00:00Energy demand is surging as AI and data centers strain the grid.
00:04Now one founder is looking to space for a solution.
00:07Beiju Bhatt joins us now.
00:08He's the co-founder of Robinhood and founder and CEO of Aetherflux.
00:12Thanks a lot for joining us.
00:13Really good to have you.
00:14First and foremost, give our viewers, Beiju, a bit of an introduction to your work at Aetherflux for anyone a bit unfamiliar.
00:21And what are the main problems that you are setting out to solve?
00:25Yeah, thanks so much for having me on.
00:26Yeah, so I founded a company called Aetherflux a little while ago.
00:31And the mission of the company is to build a power grid in orbit.
00:35And the idea here is that the United States and the world as a whole are facing a massive energy crunch right now.
00:42And as we sort of venture into the era of AI, it's imperative that we find alternative sources of energy, different ways to power the compute in space.
00:51So I started the company with the idea of harnessing the power of the sun in orbit.
00:56And this idea of collecting electrical power in low Earth orbit is this kind of novel idea behind what we're doing.
01:04And the reason this is interesting is because in space, you have the ability to make solar farms as big as you want to.
01:12There's abundance of space.
01:13And also, you can put satellites in orbit where they're nearly continuously illuminated.
01:18So thereby providing power sort of as AI and other applications consume them.
01:26And I'll add one final thing, which is that a few, like about a month and a half ago or a month ago, we announced that a new initiative called Galactic Brain, which is a concept that puts the actual chips behind artificial intelligence directly in orbit.
01:41So right next to the source of power, so to speak, and is able to do the compute there and pass the data back down to Earth.
01:49So I got lots of cool questions or lots of questions because this sounds very cool.
01:56First and foremost, what are you joining us in front of?
01:59What is behind you for your shot?
02:00Yeah, so I'm at our office in San Carlos, California, and directly behind me is our clean room where we're in the stages, the sort of final stages of subcomponent assembly and integration into our first satellite that goes into space this summer.
02:16And these are some big old liquid nitrogen tanks that are powering different things in our clean room behind us.
02:23So I think these are for some of the space simulation TVAC chambers that we've got.
02:28So is the main advantage of going to space rather than expanding Earth-based supply, the fact that essentially if you're positioned the right way, there really are no shadows?
02:40You could take advantage of the solar energy and other things, or are there other advantages as well?
02:44There are other advantages in that right now, if you look out into the future, and in particular as new data centers, new energy streams come online, there's a massive delay.
02:56In some cases, five years, seven years.
02:59And what this technology allows us to do as we get it up to scale is say once the chips are available, we can directly integrate them into satellites.
03:08And once they go into space, they're effectively turned on as soon as they're illuminated by the sun.
03:14And so said differently, one of the major advantages of this at scale is going to be the speed of deployment of new chips.
03:21So as the industrial forces are manufacturing more and more chips for AI, we see a world where the limiting factor is actually going to be, where do you plug these things in?
03:32And if you do have a place to plug them in, how long are you going to wait to do that?
03:35And our approach actually just says, if you can turn them into satellites that will power them, the main step there is simply getting it into space via a rocket.
03:45And once it's there, it's on.
03:47Feiju, I'll be better to understand how much of today's energy crunch is being driven by AI and data centers.
03:54We're hearing a lot more about the overall, not just compute requirements, just how much energy here it's actually required.
04:01Yeah, I mean, it's kind of a it's almost an unfathomable, an unfathomable amount of energy that's being consumed here.
04:08And as we look out sort of even in the next three to five years, right, we see this as putting a lot of strain on energy grids, has the potential to result in energy price increases for everyday consumers.
04:22Because as these super hungry, energy hungry applications come online, they're plugging into the same grid that you and I are using, in many cases, to power homes, to power other applications.
04:34And so if we if we think into the future, say, hey, we need gigawatts of power for these data centers, you know, the energy production in the United States wasn't really planned for this level of scale this quickly.
04:48And so there's a lot of really interesting stuff that's going on in this space from things like, you know, reinvestment in nuclear power plants to the things that we're working on, which we think have a real long term plan once we get the technology online and demonstrating the capability.
05:06Paju, if I'm not mistaken, you are targeting a satellite launch for 2027.
05:10What will that prove?
05:11What do you hope to accomplish?
05:14Yeah, so we have two satellite launches that we're targeting this year.
05:17Those are actually working on a different application that beams power down from space to Earth.
05:23We're also targeting beginning first half of next year to put the actual sort of a cluster of interconnected GPUs that you would use for applications like inference interconnected onto one satellite that goes into orbit.
05:39So that's we think what one of the first applications of artificial in space is artificial intelligence in space is going to be.
05:45There's the, you know, as our president would say, big, beautiful data centers that are built for training purposes.
05:53Those, I think, are going to continue to live on planet Earth for the foreseeable future.
05:58But the actual interaction with the models, the using of artificial intelligence, that's where we see the majority of the energy use going in the years to come.
06:08And those applications are actually pretty well suited to put on satellites in space.
06:13Man, this is fascinating.
06:14I listen to interviews like this and I think, man, this is stuff out of sci-fi.
06:18Like, Rod Sterling at the Twilight Zone could never imagine this stuff.
06:21And you're reminded that technology is just like the future is already here.
06:24Before I let you go, give me some sense of what success looks like even 10 years out.
06:29I know that's a big, ambitious time horizon.
06:32But with an eye to the future, what are you most looking forward to?
06:36Yeah, I think there's a degree of inevitability to this idea in that as we build more and more compute applications,
06:44the demand for energy for artificial intelligence, there's no limit to this as we see it at this point in time, right?
06:50Like, the projection of the future just keeps getting bigger and bigger.
06:55And so having waste to harness energy that are both, you know, not problematic for people that live on planet Earth,
07:03but continue to let the energy demand scale the technology.
07:06I think space is a place that has a very natural application of this.
07:11And there's a confluence of a lot of things happening at the same time, from access to low-Earth orbit becoming more ubiquitous,
07:18that we think that, you know, it's hard to see a world where this doesn't become a part of the mix in the future.
07:24So very excited to see it scale to small scale first and then bigger and bigger over time.
07:30Beiju Bhatt, founder, CEO, Aetherflux.
07:32What a cool conversation.
07:34Had a very cool time.
07:36Best of luck for the road ahead.
07:37And we'll continue to check in.
07:38Join us again as progress continues forward, Beiju.
Be the first to comment