YT YouTube2Text

YouTube2Text → Transcripts → Elon Musk STUNS In New Interview With Predictions & Opens Up (With Timestamps)

Elon Musk STUNS In New Interview With Predictions & Opens Up (With Timestamps) — Transcript

by Matt Pocius on Tesla Stock & Money · 10,026 words · 1,453 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:00When I was preparing for this
0:02conversation, one of the things I just
0:05visualized was ET. So ET going across
0:09the moon, you know, with a boy on a
0:11bicycle and a a contraption to
0:14communicate to call home. Obviously a
0:18predecessor to Starlink. So I guess that
0:22didn't work.
0:26So, so uh so Elon has uh uses a
0:31principle uh you know first principles
0:33about everything should be done the
0:35right way not the fastest way not the
0:38quickest way not the cheapest way the
0:40best way and everything looks at all
0:42over again. So if someone makes an
0:44investment on his behalf uh for one of
0:46his businesses uh it's always viewed in
0:49the terms of is this the best we can do
0:52best we can do the fastest we can do uh
0:55the most effective so uh so one thing I
0:59would say is that he's a builder he's
1:01probably the only builder I'm aware of
1:04United States who builds stuff but
1:06builds stuff that's not just regular
1:08stuff like 10 million square foot
1:10buildings 5 million 20 million And now
1:13he's got a 100 million square foot
1:14building for Terraab that he's thinking
1:17about or going to build and uh and and
1:20so the question and no one else has done
1:23this. No one. And uh anything like it.
1:27And so the question is of course and
1:29when you're doing such a thing then you
1:31get scaling benefits. You get to buy
1:34things uh you produce things uh much
1:37more effectively than you could if you
1:38were in a smaller entity. So question is
1:41how do you manage to do this and why
1:43didn't anyone else ever try to do
1:44anything like this? We're trying to get
1:46silicon to come back to United States
1:48again. We need it. We need power. We
1:50need silicon.
1:52No one's doing this except for you. How
1:54come?
1:54>> Well, that's actually that's not quite
1:57accurate in that TSMC and Samsung are
2:00building fabs in the US. So TSMC is
2:03building large collection of fabs in
2:05Arizona. Some of them are, I believe,
2:07operational.
2:08And Samsung has some operational fabs in
2:11Texas. Um, and is is building or has
2:14built, I should say, a fab for for Tesla
2:17that will be dedicated to the uh uh AIS5
2:20and and AI6 uh chip designs that Tesla's
2:24done. So, but relative to what's being
2:28done in Asia, the amount of chip
2:31fabrication in the US is very small. And
2:33even if you take the best case
2:35assumptions of all the chip fabs that
2:37that have been announced that will be
2:40built in in America, that's still a tiny
2:42fraction perhaps on the order of 10%
2:45optimistically 10% of what's in Asia. So
2:50uh as I think most people know, Taiwan
2:52does an incredible amount of heavy
2:53lifting on chip production. Uh TSMC is
2:56an amazing company u truly incredible
2:59that the sheer number of a variety of
3:01chips and the magnitude and the volume
3:03is incredible for that that TSMC does.
3:06Same thing for Samsung and and and I
3:09think Micron does does a a great job too
3:12on the memory side. So um but the but
3:16it's overwhelmingly built in Asia. Um
3:20the
3:22so so my my my concern really is that
3:26there is always like some geopolitical
3:28risk in that region. Um and I think that
3:32that risk may come to fruition at some
3:34point in which case uh if we don't have
3:38all the pieces necessary to build chips
3:40at scale in America then um we may not
3:44have any chips at all or very few the
3:4610% number type of thing. Um and and
3:51then there's the the the maybe looking
3:53ahead, you know, past any sort of
3:57medium-term geopolitical bumps in the
4:00road, uh we're we're simply even if you
4:04if you exclude geopolitical crisis
4:06issues, the rate at which we're making
4:09chips is simply too slow.
4:12So if you if you want to get to say a
4:14terowatt a year of chip production which
4:17is our goal that's the reason it's
4:18called the the terapab it's sometimes
4:21people when I say terap there's you know
4:23three possible meanings to terapab one
4:24is terror t e r r o r [laughter]
4:29which I think sometimes people like why
4:31would you want a terapab um oh and then
4:34there's t e r r a which just means earth
4:37um but it's actually t e r a uh as in
4:40terowatt Um, so but arguably all three
4:44apply. So
4:48[clears throat and laughter]
4:50it's terrifyingly large. Um, so
4:54so so the rest of the world is making
4:56we'll make this year around 20 GW of AI
4:59chips, maybe 25. Um, and then next year,
5:04depending on who whose projections you
5:06believe, it's maybe around 35 to 45 GW
5:10of of AI chips next year. Um, and um,
5:17but but if if you want to do a a
5:19terowatt, that's now you need you need
5:2130 times what current global production
5:25will be roughly next year
5:27>> for just us.
5:29We're building that for us.
5:32Well, yes, we're building
5:33>> I mean we could sell it to other people
5:35too, but basically we're building 30
5:37times what the annual production is in
5:39one plant, which you think is going to
5:41do as val as big production as the GDP
5:45in the world as a GDP for the 20th
5:47largest country. One one plant.
5:50>> Yeah, it might even make the top 10. Uh
5:53so um now no obviously you know a plan
5:58like this uh I think um if it if it if
6:02it fails it'll be crazy and if it
6:04succeeds it'll be audacious. Uh so it
6:07sort of depends on the outcome. I think
6:09it will succeed at least to some degree.
6:10Um
6:11>> hope so.
6:12>> So
6:13yeah. Yeah. I mean
6:14>> it's going to cost a trillion dollars.
6:20probably more than that um over time.
6:24Um but we won't need that money
6:26immediately. We'll be built in phases.
6:28Um and and we are looking at uh ways to
6:32change the manufacturing process for for
6:34chips to uh improve the efficiency of of
6:37chip production as we've done in many
6:40other arenas. Um
6:42you know like just one example like like
6:44right now u the amount of time that a
6:47wafer in a chip app spends simply being
6:49transported and waiting uh is more than
6:53half the time. So more than half of what
6:55a chip is doing in a chip pab is
6:59nothing. It's it's either being moved
7:01around or or waiting for a machine. So
7:05uh and and I think I think there's
7:09basically a lot of a lot of potential
7:11for improvement in chip fabrication.
7:13>> Why don't these people do it themselves?
7:16If they're if they have plants just sit
7:18with chips half the time just not
7:20moving. What's to stop them from
7:22improving their own systems?
7:25>> Well, no. Actually, if you if you visit
7:26a chap a fab, you'll see that the the
7:29the f the the wafers are moving a lot.
7:32they're zipping all over the place. Um,
7:34and and waiting, but but but the the
7:37amount of time that that atoms are being
7:39changed on the wafer in a way that is
7:40useful is is is less than half the time.
7:43Um, and I'd say sometimes it's maybe 20%
7:46of the time that that that something
7:48useful is happening to the atoms on that
7:50wafer. Um the the the reason that
7:54they're they that they don't want to try
7:56something adventurous or or with with
7:59chip apps is that these things cost a
8:01tremendous amount of capital and and and
8:04if if they have a recipe that works.
8:06They just they simply don't want to risk
8:08u something new. So the the it generally
8:12the preference in the chip industry is
8:14to repeat uh what they did last time
8:17with minor changes and and not uh take
8:21any risks.
8:23So we we we're building a research lab
8:28in Texas just in Austin at the at
8:32Gigafactory Texas. That research lab is
8:35going to try many different approaches
8:37to chip fabrication. Um and and many of
8:41those approaches won't work, but some of
8:44them I I'm sure will work. And then we
8:47will translate those improvements to the
8:49the tariff itself, the production
8:50tariff, which is um in uh Grimes County
8:54in Texas near Texas A&M. Um and and
8:58that's that's that's where we'll scale
9:00up production um of our AI chips. the I
9:04think one of the things that's going to
9:06be interesting about the research lab in
9:08Austin is that I don't wonder if we will
9:10have the ability to make um uh a do do
9:16logic memory packaging and do the
9:19lithography
9:21uh uh so so that we can have we can
9:25design new chips and and in under one
9:28roof make the make the whole chip test
9:30it make the whole chip test it with a
9:33very fast iteration cycle.
9:36Um something that would normally take a
9:38year, I think we we may be able to
9:40reduce to a month.
9:43Uh so
9:45for any given technology endeavor, the
9:48the pace of progress is determined by
9:50how many it how many iterations can you
9:52do and how much progress can you make
9:54between those iterations.
9:56Um so if if [clears throat] you can
9:59improve the speed of iteration and the
10:02and the and like how many iterations can
10:04you do per year or per unit time and the
10:06progress between each iteration in any
10:08given technology you will uh you will
10:10accelerate your progress at a rate far
10:13greater than anyone else. If if you are
10:16you know if you're able to iterate in
10:18order order of magnitude faster you as a
10:21first approximation will advance your
10:23technology order of magnitude faster.
10:24So, so most people are there's a
10:29tremendous amount of inertia in in
10:31business because if someone makes an
10:35effort uh to spend money and it's not
10:38successful uh then that's a black mark
10:40on them and if it is successful and
10:43they're trying to get the top job of the
10:45business then it just means that they
10:47have to work harder. So if they're
10:49successful, they work harder and if it's
10:51if it fails, uh they're finished for
10:55their opportunity for the top job. Uh
10:57Elon has established a culture in his
11:00business business is where it's always
11:04about come to me with ideas and then if
11:07it's success if you if if you think it's
11:09successful and I believe that's the
11:11case, then we'll throw away what we've
11:13done before and start all over again.
11:14It's always about doing the best thing.
11:17One of the things that's interesting to
11:18me about your culture is that you
11:22introduced me to uh key uh key luck uh
11:26the person who's in charge Intel who's
11:29now going to be uh have an interest in
11:32terap and so I spent some time with him
11:35and he comes to visit me and I said so
11:38so what's it like to work with Elon he
11:41says well other people come to visit my
11:44facility in Oregon and they come and
11:47they walk around for an hour and a half
11:48and they leave and Elon comes and then
11:51he sits there he was there for four and
11:53a half hours and he interviews one
11:55engineer after another after another
11:57after he's questioning everything that's
11:59all concept so everything he has to do
12:01with speed and then when I went to uh
12:04Texas to Bokhica uh to visit uh their
12:08car factory that they were building
12:10there uh they intro while it was under
12:12construction uh they introduced me to
12:14the head of that facility the manager,
12:17giant man. And and of course, everyone
12:20seems like a giant to me, but he was a
12:22he was a really giant guy. And he had
12:24been previously the head of production
12:26at Ford. And so I said, "Well, so what's
12:28it like difference between Ford and
12:30Tesla?" He said, "Well, uh, when I was
12:33at Ford and I had a change, I had to go
12:36through six different levels to get a
12:38change approved. When I have to have a
12:41change here, then all I do is walk
12:42upstairs and there's Elon. And so when a
12:45plant is something under construction,
12:47Elon's not some distant guy. He's there.
12:50He even has an airirstream. He lives in
12:52an airirstream. Do you know any other
12:54executives of any big companies,
12:57potentially the largest company in the
12:58world, going to live in an airirstream
13:00to make sure that all the questions that
13:02are taking place in making a facility
13:04are going to be answered fast. So um you
13:07know it's amazing culture, amazing
13:09person um that uh and amazing people
13:12who's recruited. So this whole people
13:15who work there, they're just brilliant
13:17people. It's like harder to get a job
13:19there than get into Harvard. Of course,
13:21I don't know why people would go to
13:22Harvard anyway anymore.
13:25>> Yeah. Well, but by the way, Sreams are
13:27actually pretty nice. So I wouldn't say
13:28I'm like [applause]
13:30No, it's not that bad. Honestly, I don't
13:32kind of like my little estream. Um but
13:35it's it's it's it's it's actually parked
13:37right next to the power plant that we so
13:39we built this power plant that's
13:41producing over a gigawatt. Um and uh
13:45probably it'll you know go to one and a
13:46half to 2 gawatt of internal power plant
13:49production in about a fifth of the time
13:52that such a thing would normally be
13:54built. So built in about a year as
13:56opposed to normally be about 5 years for
13:58that. Um, and it it's not it's not as
14:01though somebody has to come to me uh
14:03upstairs. There there is no upstairs. So
14:04we have no upper floor in in the uh at
14:07our Memphis classes two thing. Um I I
14:09walk I walk the entire site uh
14:13sometimes two sometimes three summers
14:15four five times a day. So uh the entire
14:19site, the power plant and all the
14:21buildings. Uh, and
14:23uh, I also insist on doing at least uh,
14:26uh, some of the work myself manually.
14:30So, uh, so I've like I've pulled power
14:33cables the size of a an anaconda. Like
14:35they're incredibly heavy um, and long.
14:39And uh, I've I've um done fiber optic
14:43cabling. Um, I I've done electrical
14:46switches. Uh I've um worked on the on
14:50the power plant management software. It
14:52it's so I I I try to uh do at least a
14:56little bit of everything so that I'm not
14:58uh numb to the pain of the people who
15:00are actually doing it and I have some
15:02understanding of the difficulty and what
15:03can be done to improve the situation. Um
15:07so it's not as though I'm sitting in an
15:09office and somebody has to come to me.
15:10I'm walking the floor asking them
15:12questions actively.
15:15>> It's amazing. Um
15:18so let's talk a little bit about data
15:19centers which are now in the news and
15:22people
15:23>> yeah popular topic [laughter]
15:26how to how to how to make friends and
15:28influence people you know data centers
15:30are you know the pitchfork level for
15:32data centers is pretty high. Yeah, I'd
15:36say uh so so [laughter]
15:39but but but
15:42if you want to have AI, you got to have
15:44data centers. And so let's talk about
15:46them first on this planet. So so Elon
15:50believed that the uh the roadblock uh
15:55that would be in the way of data center
15:57construction would be having enough
16:00chips, enough compute actually. And so
16:04so what he did is that a lot of people
16:07we have Grock so he was at the forefront
16:10of open AI of anthropic those guys are
16:13his guys and then uh of course it
16:16doesn't like open AI anymore but uh
16:20>> yeah well I I I did start open as a non
16:23as an open source nonprofit and now it
16:25is a closed for maximum profit uh entity
16:28so it's literally the polar opposite of
16:29what I started
16:30>> right
16:31>> and then anthropic
16:33was uh founded by seven people that left
16:37uh OpenAI because they were they they
16:40didn't trust Sam Alman for good reason.
16:42Um they said this publicly as it's not a
16:44secret and and so effectively anthropic
16:47is an offshoot of OpenAI. Um so uh yeah
16:53it's I guess I should have just started
16:55OpenAI as a regular COP instead of u as
16:59a as a nonprofit. Um but uh you know I
17:03thought this would be something that
17:05would be a philanthropic thing for the
17:06world. I didn't anticipate it would one
17:09day turn into a a for-profit um that is
17:13closed source. Um
17:14>> and when you left because you were
17:16disappointed in direction they offered
17:18you an ownership interest and you said
17:20no I don't want it and so you left.
17:22>> Well yeah I mean well I I I was confused
17:25as to how you could have a commercial
17:26interest in a nonprofit. That that
17:28didn't seem to make sense to me. I I you
17:30know
17:32um you know I provided almost all the
17:34capital in the beginning uh for the
17:36company. It was my idea and I named it
17:38and recruited the key people and taught
17:40them everything I know but apart from
17:42that nothing.
17:45So to go back to data centers. So you
17:48thought that the idea about putting all
17:52this AI in operation was that we need
17:54data centers and people want to build
17:56them and it takes them ordinarily about
18:00two and a half years to build one and it
18:02costs maybe $60 billion to build it. 50
18:05or 60 billion to build a gig. And then
18:08the people who build them for
18:10hyperscalers and rent them, in order for
18:12them to get the capital, they need to
18:14borrow it or get it in some way that
18:16they have to have long-term leases, five
18:18years at least to get them. So the rents
18:21people pay in fiveyear leases or maybe
18:24$12 billion a year for $60 billion
18:27investment. What we do instead is that
18:30we build them. And we build them because
18:32we have the use for them because we're
18:34vertically integrated for our cars, for
18:38our robots, uh uh for our rockets, uh
18:41for for everything. And uh and because
18:45we need them, then we can take risk of
18:49building them without knowing that we
18:51have contracts for other people. And
18:53what happens though is we have them
18:54built and they're not. So it takes us
18:59somewhere was taken us four months. has
19:00now taken us eight or nine months to
19:02build. So it takes them two and a half
19:04years, takes us 8 months, 9 months, cost
19:06them 60 billion, cost us 35 billion. And
19:10the demand for compute is so great that
19:13people who built centers, they're dark
19:15because they don't have the power to
19:16turn them on. And uh and we'll rent them
19:19to compute until we need it for
19:21ourselves for Grock and we'll rent it to
19:24them for somewhere around 50 billion a
19:26year. So, we're getting 50 billion a
19:29year for $35 billion investments. So,
19:32the revenues for SpaceX are about to
19:34explode.
19:36Going up.
19:37>> Yes, they are exploding. It's uh I I
19:39think yeah, we Q2 to Q3 revenue increase
19:43was I think over 80% this year. Uh I I'm
19:48not sure what the exact numbers will be
19:49Q3 to Q4, but but probably something
19:51like 70% uh Q3 to Q4. And I would I
19:55would say maybe it's something similar
19:57you know Q4 to Q1. Um the like it it is
20:02extraordinarily difficult to uh bring on
20:05compute at scale. Um
20:09it's it's and I think there there's a
20:11lot of speculators out there who think
20:12you can simply order some GPUs uh hand a
20:16purchase order to some company to build
20:19the data center and just uh rake in
20:21cash. Um, and the reality is it's
20:24insanely hard. Um, so there there's a
20:27meme uh that I think is quite funny u
20:30which I think it's called the Jose meme
20:32where there's you may have seen it
20:34before that there's one guy digging the
20:36hole in the road and there's like eight
20:38people watching him dig the hole. Um so
20:43now in that meaning like there's one
20:45there's one person doing actual work and
20:47one and then eight people watching the
20:48guy do the work. Um now and unless you
20:52increase the number of hoses in in the
20:54country then uh you the amount of work
20:56done does not increase uh and and so
20:59adding more people who who simply you
21:01know [snorts] issue purchase orders to
21:02Jose does not result in more things
21:05occurring only if there are more Jose's
21:07more work so I aspire to be Jose um and
21:12SpaceX wants to be is Jose we do actual
21:14work this is a this is a really big deal
21:19and and um and and with the massive
21:22expansion and construction in the United
21:23States, the frankly it's not as though
21:26there's been this massive influx of
21:27people who do actual work at at a
21:29physical level, but but there have been
21:31there's been a massive increase in the
21:32number of people who issue purchase
21:33orders. So that means that the people
21:35doing actual work are spread incredibly
21:36thin. Um and and so you have a large
21:40number of partially completed uh data
21:42center projects throughout the country.
21:45Um because the the at this point the
21:47peanut butter is spread so then you can
21:49barely taste it. Uh the the the
21:51difference is that at SpaceX we actually
21:53are used to doing incredibly challenging
21:56physical work in terms of building giant
21:58rockets and uh and and vast launch
22:02complexes. Um and and uh and and Tesla
22:06is also very good at large scale
22:08construction and and probably the two
22:11best companies at physical hardware on
22:13earth outside of China are SpaceX and
22:15Tesla. I don't even know who would be
22:17number three.
22:19Um and and you know I think even
22:22Tesla might be number the best in the
22:24world even including China where China
22:26is incredibly good. It's worth noting
22:29that, you know, the Tesla Model Y is is
22:31the bestselling U EV in China.
22:33>> Well, you taught them a lot in China.
22:35>> This is like this is like, you know,
22:37winning a land war in Asia. Classically
22:38difficult. Uh and and yet we we kind of
22:42are actually um in that Model Y
22:45bestselling car in China. Think about
22:47that for a second. And let me tell you,
22:49Chinese car companies are super good.
22:52>> Yeah. The person from Ford went there.
22:53He said he was blown away by the quality
22:55of the cars in China. Uh yes, partly
22:58>> China's awesome. I love China, by the
23:00way.
23:02It's it's inspiring to go there.
23:04>> Well, most people who build in China,
23:06they have to have Chinese ownership of a
23:09significant portion of the operation in
23:11China. And they were so anxious. Ch 49%
23:15is what you're allowed to own in China
23:16normally. And Tesla, they wanted us to
23:20be there so badly that they let us go
23:22100% ownership for us.
23:25>> Yes. Um that took quite a few years of
23:27meeting with with with Chinese officials
23:30and I actually you know some of the
23:32people in the government actually you
23:33know they actually literally yelled at
23:34me um like spitfing type of thing that I
23:38that we had to get sort of you know
23:41married to a Chinese partner and I'm
23:42like well we're just a baby company and
23:44I don't think babies should get married
23:46you know cuz that would look funny going
23:47down the aisle with a baby and
23:50uh and they and they're like you're not
23:51a baby and I'm like no we're we're a
23:53baby and and you know babies should
23:56just, you know, not get married. It's
23:57too young. So, uh, then eventually they
24:00they they said, "Okay, fine. We'll we'll
24:02change the law and and then you can own
24:03the factory." And and mostly it was like
24:07we just need to be able to move fast,
24:09you know, and and make decisions and and
24:10if if you've got to always ask someone
24:12else for permission to do the decisions,
24:14then how you going, you know, it's
24:16difficult to innovate um if you don't
24:18actually have your hands on the steering
24:19wheel. Well, these days you don't need
24:21your hands on the steering wheel because
24:22of self-driving, but that metaphor is
24:24not going to make any sense in the
24:25future.
24:26>> The uh so so we're talking about
24:29revenues going up a lot for SpaceX. It
24:32was I think 15 or 16 billion last year
24:35and this year it's going to be ending
24:37the year 26 at uh at a hundred billion
24:41dollars or so of annualized revenues.
24:45And then next year we think it's going
24:47to be between 500 billion and a
24:50trillion. So we've gone from 15 billion
24:54to somewhere around a hundred billion
24:55annualized to 500 billion to a trillion.
24:59And that's just really just getting
25:00started. And what's really cool about
25:03selling renting renting our data centers
25:07to competitors
25:09that basically they're paying us 50
25:11billion for something cost us 35 and
25:13we're going to have collections of cash
25:16coming in for renting data centers that
25:18we still own. So we got them for free,
25:21right?
25:22>> Uh well um I mean
25:25>> sort of
25:26>> I think I think just just to comment on
25:27a few of those things. So um we we are
25:30tracking to um have an annual rate of
25:33revenue
25:35uh that is in excess of hundred billion
25:38in December. Um obviously December is
25:40pretty close so I have good incent. So
25:42unless unless there's some sort of you
25:44know natural disaster or something we we
25:47will we'll meet that target. Um and and
25:50then based on the plans that we have
25:52next year, um it seems quite likely that
25:55our annual recurring revenue in December
25:58of next year will will exceed 500
26:00billion. Now um this is just my this is
26:04this is not a promise. This is just this
26:06is what I think will happen. That this
26:08is my best guess. I just sort of want to
26:10distinguish between this is you know
26:12like a promise to do something uh versus
26:15what I think will occur. What I think
26:18will occur is an an ARR in December of
26:22next year of of 500 billion. And and and
26:25and I and I simply said a trillion is
26:27not out of the question. Um meaning it
26:30it was previously out of the question
26:32and it is no longer out of the question.
26:33It's maybe 2% or something like
26:36likelihood. But but it's interesting
26:38that that it's not 0%. That's the
26:40interesting part. Um and and the
26:42probability is increasing with time. Um
26:45so um I mean my as a sort of
26:51what my biological neural net predicts
26:54for
26:56digital super intelligence in terms of
26:58of its effect on the global economy is
27:00about a 20 to 30% uh increase this is in
27:04the abs without taking robotics into
27:07account um I think I think around 20 to
27:1130% increase which is which would be
27:12roughly
27:14call it a $25 trillion increase in the
27:16global economy.
27:18And this will happen very fast.
27:21>> So that's $130 trillion global economy.
27:24And you think that digital AI is going
27:26to increase that by 25 30%.
27:29>> 20 to 20 to 30%. Uh yes, that that's
27:33that that's my guess. Okay. So well,
27:35what's your guess for how much would
27:36increase the economy? I think probably
27:3720 to 30%. Um I mean I think like and it
27:42could be higher than that. Um but it
27:44probably doubles the productivity of
27:45white collar work.
27:47>> So that's just digital. That's not
27:49physical. That's digital. Digital AI.
27:53>> Physical is something that we're going
27:54to be dominant in.
27:57>> I think
27:58>> yes. Well, the way to think about
28:00physical AI in the form of robotics is
28:02that um you like any given economy is
28:07the number of people in it times the
28:08average productivity per person. So if
28:11you have a humanoid robot that is
28:13extremely productive and can work 24/7.
28:16So it's working 168 168 hours a week. Um
28:20it doesn't take vacation or bathroom
28:21breaks or need to eat food or or
28:23anything like that. And um it's it's uh
28:28this the speed at which you can operate
28:29is faster than than human speed. Uh you
28:32can expect roughly a five you know a
28:35productivity per week that is uh on
28:38average five times better than a person
28:41roughly because a person's like maybe
28:43working 40 hours a week and and and does
28:45need to take breaks obviously um and
28:47doesn't move as fast as a robot. So, so
28:50I say a minimum productivity expectation
28:52for human or robot would be five times
28:54that of a human then um of of the you
28:57know say take the population of the
28:59United States say how many people are
29:00either in school or retired well that's
29:02about half of the people about half of
29:04people are either in in school or
29:07retired and then of those who are
29:08working age not everyone works um or is
29:12engaged in in highly productive
29:13employment um so so then you're you're
29:16left with you know perhaps um around I
29:20don't know 120 million people in America
29:23would that have a net uh positive output
29:26of goods and services. Um so so then you
29:29say well how many robots then would be
29:31needed to match that productivity? Well,
29:33I guess you you'd need um you know about
29:3720 to 30 million robots
29:39to to match the goods and services
29:42output of the United States.
29:46And we would build those. It's less
29:50expensive to build a robot than to build
29:52a car.
29:54>> Yes. Initially, I'd say about the same.
29:58Uh but then over time, uh it would be
30:00much less.
30:00>> So 30,000 cost for a robot on the way to
30:0320.
30:04>> Yeah.
30:05>> And we would rent those or we would sell
30:07them or what would we do?
30:10>> Do we have an idea yet?
30:11>> I think I think probably rent them.
30:16um to stay on the data centers. Then if
30:20people are really upset about data
30:22centers in their communities, is there
30:25any way that since these are so
30:28profitable for us? Why wouldn't we just
30:31say, "Okay, here your utility bills in
30:35your community will stay flat and we'll
30:38subsidize it and we will use that
30:41subsidy and we will also give you $50
30:44million for your schools and hospitals."
30:47Do we have any programs like that where
30:50we have gone to communities and say yes,
30:53we are going to help you in your
30:55communities uh and we'll give you more
30:57jobs. We're really a job creating
30:59company. No one creates jobs the way we
31:01do. Is there anyone any way I don't
31:04think we've done I don't think our
31:05industry's done a good job about
31:08explaining how this is the benefit of
31:10people not harming them.
31:14>> Yeah. Well, actually locally um in in in
31:17the Memphis area I think we're actually
31:18quite popular. Um it's it's actually
31:21people it's oddly the popularity is
31:23there where where the lack of popularity
31:26is where is where there aren't data
31:28centers and they think the data center
31:30will have a bad effect. Um but but in
31:33Memphis uh you know Memphis is booming
31:36because of of SpaceX. Um the restaurants
31:39are full, the hotels are full. Uh we're
31:42we're paying at times twice the the
31:45prevailing wage that people were earning
31:47before. Why don't people write about
31:49this? Why aren't there stories about
31:50this?
31:52>> Why is this not
31:52>> There are some there are some stories
31:54that affect We've also uh helped out
31:56with local schools and uh you know it's
32:00and and and the tax revenue that uh the
32:03county and the city are earning is is
32:05tremendous. Uh so um
32:09so I so I I think I think this maybe
32:11this this popularity thing is a
32:13temporary concern because because there
32:16actually is tremendous prosperity that's
32:18brought by the data centers um and um
32:21and if you talk to the uh the people in
32:24town like there the they all feel it
32:28because their businesses are booming.
32:31So, so then go to data centers in space,
32:34the obvious solution, so people don't
32:37have to worry about them in their
32:38communities anymore and if they want if
32:41they don't want to. And and for us, it's
32:43less expensive to put them in space. And
32:46every time we put up new generations,
32:48they become more and more powerful than
32:50they had been before. Isn't that And
32:53everyone says, "Well, it's 10 years
32:55away." And Elon says, "No, we're
32:59starting. You know, Google has already
33:00tested with our sat with a satellite
33:02that we've run for them. It's been
33:04effective and so we're going to be
33:05starting to put them up in 27 or 28. How
33:09long before impact?
33:11>> Yeah. So, yes, there's there there's
33:14currently um people have tested u NVIDIA
33:19GPUs. So there's there's a satellite
33:20with an Nvidia GPU up there that we
33:22launched and and there's a a satellite
33:24we recently launched with a Google TPU
33:26that and and these are operating to the
33:29best of my knowledge that they're
33:30working fine in space. U we expect to
33:33start launching the first Star Mind uh
33:36computers uh which we've co-designed
33:38with Nvidia uh probably I'm just
33:41guessing here but in the third quarter
33:43of next year. Um and we'll we'll start
33:45off slow but but then it will ramp to
33:47significant volume probably in in 28 and
33:49then and then very very large volume in
33:5229. Um and um and this this really in
33:57order to get to the terowatt per year
33:59scale or even I'd say maybe the 200
34:02gawatt per year scale, you really need
34:04to go to space because they're just
34:07there there isn't enough uh power on the
34:10ground to to realistically get to those
34:12levels per year. Um just as a frame of
34:15reference the power output the average
34:18power the average power consumption in
34:20the United States is just over 500 GW.
34:23Um so if you look you know the peak
34:27power capable in United States is
34:28roughly 1.2
34:31uh terowatt but average power
34:33consumption is just over 500 gaw. So,
34:36so, so you can really see that like if
34:38if you're doing say 200 gawatt a year of
34:41AI, that would be like doing a 40%
34:44increase in the entire US power grid
34:46every year,
34:48which is simply not feasible.
34:52Um, and and when you when you take solar
34:54panels and you and you put them in
34:56space, uh, because it's essentially it's
34:59always sunny in space and the solar
35:01panels can always face the sun directly.
35:03they're not they're not not off angle to
35:05the sun. Um
35:08uh the uh and and you don't have atmosph
35:10at atmospheric attenuation of of solar
35:13power. Um so you get 30% more solar
35:16power in space. It's sunny all the time.
35:18There's no seasonality. Uh you don't
35:21really need you can have you can get
35:22away with a very tiny battery pack
35:23because you don't need to last through
35:24the night. Um the the cost of of
35:29intelligence in space is dramatically
35:31less than the ground. once you have a
35:34low cost per turn to orbit. So Starship
35:37because it will be the first fully
35:38reusable rocket um will um lower the
35:42cost of access to space by a factor of
35:45100 or maybe more. Um I I I think
35:49actually in the limit the
35:52I mean maybe within the say probably 2
35:54or 3 years the cost of a ton to orbit
35:57will be less than the cost of air
35:59freight from California to Australia.
36:02So NASA said that in 2010 the cost to
36:06get to orbit was $18,500
36:10a kilogram. That was because you use a
36:12rocket one time. And then when we came
36:15up with a rocket Falcon 9 you could use
36:17over and over again, it became I think
36:20it was $2,500
36:22and then it went to 1500
36:24>> and now it's going to go to 100 or 15 or
36:2750. I think we get to um under $100 a
36:30kilogram to orbit.
36:33So now it's it's important to just
36:37appreciate how profound
36:39full reusability is. Um so with with
36:43Falcon 9 we have I I guess probably
36:46around 80% reusability. So the booster
36:49is you know comes back and lands. You've
36:51probably seen the videos of the Falcon 9
36:53booster landing and then the fairing or
36:55nose cone that contains the satellites
36:58uh that that those those fairing halves
37:01land by parachute in the [clears throat]
37:03ocean and we refly those but we do not
37:06refly the upper stage. Uh so we're
37:08unable to recover the upper stage of
37:10Falcon 9 and the cost of the upper stage
37:12is roughly the cost of a mediumsiz jet.
37:15So that means the the minimum cost per
37:17flight for Falcon 9 is is a mediumsiz
37:20jet. Now, that's still much better than
37:21anything else. Uh, but it's safe to say
37:24that there wouldn't be a lot of air
37:26travel if every time you flew somewhere,
37:28you had to throw away a mediumsized jet.
37:30You like be like, I think I'll drive,
37:33you know. [laughter]
37:34Um, so um,
37:39so now with with Starship, the entire
37:42system is designed for reusability.
37:46So the and and and and not just
37:49reusability but but extremely rapid
37:51reusability. The the reason to to to
37:54catch the booster and the ship with the
37:56tower arms we call the we call as the
37:59chopsticks. Um it's kind of like Mr.
38:01Miyagi and Karate Kid, you know,
38:02catching the fly, but just make it
38:04bigger. Um and um the the reason for
38:08that is is is that the rocket takes off
38:10from the launch pad. the So the rocket
38:12is placed on the launch pad by the tower
38:15arm and then it is caught by those same
38:17tower arms and placed back on the launch
38:19pad when it returns from space.
38:22So you don't just have full reusability,
38:25you have extremely rapid reusability.
38:27You really couldn't get faster
38:28reusability than landing back where you
38:31took off.
38:32Um and um this in principle would would
38:37allow us to uh refly the the rocket, the
38:40giant booster, the super heavy booster
38:42um once an hour,
38:46>> 10,000 launches a year.
38:50>> Yes, probably more in the in the end,
38:51but yes, uh more than a launch an hour.
38:56So when I was young, people would line
38:58up to watch every now very rarely a
39:02rocket. And we can do one an hour.
39:05Doesn't even make newspapers anymore.
39:08>> Yes. Well, this will be quite quite I
39:11mean, you really want to watch this one.
39:13This one's cool. Um, I mean
39:17it's it's the largest uh flying object
39:21ever made and and it's the highest
39:23thrust object ever made by humans of any
39:26kind whatsoever.
39:29Um, so yeah, this thing weighs over
39:325,000 tons on liftoff and it's going
39:34straight up,
39:38you know, fast.
39:40Um and and and then that then that then
39:43that giant thing comes back, you know,
39:45fly comes back from space, you know,
39:48like 5 minutes after liftoff that that
39:50booster is coming down one way or
39:52another. It's going to, you know,
39:54hopefully in a good way, but it's coming
39:55down one way or another. So excitement
39:57is guaranteed. Success success is not
39:59always guaranteed, but excitement
40:00certainly is.
40:01>> It's very cool when you see it come back
40:03and lands in the same place where it
40:05took off.
40:07>> Yeah.
40:07>> Chopsticks.
40:09>> Yeah. It looks crazy. Um,
40:13you know, when when I first proposed
40:14this to the team that they looked at me
40:16like I was lost my mind, but I'm like,
40:18look, I think this can be done, you
40:19know. Um, there's there's nothing that's
40:22preventing it from happening from a
40:23physics standpoint. And and the only
40:26thing I can think of that would be
40:27slightly better than catching it with
40:29the tower arms would be literally
40:30landing in the on the launch pad itself.
40:33But then, you know, if you do if you get
40:36something wrong, then you damage the
40:37launch pad and then that would take a
40:39lot of work to fix. So, if we if we just
40:42catch with the arms, that's that's a
40:44little better. Um cuz we don't smash the
40:46launch pad necessarily if if something
40:48goes wrong. Um but it's that's a
40:51difference of minutes uh between
40:53catching with the arms, catching with
40:54the and simply landing in in the actual
40:57launch pad to took off from. Um
41:00now the the the much harder than the
41:03than the booster is catching the ship.
41:06Um and the you know the ship we've got
41:09to have also there's a lot of things
41:11that have to be done that have never
41:13been never never been done before. No
41:14one has ever made a truly reusable
41:17orbital heat shield. So the the the heat
41:21like the shuttle heat shield required 9
41:23months of repair between flights.
41:27Um so even though it did the shuttle
41:30orbiter portion did come back you know
41:32they'd lose the main tank which is the
41:34primary airframe or the primary sort of
41:36primary structure really to which
41:38everything attack on every flight but
41:40they'd get the orbiter back but it would
41:42take 9 months of refurbishment of the of
41:44the heat shield before they could fly
41:45again. Um so no one has ever made a
41:50fully reusable orbital heat shield. Um
41:52it's an incredibly difficult thing to
41:53do. Um, and it's something that we've
41:55been working on for a decade. Basically,
41:58it's so hard. One of the hardest
42:00technical problems I think humans have
42:03ever tried to solve. Um, but we've got
42:06an incredibly talented team that has has
42:09worked
42:11incredibly hard for a very long time and
42:13we've been able to iterate. Go back to
42:15iteration. We've had many be able to do
42:17many iterations whereas the the shuttle
42:19couldn't really iterate. they they they
42:21always have people on board the shuttle,
42:23so they couldn't take any risks um
42:26because somebody might get killed.
42:28Whereas with Starship, we we we're
42:30taking many risks um in order to explore
42:34the the physics of what's needed to
42:38achieve full reusability of the heat
42:40shield and and the entire vehicle. Um,
42:43so we've we've actually achieved sort of
42:46precise uh soft landings of the ship uh
42:49several times out out to sea in the
42:52ocean. Um where where the ship if there
42:56had been a tower there uh the ship would
42:59have been caught.
43:01So that's maybe an interesting point to
43:03to make that that if it the ship thought
43:07it was landing with sort of in a with a
43:09simulated tower and had a tower been
43:11there it would have been caught. Now we
43:13want to make very sure that the that
43:15everything's going to be everything's
43:17going to work when we bring it back to
43:19the actual tower so that we don't smash
43:21the tower or you know rain debris on
43:23people's heads which will definitely
43:24affect our popularity. So, we want to be
43:29highly confident in the reliability of
43:30of the ship. And I think we're we're
43:32finally there where where we think u it
43:36is worth uh with
43:39with with ship 42. Um which hopefully is
43:43prophetic. You know, 42 is kind of a
43:46special number. Uh that we we can try to
43:49catch up with the tower. So, that's what
43:50we're going to do in about a week from
43:54now.
43:55We're all rooting for it.
43:59[applause]
44:03Um,
44:05so, so starship
44:08is the great unlock here. Just as an
44:11aside, I was watching the young person
44:16who was the heading the starship
44:19uh operation
44:21and he was speaking to a group of people
44:24in the audience wearing sport jackets
44:26and suits and white shirts and he came
44:29on and he was wearing engineer boots and
44:32blue jeans and a plaid shirt and
44:35unshaven and really long hair and he
44:38starts talking and he said, "F this and
44:40f that and f this.
44:43And then uh and I'm looking at him, I'm
44:46saying, "Oh my god, that's me when I was
44:4740.
44:50Really good guy." Uh [laughter]
44:55>> yeah,
44:55>> drives motorcycles.
44:58Um, so, so this is the great unlock and
45:02but all the things that Elon does, we
45:06might not make sense to you at the very
45:09beginning when you hear about them and
45:12then you say, "Oh, wow. I see. He
45:14probably was thinking about that all
45:16along." And so when you think about uh
45:20you reusable rocket, you know, when you
45:22talk about so much cost to get to space,
45:24the only ones who can go to space that
45:26was $18,000,
45:28the only ones who go to spa space are
45:30governments to protect their countries
45:32or it can be a television company who
45:34spends a $2 billion on a satellite so it
45:36broadcasts to one place, no one else.
45:39And so at bringing down the cost so
45:41dramatically then all of a sudden you
45:43enable Starlink and Starlink you have
45:4710,000 satellites
45:49now on the way to 20,000 and uh and and
45:53that's what I thought and uh now the
45:57says no it's not going to be 20,000 it's
45:59going to be 100,000. Why is it going to
46:01be 100,000? because we thought in the
46:04past that the revenues he'd be
46:07generating from Starlink were going to
46:09be from consumers and government and
46:11enterprise. Uh and and now though he
46:15says that agentic is going to be
46:18multiples of what we thought about
46:20before and every time you
46:25>> Oh yeah. Yeah. Well, um my understanding
46:28is that actually a gentic uh internet
46:31traffic already exceeds uh human traffic
46:34if I read that correctly from Cloud
46:36Cloudflare. Um and um and I I believe
46:40CloudFare uh predicted
46:43uh that in 5 years a traffic will be a
46:48thousand times more than human traffic.
46:50And I if anything I think that is
46:53underestimating the amount of agentic AI
46:55traffic
46:57>> 1,000 times in how many years? Five.
47:00>> Yes.
47:02>> So, so we've been thinking about
47:04Starlink as being a business that could
47:07do a trillion dollars in revenues and
47:10make seven or $800 billion in IBIDA in
47:13uh in 10 years and uh and be worth 14 or
47:1815 trillion. all of SpaceX is now worth
47:21less than two trillion. So that was one
47:24ex one one one opportunity. Uh but now
47:28uh I think we're probably really light
47:31on how much SpaceX how much Starling uh
47:35can be worth. But because you're adding
47:38so many
47:40so many more satellites now we got space
47:42centers data centers.
47:45>> Yeah. Yeah, I mean Starlink revenue is
47:47growing very rapidly by any sort of
47:49normal business standards. You know,
47:50like you're roughly doubling revenues
47:53per year um at large scale. And you
47:57know, I remember when that used to be a
47:58big deal. Um it wasn't that long ago uh
48:01that you know, doubling revenues with
48:03high gross margin and positive cash flow
48:06every year would normally be an
48:10impressive thing. Um although I have to
48:12say with AI, it does exceed that.
48:16So, uh with AI you have order of
48:18magnitude per year uh growth potential.
48:21So
48:23um
48:26and and uh I mean if you think about
48:29like things I just said because I say a
48:30bunch of things I've said sound
48:31extremely crazy uh but but I think we'll
48:35get there. You know sometimes I'm a
48:37little optimistic on the time frame but
48:38I always deliver in the end. Um, so
48:43you know there's um and and and and also
48:47I'm not always late. I'd like to say
48:49that [laughter]
48:50because the the the press always reports
48:52on when I'm late, but they never report
48:53on when I'm early. Um and and uh and
48:56actually I'm early quite often. Um so uh
49:00but the press, you know, is a sort of
49:03they don't write positive articles. They
49:05just ignore the times where I'm early
49:06and only print the times when I'm late.
49:07So it seems like I'm always late, but
49:08I'm that's actually not the case. my my
49:10predictions I tra I I aim for my
49:12predictions to be at the 50th percentile
49:13which means that they are basically half
49:16half rung half right um with respect to
49:19the predicted time frame um and that's
49:21that's about how things have turned out
49:23um so um
49:27you know we have 11,000 satellites up
49:30there right now um primarily the version
49:34two satellite we've just started
49:35launching the version three satellite
49:37version three satellite has over 10
49:40times the capability of a version two
49:41satellite and will be launching 10 times
49:43as many of them. So the amount of
49:44bandwidth that stallink will produce um
49:47in I don't know two two or 3 years will
49:50be 100 times more bandwidth than it
49:51currently offers
49:53and um
49:56yeah so I think the you say like what's
49:58the revenue per you know gigabit per
50:01second type of thing it's going to be
50:04it's going to be high I mean and then
50:06you say how many gigabits are you
50:07transferring it's very big number so and
50:11and with massive demand for you know by
50:14by AI and robots far that far exceeds
50:16human demand by like it's literally a
50:18thousand um and just the starting
50:20communications revenue will be something
50:22staggeringly high um but if you look at
50:25the the uh the AI stuff um if if we do
50:30say 100 gawatt a year to space of of AI
50:34compute say well what's the value per
50:37gawatt of that compute
50:40um
50:42well if if if Even if it went down to
50:44$10,
50:45uh you that would be a trillion dollars
50:47a year
50:49>> extra.
50:49>> Um
50:51and cumulative it would be so every year
50:53would be another trillion dollars
50:56because theat satellites don't vaporize.
50:57So it's like so you'd be stacking a
50:59trillion dollars a year of revenue per
51:01year
51:02>> and every five years
51:04>> $10 and if if if it if if it was say at
51:06at 50, you know, you'd have five
51:09trillion. So how are how are telos going
51:13about it every year?
51:14>> How are telos ever going to meet demand
51:18for the service
51:20uh with the infrastructure that they
51:24have? And if they have spectrum for
51:26example, which is a public good, why
51:30does the government, why would the
51:32government allow that to take continue
51:34to be the case when there's a need for a
51:38lot more capacity than they could
51:40possibly provide?
51:44>> Yeah. Well,
51:47um,
51:48>> should we pass that question?
51:53I think SpaceX will be able to get very
51:55far with the spectrum that it's
51:56acquired. Uh but you know, at certain
51:59point if if we're able to make much
52:01better use of the spectrum than um than
52:04say a legacy telco is, we could just
52:07acquire that legacy telco and use that
52:09spectrum, you know, 10 times more
52:11effectively.
52:14>> [applause]
52:16>> Um
52:18so and and and we have we we've been
52:22thinking that for the data center so
52:25we're thinking about 10,000 to to
52:28100,000 for Starlink ultimately and
52:31every time every 5 years when we replace
52:33the satellites that exist we're
52:35replacing them with 10 times the
52:37capacity of what we're taking down
52:38deorbiting.
52:42So
52:44that's your that those are thou
52:46thousands of you know huge amounts of
52:47capacity we're going to be providing.
52:50>> Um
52:51>> yes um
52:54like we we still see opportunity for
52:55order of magnitude improvements in in
52:57the satellites capability. So um so I
53:01think Starink version 4 will have 10
53:03times the capability of version three
53:06uh especially with the longdistance
53:07laser links. So the you know our current
53:11laser links are um around 100 Gbits a
53:15second. Uh our next generation laser
53:18links will be 10 terabits a second. 100
53:21times better.
53:24>> So
53:25>> um and then we'll we'll put 10 of them
53:26on. So you you know uh to you know
53:31>> so we get
53:32>> get ahead of that for a second.
53:33>> So we get a little short of time but
53:36there's a couple things more. So when
53:38people worried about data centers and
53:41about AI,
53:43um people in China aren't worried about,
53:46you know, about having too much data
53:48centers and too much power. So in China,
53:52they have about three times the amount
53:54of power that we do. And they're adding
53:56as much power a year, what, 300
53:58gigabits, compared to our 550 that we
54:02have in total. They're adding 300 a
54:05year. Um
54:08I'm not sure if that's a steady state
54:09power number but there are certainly
54:10increase I don't think it's quite that
54:12high. I think it's a little lower than
54:13that. Um
54:16and and
54:18but but it's it's it's it's it is a
54:20massive number. Um so the so China at
54:22this point I think is roughly three
54:24times the power output of the United
54:25States and it's trending to four to get
54:28to at least four times the the total
54:30power output of the United States. Um
54:33and industrial capacity to first
54:35approximation is proportionate to your
54:38electricity output.
54:40So, uh, and then chips, they're behind
54:45>> for now
54:46>> and but they're they're moving forward
54:49fast. Yeah.
54:50>> And and uh one executive
54:53uh at SpaceX a number of years ago told
54:56me that uh she thought that the uh only
55:02thing standing between my grandchildren
55:05speaking Mandarin and uh uh with SpaceX.
55:11Do you believe that?
55:13>> Well, I think there's some merit to
55:14learning Mandarin, you know, either way.
55:17Um
55:19it's you know a very interesting
55:21language. Um but um
55:25yeah I I think there's in terms of uh
55:29being of competing with China uh you
55:32know from an like an electricity
55:36generation standpoint and um just
55:38overall sort of how much digital
55:40intelligence have you harnessed for
55:42economic activity. I think the only way
55:44to be competitive with China is to take
55:48take the intelligence to space. Um so
55:52there there we can sort of bring on a
55:55few hundred gigawatts a year. Um and and
55:58like potentially you know a terowatt a
56:01year. Um so if if if we if we're able to
56:05do a terowatt a year of intelligence to
56:07space then um in that scenario unless
56:12China can do a terowatt a year to space
56:14then we will have uh e an economic
56:18output in excess of China.
56:21So but that that's a very big number of
56:24course that that's that's saying we do
56:26twice as much uh
56:29digital super intelligence per year as
56:33there is currently the average power
56:36usage of the United States. So average
56:38power usage of the United States is 500
56:40g a terowatt is 1,000 gawatt. So if we
56:43did a terowatt per year, that would be
56:45doing twice the current average power
56:47usage of the United States to space with
56:49digital super intelligence every year.
56:51These are like staggering numbers
56:53basically.
56:55>> So two more questions. Uh why do we need
57:00so much power? Why do we need so much
57:02chip? Why what's the what's the demand
57:05that's going to come for that
57:08intelligence
57:09for that power?
57:12and chips.
57:16>> The the demand will be
57:18and this is going to sound somewhat
57:21circular but the demand for intelligence
57:24for digital super intelligence will come
57:25from digital super intelligence
57:29>> which is for
57:31medical purposes which is for I mean
57:34what does that mean?
57:37Um it means that super intelligence will
57:40saturate all human needs by far. Uh
57:43there will be universal high income,
57:45universal extremely high income. Um
57:49universal medical care that is better
57:51than the best medical care available on
57:53earth today. Um
57:57where you could wish for a castle uh or
58:01a whatever you want basically. Um and
58:04and the digital super intelligence can
58:06make it for you as as a side project.
58:10Um so like I guess if you zoom out a lot
58:14you sort of think well okay as
58:17civilizations progress you can think of
58:19the progression of civilization as how
58:22much energy has been harnessed by that
58:24civilization.
58:26And um and so so currently
58:32ju just just to show you how much room
58:33for growth there is in our solar system.
58:36Um even if we increase the amount of
58:39power harnessed by our civilization by a
58:42factor of a million, we would still be
58:45using much than a millionth of the sun's
58:49energy.
58:54That's a lot of room for growth.
58:57[snorts] Um,
58:59and and you you'd obviously, you know,
59:03completely saturate anything anyone
59:05could ever wish for on Earth. Um, and
59:07that and and at that point you it's
59:08really the digital super intelligence
59:10serving the digital super intelligence.
59:12Um
59:15and and and our economy will seem quaint
59:19uh quaint and and like and and our money
59:22system will will seem like like you know
59:24exchanging seashells for currency.
59:27That's what our monetary system will
59:28seem like to super intelligence. Um you
59:32know technically that's thing that used
59:33to happen. Well, I mean, we used to
59:35money used to be like shiny metal that
59:37people would give to each other. Like,
59:39I'll give you this shiny silver piece of
59:41metal and and then and and then you have
59:42like some shiny gold pieces of metal and
59:45um and you have to carry these shiny
59:47bits of metal from one place to another
59:49and in order to do transactions
59:52um which is really ownorous, frankly,
59:53and and easy to rob um you know, if you
59:57if you got gold coins or silver coins,
59:58but that used to be how people would uh
1:00:01you know, pay for things. um must be
1:00:04incredibly difficult to buy things long
1:00:06distance, you know, uh because you
1:00:09you've got to transport all the silver
1:00:10and gold over the ocean and the ship
1:00:12sinks and you don't even know what's
1:00:13going on. Um it's not like you can call
1:00:16them on the your cell phone or something
1:00:17like that like
1:00:19>> So, [laughter] so, so obviously it'd be
1:00:21fun for me to talk to you forever and uh
1:00:24but I'm out of time and so have just one
1:00:26last question. So, when you're 83 like I
1:00:29am, how do you want people to think
1:00:31about you? So you're working like these
1:00:34crazy hours you're making, you know,
1:00:36nobody would work the way you work to
1:00:40just accumulate money, I think. And so
1:00:43you think that whatever you have is
1:00:45going to be highly taxed at the end, but
1:00:47you have a lot of money ultimately going
1:00:49to foundation, your foundation.
1:00:52>> What is what how do you want people to
1:00:54think about you?
1:00:58How will you want people to think about
1:00:59you when you're 83?
1:01:03Well, I guess I'd like I I would like
1:01:06there's perception and reality. I mean,
1:01:07I would like the reality, whatever the
1:01:09perception is, but I'd like the reality
1:01:11to be that I made a significant
1:01:13contribution in the furtherance of
1:01:14civilization.
1:01:16Uh that that that you know I Yeah.
1:01:20[applause]
1:01:25>> Thank you so much for being here today.
1:01:28[applause]
1:01:29Thank you. Heat.
1:01:34Heat.
1:01:36[applause]

About this transcript

This page contains the full transcript of Elon Musk STUNS In New Interview With Predictions & Opens Up (With Timestamps) by Matt Pocius on Tesla Stock & Money, generated from the public captions YouTube serves with the video. The transcript has 10,026 words across 1,453 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.

What you can do with it

Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to choose a caption track available on the video.

Free YouTube transcript tool

YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, caption language selection, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.