00:04Jensen, this is such an honor,
00:06thank you for being here.
00:07>> I'm delighted to be
00:11I decided we'd start talking about
00:13the time when you first left.
00:15You joined LSI Logic, and that was
00:17one of the most exciting companies
00:19at the time, you were building
00:21a phenomenal reputation with some
00:23of the biggest names in tech.
00:24And yet you decided to leave to
00:29>> Chris and Curtis.
00:33I was an engineer at LSI Logic and
00:36Chris and Curtis were at Sun.
00:38And I was working with some of
00:40the brightest minds in computer
00:42science at the time, of all time,
00:44including Andy Bechtolsheim and
00:47others, building workstations, and
00:49graphics workstations, and so
00:54And Chris and Curtis said one day
00:57that they'd like to leave Sun, and
01:00they've like me to go figure out
01:02what they're going to go leave for.
01:07And I had a great job, but they
01:10they insisted that I figure out
01:13with them how to build a company.
01:17And so we hung out at Denny's
01:19whenever they dropped by, which is,
01:23by the way, my alma mater,
01:28My first job before CEO was
01:32I did that very well.
01:34>> [LAUGH] >> [LAUGH] >> And so
01:37anyways, we got together.
01:41the microprocessor revolution,
01:441992 when we were getting together.
01:47The PC revolution was just getting
01:49going, Windows 95, obviously
01:51which is the revolutionary version
01:53of Windows, didn't even come to
01:56Pentium wasn't even announced yet.
01:58And this is all right before the PC
02:01revolution, and it was pretty clear
02:04that the microprocessor was going
02:07to be very important.
02:08And we thought why don't we build
02:11a company to go solve problems that
02:14a normal computer that is powered
02:16by general purpose computing can't?
02:19And so that became the company's
02:21mission, to go build a computer,
02:23the type of computers and
02:25solve problems that normal
02:26computers can't, and
02:28to this day we're focused on that.
02:30And if you look at all the problems
02:33in the markets that we opened up as
02:36it's things like computational drug
02:39design, weather simulation,
02:44These are all things that we're
02:45really, really proud of.
02:46Robotics, self-driving cars,
02:49autonomous software we call
02:51artificial intelligence.
02:54And then of course we drove
02:56the technology so hard that
02:59eventually the computational cost
03:03went to approximately zero.
03:07And it enabled a whole new way of
03:08developing software,
03:09where the computer wrote
03:10the software itself, artificial
03:12intelligence as we know it today.
03:14And so that was the journey.
03:18>> Thank you all for coming.
03:19>> [LAUGH] >> [LAUGH] >> Well these
03:23applications are on all of our
03:26But back then the CEO of LSI Logic
03:28convinced his biggest investor, Don
03:31Valentine, to meet with you, he's
03:33obviously the founder of Sequoia.
03:36Now I can see a lot of
03:37founders here edging forward in
03:40But how did you convince the most
03:42sought-after investor in
03:44Silicon Valley to invest in a team
03:46of first-time founders building
03:50a market that doesn't even exist?
03:52>> I didn't know how
03:53to write a business plan, and so
03:56I went to a bookstore, and
03:58back then there were bookstores.
04:00And in the business book section,
04:03there was this book and
04:05it was written by somebody I knew,
04:10I should go find it again,
04:12but it's a very large book.
04:14And the book says how to write
04:16a business plan, and that was
04:19a highly specific title for
04:21a very niche market, and
04:23it seems like he wrote it for like
04:2514 people and I was one of them.
04:28And so I bought the book, I should
04:31have known right away that that was
04:33a bad idea because Gordon is super,
04:37And super smart people have a lot
04:40to say, and I'm pretty sure Gordon
04:43wants to teach me how to write
04:45a business plan completely.
04:48And so I picked up this book,
04:50it's like 450 pages long,
04:52well I never got through it,
04:55I flipped through it a few
04:57pages then I go you know what,
04:58by the time I'm done reading this
05:00thing I'll be out of business,
05:02I'll be out of money.
05:04And Lori and I only had about six
05:08We had already Spencer, and
05:10Madison, and a dog, and so
05:11the five of us had to live off of
05:13whatever money we had in the bank.
05:15And so I didn't have much time,
05:18and so instead of writing
05:20the business plan I just went to
05:22talk to Wilfred Corrigan.
05:24And he called me one day and
05:25said hey, you left the company, you
05:27didn't even tell me what you were
05:28doing, I want you to come back and
05:30And so I went back and
05:31I explained it to Wilf.
05:32And Wilf at the end of it, he said
05:35I have no idea what you said, and
05:39that's one of the worst elevator
05:42pitches I've ever heard.
05:47then he picked up the phone, and
05:48he called Don Valentine.
05:50And he called Don, and he says Don,
05:53I'm going to send a kid over,
05:55I want you to give him money, he's
05:58one of the best employees LSI Logic
06:01And so the thing I learned is you
06:05can make up a great interview,
06:09you can even have a bad interview.
06:14But you can't run away from
06:16your past, and so have a good past,
06:19try to have a good past.
06:21And in a lot of ways I was serious
06:23when I said I was a good
06:24dishwasher, I was probably Denny's'
06:28>> [LAUGH] >> I planned my work, I
06:30was organized, I was mise en place.
06:33And then I washed the living
06:35daylights out of the dishes.
06:36And then they promoted me to
06:39I was certain I'm the best
06:41busboy Denny's ever had, I never
06:43left the station empty handed,
06:46I never came back empty handed,
06:48I was very efficient.
06:49And so anyways eventually I
06:52became a CEO, I'm still
06:55working on being a good CEO.
06:58>> You talk about being the best,
07:00you needed to be the best among
07:0289 other companies that were funded
07:04after you to build the same thing.
07:07to nine months of runway left,
07:09you realized that the initial
07:11vision was just not going to work.
07:14How did you decide what to do next
07:16to save the company when the cards
07:18were so stacked against you?
07:21>> Well we started this company for
07:23accelerated computing, and
07:25the question is what is it for,
07:27what's the killer app?
07:29And that came our first great
07:33this is what Sequoia funded.
07:38decision was the first killer
07:40app was going to be 3D graphics.
07:42And the technology was going to be
07:443D graphics, and the application
07:46was going to be video games.
07:49At the time 3D graphics was
07:51impossible to make cheap,
07:53it was million dollar image
07:55generators from Silicon Graphics.
07:58And so it was a million dollars And
08:00it's hard to make cheap.
08:02And the video game market was $0
08:06So you had this incredible
08:08technology that's hard to
08:10commoditize and commercialize,
08:12and then you have this market that
08:15That intersection was
08:17the founding of our company.
08:19And I still remember when,
08:21at the end of my presentation,
08:23Don was still kind of, one of
08:25the things he said to me, which
08:28made a lot of sense back then,
08:30it makes a lot of sense today,
08:32he says, startups don't invest in
08:35startups, or startups don't partner
08:39And his point Is that in order for
08:43we needed another startup to
08:46succeed, and that other startup was
08:50And then on the way out,
08:52he reminded me that Electronic
08:55Arts' CTO is 14 years old and had
08:58to be driven to work by his mom.
09:02he just wanted to remind me that
09:03that's who I'm relying on.
09:05>> [LAUGH] >> [LAUGH] And
09:06then after that he said if you lose
09:09my money, I'll kill you.
09:12And that was- >> [LAUGH]
09:13>> That was kind of my memories of
09:18we created something,
09:20we went on the next several years
09:23to go create the market,
09:25create the gaining market for PCs.
09:28And it took a long time to do so,
09:30we're still doing it today.
09:31We realized that not only do you
09:34have to create the technology and
09:36invent a new way of doing computer
09:38graphics so that what was $1
09:40million is now $300, $400,
09:43$500 that fits in a computer.
09:46go create this new market.
09:47So we had to create technology,
09:49The idea that a company would
09:51create technology, create markets,
09:53defines NVIDIA today.
09:55Almost everything we do,
09:56we create technology,
09:58That's the reason why people say,
10:01people call it a stack,
10:02an ecosystem, words like that, but
10:05that's basically it.
10:06At the core, for 30 years,
10:08what Nvidia realized we had to do
10:10is in order to create
10:11the conditions by which somebody
10:13could buy our products, we had
10:15to go invent this new market.
10:17And it's the reason why we're early
10:19in autonomous driving.
10:20It was the reason why we're early
10:22It was a reason why we're early in
10:24just about all these things,
10:26including computational
10:28drug design and discovery,
10:30all these different areas we're
10:32trying to create the market while
10:35we're creating the technology.
10:37And then we got going, and then
10:40Microsoft introduced a standard
10:43called Direct3D, and that spawned
10:46off hundreds of companies.
10:49And we found ourselves,
10:50a couple years later, competing
10:52with just about everybody.
10:53And the thing that we invented,
10:55the company, the technology we
10:57invented 3D graphics with,
10:59consumerized 3D with, turns out to
11:01be incompatible with Direct3D.
11:04So we started this company,
11:05we had this 3D graphics thing,
11:07million dollar thing, we're trying
11:08to make it consumerized, and so
11:10we invented all this technology.
11:11And then shortly after it became
11:15And so we had to reset the company,
11:17or go out of business.
11:19But we didn't know how to build
11:22it the way that Microsoft had
11:26And I remember a meeting on
11:28a weekend, and the conversation
11:32was, we now have 89 competitors.
11:35I understand that the way we do
11:38it is not right, but we don't know
11:41how to do it the right way.
11:46there was another bookstore.
11:50>> [LAUGH] >> And the bookstore
11:52is called Fry's Electronics,
11:55I don't know if it's still here.
11:57And so I think I drove Madison, my
12:00daughter, on the weekend to Fry's,
12:04and it was sitting right there,
12:07the OpenGL manual, which would
12:10defined how Silicon Graphics did
12:15And so it was it was right there,
12:17it was like $68 a book.
12:18And so I had a couple hundred
12:20dollars, I bought three books,
12:21I took it back to the office and
12:23I said, guys, I found it,
12:24our future, and I handed out,
12:26I had three versions of it.
12:28had a big nice centerfold.
12:31The centerfold is the OpenGL
12:33is the computer graphics pipeline.
12:35And I handed it to the same
12:38geniuses that I founded the company
12:43And we implemented the OpenGL
12:45pipeline like nobody had ever
12:47implemented the OpenGL pipeline,
12:49and we built something the world
12:53a lot of lessons are right there.
12:56That moment in time for our company
12:59gave us so much confidence.
13:02And the reason for that is you can
13:04succeed in doing something,
13:07inventing a future, even if you
13:09were not informed about it at all.
13:12And it's kind of my attitude about
13:16When somebody tells me about
13:18something and I've never heard of
13:21it before, or if I've heard of it,
13:23don't understand how it works at
13:26all, my first thought is always,
13:30And it's probably just a textbook
13:33You're probably one archive paper
13:35away from figuring this out.
13:38I spent a lot of time reading
13:40archive papers, and it's true.
13:44Now, of course, you can't learn how
13:46somebody else does something and
13:48do it exactly the same way and
13:49hope to have a different outcome.
13:51But you could learn how something
13:53can be done, and then go back to
13:56first principles and ask yourself,
13:58given the conditions today,
14:00given my motivation,
14:02given the instruments, the tools,
14:04given how things have changed,
14:06how would I redo this?
14:10reinvent this whole thing?
14:12How would I build a car today?
14:14Would I build it incrementally from
14:17How would I build a computer today?
14:19How would I write software today?
14:21Does that make sense?
14:22And so I go back to first
14:24principles all the time,
14:25even in the company today and
14:27just reset ourselves,
14:28because the world has changed.
14:31And the way we wrote software in
14:32the past was monolithic and
14:34it's designed for supercomputers,
14:36but now it's disaggregated, so
14:39And how we think about
14:41how we think about computers today,
14:43just always cause your company,
14:45always cause yourself to go back to
14:47first principles, and it creates
14:49lots and lots of opportunities.
14:50>> Yeah, the way you applied this
14:54You get all the momentum that you
14:56need to IPO and then some more,
14:58because you grow your revenue nine
15:00times in the next four years.
15:02But in the middle of all of
15:05you decide to pivot a little bit
15:07the focus of innovation happening
15:09at NVIDIA based on a phone call you
15:12have with this chemistry professor.
15:14Can you tell us about that phone
15:17how you connected the dots from
15:19what you heard to where you went?
15:22>> I remember at the core of
15:23the company was pioneering a new
15:25way of doing computing.
15:27was the first application.
15:29But we always knew that there would
15:31be other applications.
15:32And so image processing came,
15:34particle physics came, fluids came,
15:37All kinds of interesting things
15:39that we wanted to do.
15:40We made the processor more
15:42programmable so that
15:44we could express more algorithms,
15:49we invented programmable shaders,
15:51which made all forms of imaging and
15:54computer graphics programmable,
15:56that was a great breakthrough, so
15:59On On top of that we tried to
16:02look for ways to express more
16:05sophisticated algorithms that could
16:08be computed on our processor which
16:11is very different than a CPU.
16:14And so we created this thing called
16:17CG, and so I think it was 2003 or
16:21It predated CUDA by about three
16:26The same person who wrote the
16:27textbook that saved the company,
16:29Mark Hilgard, wrote that textbook.
16:32And so CG was super cool,
16:34we wrote textbooks about it.
16:38We started teaching people how to
16:40use it, we developed Tools and
16:42such, and then several
16:43researchers discovered it.
16:46Many of the researchers here,
16:47students here at Stanford,
16:49Many of the engineers that then
16:52became engineers at NVIDIA were
16:58A doctor, a couple of doctors at
17:01Mass General picked it up and
17:03used it for CT reconstruction.
17:06So I flew out and saw them and
17:07said, what are you guys doing with
17:10And they told me about that.
17:11And then a computational
17:15quantum chemist used it to
17:19express his algorithms.
17:23And so I realized that there's some
17:26evidence that people might want to
17:29And it gave it gave us gave us
17:32incrementally more confidence
17:34that we ought to go do this.
17:37That this field, this form of
17:39computing could solve problems that
17:42normal computers really can't and
17:44reinforced our belief and
17:47>> Every time you heard something
17:50new, you really savored that
17:53that seems to be a theme throughout
17:55your leadership at NVIDIA.
17:57It feels like you make these bets
17:59so far in advance of technology
18:01inflections that when the apple
18:02finally falls from the tree, you're
18:05standing right there in your black
18:07leather jacket waiting to catch it.
18:09>> [LAUGH] >> How do you find
18:12>> It always seems like
18:13>> [LAUGH] >> Always does seem
18:15like a diving catch.
18:17You do things based on core
18:20We deeply believe that we
18:22could create a computer that
18:24solves problems, normal processing
18:27can't do and that there
18:29are limits to what a CPU can do.
18:32There are limits to what general
18:33purpose computing can do, and
18:35then there are interesting problems
18:36that we can go solve.
18:39the question is always, are those
18:41interesting problems only, or can
18:43they also be interesting markets?
18:46Because if they're not interesting
18:47markets, it's not sustainable.
18:49And NVIDIA went through about
18:51a decade where we were
18:53investing in this future.
18:56And the markets didn't exist.
18:57There was only one market
18:58at the time, was computer graphics.
19:0115 years the markets that fuels
19:04NVIDIA today just didn't exist.
19:07And so how do you continue with all
19:10of the people around you.
19:14NVIDIA's management team and
19:16all of the amazing engineers.
19:18They're creating this
19:21All of your shareholders,
19:22your board of directors,
19:24all your partners, you're
19:25taking everybody with you and
19:27there's no evidence of a market.
19:30That is really, really challenging.
19:33The fact that the technology can
19:34solve problems and the fact that
19:36you have research papers that
19:38are used, that are made possible
19:39because of it are interesting, but
19:41you're always looking for
19:45before a market exists,
19:47you still need early indicators of
19:51We have this phrase in the company,
19:54there's a phrase called
19:55key performance indicators.
19:58Unfortunately, KPIs are hard to
20:02I find KPIs hard to understand.
20:06A lot of people, when we look for
20:08KPIs, you go gross margins.
20:11That's not a KPI, that's a result.
20:15You're looking for something that's
20:18an early indicators of future
20:19positive result, okay,
20:21and as early as possible.
20:24that is because you want that early
20:26sign that you're going in the right
20:29And so we have this phrase that's
20:32early indicators, E-I-O-F-S,
20:35early indicators of future success.
20:37And it helps people because I was
20:40using it all the time to give
20:43the company hope that hey, look,
20:46we solve this problem.
20:49We solve that problem,
20:49we solve this problem.
20:50The markets didn't exist, but
20:52there were important problems and
20:54that's what the company is
20:55about to solve these problems.
20:57We want to be sustainable.
20:59And, therefore, the markets have to
21:00exist at some point.
21:02But you want to decouple the result
21:06from evidence that you're doing
21:10the right thing, okay?
21:12And so that's how you kind of solve
21:15this problem of investing into
21:17something that's very, very far
21:19away and having the conviction to
21:21stay on the road is to find as
21:23early as possible the indicators
21:25that you're doing the right things.
21:28And so start with a core belief.
21:30Unless something changes your mind,
21:32you continue to believe in it and
21:35early indicators of future success.
21:37>> What are some of those early
21:39indicators that have been used by
21:41Product Teams at NVIDIA?
21:48long before I saw the paper I met
21:51some people that needed my help on
21:54this thing called deep learning.
21:56At that time I didn't even know
21:57what deep learning was.
22:00And they needed us to create
22:02a domain specific language so that
22:04all of their algorithms could be
22:06expressed easily on our processors.
22:10And we created this thing called
22:13cuDNN, and it's essentially the SQL
22:16in storage computing.
22:18This is neural network computing.
22:21And we created a language, if you
22:23will, domain-specific language for
22:26them, kind of like the OpenGL of
22:30And so they needed us to do that so
22:32that they could express
22:34And they didn't understand CUDA,
22:36but they understood their
22:38And so we created this thing in
22:40the middle for them.
22:41And the reason why we
22:42did it was because even
22:44though there were zero,
22:45these researchers had no money, and
22:48this is kind of one of the great
22:50skills of our company, that
22:52you're willing to do something,
22:54even though the financial returns
22:57are completely non-existent or
22:59may be very, very far out even if
23:04is this worthy work to do?
23:07Does this advance a field of
23:08science somewhere that matters?
23:10Notice, this is something that I've
23:13been talking about since
23:15the very beginning of time.
23:17We find inspiration not from
23:19the size of a market, but
23:20from the importance of the work.
23:24Because the importance of
23:25the work is the early
23:26indicators of a future market.
23:28And nobody has to do a business
23:33Nobody has to show me a P&L,
23:35nobody has to show me a financial
23:39The only question is,
23:40is this important work?
23:41And if we didn't do it,
23:42would it happen without us?
23:45Now, if we didn't do something, and
23:47something could happen without us,
23:48it gives me tremendous joy,
23:51And the reason for that is,
23:53the world got better.
23:54You didn't have to lift a finger.
23:57That's the definition of Ultimate
24:00laziness, and in a lot of ways,
24:03you want that habit, and
24:05the reason for that is this.
24:07You want the company to be lazy
24:08about doing things that other
24:10people always do, can do.
24:12If somebody else can do it,
24:14We should go select the things that
24:17the world would fall apart.
24:18You have to convince yourself
24:21That if I don't do this,
24:25And if that work is hard and
24:27that work is impactful and
24:29important, then it gives you
24:33And so our company has been
24:34selecting these projects.
24:36Deep learning was just one of them,
24:38and the first indicator of
24:40the success of that was this fuzzy
24:42cat that Andrew Ng came up with.
24:44And then Alex Kruszewski detected
24:48cats, not all the time, but
24:50successfully enough that it was,
24:53this might take us somewhere.
24:56And we reasoned about the structure
24:59of deep learning, and
25:00we're computer scientists and
25:02we understand how things work, and
25:05so we convinced ourselves this
25:07could change everything.
25:09And anyhow, but that's an example.
25:11>> So these selections that you've
25:13made, they've paid huge dividends,
25:15both literally and figuratively.
25:17But you've had to steer the company
25:19through some very challenging
25:21times, like when it lost 80% of its
25:23market cap amid the financial
25:25crisis because what Wall Street
25:27didn't believe in your bet on ML.
25:29In times like these,
25:31how do you steer the company and
25:34keep the employees motivated at
25:38>> My reaction during that time is
25:41the same reaction I had about
25:44you asked me about this week.
25:46My pulse was exactly the same.
25:50This week is no different than last
25:51week or the week before that.
25:53And so the opposite of that, when
25:56you drop 80%, don't get me wrong,
26:01when your share price drops 80%,
26:05it's a little embarrassing, okay?
26:09And you just want to wear a t-shirt
26:13that says, it wasn't my fault.
26:18even more than that, you just don't
26:20want to get out of your bed,
26:22you don't want to leave the house,
26:24all of that is true,
26:25all of that is true.
26:27But then you go back to just doing
26:30your job, woke up at the same time,
26:32prioritize my day in the same way.
26:35I go back to what do I believe,
26:37you gotta always gut-check back to
26:41What do you believe?
26:42What are the most important things?
26:45And just check them off.
26:47Sometimes it's helpful,
26:49the family loves me, okay, check,
26:53double-check, right?
26:55And so you just gotta check it off
26:57and you go back to your core, and
26:59then go back to work, and
27:00then every conversations go back to
27:03Keep the company focused
27:05Do you believe in it,
27:06did something change?
27:07The stock price changed, but
27:09did something else change?
27:13Did all of the things that we
27:15assumed that we believed that led
27:19did any of those things change?
27:22Because if those things change,
27:23you've gotta change everything.
27:24But if none of those things change,
27:26you change nothing, keep on going,
27:28that`s how you do it.
27:30>> In speaking with your employees,
27:31they say that you- >> I try to
27:34>> [LAUGH] In speaking with
27:36they've said that your leadership-
27:39>> Including the employees.
27:41>> [LAUGH] >> I'm just kidding.
27:42>> [LAUGH] No, the lead leaders
27:44have to be seen, unfortunately,
27:47that's the hard part.
27:48I was an electrical engineering
27:51student and I was quite young when
27:56When I went to college,
27:57I was still 16 years old and so
27:59I was young when I did everything.
28:03And so I was a bit of an introvert,
28:06I don't enjoy public speaking.
28:09I'm delighted to be here, I'm not
28:10suggesting- >> [LAUGH] >> But it's
28:12not something that I do naturally.
28:15And so when things are challenging,
28:20it's not easy to be in front of
28:23precisely the people that you
28:30that is because could you
28:31imagine a company meeting,
28:33our stock prices dropped by 80%?
28:35And the most important thing I have
28:37to do as the CEO is this,
28:39to come and face you, explain it.
28:42And partly, you're not sure why,
28:45partly, you're not sure how long,
28:50you just don't know these things.
28:54But you still gotta explain it,
28:56face all these people,
28:58and you know what they're thinking.
29:01Some of them are probably thinking
29:04Some people are probably thinking
29:05you're an idiot, and
29:06some people are probably thinking
29:08And so, there are a lot of
29:10things that people are thinking and
29:12you know that they're thinking
29:13those things, but you still have to
29:15get in front of them and
29:18>> You may be thinking of those
29:19things, but yet, not a single
29:21person of your leadership team left
29:23during times like this.
29:25And in fact- >> They're
29:26>> [LAUGH] >> That's what I keep
29:29>> [LAUGH] >> I'm just kidding,
29:31I'm surrounded by geniuses.
29:33I'm surrounded by geniuses, yeah,
29:36other geniuses, unbelievable.
29:39Nvidia is well known to have
29:41singularly the best management team
29:44This is the deepest technology
29:46management team the world's
29:49I'm surrounded by a whole bunch
29:51of them, and they're just geniuses.
29:53Business teams, marketing teams,
29:55sales teams, just incredible.
29:57Engineering teams, research teams,
30:03>> Your employees say that your
30:04leadership style is very engaged,
30:06you have 50 direct reports.
30:08You encourage people across all
30:10parts of the organization to send
30:12you the top five things on their
30:15mind, and you constantly remind
30:17people that no task is beneath you.
30:19Can you tell us why you
30:21purposefully designed such a flat
30:23organization, and how should we be
30:26thinking about our organizations
30:28that we design in the future?
30:31>> To me, no task is beneath me
30:32because, remember, I used to be
30:34a dishwasher, and I mean that, and
30:36I used to clean toilets.
30:37I mean, I cleaned a lot of toilets,
30:38I've have cleaned more toilets than
30:40all of you combined.
30:40>> [LAUGH] >> And some of them,
30:44you just can't unsee.
30:50>> I don't know what to tell you,
30:54that's life, and so you can't show
30:58me a task that's beneath me.
31:02Now, I'm not doing it only because
31:05of whether it's beneath me or
31:10If you send me something and
31:12you want my input on it and
31:14I can be of service to you, and
31:17in my review of it, share with you
31:19how I reasoned through it,
31:21I've made a contribution to you.
31:24I've made it possible for you to
31:26see how I reason through something,
31:29and by reasoning, as you know,
31:32through something empowers you.
31:34You go, my gosh, that's how you
31:35reason through something like this.
31:38It's not as complicated
31:39This is how you reason through
31:41something that's super ambiguous.
31:43This is how you reason through
31:44something that's incalculable.
31:46This is how you reason through
31:48something that seems to be very
31:50scary, this is how you see,
31:53And so, I show people how to reason
31:56through things all the time,
31:59how to Forecast something,
32:02how to break a problem down.
32:04And you're empowering people all
32:07And so that's how I see it.
32:08If you send me something you want
32:10me to help review it,
32:12And I'll show you how I would
32:16In the process of doing that,
32:18of course I learned a lot from you.
32:20You gave me a seed of a lot of
32:22information, I learned a lot.
32:24I feel rewarded by the process.
32:27It does take a lot of energy
32:28sometimes because in order to add
32:30value to somebody and they're
32:31incredibly smart as a starting
32:33point, and I'm surrounded by
32:35incredibly smart people, you have
32:36to at least get to their plane.
32:38You have to get into their
32:40And that's really hard.
32:44And that takes just an enormous
32:45amount of emotional and
32:47intellectual energy.
32:48And so I feel exhausted after I
32:51work on things like that.
32:53I'm surrounded by a lot of
32:55A CEO should have the most
32:57direct reports by definition
32:59because the people that
33:00report to the CEO requires
33:02the least amount of management.
33:04It makes no sense to me that
33:07few people reporting to them.
33:10one fact that I know to be true.
33:12The knowledge, the information of
33:15a CEO is supposedly so valuable,
33:19You can only share it
33:21with two other people, or three.
33:24And their information is so
33:27incredibly secretive, that they can
33:30only share it with a couple more.
33:32Well, I don't believe in
33:35a culture, an environment, where
33:38the information that you possess
33:41is the reason why you have power.
33:45I would like us all to contribute
33:47to the company and our position in
33:50the company should have something
33:52to do with our ability to reason
33:54through complicated things, lead
33:57other people to achieve greatness,
34:00inspire, empower other people,
34:02support other people.
34:04Those are the reasons why
34:05the management team exists in
34:07service of all of the other people
34:09that work in the company,
34:11to create the conditions by which
34:13all of these amazing people,
34:14volunteer to come work for you
34:16instead of all of the amazing high
34:18tech companies around the world.
34:20They elected, they volunteered to
34:22work for you, and so
34:23you should create the conditions
34:25by which they could do their
34:26life's work, which is my mission.
34:29You've probably heard it,
34:31I've said that pretty clearly and
34:35What my job is is very simply to
34:36create the conditions by which you
34:38could do your life's work.
34:40And so how do I do that?
34:42What does that condition look like?
34:43What that condition should result
34:45in a great deal of empowerment,
34:47you can only be empowered
34:48if you understand the circumstance.
34:52You have to understand the context
34:53of the situation you're in,
34:55in order for you to come up with
34:57And so I have to create
34:58a circumstance where you understand
35:01the context, which means you have
35:04And the best way to be informed is
35:07for there to be as little layers of
35:10information mutilation, right?
35:15And so that's the reason why it's
35:17very often that I'm reasoning
35:20like in an audience like this.
35:22I say, first of all,
35:23this is the beginning facts.
35:25These are the data that we have,
35:26this is how I would reason
35:28These are some of the assumptions,
35:30these are some of the unknowns,
35:32these are some of the knowns.
35:34And so you reason through it, and
35:36now you've created an organization
35:37that's highly empowered.
35:39NVIDIA is 30,000 people,
35:40we're the smallest large company in
35:43We're a tiny little company, but
35:45every employee is so empowered and
35:47they're making smart decisions on
35:49my behalf every single day.
35:51And the reason for that is because
35:55they understand my condition,
35:58I'm very transparent with people.
36:02And I believe that I can trust you
36:04with the information.
36:05Oftentimes the information is hard
36:09the situations are complicated, but
36:12I trust that you can handle it.
36:14A lot of people hear me say,
36:18you can handle this.
36:19Sometimes they're not
36:21They just graduated.
36:24[LAUGH] I know that when I first
36:26graduated, I was barely an adult.
36:29But I was fortunate that I was
36:31trusted with important information.
36:35So I want to do that.
36:36I want to create the conditions for
36:39>> I have I do want to now address
36:42on everybody's mind, AI.
36:44Last week, you said that generative
36:47AI and accelerated computing
36:48have hit the tipping point.
36:50So as this technology becomes
36:53what are the applications that you
36:55personally are most excited about?
36:58>> Well, you have to go back to
36:59first principles and ask yourself,
37:00what is generative AI?
37:04What happened was we now have
37:06the ability to have software that
37:10can understand something.
37:13we digitized everything.
37:14That was, for example,
37:18But what does it mean?
37:20That sequence of genes,
37:22We've digitized amino acids.
37:24But what does it mean?
37:26And so we now have the ability,
37:31we digitize images and videos,
37:33we digitize a lot of things.
37:35But what does it mean?
37:36We now have the ability through
37:39stunning a lot of data.
37:41And from the patterns and
37:42relationships we we now understand
37:45Not only do we understand what they
37:46mean, we can translate between them
37:48because we learn about the meaning
37:50of these things in the same world.
37:52We didn't learn about them
37:54So we we learned about speech and
37:56words and paragraphs and
37:58vocabulary in the same context.
38:01And so we found correlations
38:02between them and they're all
38:04registered, if you will,
38:06registered to each other.
38:07And so now, not only do we
38:09understand the modality,
38:11the meaning of each modality,
38:13we can understand how
38:14to translate between them.
38:16And so for obvious things, you
38:18could caption the video to text,
38:20that's captioning, text to images,
38:23mid-journey, text-to-text, ChatGPT,
38:28we now know that we understand
38:30meaning and we can translate.
38:33The translation of something is
38:35generation of information, and
38:38all of a sudden, you have to take
38:40a step back and ask yourself, what
38:42is the implication in every single
38:45layer of everything that we do?
38:47And so I'm exercising
38:49I'm reasoning in front of you the
38:52same thing I did a 15 years ago,
38:55when I first saw AlexNet some 13,
38:5814 years ago, I guess,
39:00how I reasoned through it.
39:10But then, most importantly,
39:13What does it mean to
39:14every single layer of computing?
39:15Because we're in the world of
39:17And so what it means is that that
39:19the way that we process information
39:21fundamentally will be different
39:23That's when a video builds,
39:27The way we write software will be
39:28fundamentally different in
39:30The type of software will be
39:32able to write in the future will be
39:33different, new applications.
39:36the processing of those
39:38applications will be different.
39:40What was historically
39:42a retrieval-based model
39:44where information was pre-recorded,
39:48if you will, almost.
39:50We wrote the text, pre-recorded,
39:52and we retrieved it based on some
39:54recommender system algorithm.
39:57some seed of information will be,
40:00Be the starting point.
40:01We call them prompts,
40:03as you guys know, and
40:04then we generate the rest of it.
40:06And so the future of computing will
40:08be highly generated.
40:10Well, let me give you an example of
40:13For example, we're having
40:14a conversation right now,
40:16very little of the information
40:18I'm conveying to you is retrieved,
40:21most of it is generated.
40:24It's called intelligence.
40:27And so in the future,
40:28we're going to have a lot more
40:29generative, our computers will
40:31perform in that way.
40:32It's going to be highly generative,
40:34instead of highly retrieval-based.
40:36Then you go back and
40:37you're going to ask yourself, now,
40:40you've gotta ask yourself,
40:41what industries will be disrupted,
40:45about networking the same way?
40:47Will we think about storage the
40:48same way? Would we be abusive of
40:50Internet traffic as we are today?
40:53Probably not, notice we're having
40:55a conversation right now, and I'm
40:57to get in my car every question.
41:00So we don't have to be as abusive
41:02of transformation, information,
41:05transporting, as we used to.
41:07What's going to be more?
41:08What's going to be less?
41:10What kind of applications, etc?
41:12So you can go through the entire
41:13industrial spread and ask yourself,
41:15what's going to get disrupted?
41:16What's going to get big different?
41:17What's going to get nude, so on so
41:21starts from what is happening.
41:23What is generative AI?
41:26Foundationally, what is happening?
41:27Go back to first principles with
41:30There was something I was going to
41:31tell you about organization,
41:32you asked the question and
41:33I forgot to answer it.
41:34The way you create an organization,
41:36by the way, someday, don't worry
41:39about how other companies or
41:42you start from first principles.
41:45an organization is designed to do.
41:47The organizations of the past,
41:50where there's a king, CEO, and then
41:54you have all the royal subjects,
41:57the royal court, and
42:01keep working your way down,
42:03eventually, there are employees.
42:04But the reason why it was designed
42:06that way is because they wanted
42:08the employees to have as little
42:10information as possible,
42:11because their fundamental purpose
42:13of the soldiers is to die in
42:15the field of battle.
42:16To die without asking questions,
42:21I only have 30,000 employees,
42:23I would like none of them to die.
42:25[LAUGH] I would like them to
42:27question everything.
42:30And so the way you organize
42:31in the past and the way you
42:32organize today is very different.
42:33Second, the question is what is in
42:37An organization is designed so
42:40that we could build whatever it is
42:46why are we organized the same way?
42:50Why would this organizational
42:52machinery be exactly the same
42:53irrespective of what you build?
42:55It doesn't make any sense.
42:57You build computers,
42:58you're organize this way.
42:59You build health care services,
43:01you're built exactly the same way.
43:03It makes no sense whatsoever.
43:05And so you got to go back to first
43:06principles, just ask yourself,
43:07what kind of machinery?
43:11What are the properties of this
43:14What is the forest that this animal
43:19What is characteristics?
43:21Most of the time, you're trying to
43:23squeeze out the last drop of water.
43:25Or is it changing all the time,
43:28being attacked by everybody?
43:31You're the CEO, your job is to
43:33architect this company.
43:34That's my first job,
43:35to create the conditions by which
43:37you can do your life's work, and
43:39the architecture has to be right.
43:40And so you have to go back to first
43:42think about those things.
43:43And I was fortunate that when I was
43:4529 years old, I had the benefit of
43:47taking a step back and
43:49asking myself, how would I build
43:51this company for the future,
43:52and what would it look like?
43:54And what's the operating system,
43:56which is called culture?
43:57What kind of behavior do we
44:00And what do we discourage and
44:03not enhance, so on and so forth,
44:06>> I want to save time for
44:07audience questions, but
44:09this year's theme for View from
44:10the Top is Redefining Tomorrow.
44:12And one question we've asked all of
44:15our guests is, Jensen,
44:17as the co-founder and
44:20if you were to close your eyes and
44:23magically change one thing about
44:26tomorrow, what would it be?
44:29>> Were we supposed to think about
44:32>> [LAUGH] >> I'm going to
44:37give you a horrible answer.
44:43I don't know that it's one thing.
44:45Look, there are a lot of things we
44:51of things we don't control.
44:54a unique contribution,
44:56live a life of purpose.
44:58To do something that nobody else in
45:01the world would do or can do to
45:03make a unique contribution.
45:04So that in the event that after
45:08you're done, everybody says,
45:11the world was better because you
45:16And so I think that to me,
45:18I live my life kind of like this,
45:21I go forward in time and
45:26So you asked me a question that's
45:28exactly from a computer vision pose
45:30perspective, exactly the opposite
45:34I never looked forward from where
45:36I am, I go forward in time and
45:43I would look backwards and
45:44kind of read my history,
45:45we did this and we did that way,
45:47and we broke that prom down.
45:49Does that make sense?
45:50And so it's a little bit like how
45:53you guys solve problems.
45:55You figure out what is the end
45:57result that you're looking for, and
45:59you work backwards to achieve it.
46:01And so I imagine Nvidia making
46:03a unique contribution to advancing
46:06the future of computing,
46:07which is the single most important
46:10instrument of all humanity.
46:12Now, it's not about our
46:14self-importance, but
46:15this is just what we're good at,
46:17and it's incredibly hard to do.
46:19And we believe we can make
46:20an absolute unique contribution.
46:21It's taken us 31 years to be here,
46:23and we're still just beginning our
46:26And so this is insanely hard to do.
46:29And when I look backwards,
46:31I believe we're going to be
46:33remembered as a company that
46:35kind of changed everything.
46:38Not because we went out and
46:39changed everything through all
46:41the things that we said, but
46:42because we did this one thing that
46:44was insanely hard to do.
46:45That we're incredibly good at
46:46doing, that we love doing,
46:48we did for a long time.
46:49>> I'm part of the GSB LEAD,
46:51I graduated in 2023.
46:53So my question is, how do you see
46:56your company in the next decade?
46:58What challenges do you see your
47:00company would face and
47:01how you are positioned for that?
47:04can I just tell you what was going
47:06As you say, what challenges,
47:09the list that flew by my head-
47:12>> [LAUGH] >> Was so
47:14large that I was trying to
47:16figure out what to select.
47:19>> [LAUGH] >> Now, the honest truth
47:23is that when you ask that question,
47:25most of the challenges
47:28me were technical challenges.
47:31And the reason for that is
47:33because that was my morning.
47:35If you were chosen yesterday,
47:37it might have been market creation
47:41There are some markets that, gosh,
47:43I just desperately would love
47:46Can't we just do it already?
47:49But we can't do it alone.
47:50Nvidia's a technology platform
47:52company, we're here in service of
47:55a whole bunch of other companies so
47:57that they could realize,
47:59if you will, Our hopes and
48:01dreams through them.
48:03And so, some of the things
48:05that I would love, I would love for
48:07the world of biology to be at
48:09a point where it's kind of like the
48:12world of chip design 40 years ago.
48:15Computer aided in design, EDA,
48:17that entire industry,
48:19really made possible for us today.
48:22And I believe we're going to make
48:24possible for them tomorrow,
48:27computer-aided drug design,
48:29because we are are able to now
48:31represent genes, and proteins, and
48:34Very, very close to be able to
48:36represent and understand the
48:37meaning of a cell, a combination of
48:39a whole bunch of genes.
48:40What does a cell mean?
48:44what does that paragraph mean?
48:46Well, if we can understand a cell
48:49like we can understand a paragraph,
48:51imagine what we could do.
48:54And so, I'm anxious for
48:56kind of excited about that.
48:58There's some that I'm just
49:00excited about that I know we're
49:01around the corner on.
49:03human-oriented robotics.
49:05They're very, very close,
49:07And the reason for that is because
49:09if you can tokenize and understand
49:11speech, why can't you tokenize and
49:13understand manipulation?
49:14And so these kind of computer
49:17once you figure something out,
49:18you ask yourself, well,
49:19if I do that, why can't I do that?
49:21And so I'm excited about those kind
49:24And so that challenge is kind
49:26of a happy challenge.
49:28Some of the other challenges of
49:30course are industrial, and
49:35you've heard all that stuff before.
49:39the social issues in the world, the
49:42geopolitical issues in the world.
49:44Why can't we just get along with
49:45things in the world?
49:46Why do we have to say those kind of
49:48things in the world?
49:49Why do we have to say those things
49:51and amplify them in the world?
49:53Why do we have to judge people so
49:56you guys all know that.
49:57I don't have to say those
49:59>> My name's Jose, I'm a class with
50:02the 2023 from the GSB.
50:03My question is, are you worried at
50:05all about the pace at which we're
50:09And do you believe that any sort of
50:11regulation might be needed?
50:14>> Yeah, the answer is yes and no.
50:17The greatest breakthrough in modern
50:21AI, of course deep learning, and
50:24it enabled great progress.
50:27But another incredible breakthrough
50:29is something that humans know and
50:31we practice all the time.
50:33And we just invented it for
50:35language models called grounding,
50:37reinforcement learning human
50:40I provide reinforcement learning
50:42human feedback every day.
50:45And for the parents in the room,
50:47you're providing reinforcement
50:49learning human feedback all
50:52Now, we just figured out how to
50:54do that at a systematic level for
50:56artificial intelligence.
50:58There are a whole bunch of other
51:01technology necessary to guard rail,
51:07how do I generate tokens that obey
51:09the laws of physics?
51:11Right now, things are floating in
51:14space and doing things, and they
51:18don't obey the laws of physics.
51:21That requires technology.
51:22Guard railing requires technology,
51:24fine tuning requires technology,
51:26alignment requires technology,
51:27safety requires technology.
51:29The reason why planes are so safe
51:31is because all of the autopilot
51:32systems are are surrounded by
51:34diversity and redundancy, and
51:36all kinds of different functional
51:38safety, and active safety systems
51:41I need all of that to be invented
51:45You also know that the border
51:47between security and artificial
51:49intelligence cybersecurity and
51:51artificial intelligence is
51:52going to become blurry and blurry.
51:54We need technology to advance very,
51:56very quickly in the area of
51:58cybersecurity, in order to protect
52:00this from artificial intelligence.
52:02in a lot of ways we need technology
52:05to go faster, a lot faster, okay?
52:09Regulation, there's two types of
52:11There's social regulation,
52:13I don't know what to do about that.
52:14But there's product and services
52:16regulation, we know exactly what
52:18to do about that, okay?
52:19So, the FAA, the FDA, the NHTSA,
52:22you name it, all the Fs and
52:25all the Ns, and all the FCCs,
52:28they all have regulations for
52:31products and services that have
52:34particular use cases.
52:36Bar exams and doctors, and so on,
52:40You all have qualification exams,
52:42you all have standards that you
52:44you all have to continuously be
52:46certified accountants, and so
52:49Whether it's a product or
52:50a service, there are lots and
52:52lots of regulations.
52:54Please do not add a super
52:55regulation that cuts across of it.
52:57The regulator who is regulating
52:58accounting should not be the
53:00regulator that regulates a doctor.
53:05I love accountants, but if I
53:07ever need a open heart surgery,
53:09the fact that they can close books
53:12is interesting but not sufficient.
53:14>> [LAUGH] >> And so
53:16I would like all of those fields
53:19that already have products and
53:22services, to also enhance their
53:25regulations in context of AI, okay?
53:28But I left out this one very big
53:30one, which is the social
53:34And how do you deal with that?
53:36I don't have great answers for
53:38enough people are talking about it.
53:40But it's important to subdivide all
53:41of this into chunks,
53:43So that we don't become super
53:45hyper-focused on this one thing,
53:47at the expense of a whole bunch of
53:49routine things that we could
53:51And as a result, people are getting
53:53killed by cars and planes, and
53:54it doesn't make any sense.
53:55We should make sure that we do
53:57the right things there, okay?
53:58Very practical things.
54:00May I take one more question?
54:01>> Well, we have some rapid fire
54:03questions for you as view
54:05from the [INAUDIBLE] division.
54:07>> Okay. >> [LAUGH] >> I was trying
54:09>> [LAUGH] >> Okay, all right,
54:15>> Well, your first job was
54:16at Denny's, they now have a booth
54:18What was your fondest
54:19memory of working there?
54:19>> My second job was AMD,
54:22>> [LAUGH] >> Is there a booth
54:25dedicated to me there?
54:26>> [LAUGH] >> I'm just kidding.
54:29>> [LAUGH] >> I loved my job there,
54:34it was a great company, yeah.
54:36>> Yeah, if there were a worldwide
54:38shortage of black leather jackets,
54:40what would we be seeing you
54:43>> [LAUGH] >> No, I've got a large
54:46reservoir of black jackets.
54:48>> [LAUGH] >> I'll be the only
54:50person who is not concerned.
54:52>> [LAUGH] You spoke a lot about
54:54textbooks, if you had to write one,
54:56what would it be called?
55:02>> I wouldn't write one.
55:02>> [LAUGH] >> You're asking me
55:05a hypothetical question that has no
55:08>> [LAUGH] >> [LAUGH] That's fair.
55:11And finally, if you could share
55:13one parting piece of advice to
55:14broadcast across Stanford,
55:24belief, Gut-check it every day.
55:31Pursue it with all your might,
55:35pursue it for a very long time.
55:40Surround yourself with people you
55:43love, and take them on that ride.
55:45So, that's the story of Nvidia.
55:48this last hour has been a treat,
55:52thank you for spending [INAUDIBLE].
55:56>> Thank you very much.
![Welp, Here's what the Fed *Just* Said [Fed FOMC Minutes]](/_next/image?url=https%3A%2F%2Fi.ytimg.com%2Fvi%2Fl8u23oq1E9E%2Fhqdefault.jpg&w=3840&q=75)
