Solving the AI energy crisis

So we are here with Greg Osuri. Don’t want to trigger the other thing near me, so I’m not going to say it more than that, Greg… We had a fantastic conversation a few months back to sort of prime the pumps to this conversation, really about powering the future of AI: data centers, megawatts, waterfalls… All the crazy stuff that goes into building data centers and all that good stuff of the future of what we’re doing. But recently you testified in front of Congress around this. What is the burgeoning topic here? What is the concern?

Yeah, so… I mean, AI is taking an incredible, I guess, place in all our lives. And while there is an amount of progress that’s being made, from let’s say solving cancer, to securing the homeland, all that progress comes with an enormous cost, and that’s in energy. So the amount of energy we need is reaching a point of what we call a crisis. And President Trump declared energy as a national emergency recently. So I was invited to testify before Congress on this particular subject, as to how much energy is being consumed by AI and what are possible scenarios in the future if we don’t do something about it, and what are some of the solutions we think we can try out nationally, as well as internationally, to solve this crisis.

So yeah, it’s a great opportunity, and I think my testimony covered a pretty wide breadth. I went into the report from the Department of Energy, the official sanction report by U.S. Congress, which was conducted by Lawrence Berkeley Lab, and happy to get into it if you want.

Well, I’m interested in just how much energy we’re going to need, or how short we’re going to come up. I did read a 40% figure, I think, in your testimony…

So the Department of Energy - it’s a government organization. It’s known for its conservative estimates; not real, actual estimates we call them. Their conservative estimates amount to about 500 plus, I think, 508 terawatt hours by 2028. For context, that’s about 12% of U.S. energy consumption for AI. The problem is - and that’s a huge jump from 2026, I believe, and we consumed about 4.4% of energy. So it’s an incredible jump in terms of percentage consumption.

And number two is, the problem is the U.S. grid grew about 1.2% in the last 20 years. So the challenge is the grid is not growing fast enough to service AI workloads. Gartner, which is an analyst, predicts by 2026 - that’s next year - about 40% of AI data centers will be out of power, because we will be forced to a position where we have to make a decision as to who gets the energy. Is it the homes, or is it going to be data centers? I mean, obviously, we’re going to prioritize homes over data centers.

The challenge is, in the West our grid is extremely hard to upgrade. And it’s also extremely hard to produce energy, considering we have exhausted all the natural resources for the most part, and we don’t have enough hydro, we don’t have enough dams; we don’t have enough sort of like naturally generating energy, so we have to either build nuclear plants, or we have to look for alternatives. Nuclear plants give about a gigawatt of energy per plant. The only problem is it took about 14 years to build the last one. So we can’t build them fast enough to address the immediate need. And even if we do build nuclear plants, we can’t transmit them, because the transmission infrastructure in the States is extremely hard to upgrade. I mean, we had electricity – we’re one of the first countries… In fact, the first country to have electricity. So when we designed the grid, it was designed to be – they didn’t quite have the scale in mind when they designed the grid. So if you want to upgrade the lines, there are two ways to do it. Number one is - well, put an additional line, power line; and no one wants additional power line running through their backyards because, and that’s real estate values, it’s going to plummet. Or you’ve got to do interruptive. That means stop energy, upgrade the lines… These are literally copper wires. Like, you have thick copper wires. Replace them by interruption, and nobody wants their power to be interrupted. So you can’t put new lines, you can’t upgrade existing lines. So how do you upgrade the grid?

[00:08:38.05] China is a communist country. It can just basically say “You have no option.” America and the West in general, or Western democracies, it’s very, very hard to upgrade lines, because it has to do with property rights. So the third option is to underground. And that seems to be a feasible option, but the cost is so high that you have to literally justify how you’re going to dig a line under the ground and upgrade infrastructure there. So either way, it’s extremely challenging to upgrade infrastructure in the U.S.

So you can say “Oh, maybe we can do renewables.” Well, renewables, if you think about it, solar is terrible at utility scale.

And if you do utility solar, you end up with the challenge of transmitting energy. So you can’t have large swathes of solar panels… And the expense of storage, which is batteries, is super-expensive… And you need more energy to cool the batteries, so the efficiency goes down… So solar is not really an option right now at utility scale. Solar is good at residential scale. And you really don’t have – wind is not practical, because it’s very location-specific. And if you do have wind, you’re not able to transmit… So now we’re in a limbo as to what to do with the energy problem. So there’s no easy solutions, let’s put it that way, and it’s not stopping.

So right now, the only solution is burning fossil. So if you look at the XAI data center that trained Grok 4, it has about 150 megawatts of capacity in Memphis, Tennessee. They consume about seven megawatts from the grid. The rest, they’re burning gas, LNG, to power data centers. Well, the effects of that is increasing respiratory conditions in the Southern Memphis area. There’s a new lawsuit that’s, I guess – I’m not sure what the status of it is. I think it’s [unintelligible 00:10:45.24] Center has sued xAI about health issues that are coming out of these data centers.

We just heard yesterday Sam Altman tweeted out saying that they have acquired 4.5 gigawatts of capacity in Abilene, Texas, in partnership I believe with Oracle. For context, 4.5 gigawatts is about five nuclear facilities, equivalent. There’s no way in hell they’re getting that from nuclear, because we don’t have free nuclear plants in America. I mean, we have about 96 reactors that are fully consumed 93% of the time. There’s no nuclear reactors energy. So where do you think they’re getting the energy from? Burning fossil. That much amount of LNG will produce anywhere between 2.5 to 2.7 million tons of carbon, net new carbon, into the environment. And that’s equivalent to about two thirds of all the CO2 emissions from Vermont. So we can shut down Vermont…

Two thirds?

In 2023.

That’s more than half, right?

It’s more than half.

Just checking. That’s more than half?

Yeah.

That’s a lot.

So over 66% of CO2 –

So what are you gonna pick, Vermont or OpenAI, right?

I’ll choose ChatGPT…

[laughs]

[00:12:13.29] I mean, we’re talking about Vermont being a very agricultural-heavy state, with a lot of methane and all that… Just to give you context, this is one data center we’re talking, one company. So put that with the heavy competition between - what’s his name… Mark Zuckerberg tweeted out - or Facebooked out - saying that they’re going to build a data center bigger than Manhattan. I mean, I don’t know if you saw that tweet…

I did not see that.

Gosh…

Everybody is competing. Yesterday Elon Musk said they’re going to – in fact, they’re going to get about 5 million GPUs by 2028, I believe he said. Or next five years, so maybe 2030. 5 million GPUs - he said 5 million [unintelligible 00:12:56.02] and that’s five times one kilowatt, right? 1.2 kilowatt per GPU. Just to give you a brief idea as to how much compute or how much energy we’re looking at. Oh, sorry, 50 million. These numbers are so big…

Not 5 million, 15. 15.

50.

Oh, 50 million. So 10x what you earlier said.

Yes. Apologies. I mean, these numbers are so large that it’s hard for me to fathom.

Yes, it’s hard to actually fathom. 50 million GPUs…

For context, NVIDIA made about 2.5 million H100s in 2024. So… Not financial advice, but NVIDIA, I believe, is going to be extremely beneficial.

You think the stock is high now…

Wait till you see the GPU demand that’s spiking. And you need that.

That GPU order coming from XAI.

And you asked me why do you need that - it’s because historically speaking we looked at data all the way from 2013, and looked at the amount of energy we need to train state-of-the-art models. The amount of energy we need - the installed capacity required to train state-of-the-art models all the way from ChatGPT 3s, even ChatGPT2, to Grok I believe 4, is doubling every two years. So right now, the OpenAI data center that’s operational in Texas draws about 300 megawatts of capacity. So doubling that means by 2030 we’ll need a nuclear reactor at the minimum to train the state-of-the-art model. So be it ChatGPT 6 or 7 is going to be a nuclear-powered data center.

When you say the one in Abilene, are you talking about the Stargate megafactory that’s being built?

Correct.

So that’s not online yet, right? That’s gonna come online I think in like the next 18 months; it’s literally dirt and being built 24/7. It’s an intense scenario there.

Yeah. Soon.

I mean, they’re working their butts off. They’re literally working around the clock. It’s a 24/7 operation to build this thing.

The XAI data center, they brought it online in 20 days.

Okay.

At Elon Musk speed. So that’s a new precedent in terms of speed. And Jensen Huang was just on stage, he was like unbelievable at the rate Elon moves in terms of building this stuff… There’s a rumor that I heard - since they don’t have generators in America that can produce that much amount of energy, he’s literally buying a generator from God knows which country and shipping that to America to produce energy. I mean, because Elon thinks from first principles. So there’s all kinds of crazy things. The development is so rapid in the last one month or so… It’s really hard to keep up.

I haven’t quite sat down and looked at all these numbers people are claiming that they want to reach, and really do an analysis, because these are brand new numbers that came out yesterday… So you have competition between Meta, Google, xAI, OpenAI, Anthropics of the world… And these are just early players. And these are just American players.

[00:16:06.24] I mean, forget the Chinese and the rest of the world. So it’s not slowing down, the demand for GPUs and demand for energy. Now, if we don’t do something about it, we will start seeing rolling blackouts by next year. We will start seeing energy crises similar to the oil crisis in the 1970s, or even much larger scale right now. And so there’s all kinds of concerns, and that’s my testimony about it.

Yeah.

Sorry [unintelligible 00:16:36.06] but I do have some hope… There’s some hope though.

What I think is interesting though is that – one of the things you mentioned in your testimony, and I just glossed through some of the details from that, was that while you said all those things and while you mentioned all those human nature feats, we must somehow navigate, because there’s no putting this genie back in that bottle. It’s just not gonna go back there, right? You’re arguing that this isn’t just another tool. This is becoming foundational, like electricity, or internet. This is a foundational thing, so we can’t not solve this problem. And I know we have politics, and countries, and political powers and stuff like that… And I just wonder, how do we unite humanity around this movement? Is it possible? Is the geopolitical powers just too strong and too inable to share the information and think humanistically, versus China versus US, or Europe versus US, or whatever the comparisons may be? How do we get to a place where it’s like “Humanity needs to solve this problem, because everyone across the entire globe uses electricity”? No one argues about that. We’ve figured it out. We’re still figuring it – we’re figuring it out every day, or every year, iterating towards how we do better electrical grids. As you mentioned, the challenge to even upgrade. How do we unite around this as a human race, versus a country, or geopolitical?

The two challenges to overcome, the way I see it - the first challenge is people are fatigued with the climate activist agenda. So anything that’s related to climate or carbon emissions… Like, I’m a pragmatist. See my testimony. I was like “Well, this is the situation. We can’t sacrifice progress for other things.” But it doesn’t matter what side of the political angle you are, if you don’t like dirty air. Because it’s literally dirty air. Like, if you’re burning coal, or if you’re burning fossil, the air gets polluted, and nobody likes dirty air. That’s my concern. Others… Carbon emissions - questionable and whatnot.

So that’s a challenge. How do you rally politically a crowd that’s fatigued with this agenda, the climate agenda, over the last decade or so? That’s one. And second is we have right now a very challenging national security posture when it comes to AI. China is outbuilding, and they’re doing a phenomenal job with renewables and whatnot. I mean, don’t quote me on these numbers, but somewhere I saw a stat where they’re building more solar every year than the entire U.S. grid; or they will be, in a few years. So they’re building incredibly well, and they had this vision about 10 years ago, to build this infrastructure, and the U.S. fumbled because of the climate agenda.

And we have these national security hawks to convince. So U.S. needs to have a leg up. The only way to do that is through getting more energy in whatever ways we can get more energy. So how do we unite people, spread the message? I mean, I’m doing my part…

Yeah.

[00:20:05.02] …but also, show hope. So, I mean, it’s not all dark days ahead. I mean, we do have some hope. I was in ICML conference last week, which is the most prestigious AI conference… And for the first time, I’ve noticed people were a lot more cognizant towards the energy crisis. And I was fortunate to give a talk there… Usually, you only give a talk if your paper gets accepted, and the papers have to be peer-reviewed, academically rigorous papers… I was the only one that gave a talk without a paper…

Wow.

But the general feeling is the reason why you need so much energy is because training is very co-location specific. That means you need large swathes of GPUs located as close as possible, connected with the highest possible connectivity to do training. And that’s just getting bigger and bigger and bigger, the bigger datasets we get. That makes sense, right?

So you need very, very large quantities of energy in a concentrated manner. So you can’t do distributed AI. You can’t train half the AI in one data center, another half in another data center. That was about a year ago. Now, over the last year, we saw incredible progress in distributed training. It’s not quite as good as the non-distributed co-located training, but the results are very, very promising. The three main areas that we are seeing progress in - the first is low communication algorithms, we call it. What that means is, by reducing the amount of communication required by different layers of AI when you train, you can reduce the need for connectivity. So you can have multiple GPUs in different data islands, we call them, and somehow like train them. So this method was proposed by Google DeepMind, which is the most advanced AI lab in the world, arguably. I mean, they created transformers, which is a GPT thing.

DeepMind proposed a framework called DiLoCo, which another company called Prime Intellect implemented that and actually was able to train a 32-billion parameter model. For context, LLaMA, which is the first open source model, was about 15 billion parameters. So it’s a little bigger than the first open source model. Still not state of the art…

Another context is that GPT-3 was a 150-billion parameter model. So it’s like far behind GPT-3. GPT-3 was the first model we used for ChatGPT, which was [unintelligible 00:22:46.09] So we are far behind, but 32-billion parameter fully distributed is a great, great outcome.

Another company called News Research, which also has another algorithm called Distro, similar - they were able to train a 15-billion parameter model. Now they’re training a 45-billion parameter model, fully distributed. Another promising company called Pluralis - I’m very excited about this company… They’ve figured out how to do the second level of optimization, which is fault tolerance, and asynchronous training. I don’t want to get too deep in the weeds as to why this is important, or how it works rather… Why it’s important is using what they’ve figured out - which has been a very challenging problem, to do asynchronous machine learning training. Asynchronous means it lets no – so machine learning training, or training right now is heavily homogenous. That means if you’re using H100s, you have to use H100s across the board. If you use any less powerful GPUs, that one GPU will hold up the entire training run. It’s called straggler nodes.

[00:23:55.18] These guys figured out how to do asynchronous training, where you can have slower GPUs mixed with faster GPUs, distributed across the world, using something called a swarm training mechanism. They can all somehow come together and train a model with 1% or less loss in accuracy… Which is very, very good in terms of results. So it’s very promising. Now, that’s not in code yet.

And another company called Gensyn, which is also – they’ve raised quite a bit… They’ve figured out how to do another fault-tolerance mechanism similar to what Pluralis did. And there’s other companies that are doing like zero-knowledge proofs for gradient synchronization… So what that means is you don’t have to verify all the gradients, so the gradients can come from – and you can verify all the gradients without using zero-knowledge proofs and whatnot.

So the point being, there is a lot of work that indicates distributed training will become a reality by end of the year. By end of the year, we will produce a model as good as GPT-3. By end of next year, or maybe going into 2027, there’s a good chance that we’ll be able to produce a state-of-the-art model fully trained distributed; or decentralized, rather. And decentralization on top of distributed models give you incentives. If you all know, Bitcoin is the largest supercomputer in the world. Why? Because there’s an incentive structure to contribute a computer to Bitcoin. Similarly, if there’s an incentive structure to contribute your compute to train a model, there is a possibility we may actually out-compete OpenAI’s of the world, because now you have the public. And there’s lots of GPUs around. Maybe I have a 4090 here, a gaming machine… I have a Sony PlayStation that I barely use, that’s got powerful enough GPUs… And there’s so many GPUs out there that were discarded by OpenAI’s, because they’re not the latest and the greatest… There’s older GPUs that are functional, they’re still around… And the best part is we can distribute GPU now. We tap into energy when they’re available. It’s not like we don’t have enough energy. We just don’t have enough concentrated energy.

What that means is, during the day in California – I mean, California I think roughly takes about 60 gigawatts of energy. The installed capacity. It is the largest consumer of energy in US. During the day, California produces, or captures rather significantly higher amount of energy than we can use from solar. A lot of solar in California. In fact, the energy is so much that it cannot transmit back to the grid, or else it’ll break the grid. Because our grid can’t handle load. So they had to literally waste energy, or pay you to use energy during the day.

That’s right. I like that.

However, the moment the sun sets… You don’t think of these things like actual –

Everything flips.

…actual electrons moving, right? They have to turn on all the non-renewable sources during the night, be it nuclear, be it coal, be it LNG, be it whatever source, to power homes within a matter of seconds. If they don’t do that, you will have a rolling blackout, and it will destroy the grid. Our energy grid or grid in general is so fragile, if you think about the actual operations that happen… And if you study the grid as much as I’ve studied, you would want to go full sovereign. I’m building a house in Texas that is completely off grid… Because it’s so fragile. I mean, I can’t emphasize how delicate this world works. Everything’s just in a little balance; a little extra thing will destroy the grid.

[00:28:02.13] There’s also a lot of money to be made if you can actually supply energy right when the cutoff happens, from like renewables to non-renewables, in the evening. There are companies like Allied Energy, which is a [unintelligible 00:28:13.29] company, and there’s another company called Base Power, based in Austin, started by Zach Dell, which is Michael Dell’s kid… There’s a lot of really cool companies that are coming out now, that can do arbitraging for energy. There’s lots of stuff that we’re working on with Akash as well, that I’m happy to get into, to solve this crisis.

Such a big topic, man. That’s why I was telling you, Jerod… When I sat down with him, I was like “Oh, my gosh”, he almost made my head explode. I think we talked for like two hours, didn’t we? It was a deep conversation, I feel like. I took a ton of notes… I was like “I cannot wait to talk to you on this podcast.”

Tell me if you – you’ve probably read this book from Thomas Friedman… It was back in, I think 2008. And at the time, I thought “This is never gonna happen.” Like, I can see the world flattening. So Thomas, before that I believe had a book called “The World is Flat.” And then this is the second book, as a follow-up to this idea of the world flattening. It was talking about the employment structure globally; how offshore - I think they called it offshore - hiring folks from India or different places throughout the world… If you’re in like the US, to hire a team from India… The world was flattening from an employment standpoint. You can work from India, or China, or Europe, and anywhere you want in the whole world you can work. And it was a flattening of the employment structure.

And then it was this book that followed up with that, “Hot, Flat and Crowded.” And in there he talked about all this different stuff with this energy, and I was just like “This is never gonna happen in our age.” At least I knew it was gonna progress, but I never thought that it would accelerate… But does that book ring a bell to you? Have you read that book?

I think I’ve heard about the book… But I think I have an idea where they’re going with the book.

I mean, it’s kind of going there, though. I mean, if you go back and listen to this… This is 2008. And it’s so pertinent now. It’s like hot, in terms of - obviously, like, it’s pretty hot. Air conditionings can’t have the power to be consumed. AI wasn’t discussed in this book at all, but it did talk about the fragility of the energy network, which if we’re trying to build the next evolution or the next major innovation for humanity, as you’ve said, it’s gonna take all this energy. But there was so many precursors mentioned in that book that I just never thought I would be alive to see this day, essentially, to see it really be hot, flat and crowded.

And the water problem too is another big problem I think we’re not talking about too much, too. This water and energy – I mean, these both are like the essence of life. If you have these both, you can do a lot. And then some of the problems that I look at… The more I study the grid – I look at California, for example. I love California, but I think the state of California, the government is one of the dumbest people alive. It just breaks my brain as to the dumb decisions they made over the last 20 years in the name of climate agenda, in the name of whatever, that destroyed the state’s ability to be reliable in the next 10 years. It’s just – my God, okay?

What are some examples? What do you mean? What decisions have they made?

Water.

Water. Yeah.

The water, I mean…

They’ve diverted water. Didn’t they have like a lot of water diversion that like totally changed ecosystems? They messed with things to a degree where it was like playing God with landscape, and geography, to the point that it’s done major, major change. That’s one thing I’m aware of, at least…

[00:31:51.28] Yeah. It was a human catastrophe what happened with the LA fires recently. That should have not happened at all; if they didn’t divert water the way they diverted, for reasons that are just dumb… You know? And also, California – I mean, earth is 70% water. If we think about it… Why are we not using the water? Well, it’s saltwater.

Saltwater.

Well, how do you use saltwater?

Desalinate.

You can desalinate. Well, what does it take to desalinate?

More energy.

About 8 to 12 watts of energy per gallon to desalinate, on average, to do reverse osmosis. Well, don’t we have excess energy? We do have excess energy that produces during the day. Why are we not building desalination plants, question mark? Oh, there was a proposal, but environmental concerns that desalination will produce too much salt. We don’t know where to store the salt. Well, how about digging a hole, and putting the salt there?

Put it back in the sea.

Or how about export the salt out to – throw it back in the sea. That will destroy the reef.

Okay, sorry. Bad idea.

Imbalancing. But there’s so much – if you go down like four levels deep and you ask questions, it’s just almost embroiled in someone doesn’t like, because they think it’s gonna be bad for environment. So, well, is dying and fires okay for the environment?

I mean, even the water there, you’ve got the challenges in California with that… And I don’t know – I know you’re in Austin, like I am, and then if you’re here in Texas, or at least close to Austin, you know what happened in Hill Country with this major flood that just happened… The whole world’s aware of it now. I mean, in moments, like overnight - I’ll come close to crying - we have friends who died in this flood; really close friends, whose kids are not coming back. And it’s a shame, because you’ve got – I don’t know how much of this is about water diversion, or being stupid, or making poor political decisions with different choices you have to make, but… These are things that don’t have to happen. That water [unintelligible 00:33:52.20] was just insane. It was a major flash flood. Was it because of water diversion, or things over time? I don’t know. But what I can tell you - and you know this - is that we are constantly… And I don’t know how much in Austin you are, but over here in Dripping Springs, west of Austin, we’re constantly in a state of drought, up until this flood, basically. It reinvigorated the aquifers… It’s not back to normal, but Lake Travis is actually a lake again. It’s not very little water and a few islands and boats in the land out there. It was laughably not a lake. And now it’s a lake again. So I don’t know how this plays into it, but that’s a pretty big deal, the power of water.

Yeah. I haven’t looked into the actual flash floods and the reasons… There’s a few speculations out there, but it’s very sad what happened. We lost someone from our neighborhood as well, a little girl, eight years old, at a camp. It was so sad, because they were at a little girl’s camp, and I have a little daughter, and it’s just… It hit me.

Yeah, man. I mean…

I live in Westlake right now, so we don’t have –

Then you know then, in Westlake; you know how close we are to the water challenges here. So you mentioned water challenges in California, but they’re also happening here. I mean, even our own expansion is…

Well, we can claim that Texas doesn’t have too much ocean, except for Houston… I mean, so you can make a claim that “Well, we don’t–” I mean, our sources of water is just rainwater capture. That’s what I’m doing in my home.

That’s true, yeah.

But California has an entire coast, so it’s like the dumbest problem.

Yeah, I do agree with that.

And most of the population lives in the coast, in California. So you have most of the energy production happening at the coast, and instead of burning energy, why don’t we just desalinate water? That’s what Israel does. I mean, that’s what in the Middle East, the whole – I mean, they’ve been building oases in the middle of deserts, in Saudi Arabia. Excuse my French, but it just makes me so mad, looking at the things people could do with energy, in the desert, versus - California is one of the most fertile places on planet Earth, and the most beautiful places.

[00:35:59.20] Either way, there are a lot of problems I think with enough motivation can be solved. I hope California state will take energy more seriously. And one of the worst things they did was to ban all fossil vehicles from 2030, without thinking about how they’re gonna upgrade the grid. Along with the AI challenge now, the EV challenges are catching up. So states that have EV mandates in terms of production mandates are going to suffer the most with AI. And I think it’s just gonna get worse and worse, the energy problem.

So I read a couple of weeks ago about micro-reactors. Is this something on your radar, micro-reactors?

Oh yeah, I invested in a bunch. Looked into a bunch, too.

So this seems like – okay, it doesn’t seem like there’s one big solution to all these problems. It seems like perhaps there’s this litany of smaller solutions that all work together. Even your description of the companies trying to do swarming, trying to do this other stuff… What if we could decentralize and incentivize your average Joe or your small business that has some GPUs to make them available to the network, so that you can use them and draw energy when it’s not a peak time etc? Like, all these things together might approximate a solution. And I’m curious if micro-reactors, or - there’s also small nuclear reactors, which I think are bigger than micro-reactors… I’m not an expert here, Greg. These things might be additionally – like, onsite. Like, we’ve got a new plant, we’ve got a new reactor for that plant. And maybe they’re small enough… These are like semi-truck size, I believe…

Yeah.

…to actually solve a little slice of the problem.

Yeah. I mean, I’m very pro next generation energy generators. So I’ve looked into quite a few, I’ve invested in quite a few… I wouldn’t consider myself an expert. I mean, I don’t have nuclear physics as my subject expertise, but I’ve looked at it from a progress standpoint. Progress has been extremely slow on these things. There are very few companies - there’s NuScale, there’s GE, there’s Westington, I believe… There are a bunch of them. And the progress has been extremely slow. There’s enormous regulation for nuclear. I mean, right now – so there’s something called Interconnect Network in America. So Interconnect Network is the network maintained by the Department of Energy. Whenever there’s a new energy source, you have to register that with the Interconnect, and get an Interconnect approval. It doesn’t matter if you’re using that energy locally, or whatnot.

So the Interconnect queues for nuclear is about 10 years. Now, you can deregulate, but nobody wants to deregulate nuclear, you know? Even though the first nuclear reactor built in America took about 14 days, the last one took about 14 years. Most of them is just being regulated, because it’s a very dangerous, very challenging…

It seems like the sentiment around nuclear - now, I’m in the Midwest, so maybe I have that color of sentiment, versus elsewhere in the country… It seems like the sentiment towards nuclear is changing, because I definitely grew up in the ‘90s and early aughts where it was like “Nuclear, bad. Scary. Big mushroom cloud in the sky.” And that’s definitely not what your average Joe and Jane think today about nuclear. So you’d think maybe the – what do they say, politics is downstream of culture… You’d think that some of that stuff would start to move the needle.

[00:39:32.09] Absolutely. I mean, nuclear is a lot safer now. I mean, nuclear is my favorite source of energy. It’s very clean, it’s very safe… We’ve figured out a lot of uses, even with the nuclear waste disposal… I invested in another company that does isotope fuels… But the challenge of nuclear is – it’s great in every way. I mean, if we can solve the regulation problem and if we can solve the public perception problem, I think we have a winner there. But the fuel enrichment is very, very hard. It’s very hard to get fuel. Africa has – there are parts of the world that has fuel, but all that fuel is controlled by China.

So China has done a phenomenal job over the last 20 years in controlling rare earth minerals, including nuclear fuel. They control the entire supply chain. They bought companies that do this. They bought the mines. They bought the ports in Africa. They bought the whole nine yards for nuclear… Because US and the West fumbled on nuclear, the nuclear producers, the supply chain, went towards the buyer, which is China. So there’s a geopolitical – I mean, look at Trump talking about rare earth minerals, talking about making deals with or liberating Greenland, or deals with Ukraine - all that mostly is for rare earth minerals, including nuclear. Even if you have a nuclear reactor, getting fuel is next to impossible in America. We don’t have enrichment facilities in America no more. We used to.

So we have to rebuild our nuclear capabilities, energy capabilities… I don’t know how long that’s gonna take. But with the latest advancements… We have isotope reactors, which - one reactor is based on uranium-233, not U-235. So it’s basically thorium. So thorium is a side product, a waste, a nuclear waste, that has a half-life of 400 years. It’s still a very, very good source of energy. So if you can put them in a battery - a company I know is doing that - you can effectively create nuclear batteries that’ll last 400 years. They can be placed anywhere in the world, without worrying about waste.

So there’s all kinds of new things that are coming up in the nuclear space. Fortunately, there’s more investment from DOE that’s going into the nuclear space now. This current administration has been very pro-nuclear as well.

So there could be good news in the next four years, depending on how the market goes. I’ve seen an increased amount of investment, and I know my friends are investing quite a bit in energy… And SoftBank alone… I mean, we had SoftBank, Javier, who’s one of the key persons in the Stargate program at Akash Accelerator Conference in New York… He had a lot to say about nuclear, and nuclear-related energy sources. So all in all I’m very pro-nuclear, but I’m not very bullish on the regulatory landscape to move fast enough to address the need.

Right. To avoid the impending doom, the crisis which will hit here soonish.

Very quickly.

Two years? You said a year, or two years?

2028, that’s what the DOE talks about. 2026 is when we’ve gotta make choices.

Rolling blackouts.

By ‘28 you’re gonna see a bigger crisis. 2030 I think is gonna be even harder. I’m more hopeful on the offshore wind. Wind is the cheapest form of energy, hands down. If you can get your wind, proper wind… I mean, it depends on how windy an area. If we look at Des Moines, Iowa, which hosts the largest wind-based data center, which is owned by Google… It has about 500 megawatts. It’s almost half a nuclear reactor, all from wind. And that’s a [unintelligible 00:43:21.27] and we don’t know whether we can do that again. We have that location good enough to be able to replicate what Google did in Iowa. But offshore wind –

How many windmills does that require?

[00:43:36.00] I haven’t looked into it, but it’s quite a lot. I mean, 500 megawatts is a lot. So offshore wind is very promising, actually. There’s a lot of wind on the ocean, if we can tap into it. It’s very challenging. I don’t think we’re gonna see an offshore windmill come alive, at least by 2030. So there’s a lot of promise. There’s quite a lot of using waves as energy form as well, because the ocean waves are very, very good in terms of capturing energy… There’s all kinds of cool things people are working on, so I’m hopeful that we may be able to find a solution, but I’m not hopeful that we’ll do that fast enough.

In the meantime–

I think decentralized AI is the only way.

That’s the only way.

I mean, I’ve looked at every possibility. I’ve looked at every possibility, lay it out, and there is no magic solution to get more energy in America in the next four years. Guaranteed.

Well, let’s say that you decentralize - I’m not saying you, Greg, but you, the royal you, people working on decentralization. Get that working. Don’t you think there’s still gonna be massive centralized players that just say “Nah, we’re just gonna keep buying up as much energy as we can, and citizens be damned?” I don’t know…

No, there will be. I mean, there will be a lot of players that’ll go after it, there will be a lot of protests, energy protests that are going to be after this… I think there’s going to be a political – when people start getting sick and dying, like in Alabama and Texas… I’m really concerned about the amount of – I mean, a 4.5 gigawatts facility will produce as much exhaust, equivalent to exhaust coming from 2 million cars.

In one place.

In one place.

Crazy idea, okay? This is stupid, it won’t work… But reptilian brain says this. It says “Can’t we just build this massive flex pipe, I don’t know, that goes up into space, and exhausts it all into space, because space is just forever?” I mean, it’s a dumb question, but is there a possibility to exhaust out of the atmosphere?

Well, there’s easier solutions. You can do carbon scrubbing. Like, you can recapture scrub… I mean, there are possible solutions. I don’t know.

So it’s a dumb idea, for sure. Okay. Science fiction.

I mean, it’s possible and it’s not. I mean, think of first principles. That’s something Elon would think. Like, remove all the limitations and think as big as possible. Carbon recapture could be a solution. I don’t know how effective and how it would work at a 4.5 gigawatt scale, but that’s a lot of carbon. I don’t even know the carbon footprint of producing the carbon capture devices. But no one is talking about them, no one is even doing it, because I would imagine the cost is so prohibitively expensive to do these things, that it’s going to outweigh the $500 billion that you have for building your data centers.

Well, if you spent 1 billion on the pipe…

1 billion is nothing.

I’m saying, yeah. Like, 1 billion is nothing.

1 billion used to be a lot of money.

Comparative. I mean, it used to be. If you spent 1 billion on this pipe, this tidally locked aircraft that orbits Earth, that keeps the thing connected… I don’t know. Somebody can do that. Exhaust it out there.

Yeah, but I think it’s gonna be a real problem – I don’t think it’s gonna be a solution by the time they’re done building it. And when people get 2 million –

Especially if they’re building that thing in two weeks or three weeks, or however long they take now to build one of these things.

Yeah. The 2 million worth cars exhausts in Albany, Texas is not going to make any friends in the area… It’s already destroying – and think about the water. The water for cooling. That big of a facility needs a lot of water, because you –

And you need to dump it somewhere. Where are they dumping their water?

Exactly, we don’t know.

I’m sure it’s not in a good place.

I tweeted out recently, they’re talking about this one community in Texas, I think it’s in Albany… They’re talking about how their water they’re drinking is so toxic now. It’s getting bad. It’s getting really bad. To a point we’ll start experiencing these things every day.

They should train these things in Antarctica, you know? No one’s living there, and you don’t have to cool, because it’s already cold.

[00:47:58.20] Well, you have a bandwidth problem in Antarctica. You can’t get latency…

Solve that, Greg. Solve the bandwidth problem. It’s just one pipe, you know? [unintelligible 00:48:04.06] out of the atmosphere…

I’m very, very bullish on the satellites. I have a Starlink here at home that I use as my backup internet… It gives me about 500 megabytes, which is not bad. If they put more satellites, more density of satellites that covers pretty much the entire Earth, then we can have good enough energy, a good enough – what do you call… Bandwidth to source from Antarctica. That could be a solution… But I don’t know. It’s looking bad.

Not that there won’t be any unintended consequences of having so many satellites in the sky.

Correct. That’s another problem.

We’ll have a new ozone layer around us.

We saw WALL-E, right? We saw when the camera zoomed into the Earth on the movie WALL-E, the sky was lit with satellites, and it’s basically space junk at that point.

Right.

We’re going there. And these are low Earth orbit, LEO satellites. They’re not in the geostationary orbit. So you need a lot of them, very close, because for the latency, right? I mean, you do have geostationary internet - super-expensive. I mean, they’re used for Antarctica, basically. Antarctica doesn’t have LEO coverage. So you can’t put a Starlink in Antarctica. It doesn’t work. You have to put another company - I forgot; they do geostationary. Much, much higher. [unintelligible 00:49:17.29] but you get some internet in Antarctica.

Right.

So you know, possible, but it’s just – a lot of this stuff is just very, very complicated.

Yeah. So if you add it up - back to the decentralization concept… If you took all of the wasted energy - I’m thinking mostly of these solar panels in California during the daytime situation - and you just like extrapolate that around the world, and say “How much energy do we have captured that goes unused?” And you add all that up and you put GPUs, you plug GPUs right into those suckers, and you hook them up to the grid, the electrical grid and the internet grid, and allow people to lease them, or rent them, or whatever… Does that solve the problem? I mean, do we get there?

Absolutely.

It does.

I mean, I know for sure we produce a lot more energy in the US than we consume during the day. I know for sure.

In solar.

In solar. Solar, wind, a whole lot of things. We have a lot of sources in the US. I mean, there’s been enormous investment that went into renewables over the last few years, or the last 20 years… And that is a phenomenal solution. That’s what we’re going towards. So why burn energy, but instead use that energy to train AI where it’s windy in Kansas; move the AI workload there. When it’s sunny in California, move the AI workload there. The AI workload should be elastic, and it has to be asynchronous. It has to be async. So that’s why I’m particularly excited about the progress with distributed training, because now we can have elastic workloads that can move where – there’s energy-aware schedulers. So at Akash, we’re developing one of those… I’m sorry, I forgot to tell you what we’re doing here at Akash. We are developing energy-aware schedulers that will pick the lowest energy - so we’ll start advertising the cost per energy now.

Okay.

There are lots of data centers in the US that can give you one cent… I mean, economics-wise, right?

Yeah. Give me the cheapest thing I can get.

Right. The cheapest thing. It’s actually beneficial, not only for the environment, but for your pocket. So that’s where capital is coming into play. Wind will give you a one cent per kilowatt hour, which is the cheapest possible energy you can get. Solar, depending on the setup, depending on the battery cost and a whole lot of things… Solar, it can get very, very cheap. And if you can arbitrage, you can get like almost zero.

LNG, natural gas will give you 10 cents a kilowatt hour. 10 times more than what you would pay for wind. Diesel will give you 32 cents a kilowatt hour. It’s the most expensive option. You use diesel as a backup generator. The grid, depending on where you are in California, I believe it’s 20 cents a kilowatt hour… If you just plug in from the grid, which is diverse sources. Texas is significantly cheaper. I think it’s like 10 cents or below a kilowatt hour. So just from a pure economic standpoint, the cheapest thing is to go where the energy is cheap.

Break: [00:52:17.21]

I like the idea of the workload moving where the energy is prominent, at the time it’s prominent, versus co-locating and like destroying, because that seems like just terribly bad. Using natural structures… And obviously these are – the sun, thankfully in our lifetime, is not gonna go anywhere, unless there’s astrophage, which is science fiction from the book Hail Mary, that’s coming out as a movie soon… It’s gonna be amazing, I love that book. Astrophage essentially ate the sun, and we had to combat it.

[unintelligible 00:55:16.17] that happens.

Yes, it’s a terrible thing. So in our lifetime, and barring the science fiction from Andy Weir, we don’t have to deal with astrophage. So the sun is going to be there, meaning that it’s producing energy constantly, and it’s the biggest source of energy.

Yeah.

So leverage that thing that’s not gonna go anywhere, and it’s natural.

If I remember correctly, I think every – I mean, the energy that hits Earth is about 177 terawatts of capacity just from sun. That’s the current capacity. U.S’es entire grid is about 1.9 terawatts. So we have like literally over hundreds of times more energy that we produce from sun. I think it could be more, I could be wrong. But the point is there is this civilization scale, so type one, type two, type three by Kardashev, levels we call them. I don’t know if you’re aware of this, but there’s a Soviet astrophysicist from the ‘60s who defined what civilizations in terms of advancements they are, compared to like extraterrestrials and whatnot. So Kardashev level one civilization, or type one civilization is a civilization that figured out how to capture all sources of energy on the home planet. Like from the planet. So anything that hits the Earth, we should be able to capture it. We’re currently at 0.7 level. So we’re not one yet. So to reach one - that means we are an advanced species - we need to capture all the energy that we’ve got, and figure out how to use it. Because even though we capture now, we don’t know how to use it. But with the AI, we can know how to use it.

[unintelligible 00:56:51.00]

Kardashev level two is when we get Dyson spheres. A Dyson sphere is when the civilization expands from its home planet and captures all the energy from its home solar system. There’s always a sun. So the best way to do that is to build a sphere around the sun, and capture the energy. So I don’t know how much deep into extraterrestrial life –

Only through science fiction, really, Dyson sphere. Dennis E. Taylor, he’s been on this podcast. He talked about Dyson sphere in his “We are Legion (We are Bob)”, the Bobiverse series. It covers the whole entire – there’s a book four, I believe… I don’t know. Anyways, they talk about it there pretty plausibly when it comes to the science. He’s a software developer, he’s grounded in science, and I think while it’s science fiction, it may give you some insight to a Dyson sphere.

Yeah, so there was some study… I mean, astronomers, they’ve found a star millions of light years away from Earth, that looked like it was – there was strobes around the star. So there was some indication this could be a type two civilization…

No way.

Right. The star was pulsating, it was not a dying star. So the only time a star pulsates is when you have –

Something blocking it.

[00:58:10.29] …it’s like being distracted. So they say it’s a Dyson sphere… Anyway, so type two is the entire solar system. Type three is a civilization that can leverage or harness all the energy for its home galaxy. So we are like type 0.7. So, I mean, in my lifetime I’d like to see us to go towards a type one, at least. That would be my dream, and that’s something that I want to work towards.

I dig that. Well, I mean, the Dyson sphere thing is gonna be – gosh, I could not imagine that in my lifetime. My kids’ kids’ kids will probably do that. For now, we’re trying to harness the existing energy we can capture through solar… Do you feel – I know you just said that’s the most cheapest way. There’s a lot of pushback to nuclear. Is there any positive ground being made towards like adopting nuclear here in the US?

Oh, yeah. So if you look at the pushback to nuclear, it’s on safety. And if you look at all the safety, all the incidents that happened since the beginning of nuclear, 100% of those issues were due to negligence by the staff. Staff forgot to check something, or put something, because they – whatever. It was all negligence. And there are very few incidents that were caused due to natural disasters - and Fukushima was the only one that we can point to, that melted because of a tsunami. Historically, it hasn’t been any natural disasters leading to nuclear meltdowns. And Fukushima - nobody died. Zero deaths, zero fatalities. And you can actually drink water now from Fukushima. It’s pretty safe.

So from a safety standpoint with AI now - and we have a lot of other automated systems that are phenomenally good - you can make a case saying that if you remove the human out of the equation to maintain a nuclear facility, safety levels are significantly higher. Right now, nuclear, for example, nuclear like gen 3 – we currently have Gen 2. So Gen 3s are the new ones that are coming out… The cost per kilowatt is significantly higher than Gen 2, because there’s so much of investment or so much of capital that needs to go to our safety protocols that recovery of capital is significantly longer for Gen 3. And if you put Gen 3 reactors, you’re not going to see cheaper energy for a while on the reactors. I remember doing the math… I think it was at least 50% higher than Gen 2s, because most of the Gen 2s are like actually now paid off.

So nuclear is very, very good, but you have to make a case on – safety is one case, and economics is the second case. Like, why would someone choose nuclear over, let’s say wind, unless there’s no option, or… Nuclear is more expensive than LNG right now. Like, from the Gen 3 at least. So you have to make a case, like, why would someone choose nuclear over “Drill, baby, drill.” It’s kind of sad, yeah.

So essentially, it sounds like the most plausible best solution long-term for all angles we’ve thought through really is solar, or capturing natural energy sources - water, wind, primarily solar - and then finding a way to decentralize and allow the workload to move elastically throughout the world, to train, to inference, to whatever, to handle this push. That’s where you’re at, at least. Is that what your argument, is that’s the best way to [unintelligible 01:01:58.07]

[01:01:59.07] That is my argument. That is my argument. And we’re not waiting for people, we are executing on the strategy. I’ve testified before Congress about a potential solution, and also, we announced this program called Star Cluster Program at Akash Accelerate… The Star Cluster Program is an effort by Overclock Labs, the company behind Akash Network, to prove out a distributed infrastructure can train AI with the least amount of loss. And we’re executing on it.

We’re partnering with a few telecoms… Don’t wanna name anyone out, but we have access to energy across America. We have access to about 2 million homes through our partners, that we can go into potentially. I mean, we are currently selecting ideal houses with cheap solar; solar and bandwidth are two big variables. And also, when you go into a home, you quite don’t have the same level of energy. And there’s also another policy side of things. It’s very, very, very frustrating. Like US homes on average have 15 amps at 120 volts. That’s your US standard. That’s a scam. I don’t understand why we have such low energy.

How so?

I looked into it, I think it was Edison, GE, they had a lot of things to do with the standard. Why? Well, because think about it… Like, 120 volts or 15 amps gives you 1800 watts. 1800 watts, you can’t even power – I mean, if you power a heater, a home heater, it’s done. That’s 1800 watts right there. You can put two 5090s, which is the most powerful GPU from NVIDIA… Or maybe you can put three 4090s max in a home, because of the limitation. If you were to go higher, it’ll break all the residential limits per se about how much energy. So you would upgrade to a commercial facility if you put over 1800 watts. And you lose all kinds of discounts, and all kinds of regulatory benefits that you have as a home.

The amperage isn’t really there though, right? You’re talking about the grid on the inside of the home, if I understand correctly. So you can have up to 100 to 200 amps of service coming to the residence. But we’re talking about, I think, the grid within. Like, we’re operating on 15 amp circuitbreakers, and maybe even 20 amp if you paid for the upgrade over. Is that what you mean by that?

Yeah. So what I meant is the interconnect speed, interconnect limits… Even if you’re making solar or producing solar locally, you still have to sell it back to the grid by law. So you still have to connect to the grid. And when you connect to the grid, you’ve gotta say how much energy you’re using, and that will break or make your discounts. It’s complicated when you have to deal with energy companies, and there’s a lot of legal implications, safety implications, and all kinds of things when you have a home. It’s different when you have a commercial facility. It’s possible to go higher, you just don’t get the same benefits.

It’s not common, yeah. Yeah, 1 in 10 homes will do that now.

Correct.

But it may become a state of the art future with homes being part of the grid.

My home in Texas is 100 kilowatts, no doubt. I’m not going after residential discounts. I don’t care, because I’m gonna produce so much energy, and I’m going to have a much bigger cluster. But an average home in US, if I were to come to your house and be like “Hey, I wanna [unintelligible 01:05:39.26] I cannot exceed over 1800 watts. If I do, you’re going to make some investments, so much investments in your home the economics won’t even work out for you. Like you. It doesn’t make sense.

Yeah. My home can handle it. We made those investments. We kind of planned ahead a little bit, but maybe not as aggressively as we probably should have. We did plan ahead, but not that much.

[01:06:02.21] So you’re trying to go into homes. I mean, that’s what you’re saying, right? Like, your goal is to put GPUs in people’s homes…

Correct.

…through their telecom relationship, or like that’s how you’re gonna actually reach the homes?

That’s one way to do that. Yeah, we have several ways, but telecoms are the best way, because –

Sure, that’s just one way you’re going about it. And you’re gonna connect those up to your network…

Yeah.

Right? Akash network.

Yeah.

And they’re gonna become resources available. So now consider me as your potential homeowner, and you’re pitching me on the idea of putting a GPU in my house. Well, why?

The package that we’re going after is you get free bandwidth, you get free energy.

I get free bandwidth. So you’re providing my internet connection…

Yeah.

And my bandwidth is now free, and I get free energy.

Yeah. We’ll put solar in your home, we’ll put a battery in your home. You don’t have to pay a thing. We will charge a battery during the day, we’ll send it back to the grid. When we make the most money, we’ll use the excess energy to produce enough energy to power the GPU. So you get free energy, you get free bandwidth, and maybe some money on top of it.

Yeah. Okay… And how many GPUs can you put in my house? You said one or two?

Depending on the GPU… So if you were to go with a 5090, which most likely you will, which is the latest consumer GPU from NVIDIA, we can put two GPUs in a home.

Okay. So to fulfill XAI’s desires, you only need 25 million homes to get that done, right? Two GPUs per house… They’re gonna do 50 million GPUs.

I think we have access to 22 million homes right now with our partners in the US.

So that’s it. You hit the number.

You’ve just gotta sell all of them. And roll them out. Not just sell them, but actually deploy. That’s a big job.

I think the way we live may change, don’t you think?

I’m just thinking of like as an N of one; as a person who owns a house, do I want that? Of course I want free internet and free energy. So yes. Do I want a GPU running in my house? They’re pretty loud. Is that a concern?

Yeah, so we have all that worked out. The [unintelligible 01:08:12.09] will give you about 30 decibels. It’s not that crazy loud.

Put it in the attic or something?

You can put it in your office. 30 or less. I mean, we’re trying to get it to 15 decibels. My child, when she cries, is about 90 decibels. [laughter] A vacuum cleaner at max produces 60 decibels of sound. So 30 is – I don’t know, I think it’s not quite, but it’s definitely not too loud. We’re trying to get it to –

It’s a concern. I mean, if you don’t consider that and you say “Sure, come on in and do it”, and then they do it, and you’re like “I didn’t think about how loud this was gonna be…” You’re gonna be calling them back, saying “Yeah, pull this thing out of my house.”

Yeah, yeah, yeah. So there are all those considerations as well. We’re trying to get it to as quiet as possible. Getting it quiet is very challenging…

Yeah, totally.

…because the fans are – cooling is a big challenge. You need fans going off… The only way to avoid fans is through liquid cooling, and liquid means the coolant is also not cheap.

So now you have a maintenance.

Have you experimented with the – I explored this because we do audio, and I’ve explored really crazy ways to make my room more better… My previous house had a really issue. My AC unit was right outside my office, and every time it turned on, I would just hear the hum. It went “Woom, woom, woom…” I’d just hear it, it was terrible. So I was thinking about ways to isolate the sound from out to in, and there’s a lot of sound isolation. If you could put it in a sound isolation chamber, essentially, designed to minimize out sound, but still have exhaust - could that solve it? Did you explore that?

Well, you can solve a lot of things, but the typical home will not have a sound-optimized setup. You are an audiophile…

No, but if you deliver a solution that’s already in that sort of – it’s engineered in the package.

Like a mini rack that’s engineered to be [unintelligible 01:10:03.07]

[01:10:04.24] Yeah. It has to be a self-contained unit. It has to look beautiful… For a home, it has to look pretty. It has to – it’s a consumer device, and it has to incorporate everything in a single board. So it’s not the easiest thing to design, but we’re on it. So we have a home aspiration… I would call it aspiration because we haven’t announced a home product yet. We’re still in design. But we are going more aggressively with the data center products first.

One of the cool things we discovered was – we have access to telco edges, edge networks, because telcos have data centers pretty much in every urban area. Even in small towns, you have data centers for telcos. That’s how Comcast and all these guys deliver the internet. So there’s quite a lot of connected energy, a lot of them are coming from renewables… So these guys have done quite a bit, which is very fascinating. They have large networks built out all over the West. So we’ll probably tap into that in the early days, in the first phase, in the next couple of months actually, to prove out that we can do distributed AI, and then optimize and optimize, come to the consumer model. A consumer is going to take at least a year for us to [unintelligible 01:11:11.06] I have a box right here… I can’t quite move the camera, but the box right here is – you can’t hear nothing. It’s pretty good.

It’s running right now?

It’s running; not at max, but at max you would probably hear it. It’s not running anything heavy, but when those GPUs go and the fans start kicking in, when the heat goes up, that’d be a problem.

Well, for all this to be feasible, those two things have to come together first. The ability to train distributed in a way that’s competitive… Right now, it’s not quite competitive yet. It’s getting there.

We don’t have – so in order for the home solution to work you need distributed training, which I think is solved, for the most part… At least nobody has put a token on it yet. We’re getting there. Why I say a token is because the business model for distributed training, almost of all the models I’ve looked at, the common thing they have is you contribute compute, you get some representation of that contribution in the form of a token… And that token will guarantee you future revenues from the model, from inference.

Okay.

For example, say you train ChatGPT. You’re part of ChatGPT training. Your machine is. And if I as a user of ChatGPT pay $200 a month, you get a cut in that revenue.

I see. So you’re not getting cash back out. You’re not getting –

Oh, it’s cash.

No, it’s future cash, right? If I were to sell you my internet or my energy, you’d give me $10, and I would give you some energy.

I mean, there are ways to give you instant liquidity and all that good stuff, but just to give you an idea…

But no one’s done that, though.

We’re getting there, we’re getting there. So Akash is another way to look at it. For example, a 5090 on Akash will get you 92 cents an hour right now… Provided it works from your home, and there’s still tech we gotta – this is work that needs to happen.

Right.

And Akash, or some company on top of Akash will do the training and say “I am betting on Gensyn”, for example… Gensyn, or Pluralis, is a company that I have visibility into, and I know for sure that this token is gonna be worth some money, because they’re training this amazing model, and there’s demand for the model… Again, you get into that nuances. I would take the risk, and I’d be like “Hey, you home computer, give me your energy for cheap, or compute for cheap, and I will do a pass through training to you. I’ll keep the token from Pluralis, I’ll take the risk and I win. I benefit.” When that becomes more mature, maybe consumers can take all the risks themselves. But there’s always going to be opportunity players like me that’s going to take advantage. I mean, I think there’s enormous money to be made in the next one year to two years in the decentralized AI space. If you look into those protocols deep enough, you’ll understand.

[01:14:16.05] And the best part is there’s zero hype on these protocols, and I think it’s an amazing time to – and they’re getting so much traction, too. Gensyn alone is right now training about 10,000 models concurrently. 10,000 models. Name a single AI lab that can do that.

So people have no idea how these things work, which is great… No idea what the progress has been made right now… Because again, to use Gensyn is not the easiest thing. You need to understand –

Yeah. So Gensyn is a company, is a network, is both…?

Gensyn is a decentralized machine learning training network.

Gotcha.

It’s founded by a bunch of researchers at Oxford, funded by [01:15:05.14]

What’s the URL for that, by any chance? Do you know that?

G-E-N-S-Y-N, Gensyn.

So it’s a handful of these decentralized compute networks that are currently being adopted by model training organizations. And basically, they’re all flying under the radar because all of the focus is on XAI, OpenAI, Anthropic, Google, DeepMind.

Yeah. And Meta, and a whole lot of other players now.

Right.

Every company – Oracle, and every company… Yeah.

What was that contract that OpenAI just signed with Oracle for like $30 billion, or something like that? Did you see that?

That’s nothing. I mean, they’re working with the data center in Texas. That’s the same thing.

That’s the same $500 million dollars they promised, right

$500 billion…

Sorry, 500 billion? My gosh…

Yeah, billion.

Oops… Not million, billion.

[unintelligible 01:15:59.27] money.

That’s a big discount. 500 billion was promised… They’re 100 billion in, I believe, and with plans to do more once this facility is there.

Funded by SoftBank. SoftBank is being very aggressive in the data center space. They’re like –

Oh, 30 billion a year. That’s the one I just read yesterday.

A year, maybe. Yeah.

OpenAI to Oracle, 30 billion a year. This is in addition to, or next to the Stargate project. It’s like a different thing that they just announced. Anyways, it’s hard to pay attention to anything when that kind of stuff’s flying at your face every day.

And that’s where we are. We are under the radar, which I really like… I love this time, because I think we’re about to blow up very soon in this decentralized AI space. People that know, know, kind of thing… And I’m getting as much exposure as possible personally, as well as corporate-wise, to this feature. And we’re going to do small experiments in the home very soon. My home will be the first home. But I think we’re going to have a few homes by next six months using the Star Cluster program.

Star Cluster, is that how you call it?

Mm-hm.

Fascinating stuff. Fascinating stuff. It seems like the Bitcoin miners have been onto this stuff for a while, haven’t they? I mean –

They’re all jumping ship now.

Are they switching [unintelligible 01:17:14.22]

A lot of them. I mean, these guys – I mean, CoreWeave was. CoreWeave was a Bitcoin miner.

CoreWeave, yeah.

We’ve met CoreWeave, Adam, at…

We sure did, yeah.

We didn’t know who they were. This was last year at KubeCon… And we just walked up to them and we were at their booth… They were kind of off to the side by the food, so we were over there talking to them, and we were trying to tell them they should advertise on our podcast… He’s like “We don’t need advertising.” I was like “What are you talking about?” He’s like “We have so much business, we can’t even keep up with it.” I’m like “Who are you guys? What are you doing?”

Yes…

“We have a bunch of GPUs, we’re just leasing them out. We can’t keep up.” I’m like “Interesting.” And then next thing I knew, they’re IPOing. So… That was crazy.

[01:17:52.09] CoreWeave used to be a provider in Akash. The CEO was in a Slack, DMing, selling his business. This was in 2018, 2019, 2020, until ’21, ’22… I mean, not ’22. ‘23 is when they started… My God, taking off like crazy. I almost thought it was a scam, because we know the people really well… But then – I mean, the scam was they were… Or at least like what seemed to be a scam was they were buying GPUs – they got a sweet deal from Nvidia, they were buying GPUs left and right… These are called the Neo Clouds. So they’re hyperscalers, which are Amazon, Google, Microsofts the world, and the Neo Clouds are these new companies that - maybe around for a little bit, or just popped out in the last few years. They are much smaller facilities, with maybe like 10 megawatts, or under 50 megawatt access to energy. They can build data centers and understand finance. So a lot of them, like CoreWeave, what they did was they bought compute from NVIDIA, used that compute as a collateral to get more debt, used that debt to buy more compute, and so on and so on. So they over-leveraged, or leverage very well, betting on the fact there’s going to be demand for compute. And if that didn’t happen, they would have an incredible downfall. So they did take risks quite a bit in terms of leverage, and they levered up all the way to billions of dollars.

And they were right about the demand.

They were right about the demand, in one year. So it seemed like these guys were unstoppable. And then they were selling – I think I remember they were selling secondaries for like a $12 million valuations, for like all the employees. So employees were getting out. I mean, all these things looked like a scam, but ultimately I don’t think it was a scam. It was actually kind of crazy. And it’s very similar to what MicroStrategy does, similar to what New York real estate developers do, leverage, leverage, leverage, you know… So it’s not a bad thing if the demand goes up.

Right.

Yeah, if the prediction goes the right way.

If the bottom falls out, then you crash pretty hard. Especially if you’re – in the case of MicroStrategy with Bitcoin. It’s like, they themselves are also driving the demand… Whereas with CoreWeave, I guess they were kind of also driving demand? It seems like they weren’t. Yeah, there’s a difference there. The demand was external.

[unintelligible 01:20:18.08]

Not that MicroStrategy is the only one driving demand, but they certainly are driving some of it. And so…

They’re a big player. Oh, Bitcoin has real demand… [laughs] But AI definitely, AI chips and energy have real demand.

It’s very, very clear now. I mean, I would even – I spend a lot more time now on AI and energy and below, that sort of like area… And it’s a very small area, and there’s a lot of stuff happening in Texas, because Texas is the energy capital of the world, especially Houston… And a lot of VCs that used to be crypto are also spending time on this space. Meltem Demirors is a good friend of mine, she does quite a lot of energy now. She’s doing exclusive energy now.

What’s her name?

Meltem Demirors. She was a co-founder with Grayscale, with Barry, with DCG. She was either the founding member, or co-founder. I don’t remember, but an early person there. Meltem Demirors now does a coin shares thing, and she started a new company called Crucible, a VC. So they are very, very good in terms of – I mean, she’s very active in this space, looking at energy, looking at below data center sort of like stuff.

And I was also pretty fascinated with the amount of disruption that’s happening in the space and not too many people are looking at. The new companies that are focusing on data center management software, which nobody touched for like the last four years. The data centers were not that thing. It was a write-off.

Right, they were boring.

[01:21:50.09] Boring. It was real estate businesses… They’re write-offs, they’re like depreciation write-offs for the most part, because that’s what real estate is… And now, oh my God, if you have a data center, you’re the hottest kid on the block. Everybody wants to talk to you.

We obviously do site hunting for data centers… We look at everything possible, where to put our workloads, and the demand is so high. There’s all these middlemen that let you get to data centers. There’s a whole ecosystem around data center world… And they’re just talking to these folks. The business is so, so crazy for these guys. And even people that build out the power infrastructure… We know a few companies, because once you put like anything above a megawatt, you need a transformer, you need a whole lot of energy infrastructure. The companies are specialized in building energy infrastructure for data centers, and you cannot even get on a call with them these days. They’re so hard to get by, the companies. So it’s a very, very profitable space, and people need to look at it a little bit more if you really want exposure to the real AI, which is the energy underneath, and the infrastructure and the picks and shovels; the energy designers, the consultants… And there are very few electricians in America. There’s a crisis for electricians, and we don’t know that. Like, there’s a lot of sociology majors, not too many electricians, so…

Sure.

So being an electrician now is very, very profitable.

Yeah, it’s hard to get one even to call you back. I know personally. So if we go back to foundational things and first principles thinking, what could change foundationally that would somehow invert this demand curve? Are there black swan events, are there changes to technology where all of a sudden it’s like “Well, data centers are not the hot kid on the block anymore”? Have you looked at all that?

Yeah, yeah. I mean, I think fundamentally it has to be technology. It has to be –

A new breakthrough.

A breakthrough of distributed training. It’s a major area of research. People thought distributed training was impossible a few years ago. Now you look at ICML, the conference - we had six papers that got approved in ICML. And getting a paper approved in ICML is a big feat. It is something people work towards their entire postdoc. They work on these papers for like 10 years to get published, you know? So it’s a big deal to get a paper published and presented at ICML. Pluralis, Gensyn, Bagel - all these companies presented. And that is an indication for me that this – and peer review. It’s a very, very amazing milestone I would call for distributed training. And that’s gonna be a fundamentally different driver… But data centers are not gonna go away. I mean, I don’t think – you can’t distribute everything.

Well, for instance, what if a plateau in advancement? What if Grok 5 isn’t any better than Grok 4, despite them putting out 500 million, or just astronomical compute? What if we just stop getting better? Would that then top the demand level?

There’s no limit to how better we can get… I mean, ESI, AGI… There are a lot of breakthroughs we still need to achieve. We’re like very, very early with AI. There’s interesting stuff – okay, there’s another sort of like architectural pattern that is emerging in AI called mixture of experts. A mixture of experts is instead of having one gigantic, large model, how about having a lot of small models that can work together? Like how humans work. We don’t have an expert in everything…

Sure.

…we have experts in areas, and we all work together as a team to achieve a task. So there is that improvement that’s happening. Or rather it’s taking a more center stage. If you look at the new ChatGPT agent mode in your ChatGPT, it’s not using the biggest model, it’s actually using smaller models to do agentic work. Grok 4 agent mode, the Grok 4 Heavy is also using agents. So the agent stuff, I think, is very, very interesting when it comes to inferencing or getting results, or getting tasks done.

[01:26:11.17] So there’s quite a lot of – I don’t know, I think there’s still like economies of scale that could work for hyperscalers, and they did work for hyperscalers over the last 20 years. It’s just this new energy shock is not working out so well for now. But if you solve the energy shock, I see no reason why the economies of scale won’t return for hyperscalers. So there’s always going to be a balance, I believe.

So if a hyperscaler, let’s say, goes to El Salvador, and say, well, there’s volcanoes that are giving basically free energy, and I can tap into that to build a, I don’t know, 500 megawatt capacity. Beat that, right? So there’s always going to be economies of scale that are going to work out better, if there’s not too much rush in getting to those facilities.

Right. Just the gold rush right now is causing the demand bubble.

It’s a crisis. I would call it a short-term to mid-term crisis. It will be long-term if we don’t do anything about it…

Sure.

But I don’t think people are sitting around, not doing anything about it. So that’s why I have high conviction over the next three to five years decentralized AI has a best shot to make a mark on the world. Like, we’ve got the best shot right now. If we don’t make progress in the next two years, I don’t think decentralized AI will take off. People are not going to use decentralized AI because it’s the right thing to do. People are going to use a solution because there’s no other option. That’s really the best way to sell a product.

People don’t care about privacy as much as you think people care about privacy. People are willing to give up comfort – I mean, people are willing to give up privacy for comfort. Or cost. So humans are irrational beings. So you just see what’s easy to use and cheap to use, and you tend to use that.

So decentralization, I feel like – if you look at decentralized finance and all kinds of things, DeFi only took off because it’s easier to use than your centralized stuff. Uniswap is significantly easier to use once you have it set up, than Binance. And it’s private. So are we going to be easier to use than ChatGPT? I don’t know. I mean, ChatGPT has done a phenomenal job with their product. They may not have the best models, but they still have a great product, and it’s getting better.

So I don’t know, you’ve gotta really think through why would DAI take a center stage… If you look at all the benefits DAI provides, those will not be the ones that people will care about. It has to be a product; it has to be a good product, that’s easy to use, i has to compete with the OpenAI’s of the world… Historically, we haven’t been great. Like, that’s the reality.

But if it helps you solve the energy crisis, there’s your linchpin.

And that’s why I said the next two years are going to be – the next two to five years are going to be very, very crucial for decentralized AI to take off.

Plug us into the scene. Where do you follow the progress? How do you know the movers and shakers in the decentralized AI scene? How do you participate? Is there a Discord? Is there a particular –

Well, you can follow me, Greg Osuri. I’m very loud on decentralized AI. My Twitter handle is @GregOsuri. Is there a single place you go? Unfortunately, the information is so dispersed right now. The best place to learn about AI is to actually hang out at AI conferences, not crypto conferences. Because crypto is so terrible at claiming everything is a decentralized AI, because 99% of the decentralized AI is junk. 99%. It’s not the agents, it’s not [unintelligible 01:30:01.20] it’s not all these people that got millions of dollars of funding… I mean, people raise 70 million, 80, 100 million even [unintelligible 01:30:09.14] They don’t matter. Like, seriously.

[01:30:15.10] The only ones that matter are the ones that got their papers accepted and presented at ICML. That’s the first bar that I consider to be high quality. Because there’s so much noise, and you’ve gotta find the signal. Well, one way to find signal is through peer review, is “What do the actual AI people that are doing mainstream AI, that are working at the OpenAI’s of the world and the Anthropics of the world, what do they think of decentralized?” So see their take. And the way to get their take is by looking at actual peer reviewed papers… That are deeply technical. I mean, if you’re trying to read one of these papers, you need to be expert-level in derivatives, calculus… It’s extremely arithmetic, a lot of these papers. Extremely arithmetic. Way above anything crypto has to offer in terms of technicality on the papers. Fortunately, you have an AI to help you read those papers. I’ve done differentials quite a bit… It’s very challenging to read these papers. But if you can use AI to read them, I highly recommend you do so.

Some of the investors that are making good investments in AI, and usually Andreessen Horowitz has done a phenomenal job, I think, with investing in decentralized AI, because they understand the space better than other things… CoinFund, I think, is remarkably good with decentralized AI. I would say in their AI portfolio, I would say 80% is good, legit. 70% to 80% is legit. And a lot of the companies that I talk about, almost all of them are CoinFund funded. It’s funny. Or at least they have their hand in. And I don’t want to name other VC names that are not good…

Name the bad ones, so we can avoid them.

No, I don’t want to do that, but… I think CoinFund is good, Andreessen Horowitz is a good signal… Let’s see, what are the companies that are doing phenomenal AI? And a lot of the traditional funds, like VC funds, that are heavy in AI, are not writing any small checks, too. They’re not writing anything less than $50 million these days. The AI checks are so large, it’s nuts. So yeah, I don’t know, the space is so early right now, getting the signal is very challenging. And the way to do that is just follow the smart people, I suppose, you know?

Yeah. I think the question Jerod’s asking is the kind of question I’m trying to ask too, which is you said there’s a ton of money to be made in the next two years. We dramatically – well, is that the right word to use? We talk to a lot of software developers; or I guess the ones that used to be software developers in the traditional sense, and now they’ve become something different. Architects, thinkers, babysitters… Pick your adjective to describe what you are. Where - and we’re talking about D-A-I, which I believe is D-E-A-I. Would that be – it’s not DeFi, it’s D-A-I.

Decentralized, yeah, yeah.

[01:33:34.08] Where are the – this is one of the areas. Where specifically – if we had to give marching orders to a hungry, changing audience that knows there’s something changing, knows they have to change their position in the value chain, knows they have to either become more resilient, become more dependent, or change just fundamentally how they thought they would ever navigate their software developer career… What can they do to capitalize on this change, this massive money-making possibility, and also just changing humanity in the positive way, fundamentally, in the next couple of years? Where can they apply their talents to capture some of these dollars, and also innovate?

If you are an engineer that has experience in distributed systems, go into AI. Have you seen how much Meta is paying to poach AI researchers? I’ve seen offers for up to a billion dollars for an engineer, one engineer, for four years. And they’re still rejected. But I’ve seen offers all the way from – I don’t know, the space is so… I’m like “What am I doing?”

“Not enough…! That’s not enough. I need more.”

So they’re offering engineers 300 million, 500 million offers for four years. If you’re an engineer that is good with AI… That’s the ceiling. Now, you can make a million if you’re good enough. I think a decent researcher now in the space is about a million dollar cost on average, per year. So as an engineer, you have a lot more offers. I would just drop everything – you need to do really deep into differential equations, and you need to brush up on your math skills. But if you break it down - I mean, it’s just multiplication. I bet you if you spend enough time, it won’t be that hard to regain confidence.

There are lots of opportunities in AI. Lots of opportunities if you’re an engineer. You can solve a distributed training, for example. I bet if you can figure out how to distribute GPUs, even centralized, across two data centers really well, you have a winner. I mean, centralized companies, when they – there’s a good chance they may go distributed, because Anthropic’s CEO, co-founder, talked about distributed training in their newsletter. I wrote an op-ed recently in – by the way, I wrote an op-ed in CoinDesk last week, and I mentioned about all these stuff that you can actually go… It’s a good starting point if you’re looking into getting understanding.

So there are frontier labs that are actually looking at distributed training as a solution. Not decentralized yet - I will get there - because decentralization is about adding incentives on top of distributed training. That’s what decentralization is. But distributed training, I think, is a great area, and we need a lot of people to focus on this stuff as an engineer.

Now, if you’re an investor, a savvy investor, if you’re looking for that edge, the niche… Energy. Look at this space. There’s so many people you can talk to that’s better than me with energy investing. I come from a usage standpoint. I obviously don’t understand the intricacies of the ecosystem, but… All the way from nuclear. There are public companies that do nuclear enrichment. I think nuclear stocks are kind of off the hook the last few months, because of this energy demand… I don’t know, that’s just – it’s not financial advice, I’m just like looking at areas… I’m looking at, personally, nuclear. I’m looking at rare earth minerals. I’m looking at data center companies, [unintelligible 01:37:10.07] of the world… I’m looking at chip companies rather… I have exposure to NVIDIA and AMDs of the world, but also new chip companies that are innovating. There’s a 99% chance this will fail, but that 1% is what I’m looking for. There’s this new company that’s doing quantum chips for AI. The guy, the founder used to work – he was a higher up at the quantum lab in Google… And they’re literally coming up with a chip. And they sent the cast to TSMC, which is the producer of all the chips, by the way…

The fab, yeah.

[01:37:50.08] The fab. So they’ll have a chip out in 18 months. So I look at deals like that. Make or break, high risk/high return investment. Look at data center management companies. There’s a whole lot of them. Not a whole, but there are a few companies that are doing really well. I mean, innovating, because that space hasn’t been touched in a long time, so there are disruptors that are popping up.

If you want a more niche-like area to focus on, there is energy arbitrage companies, Enron’s of the world, that could – imagine Enron existing in 2025, without crooks leading it… Phenomenal business model if they actually executed on them. They messed up on the other things, but there’s a new emergence of Enron’s of the world that are coming up.

Look into those areas. There’s lots of opportunities, I think. If you spend enough time, you’ll get excited. And ChatGPT, deep research is very, very good, some of the stuff. Give it what you’re looking for, ask it to give you an overview… Grok 4 is pretty good at getting you the information as well, the deep research stuff they have… I would start looking at them.

I bet if you put the Department of Energy report - it’s publicly available - into like ChatGPT, and put like deep research and be like “What companies are involved in all different layers of this energy?” and it’ll give you tons to just start your journey and start looking at them. But I highly, highly recommend – yeah, I mean, I highly recommend you look into the space deeply.

I’m going back a little bit to one thing that you told Jerod - and I guess me too by proxy, but you guys were really… I was listening in on that part. Where it was about the free internet, free energy stuff. And I’m thinking about this idea of UBI becoming resurfaced as a true plausible need or some sort of like baseline because of the mass unemployment. I mean, you’re talking about – we’ve got lots of developers out there that are fearful of losing their jobs, or building the thing that takes their job kind of thing… That’s been the concern, I suppose, to some degree. But then the way you described it was like “Okay, well, if you’re in distributed engineering or you’re doing Kubernetes, you’ve got lots of change that you can move up the value chain around.” So it seems like the job, the traditional software developer job of “I write code” is a changing – I wouldn’t say dying, but just changing. You’re still involved in the code writing, you’re just not literally writing it yourself. You’re directing it through teamwork, through yourself, whatever.

And this idea of UBI, is this – I don’t wanna get political with it necessarily, but the plausibility of the way people are making their money or being incentivized, it sounds like that’s a version of UBI, where you get your internet and you get your energy for free because you provide a value service back to the DAI infrastructure that will be future built. What are your thoughts on, I guess, UBI and concerns for folks having to have some sort of base level income? That seems like a version of UBI where “Hey, if I give you a little of my compute, or I take on this device, maybe I spend a little bit of money, but long-term, I never have to pay for internet again, or energy”? That’s kind of a version of UBI, because you probably cut off 500 bucks of my personal bill to my household…

Yeah. So when I’m giving you something, it’s always something in exchange. So you’re not getting free internet, or free energy…

Yeah. Well, it’s not a UBI necessarily, but it’s a way to augment the need for earning as an income. So you take an income I had to earn to maintain energy and internet, and now I can repurpose that same earning if I have it, or take it away if I don’t.

Yeah. So in my model, you’re leasing, you’re giving your real estate, and a room in your house, and the position on planet Earth to capture energy, in exchange for free bandwidth. You’re not literally getting free, but there’s always an exchange of value. I mean, that’s fine. No one will put something in your home unless there’s something to get out of it.

When it comes to UBI and like future of AI taking jobs, there are two things to look at. Number one is so far, the society has rewarded intelligence. Human time has value, whatever value that may be… Because you’re providing some time, and you’re giving intelligence during the time, you have some value that’s called value of labor.

[01:42:15.12] Now, value of labor - you get some price back. Question you should be asking is, will the value of labor be diminished to a point it’s no longer valuable? Number one. Number two, what happens when a world shifts to that mode where the only mode we knew so far in existence of all civilization is the value of time?

So question number one - will AI replace human beings fully autonomously? Now, I am a practitioner of AI, I use AI much more than most people do, and I’m also an engineer. I guarantee you, the state at which AI right now is, it cannot replace a human being. Not even close… Because of one main reason. That happens to be learning capability. AI, once trained, cannot learn continuously. Whereas a human, if you do something, if you work for me, if you’re doing a bad job, if you do something that I don’t want you to do, I’ll tell you what to do. Next time you’re gonna not repeat that. You can’t tell that AI. AI is not self-learning. You can talk about reinforcement learning and all these techniques, but it’s not self-learning yet. So it cannot replace a human being in any way. But a human being will be replaced by someone that knows how to use AI, for sure.

Now, is it a replacement or is it augmentation? There are two ways to look at it. As a CEO of a company, I was not coding as much as I used to code back in the day… But now with AI, I’m coding a lot more than I used to code, even back then. Why? Because the speed at which I can deliver stuff has gotten significantly lower compared to pre-AI world. So I’m able to do a lot more that I couldn’t do. Now, as a CEO, do I want to do more things? Absolutely. There’s no limit as to what kinds of things I want to do. Like I told you, I [unintelligible 01:44:21.05] a Siri replacement, because Siri sucks. Now, would I do that without AI? Probably not, because I don’t know how to do Swift. I don’t know how to do Mac apps. I don’t know any of this stuff. I mean, I could learn, but is it valuable for me to do that? Probably not.

But I was able to replace this whole thing in, I don’t know, an afternoon, because it was fun. I reconfigured my entire network on 4th of July, because I wanted to protect against Chinese IoT devices. Will I do that without AI? If I have the motivation, but probably not as quick as what I could do. So I could do a lot more as a human being that I would rather need a network engineer, probably I need to hire to do this, or I need a developer I need to hire to do this. Am I hiring more developers? I am, and we’re equipping them with AI to do a lot more, like I do.

So our production capability is just gonna be higher. We’re not going to use less developers, we’re just going to use more developers to do more things. That’s the human nature. There’s never a list of things that you’re gonna run out of doing. I mean, imagine, if you have all the time in the world, if I asked you, what would you do? Always gonna be things.

Always gonna be things. I want to build a spaceship if I have the time and the energy and the resources. I want to build so many things. As a builder, I always want to build things. The only limitation right now is time. That time, imagine taking that away. So it’s the Jevons Paradox, we call it. I’m pretty sure you heard this term…

No.

The Jevons Paradox is –

Yeah, we have.

Oh, we have?

Yeah, it’s pretty popular.

We heard it on the pod, like two weeks ago.

Oh, gosh. I forgot.

Abi Noda told us about it. Go ahead, Greg.

Oh, yeah.

[01:46:04.28] Yeah, Jevons Paradox came from this age when people thought coal would just basically replace all forms of energy, and humans won’t have work to do, because you don’t need human physical strength anymore; coal is gonna replace the humans, and we’re just going to not have any value for human beings. But that turned out to be the opposite. I mean, with coal we were able to industrialize, we were able to produce a lot more demand, and that in turn needed more supply of humans that can actually operate the machines. Same thing with tellers. When ATMs came, people thought it’s going to destroy the banks, and all the tellers are going to go out of business. Like, no, we had a lot more banks, and a lot more ATMs.

Same thing with iPhones, for example. People think “Oh, the chips are getting faster and faster. Are we going to have less iPhones?” No, we’re just going to have more iPhones, because the faster chips get, the more demanding the applications get. That’s just the nature. So as something gets more efficient, the usage is going to go up, historically, because of this paradox, and not the opposite. So yes, there will be a change –

What you’re saying is there’s been a dramatic inefficiency in producing code at scale. And now we have a way to produce code at scale, and so the inefficiency has been unlocked, and it doesn’t change our need to create more and produce more, it only exacerbates it.

Absolutely.

And it fuels it.

It doesn’t matter – I think we’re going to accelerate beyond our wildest imagination.

That’s a great argument, honestly. Or at least a great take on it… Because I think there’s a lot of doom and gloom out there to some degree, and I think we need to look at – I almost thought the traditional way to write software, like literally handcode something will become a hobby of a previously traditional software developer, that now they’ll do it because they really love to write software, not because they have to write it. It’ll become a choice maybe, not – and I’m thinking like really just maybe the – I don’t even know how to describe it, but just the randos who really just want to keep writing the actual software… Not for their jobs, but just for the fun. Just for the “I go ride a bike, I swing golf clubs” kind of thing, you know?

Yeah, I mean… So there’s going to be hobbyists that are going to do it for hobbies, and there are going to be people that – I write using a pen. I mean, I have a fountain pen, literally.

I produce content, but I haven’t written content at that kind of scale…

I use a fountain pen even today, even though I have more advanced pens. But why? Because I like using a fountain pen. I use an analog camera. Why? Because an analog camera produces pictures that a digital camera can’t. I use a lot of old things, because they’re just better in so many ways. And people like me are going to exist, right? But that’s going to be a small percentage of people like me. And I think most of the world is just going to use AI to accelerate, augment.

I’m particularly excited about removing the next inefficiency in production of goods and services, which is the human-computer interface. Right now, we’re so ancient. We use a mechanism to interact with a computer that was invented over a hundred years ago. A typewriter basically became a keyboard. Imagine being able to be plugged in using Neuralink or something, where your thoughts control the computer, accelerated by AI. Imagine the possibilities.

Not… [laughs]

[01:49:36.03] So one thing that will be most disruptive right now is people that know how to use AI is going to disrupt people that don’t use AI, hands down. We have an active policy in the company, we give X amount of time to employees that haven’t quite embraced AI, that if you don’t do it by X amount of time, you’re going to be replaced. And we’re very public about it, we’re very open about it, because you’re required to. And we give a lot of training to our staff. People that are not going to – I mean, fortunately we don’t have anyone that hasn’t adapted to AI, but we have a standing policy, and everything is going to be standard across different companies.

What’s the timeframe on that?

We’ll give six months.

Six months to your current employees?

Yeah.

And what do they have to do exactly?

They have to learn how to leverage AI to do everything they do. So the first step is obviously like, for an engineer, go to ChatGPT and – that’s an AI basic. But if you’re like level – we have four levels, basically. The first level is you know how to use ChatGPT. The second level is you know how to customize ChatGPT, or fine-tune. Third level is you know how to sort of like – for an engineer; I think we have for different persons. You know how to use different tools by running them together. Fourth level is you have a pipeline that’s so self-optimized, self-learning, that you actually just talk to it and it’ll do it for you. Like, most people are on level one, how do you use ChatGPT or Anthropic or Claude, and go to the web interface… But there’s so many things. I’m probably at somewhere between like two and three, where if I use ChatGPT, you wouldn’t even know I use ChatGPT. It’ll sound like I speak and I write. And I can easily detect AI slop. You can tell someone used ChatGPT or not.

Are you speaking right now, or?

I’m speaking right now, but…

This is the real Greg?

I just wanna make sure…

Sometimes I’ll wear different glasses, like glasses like this, which has AI in them… It has a HUD – sorry, a heads-up display. You can’t see this, but – or you may be able to see this.

Yeah, put them on. Let’s see. Put your new mask on. Become Clark Kent.

This is – it’s called Even Realities. Is it on right now? It’s not on, but… These are very early versions.

I was gonna say… It kind of just makes you look like a crazy person at first. [laughter] He’s like staring off into the ceiling…

This is an early version of AI glasses. I’m experimenting with a bunch of these glasses, and there are a lot of companies working on different variations, and not everything is ready for primetime, I would say…

Sure.

But okay, that’s me augmenting myself. Like, it has a teleprompter when I’m on stage. I’m giving like phenomenal keynotes, it’s teleprompting [unintelligible 01:52:19.04]

Oh, you actually wore that on stage? You wore that at your conference you went to?

People couldn’t tell.

They wouldn’t even notice.

They have no idea. They thought I was just a good speaker.

“This guy has great recall.”

“He’s got it all together, wow.”

Were you just reading your teleprompter on your glasses? Wow…

Secret’s out…

Secret’s out now.

I mean, I got the idea from – what’s his name? The guy who created Oculus… Palmer. Palmer Luckey.

Right.

Palmer Luckey uses these glasses when he gives keynotes. That’s where I got the idea from.

And it looks like you’re just talking extemporaneously, but you’re just reading.

Yeah. I mean –

Ron Burgundy style.

…it also requires a certain level of – Ron Burgundy… [laughter]

You’ve gotta make sure that script is right, you know? You don’t want to just read literally anything that’s on there.

But it needs a bit of practice… It takes a while for you to get good at it, too. I sometimes use it just for – I don’t read, read, read, read… Like, now I’ll use it to get some numbers… I don’t want to look at the screen… I mean, there’s some numbers that you don’t really remember, righ? And the order of things, the flow state…

There’s so many amazing things with AI now you can do. Incorporate it into your real life. Like, I have my new home building in Texas - it’s a large home, and multiple stories, soundproofed… It has cameras where possible around the campus, which is about two acres, inside and outside the house, it has sensors, infrared, as well as ultrasound sensors, for places where you can’t have cameras, like bathrooms and bedrooms and whatnot…

[01:54:00.27] The goal is to detect any living form in the facility, within the walls or outside the walls, and particularly to detect any mishaps. Like, my mother-in-law, she’s moving in with us… She’s got her own very private unit inside the house… If she falls down, for example, I wanna know. Or the AI should know.
So all the information, everything is recorded and analyzed and fine-tuned locally, using a massive GPU cluster. Nothing leaves the boundary of my home, the network of my home. It’s a fully locked down network, because a lot of this stuff is private. Every conversation. So this conversation would be recorded, would be transcribed, and so that I come back and talk to you next time, I’ll remember what I talked about. Simple things of that nature. Especially –

Yeah, you read my name off your glasses. “Hello… Jerod.”

Security. Like, for security purposes, if someone is there on the campus that’s not supposed to be there, there’s gonna be three drones that get dispatched, triangulate the subject, or the target rather… And if they still don’t comply, there’ll be like the robot dogs that’ll go in with –

This is now? This is your house now?

He’s building that.

I’m actually working on that. I’m building that system. Is that the West Lake house, or this is a different place?

The drones are currently set up?

I’m experimenting with a bunch of drones… The problem is a lot of these drones are like Chinese drones, and I just don’t like the software coming from China. So I’m trying to keep it mostly American.

What model are you using locally?

DJI is the only – DJI is the best drone right now, unfortunately.

Yeah, DJI drones are awesome. Are you using LLaMA, or what are you using locally, for fine-tuning?

DeepSeek.

DeepSeek, okay.

R1. 405B. I mean, Chinese have way better – I mean, nothing goes to China right now; obviously, it’s local. But DeepSeek is really good for agents. LLaMA is pretty good too. LLaMA 70B is pretty good for agents, too. But I’m experimenting. We don’t know. I’ve gotta load-test this stuff, really put it through chaos testing and whatnot…

Are you documenting this somehow, somewhere? You’ve got a blog? Do you have a TikTok?

I’m debating, because it’s a security posture. I can’t expose the security aspects of the system.

Sure. You can pick and choose what you’re gonna expose, though.

Probably, probably. I haven’t thought about how – all my friends that I talk to, they all want the system at their home… So I don’t know, is it a product opportunity there? I don’t know yet.

It’s a new business.

This is definitely a new business, righ? I mean, imagine a house where you can talk to the house, that’s local, running locally…

That’d be cool.

It’s responding to things… I mean, particularly about health. Like, I wanna see what I’m eating all the time. What kind of food –

Do you wanna see what you’re eating? I imagine you can see it going in your mouth…

Analyze it, I meant.

Oh, analyze.

I wanna know continuously what I’m putting in my body, and what I could do better… Like, things of that nature.

I was thinking about that recently in particular, this whole idea of what comes into my home from the grocery store… There’s a way to communicate that to intelligence, to my home.

Yeah. So all the attempts of like taking pictures of what I’m eating - that’s not gonna work, because I’m never going to remember.

It’s too hard.

I’m gonna be like picking something, and I’m talking to somebody… It has to be unobtrusive.

Totally. Dead simple.

It has to be like out of your way. That’s the basic thing. If it’s in your way, you’re not gonna do it. It doesn’t matter how much effort – I mean, I don’t know, you’ve gotta be highly motivated. So it’s simple – you can use AI to automate your everyday life so much, with the current tech. You don’t need much. It’s not gonna be that expensive, too. I mean, maybe… Sorry, I take it back. It’s gonna be expensive –

[01:57:40.08] Yeah, I was gonna say…

The GPUs will be expensive.

How about drones?

Drones are the cheaper aspects of it.

A few hundred bucks each?

That’ll be cheap. The AI, the chips to run in real time, AI that can respond with low latency to agents, that is expensive. The chips are expensive; energy-expensive.

Well, just lease some.

I’m so pro like local AI of the future, and leasing the unused compute back to the grid, an AI grid, with Akash. I don’t know, I’m very excited about this new future of like shared economy. Because everybody wants GPUs, if you want AI locally, in your homes… Say we have passed through the whole limits for whatever your grid imposes on you… I mean, I’m particularly excited about what you can do with AI. And also, the new home is very sovereign. It has rainwater collection about 70,000 gallons, rainwater capture. And it has a greenhouse onsite, that produces food, local energy production with – and even water for food.

70,000 gallons you said?

Yeah. Two massive rain capture tanks underneath the pool, they’re hidden… And it has great water filtering. [unintelligible 01:58:57.01] the plants, it produces food locally… The more you learn about the grid, the less –

And this is your home in Westlake you’re building? All this? Come on…

The permitting is taking a little bit…

I wanna visit the site with you. I wanna come and visit the site with you.

Yeah, happy to show you. I mean, the permitting has been –

I won’t take any pictures.

…they’re like “What do you want to build?” I’m like “Well…” They were blown away with my plans, and they were – a little challenging with permitting, but we’re getting through that stuff.

Nice.

I even proposed a bunker, but the bunker is gonna be challenging. [laughs]

Oh, boy… We’ll just make the house the bunker. Let’s make the house the bunker. Wow, okay.

It turns out you cannot put a real nuclear fallout bunker in Westlake, unfortunately. It’s kind of sad. You need a ranch for that stuff. It’s terrible.

Yeah. Well, just buy yourself a ranch, too.

Yeah, I was gonna say, there’s lots of land in Texas, you know…

Well, it depends on where the land is. So Westlake is livable, right?

Yeah, you’ve gotta get outside of town.

Outside of the whole Travis County.

Yeah, where nobody wants to live. Then it’s real cheap land.

Then you can put in whatever you want and nobody cares. You can put a range if you want, you know… Most people do.

The challenge is you have to get there from your house as things are exploding around you, you know?

So we looked into maybe a helicopter, a helipad… There’s options there. [laughs]

First principles thinking. “You know what? What have I gotta do here?”

[unintelligible 02:00:15.04]

On a modified Cybertruck that can bulldoze through pretty much anything…

There you go.

…in case of zombies.

Yeah. Zombies are next, my gosh… Listen, we got AI. I never thought we would. I didn’t think we’d ever have zombies, and I don’t think we ever will. But… Wow.

I have a thesis for zombies…

Oh my gosh, I bet you do… What is your thesis for zombies?

Let’s save it for the next time we have you on, Greg. We have to leave something on [02:00:49.19]

Let’s throw it to Plus Plus. Let’s end this and throw a little zombies to the Plus Plus folks.

Alright. Thanks, Greg.

It’s good stuff, man. Thanks a lot. I appreciate it.