:38:56.531 --> 00:38:58.566 align:center line:81%
an enormous amount of tokens.

00:38:59.133 --> 00:39:01.135 align:center line:81%
Now these tokens are going to be produced

00:39:01.269 --> 00:39:02.403 align:center line:81%
in new types of data centers.

00:39:02.537 --> 00:39:03.905 align:center line:81%
We call them AI factories.

00:39:05.340 --> 00:39:07.642 align:center line:81%
Back in the last industrial revolution,

00:39:08.009 --> 00:39:12.280 align:center line:79%
water comes into a machine,
you light the water on fire.

00:39:12.914 --> 00:39:14.349 align:center line:81%
Turn it into steam

00:39:15.083 --> 00:39:17.352 align:center line:81%
and then it turns into electrons.

00:39:17.619 --> 00:39:19.621 align:center line:81%
Atoms come in; electrons go out.

00:39:19.754 --> 00:39:21.389 align:center line:81%
In this new industrial revolution,

00:39:21.522 --> 00:39:24.592 align:center line:79%
electrons come in
and floating-point numbers come out.

00:39:25.560 --> 00:39:28.062 align:center line:79%
And just like
the last industrial revolution,

00:39:28.162 --> 00:39:30.832 align:center line:79%
nobody understood why 
this electricity is so valuable

00:39:30.965 --> 00:39:34.335 align:center line:81%
and is now sold, marketed kilowatt hours

00:39:34.535 --> 00:39:35.803 align:center line:81%
per dollar.

00:39:35.937 --> 00:39:39.540 align:center line:79%
And so,
now we have million tokens per dollar.

00:39:40.942 --> 00:39:44.712 align:center line:81%
That same logic is as incomprehensible

00:39:44.846 --> 00:39:47.215 align:center line:79%
to a lot of people
as the last industrial revolution,

00:39:47.315 --> 00:39:50.251 align:center line:79%
but it's going to be
completely normal in the next 10 years.

00:39:51.085 --> 00:39:56.090 align:center line:81%
Well, these tokens are going to create

00:39:56.791 --> 00:39:58.860 align:center line:81%
new products, new services,

00:39:58.993 --> 00:40:01.996 align:center line:79%
enhanced productivity
on whole slew of industries,

00:40:02.130 --> 00:40:05.066 align:center line:79%
a hundred trillion dollars
worth of industries on top of us.

00:40:05.700 --> 00:40:07.735 align:center line:81%
So this industry is going to be gigantic.

00:40:07.869 --> 00:40:11.706 align:center line:81%
In order to monetize that, transact that

00:40:11.839 --> 00:40:14.008 align:center line:81%
you're going to need Stripe, and so.

00:40:15.109 --> 00:40:17.312 align:center line:81%
(Audience applauds.)

00:40:19.814 --> 00:40:22.583 align:center line:79%
I got to tell you: this is one
of my favorite companies.

00:40:23.051 --> 00:40:26.020 align:center line:79%
The first time I met Patrick,
he had to explain Stripe to me.

00:40:26.120 --> 00:40:29.157 align:center line:79%
I was, first of all, 
it was so complicated.

00:40:30.325 --> 00:40:31.592 align:center line:81%
Because it's complicated.

00:40:31.726 --> 00:40:33.728 align:center line:79%
We tried to refine
the descriptions over time.

00:40:33.828 --> 00:40:35.863 align:center line:79%
But you got an early version.
-No, you're in a complicated business.

00:40:35.964 --> 00:40:36.998 align:center line:81%
no matter what.

00:40:37.498 --> 00:40:42.337 align:center line:79%
But nonetheless, I was so inspired by it.
Incredible what you guys have built.

00:40:42.470 --> 00:40:44.505 align:center line:79%
Are we going
to get you migrated to Stripe Billing

00:40:44.639 --> 00:40:46.374 align:center line:81%
now that we have usage-based billing?

00:40:46.507 --> 00:40:48.943 align:center line:79%
I wish I had a business
that required billing.

00:40:50.945 --> 00:40:53.748 align:center line:79%
I think your public filing suggests
you're doing a lot of billing.

00:40:54.182 --> 00:40:56.317 align:center line:81%
(Audience laughs.)

00:40:57.118 --> 00:40:59.954 align:center line:79%
-We'll follow up on it.
-All right, so.

00:41:00.755 --> 00:41:03.091 align:center line:79%
It's only 10 transactions,
just so you know.

00:41:03.224 --> 00:41:04.359 align:center line:81%
(Audience laughs.)

00:41:04.492 --> 00:41:07.795 align:center line:81%
Your economics serving us is like nothing

00:41:07.895 --> 00:41:09.263 align:center line:81%
It's like 10 transactions.

00:41:10.031 --> 00:41:12.433 align:center line:79%
Remember,
we'd happily take the 2.9%, but anyway.

00:41:12.567 --> 00:41:13.701 align:center line:81%
(Jensen laughs)

00:41:13.801 --> 00:41:15.470 align:center line:81%
We can discuss that separately.

00:41:16.504 --> 00:41:18.172 align:center line:81%
So...

00:41:18.973 --> 00:41:21.242 align:center line:79%
-Done.
-Think about this token.

00:41:21.376 --> 00:41:23.077 align:center line:79%
You can't say that.
You're a public company.

00:41:23.211 --> 00:41:24.212 align:center line:81%
(Audience laughs.)

00:41:24.379 --> 00:41:26.114 align:center line:81%
So thinking about these...

00:41:26.581 --> 00:41:29.617 align:center line:79%
token factories, I feel like
a big question right now is

00:41:29.717 --> 00:41:32.487 align:center line:79%
whether the models saturate
in the sense that, you know

00:41:32.587 --> 00:41:35.023 align:center line:79%
we demoed
the Sigma Assistant on stage earlier.

00:41:35.156 --> 00:41:36.657 align:center line:81%
And you can write some natural language

00:41:36.791 --> 00:41:38.026 align:center line:81%
and we convert that to SQL.

00:41:38.159 --> 00:41:41.429 align:center line:79%
And going from, you know,
maybe a 7 billion parameter model

00:41:41.562 --> 00:41:43.131 align:center line:81%
to a 70 billion parameter model

00:41:43.264 --> 00:41:46.000 align:center line:79%
or something like that, 
there might be a significant kind of...

00:41:46.667 --> 00:41:49.771 align:center line:79%
consequential improvement
in query accuracy for the user

00:41:49.904 --> 00:41:52.607 align:center line:79%
for the typical kind of queries
that people tend to construct.

00:41:52.740 --> 00:41:53.808 align:center line:81%
But, maybe going to a model

00:41:53.941 --> 00:41:56.177 align:center line:79%
of 10x larger than that
is sort of unnecessary.

00:41:56.310 --> 00:41:57.345 align:center line:79%
Like, at some point,
you get too good enough

00:41:57.478 --> 00:41:59.847 align:center line:79%
you can reliably convert
the natural language in SQL.

00:41:59.981 --> 00:42:01.482 align:center line:81%
I think there's a question of...

00:42:02.483 --> 00:42:04.919 align:center line:79%
for the use cases
for which LLMs are being deployed,

00:42:05.053 --> 00:42:07.855 align:center line:81%
what does that saturation curve look like

00:42:07.989 --> 00:42:11.459 align:center line:79%
and for how many use cases does
one need a trillion-parameter model

00:42:11.592 --> 00:42:12.860 align:center line:81%
or a 10 trillion-parameter model?

00:42:12.994 --> 00:42:15.797 align:center line:81%
Or do we simply reach a point where

00:42:15.930 --> 00:42:19.700 align:center line:79%
some number that is say less
than 100 billion is sufficient?

00:42:19.834 --> 00:42:21.002 align:center line:81%
Do you have any point of view on that?

00:42:21.135 --> 00:42:22.670 align:center line:81%
Or is that even...

00:42:23.404 --> 00:42:26.607 align:center line:79%
a reasonable way to look
at the question in the first place?

00:42:26.741 --> 00:42:28.776 align:center line:79%
Okay, let's break it down.
Let's reason about it.

00:42:30.144 --> 00:42:32.213 align:center line:79%
-In public appropriately.
-In almost everything

00:42:32.346 --> 00:42:34.816 align:center line:79%
every question I get, let's break it down,
let's reason about it.

00:42:35.116 --> 00:42:36.784 align:center line:81%
So let's start with an example.

00:42:37.518 --> 00:42:42.090 align:center line:81%
In 2012, AlexNet was computer vision

00:42:42.223 --> 00:42:47.161 align:center line:79%
ImageNet, image recognition, 82%
or something like that, accuracy.

00:42:48.029 --> 00:42:49.097 align:center line:81%
Over the next...

00:42:50.698 --> 00:42:53.634 align:center line:79%
almost not quite 10 years,
I think it was like 7 years,

00:42:54.302 --> 00:42:58.806 align:center line:79%
every single year,
the accuracy error reduced in half.

00:42:59.774 --> 00:43:04.512 align:center line:79%
Every year, the error reduced in half,
or otherwise known as Moore's Law.

00:43:04.645 --> 00:43:05.980 align:center line:81%
Okay?

00:43:06.114 --> 00:43:09.016 align:center line:79%
So you doubled the performance,
you double the accuracy,

00:43:10.718 --> 00:43:14.021 align:center line:79%
and you double
its believability every single year.

00:43:14.522 --> 00:43:17.058 align:center line:79%
Over the course
of seven years, it's now superhuman.

00:43:17.592 --> 00:43:19.227 align:center line:81%
Same thing with speech recognition.

00:43:19.360 --> 00:43:22.797 align:center line:79%
Same things
with natural language understanding.

00:43:23.364 --> 00:43:27.301 align:center line:79%
We want to know,
we want to believe, not know.

00:43:27.401 --> 00:43:31.873 align:center line:79%
We want to believe that the answer
that's being predicted to us is accurate.

00:43:32.006 --> 00:43:33.141 align:center line:81%
We want to believe that.

00:43:33.274 --> 00:43:35.910 align:center line:81%
And so the industry is going to chase

00:43:36.978 --> 00:43:39.147 align:center line:81%
that believability or that accuracy

00:43:40.214 --> 00:43:43.518 align:center line:81%
and double its accuracy

00:43:45.419 --> 00:43:46.654 align:center line:81%
2x every year.

00:43:46.787 --> 00:43:48.556 align:center line:79%
I believe that's going
to be the same thing

00:43:48.656 --> 00:43:50.525 align:center line:81%
with natural language understanding.

00:43:50.791 --> 00:43:54.562 align:center line:79%
And, of course, the problem space is
a lot more complicated.

00:43:55.163 --> 00:43:59.066 align:center line:79%
But I have every certainty
that we're going to double its accuracy

00:43:59.200 --> 00:44:02.637 align:center line:79%
every single year
to the point where it is so accurate.

00:44:02.770 --> 00:44:04.172 align:center line:81%
We've largely tested

00:44:04.305 --> 00:44:06.207 align:center line:79%
across many of your examples
when you interact with it

00:44:06.340 --> 00:44:08.910 align:center line:79%
that you go, “You know what?
This is really, really good.

00:44:09.043 --> 00:44:11.312 align:center line:79%
I believe the answer
that it's producing for me.”

00:44:11.445 --> 00:44:13.247 align:center line:81%
That condition is very important.

00:44:13.381 --> 00:44:14.615 align:center line:81%
The second thing is this:

00:44:18.452 --> 00:44:20.388 align:center line:81%
today's language models, today's AI

00:44:20.521 --> 00:44:22.757 align:center line:79%
and everything
that we've shown are one shot.

00:44:23.858 --> 00:44:27.328 align:center line:79%
And yet, you and I both know that
there are many things that we think about

00:44:27.461 --> 00:44:28.462 align:center line:81%
that are not one shot.

00:44:28.596 --> 00:44:30.231 align:center line:81%
You have to iterate.

00:44:30.364 --> 00:44:33.467 align:center line:79%
So how do you come up,
how do you reason about a plan?

00:44:33.601 --> 00:44:37.505 align:center line:79%
How do you come up
with a strategy to solve a problem?

00:44:38.039 --> 00:44:39.073 align:center line:81%
Maybe you need to use tools.

00:44:39.207 --> 00:44:41.475 align:center line:79%
Maybe you have
to look up some proprietary data.

00:44:41.609 --> 00:44:44.512 align:center line:79%
Maybe you have to do
some research, in fact.

00:44:44.645 --> 00:44:46.214 align:center line:81%
Maybe you have to ask another agent.

00:44:46.347 --> 00:44:48.149 align:center line:81%
Maybe you have another, ask another AI.

00:44:48.282 --> 00:44:51.018 align:center line:79%
Maybe you have to be
a human in the loop, ask a human.

00:44:51.118 --> 00:44:53.120 align:center line:79%
Trigger an event,
send an email to somebody

00:44:53.254 --> 00:44:54.956 align:center line:81%
or text to somebody, get a response

00:44:55.089 --> 00:44:58.259 align:center line:79%
before you can move on
to the next step of that plan.

00:44:58.392 --> 00:45:00.928 align:center line:79%
And so,
a large language model has to iterate

00:45:01.062 --> 00:45:02.597 align:center line:81%
and think of a plan.

00:45:03.030 --> 00:45:04.865 align:center line:81%
That's not a one-shot thing.

00:45:04.999 --> 00:45:08.402 align:center line:79%
And once it comes up with a plan,
as it traverses that graph

00:45:08.836 --> 00:45:10.738 align:center line:81%
there's a whole bunch of language models

00:45:10.838 --> 00:45:13.407 align:center line:79%
that are going
to get instantiated and initiated.

00:45:13.541 --> 00:45:17.645 align:center line:79%
So I think your future models
are going to iterate.

00:45:18.246 --> 00:45:23.251 align:center line:81%
So instead of a one-shot model,

00:45:23.417 --> 00:45:24.919 align:center line:81%
it's going to be a planning model

00:45:25.019 --> 00:45:26.754 align:center line:79%
with a whole bunch
of other models around it

00:45:26.887 --> 00:45:29.023 align:center line:79%
that are
particularly good at particular skills.

00:45:29.156 --> 00:45:31.158 align:center line:81%
And so, I think we have long ways to go.

00:45:32.426 --> 00:45:36.063 align:center line:79%
Meta garnered a lot of attention last week
for the release of Llama 3,

00:45:36.163 --> 00:45:37.898 align:center line:79%
which seems to be
the most impressive open-source model

00:45:37.999 --> 00:45:39.033 align:center line:81%
thus far.

00:45:39.166 --> 00:45:41.135 align:center line:81%
Any thoughts on open-source models?

00:45:42.303 --> 00:45:47.308 align:center line:79%
If you ask me what are
the top most important events

00:45:47.608 --> 00:45:48.643 align:center line:81%
in the last couple of years,

00:45:48.776 --> 00:45:50.711 align:center line:81%
I would tell you, of course, ChatGPT

00:45:51.045 --> 00:45:54.348 align:center line:79%
reinforcement learning, human feedback,
grounding into human values

00:45:54.482 --> 00:45:56.617 align:center line:79%
and having the technology
necessary to do that

00:45:57.084 --> 00:46:02.690 align:center line:79%
obviously a breakthrough
and democratized computing.

00:46:03.491 --> 00:46:05.793 align:center line:79%
It made it possible
for everybody to be a programmer.

00:46:05.926 --> 00:46:08.362 align:center line:79%
Everybody's now doing
amazing things with it.

00:46:08.496 --> 00:46:11.632 align:center line:81%
ChatGPT, the work that OpenAI did.

00:46:12.199 --> 00:46:14.402 align:center line:79%
Greg and Sam
and the team, really proud of them.

00:46:15.269 --> 00:46:17.405 align:center line:81%
The second thing that I would say,

00:46:17.538 --> 00:46:19.373 align:center line:81%
that it that is

00:46:19.907 --> 00:46:22.943 align:center line:81%
just as important, I would say is Llama,

00:46:23.077 --> 00:46:25.046 align:center line:81%
not Llama 1, but Llama 2.

00:46:25.846 --> 00:46:27.148 align:center line:81%
Llama 2

00:46:28.249 --> 00:46:34.288 align:center line:79%
activated just about every industry 
to jump into working on generative AI.

00:46:35.222 --> 00:46:37.892 align:center line:81%
And it opened the floodgates

00:46:38.025 --> 00:46:41.329 align:center line:79%
of every industry being able
to access this technology.

00:46:41.462 --> 00:46:44.198 align:center line:79%
Health care,
financial services,  you name it

00:46:44.332 --> 00:46:46.667 align:center line:79%
manufacturing,
you name it, customer service

00:46:46.801 --> 00:46:48.903 align:center line:81%
retail, all kinds.

00:46:49.036 --> 00:46:51.339 align:center line:81%
I think Llama 2 and Llama 3,

00:46:52.073 --> 00:46:55.042 align:center line:79%
because it's open sourced,
it engaged research

00:46:55.176 --> 00:46:57.211 align:center line:81%
and engaged startups, engaged industry.

00:46:57.345 --> 00:46:59.613 align:center line:81%
It made generative AI accessible.

00:47:00.147 --> 00:47:02.116 align:center line:81%
I think that's a very big deal.

00:47:02.249 --> 00:47:05.286 align:center line:79%
And so, I think ChatGPT
democratized computing.

00:47:06.087 --> 00:47:09.557 align:center line:79%
I think Llama democratized generative AI.
Does that make sense?

00:47:09.724 --> 00:47:13.427 align:center line:79%
And I think without it,
it's very hard to have activated

00:47:13.527 --> 00:47:15.696 align:center line:81%
all of the research on safety

00:47:15.830 --> 00:47:18.833 align:center line:79%
and all of the different ways 
of chains of thoughts

00:47:18.966 --> 00:47:22.303 align:center line:79%
and all the reasoning technology
that's now being developed

00:47:22.737 --> 00:47:24.505 align:center line:81%
and all the reinforcement learning stuff.

00:47:26.040 --> 00:47:29.110 align:center line:79%
That stuff would've been very hard
to have activated without Llama.

00:47:29.877 --> 00:47:34.515 align:center line:79%
Dario Amodei was
on Ezra Klein's podcast two weeks ago

00:47:34.648 --> 00:47:39.553 align:center line:79%
and he, as many others have,
many others in particular who are...

00:47:40.921 --> 00:47:44.525 align:center line:79%
involved
with Frontier Labs, was predicting

00:47:44.658 --> 00:47:48.429 align:center line:79%
AGI in the relatively near term,
conceivably the next couple of years,

00:47:48.562 --> 00:47:52.133 align:center line:79%
years like 2027
and so on are frequently thrown around.

00:47:52.933 --> 00:47:53.934 align:center line:81%
Thoughts?

00:47:54.702 --> 00:47:57.271 align:center line:81%
Depending on how you define AGI.

00:47:57.972 --> 00:47:59.673 align:center line:81%
Now, first of all, as an engineer

00:48:00.741 --> 00:48:04.979 align:center line:81%
you know that we can only solve a problem

00:48:05.112 --> 00:48:06.881 align:center line:81%
ultimately if you can measure it.

00:48:08.282 --> 00:48:10.985 align:center line:79%
And so, you have to express
the problem statement,

00:48:11.118 --> 00:48:15.222 align:center line:79%
the mission somehow
in some measurable way.

00:48:15.356 --> 00:48:20.428 align:center line:79%
If you told me that AGI is
the list of benchmarks we currently use,

00:48:20.561 --> 00:48:23.864 align:center line:79%
they're math tests
and English comprehension tests

00:48:23.998 --> 00:48:26.434 align:center line:81%
and reasoning tests, and you know.

00:48:27.868 --> 00:48:30.237 align:center line:81%
You got medical exams and bars.

00:48:30.538 --> 00:48:33.207 align:center line:79%
You make your list
of all of the tests that you want.

00:48:33.340 --> 00:48:35.643 align:center line:79%
It doesn't matter what it is.
Just make your list.

00:48:36.310 --> 00:48:37.878 align:center line:81%
If you make your list

00:48:38.012 --> 00:48:41.282 align:center line:79%
I am certain
we will achieve excellent results

00:48:41.415 --> 00:48:44.051 align:center line:81%
in a very nominal amount of time.

00:48:44.185 --> 00:48:46.153 align:center line:81%
And if that's the definition of AGI,

00:48:46.387 --> 00:48:48.289 align:center line:81%
I'll make a guess

00:48:48.389 --> 00:48:51.358 align:center line:79%
it's probably, definitely,
within the next five years.

00:48:51.492 --> 00:48:55.763 align:center line:79%
So all of the tests that
we currently measure these models with

00:48:56.397 --> 00:49:01.402 align:center line:79%
their accuracy
or their error rate is reducing

00:49:01.502 --> 00:49:03.537 align:center line:81%
in half every six months.

00:49:03.771 --> 00:49:06.774 align:center line:79%
So there's no reason
why we shouldn't expect it

00:49:06.907 --> 00:49:08.776 align:center line:81%
all to be superhuman pretty soon.

00:49:08.909 --> 00:49:10.244 align:center line:81%
So again, everyone in this audience...

00:49:10.344 --> 00:49:12.813 align:center line:81%
But that doesn't meet the standard.

00:49:12.947 --> 00:49:15.015 align:center line:79%
Just be clear.
That doesn't meet the standard

00:49:15.115 --> 00:49:17.384 align:center line:81%
of a normal person thinking it's AGI.

00:49:17.518 --> 00:49:18.686 align:center line:81%
Does that make sense?

00:49:18.819 --> 00:49:24.825 align:center line:79%
A on-the-street person, hey, AGI,
that's probably not what they're thinking

00:49:24.959 --> 00:49:26.794 align:center line:81%
what I defined it as.

00:49:27.194 --> 00:49:29.797 align:center line:79%
The way I defined it is
simply an engineering way

00:49:29.930 --> 00:49:32.099 align:center line:79%
of defining it
so that you can answer that question.

00:49:32.233 --> 00:49:33.834 align:center line:79%
The second way
of answering the question is

00:49:33.968 --> 00:49:37.404 align:center line:79%
when can you achieve
AGI in an undefined way?

00:49:38.239 --> 00:49:41.175 align:center line:79%
If it's undefinable, 
then how long would you know...

00:49:41.308 --> 00:49:43.143 align:center line:81%
How long would it take? Undefinable.

00:49:43.711 --> 00:49:46.113 align:center line:79%
So everyone in this audience,
again, runs a business.

00:49:46.247 --> 00:49:50.451 align:center line:79%
And so, a practical question
they/we all face is

00:49:52.219 --> 00:49:54.655 align:center line:81%
how do you know if you are

00:49:54.989 --> 00:49:57.691 align:center line:79%
in the face of the kinds
of changes you just depicted,

00:49:58.259 --> 00:50:03.097 align:center line:79%
how does one know, how can one know
whether one is responding appropriately

00:50:03.197 --> 00:50:06.033 align:center line:81%
sufficiently in the right ways, etcetera?

00:50:06.800 --> 00:50:08.068 align:center line:81%
Any advice?

00:50:08.202 --> 00:50:11.972 align:center line:81%
If you're not engaging AI actively

00:50:12.106 --> 00:50:14.375 align:center line:81%
and aggressively, you're doing it wrong.

00:50:15.643 --> 00:50:17.678 align:center line:81%
You're not going to lose your job to AI.

00:50:17.811 --> 00:50:20.047 align:center line:79%
You're going to lose your job
to somebody who uses AI.

00:50:21.081 --> 00:50:23.784 align:center line:79%
Your company is not going
to go out of business because of AI.

00:50:23.884 --> 00:50:25.452 align:center line:79%
Your company is going
to go out of business

00:50:25.586 --> 00:50:27.521 align:center line:81%
because another company used AI.

00:50:28.155 --> 00:50:29.790 align:center line:81%
There's no question about that.

00:50:30.190 --> 00:50:32.726 align:center line:79%
And so, you have to engage
AI as quickly as possible.

00:50:32.860 --> 00:50:34.328 align:center line:79%
You have to engage AI
as quickly as possible

00:50:34.461 --> 00:50:35.462 align:center line:81%
so that you could do

00:50:35.829 --> 00:50:39.033 align:center line:81%
things that you think cost too much to do.

00:50:39.133 --> 00:50:42.670 align:center line:79%
For example,
if the marginal cost of intelligence was

00:50:44.004 --> 00:50:46.173 align:center line:79%
practically zero,
there are a lot of things

00:50:46.273 --> 00:50:48.075 align:center line:81%
that you would do now

00:50:48.208 --> 00:50:49.677 align:center line:81%
that you wouldn't have done otherwise.

00:50:49.810 --> 00:50:52.513 align:center line:81%
And so, notice how often we do search

00:50:52.646 --> 00:50:56.650 align:center line:79%
and these days, notice
how often we ask questions.

00:50:57.451 --> 00:51:00.354 align:center line:79%
I mean, any random question
I'll be asking Perplexity, right?

00:51:01.121 --> 00:51:03.324 align:center line:81%
And so, why not? Just in case.

00:51:03.424 --> 00:51:05.125 align:center line:81%
Aravind just gave a talk here at Sessions.

00:51:05.259 --> 00:51:08.028 align:center line:81%
Okay. I love using it.

00:51:08.162 --> 00:51:10.631 align:center line:79%
And even if I know
the answer, I'll just ask it anyways.

00:51:12.066 --> 00:51:13.968 align:center line:79%
You know, just to see
what it comes up with.

00:51:14.768 --> 00:51:18.372 align:center line:81%
And so I think we want that to happen.

00:51:18.505 --> 00:51:21.342 align:center line:79%
We want the marginal cost
of these types of activities to be

00:51:21.475 --> 00:51:24.311 align:center line:79%
as low as possible
so that you use it in abundance.

00:51:24.411 --> 00:51:27.681 align:center line:79%
Second,
if you could use AI to be productive

00:51:29.950 --> 00:51:33.087 align:center line:81%
you know that productive companies leads

00:51:33.220 --> 00:51:34.321 align:center line:81%
to higher earnings.

00:51:34.455 --> 00:51:36.523 align:center line:81%
Higher earnings leads to more employment.

00:51:37.791 --> 00:51:39.793 align:center line:79%
More employment leads
to more social growth.

00:51:40.060 --> 00:51:43.063 align:center line:79%
So there is
a lot of reasons to want to drive

00:51:43.197 --> 00:51:44.598 align:center line:81%
productivity into companies.

00:51:44.965 --> 00:51:48.469 align:center line:79%
And apart from just changing
your manufacturing plans

00:51:48.602 --> 00:51:49.737 align:center line:81%
and your CapEx plans,

00:51:49.870 --> 00:51:52.973 align:center line:79%
how has AI changed
how NVIDIA works internally?

00:51:54.174 --> 00:51:56.477 align:center line:79%
We were one of the first
technology companies to invest

00:51:56.577 --> 00:51:58.679 align:center line:81%
in our own AI supercomputers.

00:52:00.014 --> 00:52:02.416 align:center line:81%
We can't design a chip anymore without AI.

00:52:03.584 --> 00:52:07.821 align:center line:79%
At night,
our AIs are exploring design spaces

00:52:08.422 --> 00:52:10.824 align:center line:79%
vast and wide 
that we would never do ourselves

00:52:10.958 --> 00:52:12.626 align:center line:79%
because it costs
too much money to explore it.

00:52:12.726 --> 00:52:15.696 align:center line:81%
And so we... I...

00:52:16.163 --> 00:52:18.165 align:center line:81%
Our chips are so much better.

00:52:18.666 --> 00:52:20.968 align:center line:79%
Because of an AI,
we could reduce the amount

00:52:21.068 --> 00:52:23.370 align:center line:79%
of energy used
for our chips as higher performance.

00:52:24.338 --> 00:52:28.475 align:center line:79%
Our software, we can't write
software without AI anymore.

00:52:28.609 --> 00:52:29.977 align:center line:81%
We have to explore all the...

00:52:30.444 --> 00:52:34.014 align:center line:79%
The design space
of optimizing compilers is too large.

00:52:34.748 --> 00:52:38.519 align:center line:81%
We use AIs to file bugs.

00:52:38.652 --> 00:52:42.056 align:center line:81%
So our bugs database actually tells you

00:52:42.156 --> 00:52:44.558 align:center line:81%
who's... what's wrong with the code

00:52:44.658 --> 00:52:46.326 align:center line:81%
who's likely involved

00:52:46.560 --> 00:52:49.029 align:center line:79%
and activates that person
to go fix it, you know?

00:52:50.631 --> 00:52:53.667 align:center line:81%
And so, I think I want everybody,

00:52:53.801 --> 00:52:56.370 align:center line:79%
every organization in our company
to use AI very aggressively.

00:52:56.670 --> 00:52:59.073 align:center line:81%
I want to turn NVIDIA into one giant AI.

00:52:59.306 --> 00:53:00.874 align:center line:81%
How great would that be?

00:53:01.208 --> 00:53:02.710 align:center line:81%
And then, I'll have work-life balance.

00:53:02.810 --> 00:53:05.179 align:center line:81%
(Audience laughs.)

00:53:05.279 --> 00:53:08.816 align:center line:79%
Are there any favorite examples
you've heard of businesses

00:53:08.949 --> 00:53:11.418 align:center line:79%
and maybe
in some kind of unexpected sector

00:53:11.552 --> 00:53:13.253 align:center line:81%
or some unexpected use case

00:53:13.387 --> 00:53:17.224 align:center line:79%
where you feel they kind of can serve
as a poster child for some of the dynamics

00:53:17.357 --> 00:53:20.527 align:center line:79%
you're describing
where they've really realized

00:53:20.661 --> 00:53:22.329 align:center line:81%
some of this opportunity?

00:53:23.831 --> 00:53:25.799 align:center line:81%
Well, the biggest surprise of AI

00:53:25.933 --> 00:53:27.267 align:center line:81%
that shouldn't be a surprise

00:53:28.135 --> 00:53:30.003 align:center line:81%
for a lot of people is

00:53:31.872 --> 00:53:35.075 align:center line:79%
that when we say,
“It's a large language model,”

00:53:35.209 --> 00:53:38.846 align:center line:79%
the word language doesn't mean
human language only.

00:53:38.979 --> 00:53:41.782 align:center line:79%
And it doesn't mean
English only or French only

00:53:42.816 --> 00:53:46.854 align:center line:79%
or Irish only,
that's a whole different language but...

00:53:48.188 --> 00:53:50.190 align:center line:81%
Is there a large language model for Irish?

00:53:50.657 --> 00:53:52.493 align:center line:79%
-I've tried it.
-That works?

00:53:53.293 --> 00:53:54.795 align:center line:81%
Yeah. It works well.

00:53:56.130 --> 00:53:59.867 align:center line:81%
John and I spent most of our education

00:54:00.000 --> 00:54:01.769 align:center line:81%
in Ireland being taught in Irish.

00:54:01.902 --> 00:54:06.573 align:center line:79%
So these models are
some of the first people

00:54:06.740 --> 00:54:08.876 align:center line:79%
I've had the chance
to have a dialogue with

00:54:08.976 --> 00:54:11.712 align:center line:81%
as Gaeilge, in...

00:54:12.412 --> 00:54:14.014 align:center line:79%
-Very surprising.
-Many years.

00:54:14.681 --> 00:54:17.351 align:center line:79%
And actually I've been enjoying,
have you played with Suno?

00:54:17.684 --> 00:54:18.786 align:center line:79%
-Suno?
-Suno.

00:54:18.886 --> 00:54:23.524 align:center line:81%
Suno is an app for creating music.

00:54:24.024 --> 00:54:26.260 align:center line:79%
-Synthetic music.
-Okay.

00:54:26.727 --> 00:54:29.062 align:center line:79%
-And I've been enjoying creating...
-Irish music.

00:54:29.229 --> 00:54:31.198 align:center line:81%
I, of course, tested it on that.

00:54:31.331 --> 00:54:34.067 align:center line:79%
And Celtic dubstep is
a thing that it can do.

00:54:35.469 --> 00:54:37.171 align:center line:81%
Fantastic. Okay. Makes sense.

00:54:37.437 --> 00:54:39.106 align:center line:81%
Like, if it could do that,

00:54:39.773 --> 00:54:42.442 align:center line:79%
then of course
they could learn the language of life.

00:54:43.577 --> 00:54:44.978 align:center line:81%
Of course, they could learn

00:54:45.112 --> 00:54:48.382 align:center line:79%
and if a language model
can understand sound,

00:54:48.515 --> 00:54:51.919 align:center line:79%
which is
a sequence time series, it's a sequence

00:54:52.786 --> 00:54:55.956 align:center line:81%
why can't it learn robotics articulation,

00:54:56.089 --> 00:54:57.124 align:center line:81%
which is a sequence?

00:54:57.257 --> 00:55:00.160 align:center line:79%
You just have
to figure out how to tokenize it.

00:55:00.594 --> 00:55:02.930 align:center line:81%
So the idea that all of a sudden

00:55:03.030 --> 00:55:05.666 align:center line:79%
“Oh hey, look, listen,
I could also learn SQL.

00:55:05.766 --> 00:55:08.001 align:center line:79%
I could learn ABAP,
I could learn Lightning.

00:55:08.101 --> 00:55:10.003 align:center line:79%
I could learn
all these proprietary languages.

00:55:10.103 --> 00:55:11.939 align:center line:79%
I could learn Verilog,”
I could learn, right?

00:55:12.072 --> 00:55:14.975 align:center line:79%
So, all of a sudden,
you realized, hang on a second

00:55:15.108 --> 00:55:18.378 align:center line:79%
I can put a Copilot
on top of every tool on the planet.

00:55:18.512 --> 00:55:21.481 align:center line:79%
Well, and to this point,
and you know, NVIDIA being one big AI

00:55:22.349 --> 00:55:24.551 align:center line:81%
is the future one of 100,000 models

00:55:24.685 --> 00:55:25.986 align:center line:81%
or 100 million models

00:55:26.119 --> 00:55:28.055 align:center line:81%
or is the future one of one model

00:55:28.155 --> 00:55:30.090 align:center line:79%
and there's just like a model
that does all the things.

00:55:30.424 --> 00:55:33.260 align:center line:81%
I think that it would be great to have...

00:55:35.062 --> 00:55:37.464 align:center line:81%
It would be great to have supermodels

00:55:37.798 --> 00:55:40.601 align:center line:79%
that help you reason
about things in general, but...

00:55:41.835 --> 00:55:44.137 align:center line:81%
For us, for all companies

00:55:44.238 --> 00:55:49.243 align:center line:79%
that have very specific,
domain-specific expertise,

00:55:49.376 --> 00:55:51.945 align:center line:79%
we're going
to have to train our own models.

00:55:52.079 --> 00:55:54.748 align:center line:79%
And the reason for that is
because we have a proprietary language.

00:55:54.882 --> 00:55:58.752 align:center line:81%
That difference between 99% and 99.3% is

00:55:58.886 --> 00:56:00.821 align:center line:79%
the difference
between life and death for us.

00:56:01.488 --> 00:56:04.958 align:center line:79%
So it's too valuable to us.
No different than fraud detection for you.

00:56:05.058 --> 00:56:06.426 align:center line:79%
-I was going to say...
-It's too important to you.

00:56:06.526 --> 00:56:09.429 align:center line:79%
-That's been exactly our experience.
-Yeah. It's too important to you.

00:56:09.563 --> 00:56:12.566 align:center line:81%
However good the general model is,

00:56:12.699 --> 00:56:14.568 align:center line:79%
you're going to want
to take that and fine-tune it

00:56:14.701 --> 00:56:19.006 align:center line:79%
and improve it into perfection
because it's just too important to you.

00:56:19.373 --> 00:56:22.109 align:center line:79%
So we're going
to shortly run out of time here

00:56:22.242 --> 00:56:24.611 align:center line:79%
and there's a whole bunch of questions
I haven't gotten to yet.

00:56:24.711 --> 00:56:28.548 align:center line:79%
I've exercised poor discipline
on the time management front.

00:56:28.682 --> 00:56:30.884 align:center line:81%
So there's a bunch that I think are...

00:56:31.018 --> 00:56:33.153 align:center line:81%
I was told I definitely had to ask you

00:56:33.453 --> 00:56:35.689 align:center line:79%
but there's a couple
that I really wanted to ask

00:56:35.889 --> 00:56:37.658 align:center line:81%
and it's only us up here, so.

00:56:40.193 --> 00:56:42.496 align:center line:79%
Lisa Su is
your first cousin once removed?

00:56:42.629 --> 00:56:44.564 align:center line:81%
Yes. She's terrific. She's amazing.

00:56:44.665 --> 00:56:48.302 align:center line:79%
-And then, AMD is now...
-She's the CEO of AMD, by the way.

00:56:48.402 --> 00:56:51.972 align:center line:79%
Yeah. And AMD is now one
of your competitors in the GPU space?

00:56:52.572 --> 00:56:53.640 align:center line:81%
No.

00:56:53.774 --> 00:56:55.943 align:center line:81%
(Audience laughs.)

00:56:56.076 --> 00:56:57.477 align:center line:81%
We're family.

00:56:57.611 --> 00:57:00.480 align:center line:79%
-Okay.
-We're all in the industry.

00:57:00.614 --> 00:57:02.616 align:center line:79%
-We're in the industry.
-One of your partners in the industry.

00:57:02.716 --> 00:57:04.785 align:center line:81%
Yeah. Yeah. We buy from AMD.

00:57:04.918 --> 00:57:06.320 align:center line:81%
What's going on in the water?

00:57:06.420 --> 00:57:08.522 align:center line:81%
How did we end up with two of the...

00:57:09.122 --> 00:57:12.359 align:center line:79%
arguably the two most important
GPU companies

00:57:12.759 --> 00:57:14.294 align:center line:81%
being run by close relatives?

00:57:14.428 --> 00:57:15.429 align:center line:81%
What's going on?

00:57:15.862 --> 00:57:17.197 align:center line:81%
(Audience laughs.)

00:57:17.764 --> 00:57:19.766 align:center line:81%
You got to keep it close to the family.

00:57:20.667 --> 00:57:23.170 align:center line:79%
No, I just,
I have no idea how it happened.

00:57:23.403 --> 00:57:26.506 align:center line:79%
We didn't grow up together
and we didn't know each other.

00:57:26.640 --> 00:57:27.808 align:center line:79%
That makes it
even more interesting, right?

00:57:27.941 --> 00:57:30.677 align:center line:79%
Yeah. We didn't even know each other
until she was at IBM.

00:57:31.044 --> 00:57:33.680 align:center line:81%
And her career is incredible.

00:57:33.814 --> 00:57:36.116 align:center line:81%
She's really quite extraordinary, yeah.

00:57:38.185 --> 00:57:40.520 align:center line:79%
I think 
this question requires further study.

00:57:40.620 --> 00:57:42.322 align:center line:79%
-Yeah.
-So.

00:57:43.023 --> 00:57:44.524 align:center line:79%
Okay,
you've been operating in Silicon Valley

00:57:44.624 --> 00:57:46.626 align:center line:79%
since the early '90s.
-Yes.

00:57:47.561 --> 00:57:49.863 align:center line:79%
How has Silicon Valley culture
changed in that time?

00:57:52.065 --> 00:57:53.066 align:center line:81%
Oh, wow.

00:57:55.402 --> 00:57:57.170 align:center line:79%
I haven't thought
about this in a long time.

00:57:57.270 --> 00:58:00.107 align:center line:79%
I guess in a lot of ways,
in a lot of ways, probably...

00:58:01.942 --> 00:58:03.443 align:center line:81%
Okay, here's one.

00:58:04.745 --> 00:58:07.147 align:center line:81%
When I first started NVIDIA

00:58:08.782 --> 00:58:11.118 align:center line:81%
I was 29 years old and...

00:58:13.620 --> 00:58:16.390 align:center line:81%
I was 29 years old with acne.

00:58:17.224 --> 00:58:18.692 align:center line:81%
And...

00:58:19.793 --> 00:58:22.929 align:center line:81%
you go talk to your... go recruit

00:58:23.230 --> 00:58:25.665 align:center line:81%
law firms and VCs

00:58:25.799 --> 00:58:28.368 align:center line:79%
and you know,
I got a big zit on my forehead.

00:58:29.002 --> 00:58:30.937 align:center line:81%
I don't have one today

00:58:31.071 --> 00:58:33.173 align:center line:81%
so I feel comfortable talking about it.

00:58:33.306 --> 00:58:35.108 align:center line:81%
But it could happen.

00:58:35.242 --> 00:58:36.309 align:center line:81%
And so anyways

00:58:37.377 --> 00:58:41.848 align:center line:79%
you feel rather insecure,
because most of CEOs back then wore suits

00:58:41.948 --> 00:58:45.352 align:center line:79%
and they're quite accomplished,
and they sound like adults

00:58:45.485 --> 00:58:46.620 align:center line:81%
and they use big words

00:58:46.753 --> 00:58:49.790 align:center line:79%
and they talk
about business and things like that.

00:58:50.190 --> 00:58:53.527 align:center line:79%
So when you're young,
you feel rather intimidated.

00:58:53.660 --> 00:58:55.462 align:center line:81%
You're surrounded by a bunch of adults.

00:58:56.129 --> 00:58:59.833 align:center line:79%
Well, you know,
now, if you don't have acne,

00:59:00.867 --> 00:59:02.502 align:center line:79%
I don't think you deserve
to start a company.

00:59:02.969 --> 00:59:05.238 align:center line:81%
(Audience laughs.)

00:59:06.206 --> 00:59:09.409 align:center line:81%
I just, that's one big difference.

00:59:09.843 --> 00:59:11.244 align:center line:81%
Acne.

00:59:11.344 --> 00:59:14.147 align:center line:79%
You know, the takeaway
from Jensen's speech.

00:59:14.281 --> 00:59:16.216 align:center line:81%
I just, what it means is really...

00:59:16.349 --> 00:59:19.886 align:center line:79%
we've enabled
younger people to be extraordinary.

00:59:20.454 --> 00:59:22.889 align:center line:81%
I think that the young generation of CEOs

00:59:23.123 --> 00:59:24.958 align:center line:81%
the type of things that you guys know

00:59:25.092 --> 00:59:27.627 align:center line:79%
at such a young age is
really quite extraordinary.

00:59:27.761 --> 00:59:30.530 align:center line:79%
I mean, it took me
decades to learn it, and so.

00:59:30.864 --> 00:59:31.865 align:center line:81%
Last question.

00:59:33.066 --> 00:59:36.036 align:center line:79%
That was a compliment.
See how he quickly, he changed it?

00:59:38.004 --> 00:59:40.173 align:center line:79%
I wasn't saying you have acne.
I was just say you were smart.

00:59:40.273 --> 00:59:41.908 align:center line:81%
(Audience laughs.)

00:59:43.210 --> 00:59:46.513 align:center line:79%
NVIDIA has a market cap
of roughly $2 trillion dollars

00:59:46.646 --> 00:59:48.949 align:center line:81%
and you're...

00:59:49.850 --> 00:59:51.618 align:center line:81%
and you're now within spitting distance

00:59:51.751 --> 00:59:53.620 align:center line:81%
of Apple and Microsoft.

00:59:53.753 --> 00:59:58.125 align:center line:81%
And I just checked, and they have 220,000

00:59:58.225 --> 01:00:01.628 align:center line:81%
and 160,000 employees respectively.

01:00:03.430 --> 01:00:06.099 align:center line:81%
NVIDIA has 28,000 employees.

01:00:06.199 --> 01:00:08.768 align:center line:81%
So, you know, less than a fifth

01:00:08.902 --> 01:00:10.637 align:center line:81%
of the smaller of the two there.

01:00:12.472 --> 01:00:15.775 align:center line:79%
And then, you just said,
when we were chatting backstage

01:00:15.909 --> 01:00:18.445 align:center line:81%
and I jotted this down

01:00:19.079 --> 01:00:22.716 align:center line:79%
“You can achieve
operational excellence through process

01:00:22.849 --> 01:00:25.752 align:center line:79%
but craft can only be achieved
with tenure.”

01:00:26.686 --> 01:00:29.389 align:center line:81%
And so, NVIDIA is considerably smaller

01:00:29.523 --> 01:00:32.159 align:center line:81%
than any of the other giants.

01:00:32.292 --> 01:00:34.794 align:center line:79%
And you seem to think
that tenure really matters

01:00:34.928 --> 01:00:36.530 align:center line:81%
and I guess that craft really matters.

01:00:37.164 --> 01:00:38.999 align:center line:81%
Say a little bit more there.

01:00:42.102 --> 01:00:45.739 align:center line:81%
The, I think, extraordinary thing...

01:00:45.872 --> 01:00:48.308 align:center line:79%
I think
a lot of good things could be made.

01:00:48.675 --> 01:00:51.711 align:center line:79%
Good things are made
with operational excellence.

01:00:53.580 --> 01:00:55.916 align:center line:81%
But you can't make extraordinary things

01:00:56.016 --> 01:00:58.084 align:center line:81%
through just operational excellence.

01:00:58.585 --> 01:01:01.788 align:center line:79%
And the reason for that is
because a lot of the great things

01:01:01.922 --> 01:01:04.558 align:center line:81%
in your body of work

01:01:04.691 --> 01:01:07.227 align:center line:79%
and the products that you make,
the company you created,

01:01:07.727 --> 01:01:10.163 align:center line:81%
the organizations you've nurtured,

01:01:10.931 --> 01:01:12.933 align:center line:81%
it takes loving care.

01:01:14.434 --> 01:01:16.069 align:center line:81%
And you can't even put it in words.

01:01:16.203 --> 01:01:18.438 align:center line:81%
How do you put loving care in an email?

01:01:19.005 --> 01:01:21.875 align:center line:79%
And for people to go,
“Oh, I know exactly what to do.”

01:01:22.542 --> 01:01:23.777 align:center line:81%
You can't put that...

01:01:23.910 --> 01:01:27.214 align:center line:79%
you can't put that
in a business process, loving care, and...

01:01:27.514 --> 01:01:30.850 align:center line:79%
Is love and care kind
of an NVIDIA catchphrase?

01:01:31.885 --> 01:01:35.855 align:center line:81%
Well, I use love fairly abundantly

01:01:35.989 --> 01:01:38.058 align:center line:81%
and care, I use abundantly.

01:01:39.025 --> 01:01:41.027 align:center line:79%
At Stripe,
we talk a lot about craft and beauty.

01:01:41.161 --> 01:01:43.063 align:center line:79%
Yeah, right.
You have to use these words

01:01:43.163 --> 01:01:47.567 align:center line:79%
because, that in a lot of ways,
there are no other words to describe it.

01:01:48.134 --> 01:01:49.803 align:center line:81%
You can't put it in numbers.

01:01:49.903 --> 01:01:52.272 align:center line:79%
You can't write it
in the product specification.

01:01:52.405 --> 01:01:55.242 align:center line:79%
The product specification says,
“I want you to build something great

01:01:55.375 --> 01:01:58.511 align:center line:79%
that's incredibly beautiful,
in great, great craft.”

01:01:58.678 --> 01:02:00.647 align:center line:81%
You can't specify these things, and so.

01:02:00.780 --> 01:02:02.549 align:center line:79%
But I'm sure there is
people at Stripe who think,

01:02:02.682 --> 01:02:04.317 align:center line:81%
“Patrick's always yammering on

01:02:04.451 --> 01:02:06.753 align:center line:79%
about craft and beauty,
and it's this kind of...”

01:02:06.987 --> 01:02:08.955 align:center line:79%
I never yammer.
I just want to let you know that.

01:02:09.556 --> 01:02:11.091 align:center line:81%
I don't even know what that sounds like.

01:02:11.725 --> 01:02:12.959 align:center line:79%
-Okay, well, yeah.
-Yammering on.

01:02:13.059 --> 01:02:15.695 align:center line:79%
-Go ahead. Go ahead.
-Yeah, you're more lucid than I am.

01:02:15.795 --> 01:02:18.632 align:center line:79%
I just babble.
But hey, so Patrick is always going on

01:02:18.765 --> 01:02:20.066 align:center line:81%
about this craft and beauty stuff

01:02:20.200 --> 01:02:24.671 align:center line:79%
and wants things to have
this particular ineffable character

01:02:24.804 --> 01:02:27.907 align:center line:79%
but it doesn't directly serve
some customer need and so forth.

01:02:28.041 --> 01:02:29.976 align:center line:79%
Like, customers aren't coming
to us and saying

01:02:30.110 --> 01:02:31.678 align:center line:81%
“I want the product to be more beautiful.”

01:02:31.811 --> 01:02:33.813 align:center line:79%
They're saying,
“I want it to feature X or feature Y.”

01:02:34.414 --> 01:02:37.250 align:center line:79%
And yet, we believe
that the craft and beauty really matters.

01:02:37.384 --> 01:02:39.352 align:center line:79%
It sounds
like you're getting at something similar.

01:02:39.452 --> 01:02:40.720 align:center line:81%
Why do you think it matters?

01:02:41.454 --> 01:02:45.058 align:center line:79%
Actually, your customers,
even though they didn't say it

01:02:45.425 --> 01:02:49.429 align:center line:79%
they might not have the words to say it,
but when they experience it, they know it.

01:02:50.730 --> 01:02:52.098 align:center line:81%
There's no question.

01:02:52.232 --> 01:02:53.800 align:center line:81%
Look, I...

01:02:54.868 --> 01:02:59.005 align:center line:79%
Look, Stripe's work has
beauty, has elegance,

01:02:59.105 --> 01:03:01.107 align:center line:81%
has simplicity.

01:03:01.508 --> 01:03:04.044 align:center line:81%
Simplicity is not simple as you guys know.

01:03:04.177 --> 01:03:06.746 align:center line:79%
Simplicity and simple are not
the same thing.

01:03:06.880 --> 01:03:11.551 align:center line:79%
And it has elegance
and it solves the problem

01:03:11.685 --> 01:03:12.686 align:center line:81%
but just enough.

01:03:13.753 --> 01:03:15.655 align:center line:79%
It burdens you, but not too much.
You know?

01:03:15.755 --> 01:03:18.658 align:center line:79%
And so that...
and that balance is hard to find.

01:03:19.092 --> 01:03:21.328 align:center line:79%
And you can't specify that.
You just feel your way there.

01:03:21.461 --> 01:03:24.731 align:center line:81%
And when you have a team that's with you

01:03:24.864 --> 01:03:28.635 align:center line:79%
that feels the way
they're together, in a lot of ways

01:03:28.768 --> 01:03:31.104 align:center line:81%
we've codified, we've encoded

01:03:32.105 --> 01:03:34.274 align:center line:81%
the magic of the company

01:03:34.741 --> 01:03:37.110 align:center line:81%
in a way that no words can describe.

01:03:37.544 --> 01:03:39.946 align:center line:79%
And you don't want to lose that.
You don't want to lose that.

01:03:40.046 --> 01:03:43.116 align:center line:79%
You want to take that
and take it to the next level next time.

01:03:43.450 --> 01:03:45.685 align:center line:81%
And so I don't want to reset.

01:03:45.819 --> 01:03:49.356 align:center line:79%
I don't like working
with new people for that reason

01:03:50.223 --> 01:03:52.859 align:center line:79%
because I've encoded,
I've embodied, I've deposited

01:03:52.992 --> 01:03:57.597 align:center line:79%
so much pain, suffering,
joy, knowledge, right?

01:03:57.731 --> 01:03:59.032 align:center line:81%
All that experience, life experience

01:03:59.165 --> 01:04:01.501 align:center line:79%
you've encoded it
in, all the people that you've worked with.

01:04:01.634 --> 01:04:04.371 align:center line:79%
You want to carry it on.
You want to take it to the next level.

01:04:04.504 --> 01:04:08.408 align:center line:79%
And that's really the reason
why I really deeply believe in tenure.

01:04:08.975 --> 01:04:13.546 align:center line:79%
And because of that,
small teams could do great things.

01:04:13.680 --> 01:04:16.649 align:center line:79%
And NVIDIA is kind of a small team,
we're 28,000 people.

01:04:16.783 --> 01:04:19.719 align:center line:79%
People think
we punch well above our weight

01:04:19.853 --> 01:04:20.987 align:center line:81%
because of that reason.

01:04:21.121 --> 01:04:24.290 align:center line:79%
And so, it's amazing
what you guys have done

01:04:24.424 --> 01:04:30.096 align:center line:79%
and how incredibly small you are,
7,000 people supporting a trillion dollars

01:04:30.230 --> 01:04:34.000 align:center line:81%
worth of ecosystem and industry and...

01:04:35.068 --> 01:04:37.971 align:center line:79%
and economy, and who knows
how far you guys can go.

01:04:38.104 --> 01:04:40.640 align:center line:79%
-So, I'm very proud of you.
-Jensen, thank you.

01:04:40.740 --> 01:04:42.942 align:center line:81%
(Audience applauds.)

01:04:43.410 --> 01:04:45.979 align:center line:81%
(upbeat music)