AI's REAL Bottleneck: Power, Space & The Future

Added: 2026-06-03

[00:00:00]as AI energy and space systems scale exponentially.

[00:00:03]What non-physical constraints, organizational, cultural, bureaucracy, or human, are now the real bottleneck?

[00:00:12]Is there a bottleneck?

[00:00:15]>> Um electricity generation is the limiting factor.

[00:00:20]Um >> The innermost loop.

[00:00:23]>> Yeah.

[00:00:24]Um I think people are underestimating difficulty of bringing electricity online. You know, you've you've got to get you've got to generate electricity.

[00:00:32]You've got to you need transformers for the transformers.

[00:00:35]>> Mhm.

[00:00:35]>> Um so you've got to convert that voltage to something that the computers can digest.

[00:00:40]You've got to cool the computers.

[00:00:43]So it's it's basically electricity generation and cooling um are limiting factors for AI.

[00:00:49]>> Yeah.

[00:00:50]>> Um and once you have humanoid robotics, they can address the power generation and and the the cooling stuff. Um but that that is the limiting factor and will be for at least the next 2 years.

[00:01:06]>> Isn't it amazing how divergent the Memphis version of that is from the space-based version? I mean, you have solar panels in common, but otherwise no storage, abundant amounts of energy, >> Yeah.

[00:01:19]>> but you have launch costs and you have Yeah, I mean, you can and weight suddenly matter. You don't care too much about the weight in Tennessee.

[00:01:26]Suddenly the weight is a critical factor and these two two pathways for compute have a huge divergence from here forward.

[00:01:33]>> Yeah. Um when once you get solar domestically at scale and uh if we're launching Starship at scale, then um by far the cheapest way to do AI compute will be in space.

[00:01:47]Um so once you have the once you have full and complete reusability, um the propellant cost per flight is maybe a million dollars.

[00:01:55]>> Yeah. People don't realize that. People have some >> to times ridiculous >> it's nothing. amount of expectations how much it cost.

[00:02:02]>> So so if you look at it's going to cost a million dollars to transport for 10 megawatts of of AI compute.

[00:02:08]>> Yeah.

[00:02:09]So assuming everything keeps trending the way it's currently trending, if you look at the next four years of accelerating launches.

[00:02:17]So 200 tons per launch.

[00:02:20]Yeah.

[00:02:20]>> This is where you're going, but yeah.

[00:02:22]Like if you say if they highlight to something it's probably >> more like 150 tons. But yeah, it's the right order of magnitude is at least it's in excess of 100 tons for a marginal cost per flight of around a million million dollars.

[00:02:33]>> So so what fraction of all that launched mass is data centers in space?

[00:02:38]As opposed to moon base, as opposed to launch to Mars, as opposed to satellites.

[00:02:44]>> how I mean, this is a new we weren't talking about this as a space objective even, you know, a year ago. All of a sudden data centers have become the massive driving force for opening up the space market.

[00:02:57]>> the urgent the urgent use case, too.

[00:03:00]>> I mean, I used to I used to wonder what's going to drive humanity. I I thought it was asteroid mining, right?

[00:03:04]You were focused on on Mars.

[00:03:07]>> We will actually want to mine asteroids to turn them into >> Sure.

[00:03:10]>> Uh >> You know, so we >> before you photovoltaic >> Before you >> [laughter] >> You know, not not not for anything else.

[00:03:16]Like >> I mean, if we're going to if we're going to build out Dyson Swarms.

[00:03:20]>> Yeah, just a bunch of satellites around the sun.

[00:03:22]>> Yeah, how how how long What's your time frame for Alex is another question Alex wanted to have us ask. What's your time frame for uh for humanity achieving a Dyson Swarm?

[00:03:33]Is it 50 years?

[00:03:34]>> is this?

[00:03:35]>> Yeah, I know. It's it's a matter but >> A Dyson Swarm like people think like everything's going to be covered in satellites. So I think it's not quite that that I mean, I think we You have to like what mass ends up becoming satellite. Um you know.

[00:03:50]Mercury probably ends up being satellites.

[00:03:51]>> Yes.

[00:03:52]Jupiter?

[00:03:53]>> Jupiter, yeah. Saturn.

[00:03:54]>> it's a little gassy.

[00:03:55]>> Oh, yeah.

[00:03:57]It's big, but it's got a lot of It's got a lot of rocks orbiting it.

[00:03:59]>> Do you leave Mars alone?

[00:04:01]>> Yeah, I think leave Mars alone.

[00:04:02]>> Asteroids Asteroids are are fantastic uh food source.

[00:04:06]>> Uh yeah.

[00:04:06]>> Yeah, no gravity well. Gravity well on Jupiter is a non-starter.

[00:04:09]>> already mostly differentiated into you know, carbonaceous chondrites for fuel and nickel-iron for materials.

[00:04:15]>> Gold.

[00:04:16]>> Yeah.

[00:04:16]>> A bunch of that asteroid belt probably turns into solar panels.

[00:04:20]>> Yeah.

[00:04:21]>> You know, star star power.

[00:04:23]>> So, I've known you for 20 I've known you for 26 years now.

[00:04:27]It feels to me like I want to be you know, uh it feels like you've gotten much smarter or much more capable over this last decade. Do you feel that way?

[00:04:40]Do you feel like you just have better people around you, better tools? What What's changed cuz the level of um of audacity, you know, orders of magnitude.

[00:04:52]>> [laughter] >> Orders of magnitude.

[00:04:55]I mean, >> Some say insane.

[00:04:56]>> Insanity, yeah.

[00:04:58]>> audacious.

[00:04:58]>> Yeah.

[00:04:59]>> [laughter] >> I say hope.

[00:05:01]>> Uh what's what's How do How do you feel about that?

[00:05:05]What's changed? Do you feel that way? I mean, the scope of what your ability is.

[00:05:10]>> Um >> How do you self-reflect on that?

[00:05:17]>> Well, I've I've had to solve a lot of problems in a lot of different arenas, which um you you get this cross-fertilization of of knowledge of of problem-solving.

[00:05:29]Um And if you if you problem-solve in a lot of different arenas, then like what what is easy in one arena is trivial in a it is like what what is trivial in one arena >> Yeah.

[00:05:44]>> is a superpower in another arena. It's sort of like Planet Krypton. You came from Planet Krypton. That kind of thing.

[00:05:50]So, uh you know, on Krypton, Planet Krypton, you'd just be uh normal. Um But, if you come to Earth, you're Superman.

[00:05:57]Um so, if you take, say, um manufacturing of volume manufacturing of complex objects in the automotive industry, um I had to work on solving that.

[00:06:09]Um when translated to the space industry, it's like being Superman.

[00:06:16]>> Huh.

[00:06:17]>> Um because rockets are are made in very small numbers.

[00:06:21]>> Right.

[00:06:22]>> If you apply automotive manufacturing technology to satellites and rockets, uh it's like being Superman.

[00:06:29]>> Mhm.

[00:06:29]>> Um then, if you take uh advanced material science from rockets, and you apply that to the automotive industry, you get Superman again.

[00:06:37]>> Yeah. Fascinating.

[00:06:38]>> That's came from Planet Krypton. Back Back in Planet Krypton, this is normal.

[00:06:41]>> [laughter] >> You know, it's funny how how like the knowledge ports that that was true with Tesla and SpaceX being completely separate.

[00:06:50]>> Yeah.

[00:06:50]>> But, now they actually interact cuz you know, AI ties everything together. The orbit Yeah, the convergence is crazy.

[00:06:56]Like, I don't know if you visualize these parts fitting together originally.

[00:07:00]>> No.

[00:07:00]>> No? I mean, it's >> I didn't I didn't think they At this point, things are I guess everything ultimately converges in the singularity.

[00:07:07]>> Yeah. That's what I think, too.

[00:07:09]>> You have lots of different parts of the puzzle that you get to play with.

[00:07:13]Uh >> this one part that's missing, which is the fab.

[00:07:18]>> Got it.

[00:07:18]>> You going to buy Intel?

[00:07:20]>> You get it for a fraction of uh >> That's That was the That was what the bet we made.

[00:07:26]>> billion.

[00:07:28]>> Um I think it needs to be a new fab.

[00:07:31]>> Well, I agree, but licenses, real estate, ASML machines, it's not easy.

[00:07:40]>> Just get the assets and go.

[00:07:42]>> I don't think it's easy. That's why I mean I it's not like I think it's a a simple thing to solve. I think it's a hard thing to solve, but um but it must be solved. I've come to the conclusion that um >> Would it be Would it be solely captured by you or would it be an asset for the US?

[00:07:58]>> Look, I'm just saying that we're going to we're going to hit a trip wall >> Yeah.

[00:08:02]>> if you don't do the fab.

[00:08:04]>> Yeah.

[00:08:04]>> So, we have two two choices, hit the trip wall or make fab.

[00:08:09]>> But TSMC, for whatever reason, is massively worried about overbuilding, which is insane.

[00:08:16]Um but the whole world will be stuck with a shortage [clears throat] of chips for >> No, but so >> So, so, so they are actually they're I don't know if they're right for the right reason, but they're they're right. Um >> How so?

[00:08:29]>> Because it's actually like ch- wh- what is the limiting factor at any given point in time?

[00:08:34]Um the limiting factor say if you say like like 2 3 next year, like in 9 months, 9 12 months, the limiting factor will be turning the chips on.

[00:08:46]>> Power.

[00:08:47]>> Just power.

[00:08:47]>> Yeah.

[00:08:49]>> Uh you need power and all of the equipment necessary, power and transformers and cooling.

[00:08:54]So, it's it's not like you can just sort of drop off some GPUs at the power plant.

[00:08:58]>> And you vertically integrated >> You've got to you've got to >> again within XAI, didn't you?

[00:09:02]>> Sorry?

[00:09:03]>> You vertically integrated that inside of XAI.

[00:09:06]>> We designed our own transformers.

[00:09:07]>> Yes, and your own cooling system.

[00:09:09]>> Yes.

[00:09:10]>> But they're worried that if they make more than 20 million GPUs, like they make 40 million instead of 20 million, that 20 million will not find a source of power.

[00:09:19]Or they won't be bought because no one's going to turn them >> missing that prevents them from being turned on, >> Yeah.

[00:09:25]>> um they cannot be turned on.

[00:09:26]>> Yeah.

[00:09:27]>> So, uh th- they've they've got to have a power plant with excess with enough power. So, you've got to have enough gigawatts, then you've got to convert that from probably coming out of a power plant at you know 100 to 300 kilovolts type of thing.

[00:09:43]>> Yeah.

[00:09:43]>> Um you have ultimately got to go to convert that uh down to you know, several hundred volts at the at the rack level.

[00:09:50]>> Yeah.

[00:09:51]>> Um so, if you're missing any of the power conversion steps uh you you're you you won't be able to turn them on. And then you've got to extract the heat.

[00:09:59]Um so it it it's a big shift for the data center world to move to liquid cooling cuz they've used air cooling.

[00:10:07]>> Yeah.

[00:10:08]>> Um and um you know, if the consequences of a burst pipe uh are very substantial.

[00:10:15]So, if if if you if you blow a pipe >> [laughter] >> a water pipe in a data center >> Yeah, I know. I've seen that.

[00:10:21]>> you just you just fragged a a billion dollars right there.

[00:10:24]>> Yeah.