Latest / Elon Musk Thinking / Latest: Elon Musk Unveils New Product TERAFAB!!!

Transcript

Latest: Elon Musk Unveils New Product TERAFAB!!! · Elon Musk Thinking

• 0:00
  Speaker 1: Latest interview of Elon Musk.
• 0:02
  Speaker 2: We have a profoundly important announcement to make, which is
• 0:07
  Speaker 2: the most epic chip building exercised in history by far.
• 0:14
  Speaker 1: This is really going to take things to the next level.
• 0:17
  Speaker 2: So, yeah, a level probably people aren't even contemplating right now.
• 0:21
  Speaker 2: This is not in the OUTCOLI is a sort of
• 0:25
  Speaker 2: an out of context problem. It's not in their context.
• 0:29
  Speaker 2: So we're going to adjust the context by a few
• 0:31
  Speaker 2: orders of ninet tude. Here it's a joint effort. I'm
• 0:35
  Speaker 2: pressing the button, but the button's not working. We're starting
• 0:38
  Speaker 2: to be a galactic galactic civilization. So I think the
• 0:42
  Speaker 2: future that everyone, well most people, I think would agree
• 0:46
  Speaker 2: is the most exciting is one where we are out
• 0:49
  Speaker 2: there among the stars, where we are not forever confined
• 0:54
  Speaker 2: to one planet, that we become a multi planet species,
• 0:59
  Speaker 2: like the best science science fiction that you've ever read,
• 1:03
  Speaker 2: you know, Star Trek or in Banks or Asimov or Heinlen,
• 1:11
  Speaker 2: And we want.
• 1:11
  Speaker 1: To make that real.
• 1:13
  Speaker 2: Yeah, not just fiction, turn science to fiction to science fact.
• 1:19
  Speaker 2: That's the glorious, exciting future that I certainly look forward to,
• 1:26
  Speaker 2: and it's worth considering sort of like, how would you
• 1:31
  Speaker 2: rate civilizations. You know, there's so there was a physicist,
• 1:35
  Speaker 2: I think it was Russian in the sixties, Karta chef.
• 1:39
  Speaker 2: He thought about, at a high level, how would you
• 1:43
  Speaker 2: consider any given civilization, and he said, well, if you're
• 1:48
  Speaker 2: type one, you're using most of the energy of your planet.
• 1:54
  Speaker 2: And we actually still have quite a ways to go
• 1:57
  Speaker 2: to be properly a type one. We're still using a
• 2:00
  Speaker 2: tiny fraction of the Sun's energy that reaches our planet.
• 2:05
  Speaker 2: Let's see here we are, but the Earth only receives
• 2:12
  Speaker 2: about half a billionth of the Sun's energy.
• 2:15
  Speaker 1: So the Sun is truly enormous.
• 2:18
  Speaker 2: The Sun is a ninety nine point eight percent of
• 2:21
  Speaker 2: all masts in the Solar System. So sometimes people will
• 2:25
  Speaker 2: ask me like what about you know, other power sources
• 2:28
  Speaker 2: of power on Earth, Like what about fusion on Earth? Well,
• 2:34
  Speaker 2: that is unfortunately very small, because the Sun is ninety
• 2:38
  Speaker 2: nine point eight percent of mass in the Solar System,
• 2:42
  Speaker 2: and Jupiter is about point one percent, and Earth is
• 2:46
  Speaker 2: in the miscellaneous category.
• 2:47
  Speaker 1: We are. I think it's called saying.
• 2:51
  Speaker 2: I think might have said Earth is is like a
• 2:54
  Speaker 2: tiny dust mote in a vast darkness, very very small.
• 3:00
  Speaker 1: The Sun is enormous.
• 3:03
  Speaker 2: So the way to actually scale civilization is to scale
• 3:08
  Speaker 2: power in space.
• 3:10
  Speaker 1: This is necessarily true because we actually capture such.
• 3:17
  Speaker 2: A tiny amount of the Sun's energy on Earth, because
• 3:20
  Speaker 2: we're just this tiny dust mode. Another way to think
• 3:24
  Speaker 2: of it is roughly like electricity production on Earth of
• 3:29
  Speaker 2: whole of civilization is only about a trillionth of the
• 3:33
  Speaker 2: Sun's energy, which means if you increase civilizational power output
• 3:38
  Speaker 2: by a million you would still only be a millionth
• 3:43
  Speaker 2: of the Sun's energy. I mean, it's all inspiring to
• 3:47
  Speaker 2: consider that just how tiny we are in the grand
• 3:52
  Speaker 2: scheme of things. And yeah, we often get sort of
• 3:56
  Speaker 2: caught up in the sort of these sort of squabbles
• 3:59
  Speaker 2: on Earth that are really very sort of minor things
• 4:02
  Speaker 2: in when you consider the grandness of the universe, and
• 4:08
  Speaker 2: so I think it's it is important actually to consider
• 4:11
  Speaker 2: the grandness of the universe and what we can do
• 4:15
  Speaker 2: that is much greater than what we've done before, as
• 4:19
  Speaker 2: opposed to worry about sort of small squabbles on Earth
• 4:22
  Speaker 2: type of thing.
• 4:26
  Speaker 1: Not much point in that.
• 4:28
  Speaker 2: Yeah, we want to be a civilization that expands to
• 4:31
  Speaker 2: the galaxy with spaceships that anyone can go anywhere they want.
• 4:37
  Speaker 1: That would be epic and have a.
• 4:39
  Speaker 2: City on the Moon's cities on Mars, populate the Solar System,
• 4:44
  Speaker 2: and send spaceships to other star systems. That sounds like
• 4:50
  Speaker 2: the best possible future. So to do that we need
• 4:53
  Speaker 2: to harness the power of the zone and so a terrifab.
• 4:59
  Speaker 2: While it is enormous, a terror wad of compute. Forer
• 5:02
  Speaker 2: year is enormous by our sort of civilizational standards, it
• 5:08
  Speaker 2: is still just one step along the way of being
• 5:12
  Speaker 2: even a cotdtership. You're still have a long way to
• 5:15
  Speaker 2: go to even be a coardship two level civilization, and
• 5:17
  Speaker 2: you're not even registering as a cordtership three. So it's
• 5:21
  Speaker 2: a very big thing by current human standards, but it
• 5:26
  Speaker 2: still small in the grand scheme, and it's very difficult
• 5:33
  Speaker 2: for humans. So to accomplish this very difficult goal really
• 5:37
  Speaker 2: requires a combination of efforts of SpaceX, XAI and Tesla
• 5:43
  Speaker 2: working together to create this epic Terrify project. And Tesla
• 5:50
  Speaker 2: and XAI and SpaceX have all done amazing things that
• 5:54
  Speaker 2: people did not think would be done before.
• 5:56
  Speaker 1: So there's the gig of Giga Texas bad here, there's
• 6:02
  Speaker 1: you know.
• 6:03
  Speaker 2: The Optimists robot that's being built, there's a global supercharging network.
• 6:10
  Speaker 1: There's really quite a lot.
• 6:12
  Speaker 2: And it wasn't that long ago when people thought electric
• 6:15
  Speaker 2: cars would wouldn't amount to anything, and they were. There
• 6:18
  Speaker 2: were basically no electric cars for sale when when Tesla
• 6:22
  Speaker 2: started and people said it was impossible, and now tells
• 6:27
  Speaker 2: us making two million electric cars a year. And then Xai,
• 6:31
  Speaker 2: although it's a new company now part of SpaceX, has
• 6:35
  Speaker 2: also booked the first gigaway scale compute cluster, which in
• 6:39
  Speaker 2: record time. Jensen Wrong from a video said he's never
• 6:44
  Speaker 2: seen anything built so fast in his life before. So
• 6:48
  Speaker 2: it's a great compliment from from in video. And then SpaceX. Uh, well,
• 6:54
  Speaker 2: I guess you can read it for yourself. Well you
• 6:55
  Speaker 2: already know. I mean the reusable rockets. People said that
• 7:00
  Speaker 2: reusable rockets weren't possible, and even if you did do them,
• 7:03
  Speaker 2: they weren't be economically feasible. So we did them, and
• 7:06
  Speaker 2: then we made them economically feasible, and now we've landed
• 7:09
  Speaker 2: over five hundred times. And then we did the Falcon
• 7:14
  Speaker 2: Heavy and now we're doing Starship. And Starship is a
• 7:18
  Speaker 2: critical piece of the puzzle because in order to scale
• 7:23
  Speaker 2: compute and scale power, you have to go to space,
• 7:25
  Speaker 2: which means that you need massive payload to space.
• 7:30
  Speaker 1: And Starship will enable that. So let's gives you sort
• 7:33
  Speaker 1: of just a sense of scale.
• 7:34
  Speaker 2: We've got some optimists there, optimists for scale, and optimist
• 7:41
  Speaker 2: is about five to eleven, so it gives you a
• 7:43
  Speaker 2: sense of the size of the Starship V three rocket.
• 7:47
  Speaker 2: Stasha V four will be much longer. Actually, the Stasha
• 7:50
  Speaker 2: V four will make Starship three look kind of short.
• 7:56
  Speaker 2: So we'll expand with Sasha V three to two hundreds
• 8:00
  Speaker 2: of payload to orbit from one hundred tons. We'll start
• 8:03
  Speaker 2: with V three and then you can see that just
• 8:05
  Speaker 2: the that's just a rough approximation of the the AI,
• 8:11
  Speaker 2: the mini version of the AI SAD, so that's roughly
• 8:13
  Speaker 2: one hundred kilo whites. It's showing the solar panels and
• 8:18
  Speaker 2: the radiator to scale. So for some reason there's been
• 8:23
  Speaker 2: a bizarre debate about radiators in space. It's safe to
• 8:27
  Speaker 2: say SpaceX knows how to do heat rejection in space
• 8:31
  Speaker 2: with ten thousand satellites, and over might know a thing
• 8:34
  Speaker 2: or two. So you can see the radiator is actually
• 8:37
  Speaker 2: quite small relative to the solar panels, and we'll call
• 8:43
  Speaker 2: it the mini sad.
• 8:44
  Speaker 1: Since that's just one hundred kilo.
• 8:46
  Speaker 2: Whites, we expect future satellites to probably go to the
• 8:49
  Speaker 2: megawatt range.
• 8:51
  Speaker 1: So in order to get to the terra watt.
• 8:56
  Speaker 2: Of compute per year, you need about ten million tons
• 8:59
  Speaker 2: two over per year, and at a HydroD killer what's
• 9:02
  Speaker 2: per ton. So we're confident this is feasible, like no
• 9:07
  Speaker 2: new physics or impossible things are required to get there.
• 9:11
  Speaker 2: So I'm confident that actually that SpaceX will get to
• 9:14
  Speaker 2: ten million tons toward per year, and then we're building
• 9:19
  Speaker 2: up to a terrawatt of solar, so that solves the
• 9:23
  Speaker 2: will solve the solar problem the power generation, So then
• 9:27
  Speaker 2: the key missing ingredient is therefore a terrawat of compute.
• 9:33
  Speaker 2: So this announcement is about solving the key missing ingredient.
• 9:37
  Speaker 2: To give you a sense of what we're talking about,
• 9:42
  Speaker 2: the current output of AI compute is roughly twenty gigawats
• 9:50
  Speaker 2: per year. This chart explains why we need to build
• 9:54
  Speaker 2: the terror fat because all of the rest of the
• 9:57
  Speaker 2: output from Earth is about two percent of what we need.
• 10:02
  Speaker 2: So if you add up all the fabs on Earth combined,
• 10:07
  Speaker 2: they're only about two percent of what we need. For
• 10:10
  Speaker 2: the Terra wid project or TERRAFAB project. So you know,
• 10:16
  Speaker 2: we certainly want our existing supply chain to be clear.
• 10:21
  Speaker 2: We're very grateful to our existing supply chain to Samsung, TSMC,
• 10:27
  Speaker 2: Micron and others, and we would like them to expand as.
• 10:32
  Speaker 1: Quickly as they can, and we will buy all of
• 10:35
  Speaker 1: their chips. I have said these.
• 10:39
  Speaker 2: Exact words to them, but there's there's a maximum rate
• 10:44
  Speaker 2: at which they're comfortable expanding.
• 10:48
  Speaker 1: But that rate is much less than we would like.
• 10:52
  Speaker 2: And so we we either build the terrafab or we
• 10:56
  Speaker 2: don't have the chips, and we need the chips, so
• 11:00
  Speaker 2: well build tear fab. And we're signing off with an
• 11:02
  Speaker 2: Advanced Technology fab here in Austin, and.
• 11:09
  Speaker 1: I believe Governor Abbott is in the audience.
• 11:11
  Speaker 2: I'd like to thank Governor Abbott and the State of
• 11:13
  Speaker 2: Texas for the support. So in the Advanced Technology Fab,
• 11:18
  Speaker 2: we will have all of the equipment necessary to make
• 11:20
  Speaker 2: a chip of any kind logical memory, and we will
• 11:23
  Speaker 2: also have all of the equipment necessary to make the
• 11:27
  Speaker 2: lithography masks. So in a single building we can create
• 11:30
  Speaker 2: a lithography mask, make the chip, test the chip, make
• 11:36
  Speaker 2: another mask, and have an incredibly fast recurse of loop
• 11:41
  Speaker 2: for improving the chip design. To the best of my knowledge,
• 11:45
  Speaker 2: this doesn't exist anywhere in the world where you've got
• 11:48
  Speaker 2: everything necessary to build logic memory and do packaging and
• 11:54
  Speaker 2: test it and then do the masks, improve the masks,
• 11:58
  Speaker 2: and just keep looping it.
• 12:01
  Speaker 1: So we're not just going to do conventional compute in this.
• 12:04
  Speaker 2: I think there's some very interesting new physics that is
• 12:09
  Speaker 2: potentially that I actually I'm confident will work.
• 12:12
  Speaker 1: It's just a question of one.
• 12:14
  Speaker 2: So this is going to we're really going to push
• 12:16
  Speaker 2: the limit of physics and compute, and we're going to
• 12:20
  Speaker 2: try a bunch of wild and crazy things, which you
• 12:23
  Speaker 2: can do if you've got that fast iteration move that.
• 12:28
  Speaker 2: I can't emphasize enough the importance of being able to
• 12:30
  Speaker 2: make a chip tested and then make and then change
• 12:36
  Speaker 2: the design.
• 12:37
  Speaker 1: Do another one, and have that in a single building.
• 12:41
  Speaker 2: I think that our recursive improvement with that situation is
• 12:47
  Speaker 2: probably an order of magnitude better than anything else in
• 12:50
  Speaker 2: the world. So broadly speaking, we expect to make two
• 12:56
  Speaker 2: two kinds of CHIRP, so one will be optimized for
• 13:01
  Speaker 2: edge and inference, so that'll be used primarily in optimists
• 13:05
  Speaker 2: and in the cars, but especially an optimists because I
• 13:08
  Speaker 2: expect the robots, humanoid robots to be made ten to
• 13:15
  Speaker 2: one hundred times more than the volume of cars. So
• 13:20
  Speaker 2: you know, if vehicle production vical production growth is about
• 13:24
  Speaker 2: one hundred million vehicles a year, and I expect humanoid
• 13:28
  Speaker 2: robot production to be somewhere between a billion and ten
• 13:30
  Speaker 2: billion years a year.
• 13:32
  Speaker 1: So it's a lot.
• 13:36
  Speaker 2: So yeah, tells is going to make a very significant
• 13:39
  Speaker 2: percentage of those is our goal. And then we need
• 13:46
  Speaker 2: a high power chip that is designed for space that
• 13:51
  Speaker 2: takes into account the more difficult environment in space where
• 13:56
  Speaker 2: you've got high power energy ions photons, you had electron
• 14:02
  Speaker 2: build up, it's a hostile environment in space. So you
• 14:06
  Speaker 2: want to design the chip, you want to optimize it
• 14:09
  Speaker 2: for space, and you also want to generally run it
• 14:13
  Speaker 2: a little hotter than you would normally run a chip
• 14:15
  Speaker 2: on Earth to minimize the radiator mass. So they're just
• 14:19
  Speaker 2: a bunch of constraints that you designed something differently in
• 14:23
  Speaker 2: space than you went on the ground.
• 14:27
  Speaker 1: And for the space compute, my.
• 14:30
  Speaker 2: Guess is that is the vast majority of the compute
• 14:34
  Speaker 2: because your power constrained on Earth. Why I think it's
• 14:39
  Speaker 2: probably one hundred to two hundred gigawa's a year of
• 14:44
  Speaker 2: terrestrial chips, and probably on the order of a tarrawat
• 14:50
  Speaker 2: of ships in space, just because of power distraints on
• 14:54
  Speaker 2: the ground. Is probably that's probably how it ends up.
• 14:59
  Speaker 2: Space has this advantage that it's always sunny, it's very nice.
• 15:05
  Speaker 2: So I actually think that the cost of AI and
• 15:10
  Speaker 2: deploying AI in space will drop below the cost of
• 15:13
  Speaker 2: terrestrial AI much sooner than most people expect. I think
• 15:17
  Speaker 2: it may be only two or three years before it
• 15:21
  Speaker 2: is actually lower cost to send AI chips to space
• 15:26
  Speaker 2: than it is on the ground because in space you
• 15:29
  Speaker 2: don't need much in the way of batteries because if
• 15:33
  Speaker 2: it's always sonny and the solar power, you're going to
• 15:38
  Speaker 2: get at least five or more times the solar power
• 15:41
  Speaker 2: you get in space versus the ground, because you don't
• 15:44
  Speaker 2: have atmospheric attenuation or a day night cycle or seasonality,
• 15:49
  Speaker 2: and you're always normal to the Sun, so you're really
• 15:53
  Speaker 2: maximizing the.
• 15:55
  Speaker 1: Solar power at that point.
• 15:57
  Speaker 2: And the space solar actually costs less than terrestrial solo
• 16:02
  Speaker 2: because you don't need heavy glass or framing to protect
• 16:05
  Speaker 2: it from extreme weather events. So as soon as the
• 16:10
  Speaker 2: cost to orbit drops to a low number, it immediately
• 16:15
  Speaker 2: makes extremely compelling sense to put AI in space.
• 16:21
  Speaker 1: It becomes a no brainer. Basically.
• 16:23
  Speaker 2: Moreover, as you go to space, you get increased economies
• 16:27
  Speaker 2: of scale and things get easier over time, Whereas as
• 16:30
  Speaker 2: you try to put more and more power on the ground,
• 16:33
  Speaker 2: you run out of space and you start using up
• 16:36
  Speaker 2: the easy spots, and then you get next level numbie.
• 16:41
  Speaker 2: Nobody wants the thing in their backyard. So then so
• 16:45
  Speaker 2: actually increasing power on Earth has becomes harder over time
• 16:51
  Speaker 2: and more extensive over time, but in space it becomes
• 16:55
  Speaker 2: actually cheaper.
• 16:56
  Speaker 1: And easier over time. These are very important points.
• 17:00
  Speaker 2: What you just saw there was because of course you're
• 17:04
  Speaker 2: asking what's on your mind is, well, what do you
• 17:06
  Speaker 2: do after a terrorfab?
• 17:09
  Speaker 1: Don't think small? Well yeah, good point.
• 17:13
  Speaker 2: So we you know, how do you get to a
• 17:18
  Speaker 2: cattawat is the obvious next question, and you get there
• 17:24
  Speaker 2: by having an electromagnetic mass driver on the moon with
• 17:32
  Speaker 2: robots with optimize and obviously lots of humans, and.
• 17:38
  Speaker 1: With that you can send a petalwat. You can create
• 17:42
  Speaker 1: a pedalwat of compute and send that.
• 17:45
  Speaker 2: To deep space, because on the moon has no atmosphere
• 17:49
  Speaker 2: and has one six Earth gravity, so you can you
• 17:52
  Speaker 2: don't need rockets on the Moon. You can literally accelerated
• 17:58
  Speaker 2: to escapablocity from this surface, and that dramatically drops the
• 18:04
  Speaker 2: costs once again of harnessing power and enables you to
• 18:11
  Speaker 2: go a thousand times bigger than I tell a what.
• 18:15
  Speaker 1: So, for sure, in the.
• 18:18
  Speaker 2: Future, I want to see I want to just live
• 18:20
  Speaker 2: long enough to see the mass driver on the Moon,
• 18:23
  Speaker 2: because that's going to be incredibly epic. That should hopefully
• 18:27
  Speaker 2: get us to a millionth of the Sun's energy at
• 18:29
  Speaker 2: least humbling to think about that, but a million for
• 18:34
  Speaker 2: the Sun's energy would be a million times bigger than
• 18:36
  Speaker 2: its economy, So it's good from that perspective. And then yeah,
• 18:43
  Speaker 2: you expand beyond that to the planet, to the other
• 18:45
  Speaker 2: stars and create the most exciting possible future. Then that
• 18:52
  Speaker 2: I can imagine this looks a bit like the opening
• 18:56
  Speaker 2: an Idiocracy with the mic judge unluckying an age of
• 18:59
  Speaker 2: amazing amount. So yeah, obviously the elements of that are
• 19:06
  Speaker 2: sustainable energy, space travel and an R A I and
• 19:12
  Speaker 2: robotics that bring amazing abundance to everyone. And it's really
• 19:21
  Speaker 2: the it's really the only path to amazing abundance is
• 19:23
  Speaker 2: a I R, A I and robotics, Which is not
• 19:26
  Speaker 2: to say it can't go wrong hopefully, you know, but
• 19:28
  Speaker 2: I think it'll probably go right, and it will be
• 19:31
  Speaker 2: a future that you that you love, and.
• 19:37
  Speaker 1: It's the best future I can think of at least.
• 19:41
  Speaker 2: And then we go beyond the Moon, beyond Mars, and
• 19:44
  Speaker 2: we sail through the rings of Saturn. They wouldn't it
• 19:48
  Speaker 2: be amazing if you could buy a trip to Saturn. Frankly,
• 19:52
  Speaker 2: if you just have a trip to Saturn, I think
• 19:53
  Speaker 2: you fas would just be free in the future.
• 19:55
  Speaker 1: It sounds nuts, but.
• 19:57
  Speaker 2: You know, if you've got an AI robotics economy that
• 20:00
  Speaker 2: is anywhere close to a million times the size of
• 20:03
  Speaker 2: the current Earth economy, literally any need you possibly want
• 20:08
  Speaker 2: can be met.
• 20:09
  Speaker 1: If you can think of it, you can have it.
• 20:11
  Speaker 2: So I think in Banks in his culture books has
• 20:13
  Speaker 2: it pretty much right where there actually isn't money in
• 20:17
  Speaker 2: the future and there is abundance for everyone. If you
• 20:22
  Speaker 2: can think of it, you can have it. That's it,
• 20:25
  Speaker 2: which means anyone could have a trip to Saturn. It
• 20:28
  Speaker 2: won't be you know, just a few people. If you
• 20:30
  Speaker 2: want it, you can have it. Yeah, join us on
• 20:34
  Speaker 2: this journey and help us design incredible chips and make
• 20:40
  Speaker 2: incredible ships and build a terror wat of ships, a
• 20:44
  Speaker 2: terra awadi of solar and tell million tons to over
• 20:50
  Speaker 2: it per year.
• 20:51
  Speaker 1: Thank you, thanks for listening. See you in the next episode.