00:05 good evening everyone and welcome my
00:08 name is CMAC memi I am the William Von
00:11 muffling professor of business in the
00:12 decision risk and Operations Division
00:14 here at the Columbia business school I
00:17 want to thank you all for joining us
00:18 this evening for tonight's program which
00:20 features Jensen Wong co-founder and CEO
00:23 of Nvidia Corporation as well as our own
00:25 coce mcis the dean of Columbia business
00:28 school and the David and Lin professor
00:30 of business our two speakers tonight
00:33 have much in common in fact they both
00:35 graduated from Stanford in electrical
00:37 engineering viia he has revolutionized
00:39 first the uh Graphics Processing Unit
00:41 industry and now more recently the
00:43 artificial intelligence industry he has
00:46 been named the world's best CEO by
00:48 Harvard Business review and brand
00:50 Finance as well as Fortune Magazine's
00:51 business person of the year and one of
00:53 Time magazine's 100 most influential
00:56 people our fireside chat today has been
00:59 made possible through both the David and
01:01 Lyn siin leadership series as well as
01:03 the digital Finance uh sorry excuse me
01:05 the digital future initiative and
01:07 additionally I serve on the leadership
01:09 of the digital future initiative the DFI
01:11 here at uh um CBS the digital future
01:15 initiative is cbs's new Think Tank
01:17 focusing on preparing students to lead
01:19 through the next Century of digital
01:20 transformation as well as helping
01:22 organizations governments and
01:23 communities better understand leverage
01:25 and prosper from future waves of digital
01:28 disruption now I would like to hand it
01:37 go thank you samak uh thank you all for
01:43 um this is an exciting topic uh and a
01:47 topic that is near and dear certainly to
01:52 uh and it's a topic where the school
01:55 everything that we do at the school is
01:56 changing so fast uh trying to keep up
01:59 trying to to change curricula trying to
02:02 create opportunities for our students to
02:05 actually learn about Technologies and
02:07 how they're changing the world and to be
02:09 honest prepare for the future and there
02:12 is no better person to be having to talk
02:15 about AI than Jensen hang uh Jensen
02:19 thank you so much for making the time
02:20 and coming here welcome I'm Delight I'm
02:23 delighted to be here is this on is this
02:26 on yes I just love hearing that I'm the
02:30 world can we just can we end the talk
02:33 here I think the expectation is going to
02:35 be pretty high to say something
02:39 smart well good luck with
02:43 you so I want to start uh by having you
02:50 walk us through a little bit the history
02:51 of Nvidia and then I want to talk a
02:53 little bit about that leadership thing
02:54 that we just mentioned but you've been
02:57 you launched that company 30 years ago
03:00 and you have led it through a
03:03 transformation uh different applications
03:06 different type of products um walk us
03:09 through a little bit that
03:11 Journey yeah one of one of the most
03:14 proud moments I'll start with the most
03:16 proud moment that happened
03:18 recently uh the the CEO of
03:21 Denny's uh where I used to work was my
03:23 first company and they learned that
03:26 Nvidia Not only was I a dishwasher
03:30 uh bus boy and worked my way up the
03:33 corporate ladder and became waiter at
03:36 Denny's and that they they were my first
03:40 company and uh and that I still know how
03:43 to take I still know the menu well um uh
03:47 superb Bird's excellent by the way
03:49 anybody know what a Superbird
03:51 is what kind of college student are
03:54 you Denny it's America's Diner
03:58 go and and that that Nvidia was founded
04:02 um by Chris Curtis and I out outside our
04:06 uh home in sose at one of the the
04:09 Denny's there and so they contacted me
04:12 recently and the booth that we sat at is
04:15 now you know the Nvidia booth and my
04:19 name is is uh Nvidia is this is where a
04:21 trillion dooll company was
04:24 founded and so so that's a very proud
04:27 moment um Nvidia was founded found it uh
04:30 during a time when when the the the PC
04:34 Revolution was just starting and the
04:36 microprocessor uh was was uh capturing
04:39 the INSP the imagination for of just
04:41 about the entire uh
04:43 industry and um uh the world the world
04:49 properly uh saw that the CPU the
04:51 microprocessor was going to be
04:53 revolutionary and and it really reshaped
04:55 how the IT industry how the computer
04:58 industry was formed uh companies that
05:01 were successful before the
05:02 microprocessor Revolution the x86
05:04 Revolution and companies that were
05:05 successful afterwards were completely
05:07 different we started our company during
05:10 that time and our perspective was that
05:12 general purpose Computing uh as
05:14 incredible as it is can't sensibly be
05:17 the solution for all kinds of problems
05:19 and we wanted to We believe that there
05:21 was a a way of doing Computing we call
05:23 Accelerated Computing where you would
05:26 add a specialist next to the generalist
05:30 the CPU is a generalist if your well
05:31 could do anything it could do everything
05:34 however it can't obviously if you can do
05:35 everything and anything then obviously
05:38 you can't do anything very well and so
05:41 there are some problems that we felt
05:43 that were uh not solvable not good
05:46 Solutions uh for not good problems to be
05:49 solved by the what we call the normal
05:51 computer and so we started this
05:53 accelerated Computing company the
05:54 problem is uh if you want to if you want
05:57 to create a Computing platform company
05:58 and I I don't know I know how many
05:59 computer scientists are in here but if
06:01 you want to create a Computing platform
06:02 company one hasn't been created since
06:05 1964 a year after I was born the IBM
06:09 system system 360 um beautifully
06:12 describes what a computer is in 1964 IBM
06:16 described that the system 360 had a
06:18 central processing unit IO subsystem
06:23 direct memory access virtual
06:26 memory binary compatibility across a
06:31 architecture it described everything
06:33 that we describ computers to this day 60
06:36 years later and we felt we felt that
06:39 there was a new form of computing that
06:40 could solve some interesting problems at
06:41 the time uh it wasn't completely clear
06:44 what problems we could solve um but but
06:46 we felt that we felt that accelerated
06:48 Computing has a future and so
06:49 nonetheless uh we went out to start this
06:51 company and um we made a we made a a
06:54 great first decision that frankly is
06:58 unfund to this day if somebody were to
07:01 come up to you and said uh one uh we
07:03 we're going to invent a new technology
07:05 that the world doesn't have everybody
07:08 wants to go build a computer company
07:10 around a CPU uh we want to build the
07:12 computer company around something else
07:13 connected to the CPU number one and the
07:16 killer app the killer app is a video
07:20 game a 3D video game in
07:23 1993 um and and uh that application
07:27 doesn't exist and the company who would
07:29 build this company doesn't exist and the
07:31 technology that we're trying to build
07:32 doesn't exist and uh uh so so now you
07:35 have a you have a company that uh has a
07:39 technology Challenge and a market
07:42 Challenge and uh an ecosystem Challenge
07:45 and so the odds of that company
07:46 succeeding is approximately
07:48 0% uh but nonetheless nonetheless uh we
07:51 were fortunate because because uh uh two
07:54 very important people frankly that that
07:57 um uh I had worked with uh and Chris and
08:00 Curtis the three of us had worked
08:02 with were were incredibly important
08:05 people in the in the technology industry
08:07 at the time uh called up the most
08:09 important venture capitalist in the
08:10 world Don Valentine Don Valentine at the
08:12 time and and told Don you know Don and
08:16 his name is Wolf W cor Wolf coryan the
08:18 one of the founders of the semiconductor
08:20 industry told Don give this kid money
08:23 and then figure out along the way
08:25 whether it's going to work and and uh
08:27 fortunately they did but that business
08:30 that business plan I wouldn't fund
08:32 today and uh it is just has too many
08:36 dependencies and each one of them have
08:37 some probability of success and when you
08:40 compound all of these together when you
08:41 multiply all these together you got 0o
08:42 per. uh and so so um
08:46 nonetheless uh we imagined that there
08:48 would be this this uh Market called
08:50 video games and this Market would be the
08:53 largest entertainment industry in the
08:56 world at the time it was
08:58 Zero and 3D Graphics we we uh we
09:02 postulated with would be used for uh
09:05 telling the stories of almost any SP
09:07 Sport any game and so in Virtual Worlds
09:10 you could have any game any sport and as
09:13 a result everybody would be a gamer and
09:15 so don don Valentine asked me you so how
09:18 big is this Market going to be and I
09:19 said well every G every every human will
09:21 be a gamer someday every human would be
09:24 a gamer someday also the wrong answer uh
09:27 quite frankly for starting comp so these
09:29 are horrible habits these are horrible
09:32 skills I'm not advocating them for you
09:34 um but nonetheless uh it turned out to
09:37 have been true video games turned out to
09:39 have been the largest entertainment
09:40 industry in the world uh 3D Graphics
09:42 took off and um we found the first
09:45 killer app for Accelerated Computing
09:47 which bought us the time uh to use
09:49 accelerated Computing to solve a whole
09:51 bunch of other problems which eventually
09:55 intelligence this is fantastic story so
09:58 before before we go to AI uh I would
10:01 like to ask a little bit about the
10:04 crypto period so you know gaming was a
10:07 huge uh uh obviously Journey for NVIDIA
10:13 and then at some point in time the
10:15 killer up became crypto and Mining uh
10:20 chapter yeah we uh accelerated Computing
10:23 uh can solve problems that normal
10:25 computers can't and all of our gpus even
10:27 though you use it for designing uh cars
10:30 designing buildings designing use it for
10:32 molecular Dynamics use it for playing
10:34 video games uh it has this programming
10:36 model called cuda that we invented and
10:38 Cuda is the only Computing model that
10:41 sits next that that exists today uh that
10:43 is as popular as a x86 uh it's used used
10:47 by developers all over the world and so
10:49 anyways uh one of the things that that
10:53 process uh uh parallel processing
10:55 incredibly fast and obviously one of the
10:57 algorithms that we would do very nicely
11:00 on is cryptography and so when bit
11:03 Bitcoin first came out uh there were no
11:05 Bitcoin Asic and the obvious thing is to
11:08 go find a g find the fastest
11:10 supercomputer in the world and the
11:12 fastest supercomputer that also has the
11:14 highest volume is NVIDIA gpus it's
11:16 available in uh hundreds of millions of
11:19 Gamers homes and so by downloading uh
11:22 downloading an application you could do
11:25 uh some crypto mining at your house well
11:30 this the fact that the fact that you
11:32 could buy one of our gpus one of our
11:35 computers and you plug it into the
11:38 wall and money starts squirting
11:42 out that was the day that my mom figured
11:44 out what I did for a living and so she
11:47 she uh she she she called me one day and
11:49 she said she said son I thought I
11:51 thought you were doing something about
11:52 video games I and I finally figured out
11:54 what you do you buy invidious products
11:57 you plug it in and money squirts out and
11:58 and I said that's exactly what I do and
12:00 that's the reason why that's the reason
12:02 why so many people bought it
12:05 I Bitcoins of course led to ethereum um
12:09 and uh but the idea the idea that you
12:12 would use a supercomputer use a super
12:14 processing system like like Nvidia gpus
12:17 to uh either encode or compress or do
12:21 something to to refine data and trans
12:24 transfer it transform it into a valuable
12:30 you guys know what that sounds like to
12:32 generate valuable tokens chat GPT and so
12:35 to this day really one of the things
12:37 that's that's happened um if you if you
12:40 extend the the the sensibility of about
12:43 ethereum and cryp cryp crypto mining um
12:47 uh it's kind of sensible in the sense
12:49 that all of a sudden we created this new
12:51 type of Industry where raw data comes in
12:54 you apply energy to this computer and
12:57 literally money comes squirting out and
13:00 and those the the the currency is of
13:03 course in tokens and those that token is
13:06 intelligence tokens this is one of the
13:08 major industries of the future now I'm
13:11 just going to I'll describe something
13:13 else and it makes perfect sense to us
13:14 today but back then it looks strange you
13:18 take water and you move it into a
13:19 building you apply fire to it and what
13:21 comes out is something incredibly
13:23 valuable and invisible called
13:26 electricity and so today we're going to
13:28 move data into a data center it's going
13:30 to refine it and it's going to work on
13:32 it and it's going to you know harness
13:34 the capability out of it and produce a
13:36 whole bunch of digital tokens that are
13:37 going to be valuable in digital biology
13:39 it'll be valuable in physics it'll be
13:41 valuable in it and computer all kinds of
13:43 computing areas and social media and all
13:45 kinds of things computer games and so so
13:47 all kinds of things and it comes out in
13:50 tokens so the the future is going to be
13:52 about AI factories and they'll be uh you
13:54 know Nvidia gear will be powering these
13:56 AI factories so you
13:59 we have jumped into neural networks and
14:01 I want to and we talked about parallel
14:03 Computing uh how we render Graphics
14:06 let's say on a monitor uh how we play
14:10 games how we solve cryptographic
14:16 um talk to us a little bit about how the
14:20 GPU is useful in training neural
14:23 networks uh but then what I wanted I
14:25 wanted us to do for for this audience
14:29 tell us a little bit about what it takes
14:30 to train a model like uh J GPT what it
14:34 takes in terms of Hardware what it takes
14:35 in terms of data what it terms of the
14:38 size of the cluster that you're using
14:40 the amount of money that you need to
14:41 spend because these are like huge
14:44 problems and and it's I think giving us
14:47 a glimpse of the scale will be will be
14:50 fun well everybody wants want you to
14:53 think that it's a huge problem is super
14:56 not I it's not and and let me tell you
14:59 why uh your it costs our company about
15:06 five6 billion dollar of engineering cost
15:09 to design a chip and then at one point
15:13 two years three years later I hit enter
15:15 and I send a I send an email to
15:19 tsmc and I FTP basically a large file to
15:22 tsmc and they fabit and that that
15:25 process cost our company something along
15:27 the lines of half a b billion dollar so$
15:29 5.5 billion I get a chip and that chip
15:32 of course is valuable to us uh but it's
15:35 no big deal I do it all the time and so
15:38 if somebody were to say say hey Jensen
15:40 you need to build a billion dooll Data
15:42 Center and once you plug it in money
15:44 will start squirting out the other side
15:46 I'll do it in a heartbeat and apparently
15:49 a lot of people do and the reason for
15:51 that is because who doesn't want to
15:53 build a factory for generating
15:55 intelligence now so a billion dollars is
15:57 not that much money frankly uh it we the
16:00 world spends about $250 billion dollar a
16:02 year in infrastructure Computing
16:04 infrastructure and none of is generating
16:06 money is just storing our files passing
16:08 our email around and that's already $250
16:10 billion and so the the uh one of the
16:13 reasons why we're growing so fast is
16:17 years general purpose Computing is on a
16:19 decline because it it's it's not
16:22 sensible to to invest another $250
16:24 billion do to build another general
16:26 purpose Computing data center it's to
16:28 it's too brute force in energy it's too
16:31 slow in computation and so now
16:34 accelerated Computing is here that $250
16:36 billion goes to build accelerated
16:38 Computing data centers and we're very
16:40 very happy to to support customers to do
16:43 that and in addition to that um that
16:46 accelerated Computing you now have an
16:47 infrastructure for generative AI for all
16:49 of the things that we're we're just
16:50 talking about um basically the way works
16:52 is you take a whole lot of data and you
16:55 you compress it you compress it deep
16:58 learning is like a compression algorithm
17:01 and you're you're trying to figure out
17:02 you're trying to learn the mathematical
17:05 representation of the mathematical
17:07 representation the patterns and
17:08 relationships of the data that you
17:09 you're disc you're studying and you
17:11 compress it into a neuron Network so
17:13 what goes in is uh say trillions of
17:18 bytes trillions of tokens so so let's
17:20 say a few trillion F trillion bytes what
17:23 comes out of it is 100 gigabytes
17:29 and so you've taken all of that data and
17:31 you've compressed it into this little
17:32 tiny file 100 gigabytes is like two DVDs
17:35 two DVDs you could download on your
17:37 phone and you can watch it right and so
17:39 so you could download this giant neural
17:41 network on your phone and now that now
17:44 that all of this data has been
17:45 compressed into it the data that's
17:47 compressed neur Network model is a
17:50 semantic model meaning you can interact
17:51 with it you could ask a questions and it
17:53 would go back into its memory and uh
17:56 understand what you meant and generate
17:57 text for you okay have a conversation
17:59 with you so at the core is kind of like
18:01 that it's it's not you know it sounds
18:04 magical but you know for all the
18:06 computer science scientists in a room
18:08 it's very sensible and uh don't let
18:11 anybody convince you it cost a lot of
18:12 money I'll give you a good
18:15 break everybody go build a go build a
18:18 eyes the um uh the scale if if I press
18:24 you a little bit on that scale like how
18:27 do do you need a a computer that is
18:29 essentially a data center to uh to
18:32 estimate these models 16,000 gpus is
18:35 what it took to build GPD 4 which is the
18:38 largest one that anybody's using today
18:41 dollars okay Y and that's a
18:44 check I mean it's not not even a very
18:46 big check y Don't Be
18:49 Afraid don't let anybody talk you out of
18:52 company yeah build your dreams let let
18:56 me ask you a question about the billion
18:58 dollar check uh and the growth that
19:01 you've been experiencing
19:04 um I think you you were named the best
19:07 CEO uh by hbr That's Entertainment
19:11 That's Entertainment uh I'll keep I'll
19:13 keep repeating it uh and then eventually
19:15 and then eventually we'll end with that
19:17 line but uh but in some sense um you're
19:22 leading a company right now through a
19:24 period of extreme growth hyper growth uh
19:28 uh something that most companies have
19:30 not experienced in their life um and I
19:34 want to perhaps to tell us a little bit
19:37 about what does it look like uh I mean
19:41 you know sort of doubling in size uh in
19:44 under a year or you know sort of
19:47 managing Supply chains managing
19:49 customers managing growth managing money
19:52 uh how do you actually uh do
19:56 that I love the man manag money part of
19:58 it just counting it it's
20:01 fun you just wake up in the morning just
20:03 roll around all the
20:08 cash isn't that what you guys are all
20:10 here to do my understanding
20:14 is that's the end goal what that's the
20:24 um uh let's see the the um uh building
20:29 companies is hard there there's just no
20:31 there's there's nothing easy about it uh
20:33 there's a lot of pain and sufferings a
20:35 lot of hard work if it was easy
20:37 everybody would do it and and the the
20:40 truth about all companies big or small
20:42 ours or others you know in technology
20:46 dying and and um and the reason for that
20:50 is because somebody's always trying to
20:51 lefrog you so you're always dying you're
20:53 always on the way of going out of
20:54 business and and if you don't if you
20:57 don't uh internalize that sensibility if
20:59 you don't internalize that that belief
21:01 it you will go out of business and so so
21:04 we you know I I started at Denny as you
21:07 guys know and um Nvidia was was was
21:10 built out of very unlikely odds and it
21:13 took a took us a long time to be here I
21:15 mean we're a 30- year old company and
21:17 when Nvidia fir Nvidia first founded uh
21:21 the I the PC uh Windows 95 hadn't come
21:24 out 1993 and so and that was the first
21:30 email you know we and so uh there were
21:33 no such thing as laptops or you know
21:35 smartphones none of that stuff existed
21:37 and uh uh so you you could just imagine
21:40 the world that we were started in and
21:41 the world we are we didn't have LCD
21:44 CRTs and um so the world was very very
21:47 different uh CD ROMs didn't exist I mean
21:50 just to put it in perspective all this
21:51 stuff that was the era we were founded
21:53 in and it took this long for for our
21:56 company to be uh to be recogn recognized
21:58 as having reinvented Computing for the
22:00 first time in 60 years um growing fast
22:04 growing fast is all about people uh
22:07 obviously companies is all about people
22:09 and whether you if the right systems in
22:11 place you get the right you're
22:12 surrounded by amazing people like I am
22:15 and uh uh and the company has has craft
22:18 has skills it doesn't really matter
22:20 whether you you ship a hundred billion
22:22 dollar or $200 billion doll now the
22:25 truth is that the supply chain is not
22:26 easy people think does anybody know what
22:29 a GeForce graphics card looks like can
22:31 you just show show me as a hand that
22:33 anybody knows what an Nvidia graphics
22:35 card looks like and so so you have a
22:37 feeling that the graphics card is like a
22:39 cartridge that you put into into a PC
22:41 the PCI Express slot of a PC uh but a
22:44 but our our Graphics chips these days
22:47 what is used in these deep Learning
22:49 Systems uh is 35,000 Parts it weighs 70
22:54 pounds it takes robots to build them
22:56 because they're so heavy uh it it takes
22:59 a supercomputer to test it because it's
23:00 a supercomputer itself and um and it
23:04 costs $200,000 and for
23:07 $200,000 you buy one of these computers
23:09 you replace several hundred general
23:12 purpose processors that cost several
23:14 million dollars and so for every
23:16 $200,000 you save you say for every
23:19 $200,000 you spend with Nvidia you spend
23:21 you save $2.5 million do uh in Computing
23:25 and that's that's the reason why I tell
23:27 you you know the more more you buy the
23:29 more you save and and uh and apparently
23:32 it's working out really well people are
23:34 you know really lining up and so so I so
23:39 so that that's it that's what we do for
23:40 a living and um the supply chain is
23:43 complicated we build the most
23:44 complicated computers the world's ever
23:45 seen U but you know how hard can it be
23:55 hard it's really hard uh but but at the
23:58 core of it if you're surrounded by
24:00 amazing the the simple truth is that
24:02 that it's all about people and and I'm
24:05 I'm lucky to be surrounded by a great
24:08 team you you have uh and then the CEO
24:11 says things like make it so number one
24:14 something like that yeah make it work
24:16 make it work make it so make it so the
24:19 um I want to go back uh to AI Trends and
24:24 what you think about the future but you
24:26 mentioned the word plat form earlier on
24:30 um you mentioned your software uh
24:33 environment uh so you know you have the
24:37 hardware infrastructure you have a
24:38 software environment that is actually
24:40 pervasive in training neural networks
24:42 right now um You're Building data
24:46 centers or you're creating environments
24:48 within data centers that are sort of uh
24:50 uh clusters of Nvidia uh Hardware
24:53 software and and and uh sort of telec
24:57 communic Comm unication between between
25:00 resources how important it is to be sort
25:02 of a whole platform solution versus a
25:04 sort of uh uh a hardware uh play and and
25:08 how core is that into Nvidia
25:12 strategy I unlike F first of all before
25:16 you could build something you have to
25:17 know what you're building and why why it
25:20 why what is the reason the first
25:22 principles for its existence accelerated
25:25 Computing is not a chip that's why it's
25:27 not called an accelerator accelerated
25:30 Computing is about understanding how can
25:33 you accelerate everything in life if you
25:34 can accelerate everything in life if you
25:37 could accelerate every application
25:38 that's called really fast
25:40 Computing and so accelerated Computing
25:43 is first understanding what are the
25:44 domains what are the applications that
25:46 matter to you and to understand the
25:49 algorithms and the Computing systems and
25:51 the architecture necessary to accelerate
25:53 that application and so it turns out
25:56 that general purpose Computing is a
25:58 sensible idea accelerating an
26:01 application is a sensible idea so let me
26:03 give you an example there's a uh you
26:05 have DVD decoders you play DVDs or uh
26:09 h.264 decoders on your phone it's it
26:12 does one job in one job only and it does
26:14 it incredibly well nobody knows how to
26:16 do it better accelerated Computing is
26:19 kind of this weird midle there are many
26:21 many applications that you can
26:23 accelerate so for example we can
26:25 accelerate all kinds of image process
26:27 stuff particle physics stuff we can
26:30 accelerate all kinds of things that
26:31 includes linear algebra we could
26:33 accelerate you know we could accelerate
26:35 many many domains of of applications
26:40 problem accelerating one thing is easy
26:43 generally running everything on under a
26:45 c compiler is easy accelerating enough
26:48 domains such that if you accelerate too
26:51 many domains supposedly you accelerate
26:53 every domain then you're back to a
26:54 general purpose processor right what
26:56 makes them so dumb that they can't build
26:58 just a faster chip and so on the one
27:01 hand on the other hand if you're if you
27:03 only accelerate one application then the
27:06 market size is not big enough to fund
27:09 R&D and so we had to find that slippery
27:12 middle and that that is the Strategic
27:16 journey of our company this is this is
27:18 where strategy meets reality and that's
27:20 the part that Nvidia got right that
27:23 nobody no other company in history of
27:25 computing ever got right to find a way
27:28 to have a sufficiently large domain of
27:30 applications that we can accelerate that
27:32 is still 100 times 500 times faster than
27:36 CPU and such that the economics the
27:39 flywheel the flywheel of number of
27:42 domains expanding the number of
27:44 customers expanding the number of
27:46 markets expanding the the sales which
27:49 creates larger R&D which allows us to
27:52 create even more amazing things which
27:54 allows us to stay well ahead of the CPU
27:56 does that make sense that that flywheel
27:58 is insanely hard to create nobody's ever
28:00 done it it's only been done just one
28:02 time and so that is the that is the the
28:06 the capability and in order to do that
28:08 you have to understand the algorithms
28:10 you have to understand a lot about the
28:12 domains of applications you have to
28:14 select it right you have to create the
28:16 right architecture for it and then the
28:18 last thing that we did right was that
28:20 that uh we realized that in order for
28:22 you to have a Computing platform the
28:24 applications you develop for NVIDIA
28:25 should run on all Nvidia you shouldn't
28:27 have to think does it run on this chip
28:29 is it going to run on that chip it
28:31 should run on every chip it should run
28:32 on every computer with Nvidia in it
28:34 that's the reason why every single GPU
28:36 that's ever been created our company
28:38 even though we had no customers for Cuda
28:40 a long time ago we stay committed to it
28:42 we were determined to create this
28:44 Computing platform since the very
28:46 beginning customers were not and that
28:48 was the pain and suffering it cost the
28:51 company decades and billions of dollars
28:54 getting here and if not for all the
28:56 video gamers in the room room here we
28:58 wouldn't be here uh you were our day
29:00 jobs and then at night we can go uh
29:03 solve digital biology go help people you
29:06 know with quantum chemistry go help
29:08 people with artificial intelligence or
29:10 Robotics and such and so we realized
29:13 number one that we were uh uh
29:15 accelerated Computing as a software
29:16 problem the second thing is AI is a data
29:19 center data center infrastructure
29:20 problem and it's very obvious because
29:23 you can't you can't just train an AI
29:25 model on a laptop you can't train it on
29:26 a cell phone it's not big enough of a
29:28 computer the amount of data is measured
29:31 in trillions of bytes and you have to
29:34 process that trillions of bites billions
29:36 of times and so obviously that's going
29:38 to be a large scale computer
29:40 Distributing the problem across millions
29:43 of gpus the reason why I say Millions is
29:46 16,000 inside the 16,000 are
29:49 thousands and so we're Distributing the
29:52 workload across millions of processors
29:54 there are no applications in the world
29:56 today that can be be distributed across
29:58 millions of processors Excel works on
30:01 processor and so that distribu that
30:04 computer science problem was a giant
30:06 breakthrough utterly giant breakthrough
30:09 and this reason why it en enabled
30:10 generative AI enable large language
30:12 models so we observed two things one
30:15 accelerated Computing is a software
30:16 problem algorithm problem and AI is a
30:18 data center problem and so we're the
30:20 only company that went out and built all
30:22 of that stuff and and uh the last part
30:25 that we did was a business model choice
30:28 we could have been a data center company
30:30 ourselves and be completely vertically
30:32 integrated um however we recognize that
30:35 no computer company no matter how
30:37 successful will be the only computer
30:39 company in the world and it's better to
30:41 be a platform Computing company because
30:44 we we love developers it's better to be
30:46 a platform Computing company that serves
30:48 every Computing company in the world
30:50 than to be a Computing company all by
30:51 ourselves and so we took this data
30:53 center which is the size of this room
30:55 whole bunch of wires and a whole bunch
30:57 of switches and networking and a bunch
30:58 of software we disaggregated all of that
31:01 and we integrated into everybody else's
31:03 data centers that are all completely
31:05 different so AWS and gcp and Azure and
31:09 meta and so on so forth data centers all
31:12 over the world that's an insane
31:13 complexity problem and we figured out a
31:16 way to to have enough standardization
31:18 wherever was necessary enough
31:20 flexibility so that we could accommodate
31:23 enough collaboration with all the the
31:25 world's computer companies as a result
31:27 uh nvidia's architecture is now grafted
31:30 you will if you will into every single
31:32 computer company in the world and that
31:34 that is created a large footprint larger
31:37 larger install base more developers
31:41 better applications which makes ha
31:43 customer happier customers who buy them
31:46 more chips which increases the install
31:48 base which increases our R&D budget so
31:51 on so forth the flywheel the positive
31:53 feedback system and so that's how it
31:56 works nice and easy
31:58 so one thing you haven't done and and I
32:00 wanted you to explain to us why is you
32:03 haven't invested in fabricating your own
32:07 chips um and why is
32:10 that um that's an excellent question the
32:13 reason for that is as a matter of as a
32:17 matter of strategic Choice the core
32:20 values of our company my own core values
32:23 the core values of our company is about
32:26 choosing the most important thing in
32:29 choosing how do you choose how do you
32:33 choose well everything you know how do
32:36 you choose what to do tonight how do you
32:38 choose well our company decides to
32:40 choose choose projects for one
32:45 goal my goal is to create the
32:48 environment an environment by which
32:50 amazing people in the world will come
32:54 work an amazing environment for the most
32:58 the best people in the world who wants
33:00 to pursue this field of Compu Computing
33:02 and computer science and artificial
33:03 intelligence to create the conditions by
33:05 which they will come and do their life's
33:08 work well if I say that then now the
33:12 question is how do you achieve that so
33:15 let me give you an example of how not to
33:18 that nobody that I know wakes up in the
33:21 morning and say you know
33:23 what my neighbor is doing that and you
33:26 know what I want to do I want to take it
33:30 them I can do it to I want to take it
33:33 from them I want to capture their
33:37 share I want to Pummel them on
33:41 price I want to kick them in the I want
33:46 share it turns out no great people do
33:51 that everybody wakes up in the morning
33:53 says I want to do something that has
33:55 never been done before that's incredibly
33:56 hard to do that if successful makes a
33:59 great impact in the world and that's
34:01 what nvidia's core values are one how do
34:04 we choose do something that the world's
34:06 never done before let's hope that it's
34:08 insanely hard to do um the reason why
34:10 you choose something insanely hard to do
34:12 by the way so that you have lots of time
34:13 to go learn it if something is insanely
34:16 easy to do like Tic Tac
34:19 D I wouldn't you know fuss over it and
34:23 the reason for that obviously is highly
34:24 competitive and so you got to choose
34:26 something incredibly hard to do and that
34:29 that thing that's hard to do discourages
34:31 a whole bunch of other all by itself
34:33 because the person who's willing to to
34:35 to suffer the longest wins and so we
34:38 choose things that are incredibly hard
34:40 to do and you've heard me say pain and
34:42 suffering a lot and and it's actually a
34:44 positive attribute people who can suffer
34:48 are the ultimately the ones that are the
34:50 most successful number one number two
34:52 you should choose something that somehow
34:54 you're destined to do either a set of
34:57 qualities about your personality or your
34:59 expertise or the people you're
35:01 surrounded by your scale your whatever
35:03 your perspective what you're somehow
35:05 destined to do and then number three you
35:07 better you better love working on that
35:09 thing so much because unless so the pain
35:12 and suffering is too great now I just
35:15 describe to you I just describe to you
35:17 nvidia's core values it's as simple as
35:19 that and if that's the case what am I
35:22 doing making a cell phone
35:24 chip how many companies in the world can
35:27 make a cell phone a lot why am I making
35:29 a CPU how many more CPUs do we need does
35:33 that make sense we don't need all those
35:35 things and so we naturally selected out
35:38 ourselves out of commodity
35:40 markets we naturally selected ourselves
35:43 out of commodity markets and because we
35:46 selected amazing markets amazingly hard
35:48 to do things amazing people joined us
35:52 and because amazing people joined us and
35:54 because we had the patience to and and
35:56 let them go succeed to go and do
36:00 something amazing have the patience to
36:01 let them do something amazing they do
36:04 amazing the equation is that simple the
36:07 equation is literally that simple it
36:09 turns out it's simple to say it takes an
36:13 incredible character to do does that
36:15 make sense that's why that's why it's
36:18 the most important thing to learn it
36:19 turns out great success and greatness is
36:21 all about character and no
36:23 fabrication the reason why we don't do
36:25 fabrication is because tsmc does it so
36:27 well and they're already doing it for
36:29 what reason do I go take their work I
36:32 like the people at tsmc they're great
36:33 friends of mine CC's a great friend of
36:35 mine Mark's a great friend of mine just
36:36 because I've got business I can drive
36:38 into it so what they're doing a great
36:40 job for me let's not squander my time to
36:43 go repeat what they've already done
36:44 let's go squander my time on something
36:46 that nobody has done does that make
36:48 sense nobody has done that's how you
36:51 that's how you build something special
36:53 otherwise you're only talking about
36:54 market share thinking about the future
36:58 uh what do you think um when you when we
37:01 think about this decade are these right
37:07 I I don't have an MBA and I didn't get a
37:10 finance degree I read some books and I
37:12 watched a lot of YouTubes I gotta tell
37:15 you nobody watches more you more more
37:18 business YouTubes than I do and so I you
37:21 guys have nothing on me but but are are
37:23 these right answers uh professor
37:27 you're asking the wrong
37:29 person I also didn't study business I'm
37:33 but yes there are the right answers
37:35 eh best CEO yeah right
37:40 eh what do you think about uh AI when
37:43 you're thinking about AI
37:45 applications um and where we're going to
37:47 see change in our lives let's say over
37:49 the next 3 five s years where do where
37:52 do you see that going and in in in
37:55 places where we will all potentially be
37:58 affected in our uh daily experience yeah
38:02 first of all I'm going to go to the
38:03 punchline uh AI is not going to take
38:08 jobs the person who use AI is going to
38:11 job do you guys agree with that okay so
38:16 use AI as fast as you can so that you
38:19 could stay gainfully employed let me ask
38:22 thing when productivity
38:25 increases when productivity increases
38:28 meaning we embed AI all over Nvidia
38:31 Nvidia is going to become one giant AI
38:32 we already use AIS to design our chips
38:34 we can't design our chips we can't write
38:36 our optimizing compilers without AI so
38:38 we use AI all over the place when AI
38:41 increases the productivity of your
38:42 company what happens next
38:46 layoffs or you hire more
38:50 people you hire more people and the
38:52 reason for that is give me an example of
38:54 one company that had earnings growth
38:56 productiv because of productivity gains
38:59 that said guess what my gross margins
39:00 just went up time for a
39:05 layoff so why is it that people think
39:07 about losing jobs if you think you have
39:10 no new ideas then you that's the logical
39:13 thing does that make sense if you don't
39:15 have any more ideas to to to invest your
39:18 incremental earnings then what are you
39:20 going to do when the the work is
39:23 replaced it's automated you lay people
39:26 and so join companies where they have
39:29 more ideas than they can afford to fund
39:33 so that when AI automates their work
39:36 it's going to shift of course it's going
39:38 to change the style of working AI is
39:40 going to come after CEOs right away
39:42 Deans and CEOs that we're
39:44 we're we're so toast um that was good I
39:49 think I think so for I think CEO's first
39:51 Dean second but you're
39:53 close the and so so so you join
39:57 companies where they have more ideas
39:59 more ideas than they have money to
40:01 invest and so naturally uh when earnings
40:04 improve you're going to you're going to
40:05 hire more people um uh AIS so first of
40:09 all this is the giant breakthrough uh
40:10 somehow we've we've we've taught
40:13 computers how to learn to
40:15 represent information in numerical ways
40:19 okay so you guys has anybody heard of
40:21 this thing called word Tove is one of
40:22 the best things ever word Tove you take
40:25 a word you take a you words and you
40:27 learn from the words studying every
40:29 single word its relationship to every
40:31 other word and you learn all you know
40:33 read a whole lot of sentences and
40:34 paragraphs and you try to figure out
40:36 what is the best number Vector what's
40:39 the best number to associate with that
40:41 with that word so mother and father are
40:45 numerically oranges and apples are close
40:48 together numerically they're far from
40:50 Mom and Dad dogs and cats are far from
40:53 Mom and Dad but closer probably to Mom
40:55 and Dad than they are are
40:57 from oranges and apples chair and tables
41:02 and chair hard to say exactly where the
41:04 they lie but those two numbers are close
41:06 to each other far away from Mom and Dad
41:08 King and Queen close to Mom and Dad does
41:09 it make sense imagine doing this for
41:12 every single number and every time you
41:14 test it you go son of a gun that's
41:16 pretty good you know and when you
41:18 subtract something from something else
41:19 it makes sense okay that's basically
41:22 learning the representation of
41:24 information imagine doing this for
41:26 English imagine doing this for every
41:28 single language imagine doing this for
41:29 anything with structure meaning anything
41:31 with predictability uh images have
41:34 structure because if there no structure
41:36 it'd be white noise physically be white
41:38 noise and so there must be structure
41:40 that's the reason why you see a cat I
41:42 see a cat you see a tree I see a tree
41:44 you can identify where the tree is you
41:45 can identify where the the the the the
41:47 coastline is uh where the mountains are
41:49 where the clouds is right and so we
41:51 could we could learn all of that so
41:52 obviously you could take that image and
41:54 turn it into a vector you could take
41:56 videos and turn into Vector 3D into
41:58 vectors proteins into vectors because
42:01 there's obviously structure in protein
42:03 chemicals into vectors genes eventually
42:05 into vectors uh uh we can learn the
42:08 vectors of everything well if you can
42:10 learn everything into numbers and you
42:11 know it's meaning then obviously you can
42:13 take cat word the word c
42:17 a translate it to the image C A there
42:20 this image or cat obviously could this
42:22 the same meaning if you can go from
42:24 words to images that's called M Journey
42:27 stable diffusion if you can go from
42:29 images to words that's called
42:31 captioning video YouTube videos to words
42:35 right under underneath videos and so
42:37 what if you went from uh what do you
42:40 call it if you go from uh uh say say
42:43 amino acids to proteins that's called a
42:47 priz and the reason for that is because
42:49 that's Alpha fold incredible
42:51 breakthrough isn't that right and so
42:53 this is the amazing time the amazing
42:55 time in in computer science where we
42:58 could literally take information one
43:00 kind and convert it transfer it um uh
43:05 generate it into information of another
43:07 kind and so you can go text to text uh a
43:12 large body of text PDF a small body of
43:15 text a summarization of archive which I
43:17 really enjoy right and so instead of
43:19 reading every single paper I can ask it
43:22 to summarize the paper um and it has to
43:25 understand images because under in
43:26 archive the the uh the papers have a lot
43:28 of images and charts and things like
43:30 that so you can take all of that
43:31 summarize it and so so you could now
43:34 imagine all of the productivity benefits
43:36 and in fact the capabilities you can't
43:38 possibly do without it so in the near
43:40 future go you do something like this you
43:42 can say hi I would like to uh uh design
43:46 um uh give me some options of a whole
43:47 bunch of cars I work for Mercedes I
43:50 really care about the Brand This is the
43:52 style of the brand let me give you a
43:53 couple of sketches and maybe a couple of
43:55 photographs of of the type of car like
43:56 to build it's a four-wheel uh SUV uh
44:00 four-wheel drive SUV let's say you know
44:02 so on so forth okay and all of a sudden
44:04 it comes up with 20 10 200 completely
44:08 fully 3D design CAD now the reason why
44:12 you want that instead of just finishing
44:14 the car is because you might want to
44:15 select one of them and you say iterate
44:17 on this one another 10 times and you
44:19 might finally select one and then modify
44:21 it yourself and so the future of design
44:23 is going to be very different the future
44:25 of everything will be very different now
44:28 if you gave that capability to designers
44:30 they would go insane they would love you
44:33 so much they would love you so much and
44:36 that's the reason why we're doing this
44:37 now what's the long-term impact of this
44:40 one of my favorite areas is if you could
44:42 use language to describe a protein and
44:46 you could use language to figure out a
44:48 way to synthesize protein then the
44:49 future of protein engineering is near us
44:52 and protein engineering as you know
44:53 creating enzymes to eat plastic creating
44:56 enzymes to capture carbon creating
44:57 enzymes of all kinds to grow vegetables
45:00 better um all kinds of different enzymes
45:03 could be created during your generation
45:05 and so the next 10 years is going to be
45:07 unbelievable we were the computer the
45:09 chip engineering generation you'll be
45:11 the protein engineering generation
45:14 something that we couldn't imagine doing
45:16 um uh just just a few years
45:21 okay I think we're going to open it up
45:23 for Q&A to the audience so
45:26 uh questions and maybe I'll point and we
45:28 have some mics that will be
45:31 running okay over there we'll start
45:41 there thank you for coming tonight thank
45:43 you question um so are you worried that
45:47 uh mors law business schools are
45:53 serious I understand I I I learn
45:56 understand that that the graduates
45:58 Colombia ends ends up being investment
46:00 bankers and stock Traders I'm actually
46:03 computer science is that right is that
46:05 right and one computer science you'll be
46:06 a Quant you're going to be a Quant and
46:09 so that's what I understand so so I'm
46:11 here selling stock okay and so in the
46:14 future in the future if you if somebody
46:16 ask you what stock to buy
46:18 Nvidia yeah go ahead oh yeah question
46:21 for you is uh are you worried that mors
46:23 law might actually catch up to GP
46:26 industry as it did for companies like
46:28 Intel and can you also explain the
46:30 difference between Mo's law and hang's
46:33 law I didn't I didn't phrase Wong's law
46:36 and it wouldn't be like me to do so um
46:39 the very simple thing is
46:41 this the the um uh Mo's law was twice
46:47 the performance every year and a half
46:49 approximately the easier math to do is
46:51 10 times every five years so every 10
46:53 years is about 100 times um
46:56 if that's the case if general purpose
46:58 Computing microprocessors if general
47:00 purpose Computing was increasing in
47:01 performance at 10 times every five years
47:04 100 times every 10 years why change why
47:07 change the Computing method a 100 times
47:09 every 10 years not fast enough are you
47:11 kidding me if cars would go 100 times
47:13 every five years wouldn't life be good
47:15 and so the the answer is it's in fact
47:18 Moors law is very good and I benefited
47:20 from it the whole industry benefited
47:22 from it the computer industry is here
47:23 because of it but eventually s general
47:26 purpose Computing mors law this the the
47:28 it it's not about the number of
47:29 transistors in Computing it's about the
47:32 number of transistors how you use it for
47:33 CPUs how you translate It ultimately to
47:36 Performance that curve is no longer 10
47:39 times every five years that curve if
47:42 you're lucky is two or four times every
47:46 years well the problem is if that curve
47:50 is two or four times every 10 years the
47:53 demand for computing and our aspirations
47:55 of of using computers to solve
47:58 problems our aspirations our imagination
48:01 for using computers to solve problem
48:03 it's greater than four times every 10
48:05 years isn't that right and so our our
48:08 imagination our demand the world's
48:10 consumption of all exceeds that well you
48:14 could solve that problem by just buying
48:16 more CPUs you could buy more but the
48:18 problem is these CPUs consume so much
48:20 power because of general
48:22 purpose it's like a
48:24 generalist a generalist is not as
48:26 efficient the craft is not as great they
48:29 don't they're not as productive as a
48:32 specialist if I'm ever going to have an
48:34 open chest wound I you know don't send
48:37 me a generalist you guys know what I'm
48:39 saying you know if you guys are around
48:41 just call a specialist all right and
48:50 vet you know it's a generalist look I or
48:54 the wrong specialist
48:56 so so so generalist is too brute forced
49:01 and so today it cost it cost the world
49:03 too much energy cost too much to to just
49:05 Brute Force general purpose Computing
49:07 now thankfully we've been working on
49:09 accelerated Computing for a long time
49:10 and accelerated Computing as I mentioned
49:13 is not just about the processor it's
49:14 really about understanding the
49:16 application domain and then creating the
49:19 necessary software and algorithms and
49:22 architecture and chips and somehow we
49:24 figured out a way to do it behind one
49:26 architecture that's the genius of the
49:30 done that we somehow found this
49:33 architecture that is both incredibly
49:36 fast it has to accelerate the CPU a 100
49:39 times 500 times sometimes a thousand
49:43 times and yet it is not so specific that
49:47 it's only used for One Singular activity
49:51 does that make sense and and you so you
49:54 need you need to be sufficiently broad
49:56 so that you have large markets but you
49:58 need to be sufficiently narrow so you
50:01 application that fine line you know that
50:04 Razor's Edge is what caused Nvidia to be
50:07 here it's almost impossible if I
50:10 explained it 30 years ago nobody would
50:12 have believed it and in fact very few
50:14 did to to be honest it took a long time
50:17 and we just stuck with it and stuck with
50:18 it and stuck with it and and we started
50:20 with uh seismic processing molecular
50:24 Dynamics image processing of course
50:26 computer graphics and we just kept
50:28 working on and working on and working on
50:30 it weather simulation fluid dynamics
50:32 particle physics quantum chemistry and
50:35 then all of a sudden one day deep
50:40 Transformers and then the next will be
50:42 some form of reinforcement learning
50:44 Transformers and then there will be some
50:45 reasonable some multi-step reasoning
50:47 systems and so all of these things are
50:50 are you know we're just one one
50:51 application and somehow we figured out a
50:53 way to create an architecture and solve
50:55 all that and so um will will this new
50:59 end and I don't think so and the reason
51:02 for that is this it doesn't replace the
51:05 CPU it augments the CPU and so the
51:09 question is what comes next to augment
51:10 us we just connected next to it we're
51:13 just connected next to it and so when
51:15 the time comes we'll know we'll know
51:18 that there's another tool that we should
51:19 be using to solve the problem because we
51:21 are in service of the problems we're
51:23 trying to solve we're not trying to
51:25 build a knife and make everybody use it
51:27 we're not trying to build a plier make
51:28 everybody use it we're in service of
51:30 accelerated Computing is in service of
51:32 the problem and so this is one of the
51:34 things that all of you learned make sure
51:37 right make sure that your mission is not
51:41 trains but enable Transportation does
51:44 that make sense our mission is not build
51:49 gpus our mission is to accelerate
51:52 applications solve problems that normal
51:54 computers cannot if your mission is
51:56 articulated right and you're focused on
51:57 the right thing it'll last forever thank
52:00 you okay um up there
52:07 guys that guy right there is Tony go
52:10 Tony am I Tony yeah no all right fine
52:14 I'm Tony today where's Tony Tony was
52:17 that guy in the middle
52:19 right yeah see I I met him just now I'm
52:23 kidding demonstrating my my
52:26 memory I take my chance uh I wasn't I
52:30 wasn't trying to give tony the mic I was
52:32 just demonstrating my incredible
52:34 memory poor Tony go ahead uh thanks
52:39 again um now there's a push for uh on
52:43 shuring the supply chains for
52:47 semiconductors then there are also
52:48 restrictions on the export of high-tech
52:50 to certain countries yeah how do you
52:52 think that would affect Nvidia in the
52:54 short short term but also how would that
52:57 affect us as consumers in the long term
53:00 yeah really excellent question you guys
53:02 all heard heard the question so I
53:04 repeated it relates to geopolitics and
53:05 geopolitical tension and such uh the
53:08 geopolitical tension the geopolitical
53:11 challenges will affect every industry
53:15 human we deeply we the company deeply
53:19 believe in National Security we are all
53:21 here because our countries are are
53:24 secure we believe in National Security
53:26 but we also simultaneously believe in
53:28 Economic Security the fact of the matter
53:31 is most most families don't wake up in
53:33 the morning and say good gosh I feel so
53:36 vulnerable because of the lack of
53:38 military they feel vulnerable because of
53:41 survivability and so um we also believe
53:44 in uh human rights and the ability to be
53:47 able to create a prosperous life is part
53:50 of Human Rights and as you know United
53:52 States believed in the human rights of
53:53 the people that live here as well as the
53:54 people that don't live here and so the
53:58 country believes in all of those things
54:00 simultaneously and we do too the
54:03 challenge with the geopolitical tensions
54:06 the immediate challenge is that if we're
54:07 too unilateral about
54:09 deciding that we decide on the
54:12 prosperity of others then there will be
54:14 backlash there will be unintended
54:16 consequence and and but you know I I'm
54:19 optimistic I I want to be hopeful that
54:21 the people who are thinking through this
54:23 are thinking through all the the
54:24 consequences and unintended consequences
54:27 but one of the things that that has done
54:29 is that it has caused every country to
54:32 believe to deeply internalize The
54:34 Sovereign its Sovereign
54:37 rights every country is talking about
54:39 their own Sovereign rights and that's
54:42 another way of saying everybody's
54:43 thinking about themselves and uh as it
54:47 applies to us on the one hand it might
54:50 it might restrict the use of our
54:51 technology in China and the export
54:53 control there on the the other hand
54:55 because of sovereignty and and every
54:58 country wanting to build its own
54:59 Sovereign AI infrastructure and not all
55:01 of them are enemies of of United States
55:03 and not all of them have a have a
55:06 difficult relationship with the United
55:07 States um we would help them build uh AI
55:10 infrastructure everywhere and so in a
55:11 lot of ways this weird thing about
55:13 geopolitical on it it limits limits the
55:16 market opportunities for us in some way
55:18 it opens Market opportunities for in
55:20 other ways uh for but for people for
55:23 people uh I am I am just really hopeful
55:27 I really hope not hopeful I really hope
55:32 allow that we don't allow our tension
55:36 China result into tension with
55:40 Chinese that we don't allow our tension
55:42 with the Middle East turn into tension
55:45 with Muslims does that make sense we are
55:48 more sophisticated than that we can't
55:50 allow ourselves to fall into that trap
55:53 and and so that that uh I I worried a
55:56 little bit about I worry about that as a
55:58 slippery slope um one of our greatest
56:00 sources of intellectual property for our
56:03 country as you know foreign students I
56:07 see many of them here I hope that you
56:09 stay here it is one of our C country's
56:13 single greatest Advantage if we don't
56:16 allow foreign students the the brightest
56:18 Minds in the world to come to Columbia
56:20 and keep you here in New York City we're
56:23 not going to be able to retain the Great
56:25 intellectual property of the world and
56:26 so this is our fundamental core
56:28 advantage and I I I really do hope that
56:31 we don't ruin that so as you could see
56:34 the geopolitical challenges are real and
56:36 National Security concerns are real but
56:39 so are all of the other economic Market
56:41 social technology matters technology
56:43 leadership matter matters Market
56:45 leadership matters all that stuff
56:46 matters the world is just a complicated
56:48 place and so I don't have a simple
56:50 answer for that um we will all be
56:56 so we'll take one more question over
57:01 yes but in the meantime stay focused on
57:10 study hi there so I actually started off
57:13 working as an engineer at a
57:14 Semiconductor Company had a Brion in
57:17 entrepreneurship and now I'm here as
57:19 someone like yourself that is
57:21 fundamentally a technologist and
57:22 engineer started a company very
57:24 successfully learned Finance from
57:26 YouTube videos what do you think of
57:29 mbas oh uh I I think it's I think it's
57:32 terrific uh you should be first first of
57:34 all uh you you'll likely live until your
57:38 100 and so that's the
57:41 problem what are you going to do for the
57:44 last 70 years or 60 years and so so and
57:47 this isn't something I'm telling you is
57:48 something I tell everybody you know care
57:50 about look to the to the best of your
57:52 ability uh education when you come here
57:55 your force-fed education how good can
57:57 that be after you leave like me I got to
57:59 go scour the planet for for
58:02 knowledge and and I've got to go through
58:05 a lot of junk to get to some good stuff
58:09 here in school you've got all these
58:10 amazing professors who are curating the
58:12 knowledge for you and present it to you
58:13 in a platter my goodness I would stay
58:16 here and you know pig out on knowledge
58:20 can if I could do it again I'd still be
58:23 here you know Dean and me sitting next
58:26 to each other I'm the oldest student
58:29 here I'm just preparing for the big you
58:32 know step function when I graduate just
58:34 go instantaneous to Le success you but
58:37 but my my I'm just a little kidding
58:39 about that you have to leave at some
58:40 point your parents would appreciate it
58:43 and and so so um but don't be in a hurry
58:45 I think I think learn as much as you can
58:48 uh there's no one right answer to
58:50 getting there obviously uh I have
58:52 friends who never graduated from from
58:54 college and they're
58:55 you know insanely successful and so
58:58 there there are multiple ways to get
58:59 there but statistically I still think
59:01 this is the best way to get there
59:03 statistically and so if you believe in
59:05 stat in in math and statistics H stay in
59:10 school yeah go through the whole thing
59:12 and and so I I got a virtual MBA by
59:15 working through it not because of choice
59:18 um I I was when I first graduated from
59:21 school I thought I was going to be an
59:22 engineer nobody says you know hey Jensen
59:25 here's your diploma you're going to be a
59:26 CEO you know so I I didn't know that um
59:29 so I when I got there I think go learn
59:31 it um uh MBA and there's a lot of
59:35 different ways to learn business uh
59:36 strategy matters obviously business
59:38 matters they very different things
59:40 Finance matters it's very different
59:41 things and so you got to learn all these
59:43 different things in order to build a
59:44 company U but if you're surrounded by
59:46 amazing people like I am they they end
59:48 up teaching you along the way and so
59:50 there's some things that that that that
59:53 depending on what role you want to play
59:54 play uh that's critically
59:57 yours okay and so for for a CEO um there
01:00:01 are some things that are critically uh
01:00:04 it's not only my job but it's critical
01:00:07 that I lead with it and that's that's
01:00:09 character there's something about your
01:00:11 character that about the choices that
01:00:14 you make how you deal with success how
01:00:16 you deal with with failure and and
01:00:19 normous said back um how you make
01:00:21 choices those kind of things matter a
01:00:23 lot now from a skill and craft
01:00:26 perspective the most important thing for
01:00:28 a CEO is strategic thinking there's just
01:00:31 no alternative the company needs you to
01:00:33 be strategic and the reason for that is
01:00:35 because you see the most you should be
01:00:37 able to look around corners better than
01:00:39 anybody you should be able to connect
01:00:41 dots better than anybody and you should
01:00:43 be able to mobilize remember what a
01:00:45 strategy is action it doesn't matter
01:00:47 what the rhetoric says it matters what
01:00:48 you do and so nobody can mobilize a
01:00:51 company better than the CEO and so
01:00:53 therefore the CEO's uniquely uniquely in
01:00:56 the right place to be the chief strategy
01:00:58 officer if you will and so I those two
01:01:00 things I would say are are from my
01:01:02 perspective uh two of the most important
01:01:05 things um the the rest of it has a lot
01:01:07 of you know skills and things like that
01:01:09 and you'll learn the skills um and maybe
01:01:12 if I could just add one more thing I do
01:01:15 believe that that a company is about
01:01:19 some particular craft you make some
01:01:21 unique contribution to society you make
01:01:25 and if you make something you ought to
01:01:28 it you know you should you should
01:01:30 appreciate the craft you should love the
01:01:32 craft you should know something about
01:01:34 the craft uh where it came from where it
01:01:37 is today and where it's going to go
01:01:38 someday you should you should try to
01:01:40 embody the the passion for that craft
01:01:43 and I and I hope today I I did a little
01:01:45 bit of embodying the the passion and the
01:01:48 expertise of that craft I know a lot
01:01:50 about the space that I'm in and and so
01:01:53 so you you know if if it's possible the
01:01:56 CEO should know the craft you don't have
01:01:59 to have founded the craft but it's good
01:02:01 that you know the craft there's a lot of
01:02:03 craft that you can learn and and so you
01:02:05 you want to you want to be a uh you want
01:02:07 to be expert in that field but those are
01:02:09 some of the things you can learn that
01:02:10 here um ideally uh you can learn that on
01:02:13 the job you can learn that from friends
01:02:14 you could learn that by doing you know a
01:02:16 lot of different ways to do it but stay
01:02:18 school so um before I thank the best CEO
01:02:24 uh I want to thank the digital future
01:02:26 initiative the David silen uh speaker
01:02:28 series uh but mostly thank you Jensen
01:02:31 for coming uh we all understand why you
01:02:33 were voted the best CEO now thank you
![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)
