← Blockspace: AI & Bitcoin
Blockspace: AI & Bitcoin
Trump's $1.4B Crypto Payday and Meta is Entering the AI Cloud Business
00:49:23
Loading summary
Transcript100 lines
- [00:04]
A
What's going on, y'?
- [00:05]
B
All?
- [00:05]
A
Welcome back to Blockspace Live presented by Clean Spark. Charlie Trump's financial disclosures for 2025 dropped yesterday and everyone is in a kerfuffle because the President disclosed over 1 billion in income, not holdings, from Cryptocurrency Ventures. Wild number that we couldn't have dreamt up in our wildest imaginations when he was giving that speech@bitcoin 2024. So we will be covering that as our opening SE. Following that, we have Christopher Gillette on to talk about the data center trilemma and his seminal piece on what the real bottlenecks are for getting data centers and power plants interconnected within the grid. Following that, we welcome back Thomas Brazil to talk about the SATA sell off alongside Stretch and whether or not Strategy's capital plan really addressed the problems that are shaking its stock price right now. Then we will end on Meta getting into the cloud data center business and a move that has shaken up NEO clouds on the.
- [01:05]
C
That's right. Blockspace goes live every weekday at 1pm Eastern featuring quick hits on AI data centers, some bitcoin mining, sometimes bitcoin tech markets and emerging tech. If you like what you hear, you'll love the newsletter. Newsletter blogspeaksmedia.com and if you missed the live stream, it turns into a podcast anywhere podcasts are found. This show is brought to you by CleanSpark. Nasdaq listed ticker CLSK. More on them later on in the show. Colin, my net worth is down because I made the mistake of buying bitcoin and I'm wondering who out there is on the other side of the trade. Apparently it was the president. He made money. That's where it's all gone. All of my Meme coins, all my shitters, he's got them all. So apparently.
- [01:54]
A
I don't know what to tell you, Charlie. You should have become a billionaire and then become president and then you can make another billion off of speculative crypto ventures.
- [02:03]
C
Yeah, I'm just waiting for the Coffeezilla Expose video. He's probably actually already got an Expose video on this. But so I think. Let me show, let me show my screen first. Let's do this. Every news outlet came out this morning. This is NBC. Quote, Trump's financial disclosure lists 1.4 billion in crypto earnings powered largely by Meme coins. Not surprising to me what the details. What does it say though?
- [02:34]
A
Colin I'll dive into that right now. Let me get my screen up here. I've got the full 927 page disclosure
- [02:43]
C
here, which by the way, no wonder it took so long. It's longer than like the Brothers Karamazov.
- [02:49]
A
It is longer than the Brothers Karamazov. Actually it's probably longer than most all of the Dostoevsky novels. That being said, so 927 pages here. Vance's was like 20 odd pages. Just to give you an idea of maybe. I don't know what a normal disclosure like this should look like, but I was struck by the scope and scale of all of Trump's business ventures. And we're not going to get into all of those, but this is pretty interesting if you want to see just where all of his money is being put to work. But for the purpose of our live stream, we will be focusing some specifically on World Liberty Financial and the Trump meme coin. So to recap really Quickly, Trump owns 38.25% of world liberty Holdco, which is the company that launched World Liberty Financial via the Donald Trump Marks Defi LLC and is entitled to. I didn't realize this, Charlie. 75% revenue share from the platform.
- [03:44]
D
Yeah.
- [03:45]
C
A man whose son was deleting tweets associated with World Liberty Financial. That's surprising. Yeah, he was originally what, co founder, then downgraded to Advisor. Now fully taken off the website, yet 75% interest in this company and I
- [03:59]
A
believe they were down, demoted to like ambassador after advisor and then they're just wiped entirely. But this shows you that like, you know, they're trying to distance themselves from the project but like clearly they've been involved with it from the start.
- [04:09]
D
Right.
- [04:10]
A
So as part of that business dealing they were granted Trump's LLC, which has that equity ownership, was granted 22.5 billion world liberty Financial tokens. And if we look at how much money Trump actually made from the World Liberty Financial and the World Liberty Financial token sale, it's pretty staggering. So first of all, he made 65.6 billion in equity earned sales from equity sales of World Liberty Financial stock. I, I don't. That's not a publicly. Or is it a publicly traded company? I actually, I don't think so. I don't know, but. And then he also made 236.25 million from token sales of World Liberty Financial. Now as I understand that that is separate from his allocation in the project. That's specifically from the revenue share that the company is guaranteed under the equity structure. He also made another 42.25 million from specifically virtual USD key. I don't really quite understand. I guess that's just US dollars. And then another 56 million from USDC from the token sales and 150.6 million from Ethereum and a few others. So people could buy the World Liberty Financial token in various cryptocurrencies. And so the proceeds from the token sale are listed according to which currency was used for purchasing the World Liberty Financial and the token sale. In total though, Charlie, he netted 526.8 million in income from World Liberty Financial token sales. And then if you include the equity portion of that, it's 592.4 million that he netted from World Liberty Financial including that equity stake. Now the one big caveat here is we don't know whether or not this revenue was shared in kind in the cryptocurrency that was paid for the token or, or if it was cash settled. I would have to imagine if it was in kind, which I think it probably was. The majority of that was probably sold immediately. And then some of that residual may have shown up in his cryptocurrency holdings. He holds between 5 to 25 million in ETH and over 50 million in Bitcoin. And the real kicker, if we go to page 848, this is, I think what people will draw the most ire from folks who are looking at this. 635 million from the Trump meme coin sale.
- [06:43]
C
Holy smokes, man. Over half a billion from his meme coin, which frankly, you know, I'm not surprised because it's in his likeness and it was launched in close association with, although somehow managed to be somewhat arm's length from. But that's a lot of money. And I, you know, I remember 10 years ago or eight years ago during his first term, a lot of media was like, is he even a billionaire? I remember that whole narrative. Well, he's definitely a billionaire. He's made it all from the hoi polloi buying his token.
- [07:19]
A
So, and let's go ahead and do. We'll wrap up here in a second, but let's go ahead and just do a quick temperature check on World Liberty Financial.
- [07:26]
C
Yeah, how is it doing? Oh no, it's down. It's only $1.875 billion, but down.
- [07:35]
A
Yeah, it's down 75 since it launched
- [07:38]
C
in August of last year, which frankly Colin, it wasn't. So it didn't launch in August. It actually launched like leading into his inauguration. I think this is, that's from when the coin gecko tracks it, but frankly, 75% down over the past year. And a half is pretty good considering the landscape of all meme coins. So actually it's, it's outperformed.
- [08:02]
A
Well, this isn't a meme coin, Charlie. Let's get our definition.
- [08:06]
C
It's a defi platform.
- [08:09]
A
And if we look at Trump's coin, it's, you know, what's funny is it's actually rallying today. I wonder if there was a kind of like, oh, it's in the news again. Maybe we can get a little bit of a pump. But it's down, you know, 94% since it launched on Inauguration day or the day before the inauguration. So, you know, all that being said, just an incredible, you know, windfall for the Trumps here, and last thing I want to share here, Charlie, is before we move on to our next segment, is Trump getting grilled about this? I guess this is a Forbes journalist or this is just Forbes posting this on YouTube.
- [08:48]
C
Gotta love a good Trump interview.
- [08:49]
A
Let's hear Trump's reaction to the. Here's the question about the financials. And here's Trump's reaction.
- [08:56]
B
Your financial disclosure shows you what a very lucrative year last year. What message does this send to average Americans, especially those.
- [09:04]
D
I don't get involved in my personal. We have funds that run my money. Well, I've made a lot of money before I became president. And they invest my money and I don't talk to them. I never, I don't even speak to them. So I have many people, I don't know what they call closed accounts or something. You put your money in and that's it. I don't talk to them. They're big institutions.
- [09:26]
A
So the reason why I wanted to
- [09:27]
D
show this, they run it.
- [09:29]
A
But yeah, the reason I wanted to show that specifically is, I mean, that's 100% what's happening, I would imagine. I mean, the dude has a 900 plus page financial disclosure. There's no way that he's managing this actively himself. But the, the lion's share of the income for his 2025 financial year came from cryptocurrencies.
- [09:53]
C
And yeah, his statement is. His statement is plausible just because of the size and breadth of his portfolio. If you're a billionaire like that is. I don't want to assume. I don't know too many billionaires. Not surprised about that. The fact that the majority of the income is.
- [10:10]
A
Yeah. And those were decisions made in 2024, I would imagine. Right. And probably his sons were leading most of these calls and then kind of just, you know, giving him the information and maybe working with him on how to structure the equ. But you know, the Trump meme, Coin and World Liberty Financial, the seeds for that were planted in 2024. So he probably had a much more active role in that. But what, you know, it's, what's incredible to me is obviously you're just going to do a broad stroke question here for this interview, but the, the crypto holdings or the crypto income is the whole bag here. It's, it's the vast majority of the wealth that he generated last year. And I just can't help but think back to the closer for Bitcoin 2024. Have fun with your bitcoin and your crypto and all the other things you're playing with. I mean he certainly had fun.
- [10:59]
C
Oh, he had fun. I mean he took his words literally. Okay, let me share this then because we've been, we've been punching on him kicking a dead horse. Here, here's a counterpoint, a couple observations from Trump's stock portfolio. Quote, he's not in any nuclear names that stand a Bennett from the, from the Trump's nuclear admin admin nuclear efforts, not in any of the garbage stocks that his sons have pumped and dumped on other people. That's outside of crypto and quote he's not in any rare earth stocks that stand to benefit from Trump admins rare earth efforts. His portfolio according to this Twitter poster is still mainly focused on American excellence. I see Nvidia, Tesla, MasterCard, Visa, Apple, Palantir, Goldman, Sach. So you know, take everything Colin, I say sift it through the fact that our bags are down because we're dumb. Poor bitcoin plebs who have misallocated ourselves to bitcoin instead of, instead of scamming instead of AI stocks which by the way as we're transitioning to AI, we've got in the wings Christopher Gillette who wrote a fantastic piece on Internet interconnection queues. We're going to interview Chris here in just a moment after a word from our sponsor, CleanSpark.
- [12:13]
A
We are CleanSpark, America's Bitcoin miner. A publicly traded company with the largest operating hash rate powered entirely by self operated infrastructure across four states. This is our proof of work. We are setting the standard for what's next. Learn more about the intersection of energy and bitcoin@cleanspark.com. If Bitcoin's actually the best money and
- [12:48]
C
it's the thing that people should accumulate and it's the best risk adjusted asset, I lose zero Sleep about whether or
- [12:53]
A
not that's gonna happen. I just ask the question of when is literally matrix map that you're running
- [13:00]
C
on large pieces of data.
- [13:02]
B
The bitcoin miners can absorb that energy.
- [13:04]
C
And in many ways this feels like
- [13:06]
A
a second bite at the apple to
- [13:07]
B
build a new Internet.
- [13:18]
C
All right, while Colin is hot swapping his camera battery, I'm going to bring up, I'm going to drop Colin, I'm going to bring up Christopher Gillette. Chris, welcome to the show.
- [13:34]
B
Thanks for, thanks for having me.
- [13:36]
C
Hey, sweet. And we didn't do a micro check and you sound great. So perfect. I'm going to let Colin solve his technical difficulties really quick. But you wrote a pretty seminal piece last which you published I think was it last week why American data centers can't plug in. The AI buildout is bottlenecked by energy. A lot of folks in the power markets U.S. american grid reposted this. I read through it. It's very fantastic. Validates a lot what we're seeing. Can you give me just the high level of the premise of your assessment from this piece? Sure.
- [14:19]
B
Well, the article is an overview of what I think is the main challenge to growing the power grid in the US which is the interconnection queue. So I explain what the interconnection queue is, what the problems with it are and what I think are two of the best solutions for it. Which is one, prioritizing the interconnection of flexible resources, that's flexible large loads or flexible power plants. And prioritizing the interconnection of the highest value power plants. And the reason for this is that we need to grow the power grid in the US to facilitate investment in AI but also electrification and all the other future technologies that are going to run on, on energy. And because of the growth in AI there have been a lot of capacity constraints that are beginning to materialize.
- [15:07]
C
Cool. I'm going to switch the cameras. One sec. Put you in the middle. All right, we got Colin back.
- [15:15]
A
Sorry about that. Chris, thanks for joining man. And great, great piece by the way. Really, really, really clear eyed I think and kind of gives a good top down view of like what's actually happening on the ground.
- [15:28]
C
So what again you say that the limitation, the constraint does not appear to be power generation but like the interconnection queue and there's a lot behind that. Again, this is kind of an arcane topic for those who are not, who do not study the power grid. Can you just like give me a high level view on like the numbers on data center and AI power demand looking forwards and then we can talk about where the mismatch is for interconnection.
- [15:56]
B
Sure. So there are a lot of projections of AI power demand out into the future and there's a huge band of where they sit. At the top of that of that band you have the number of large load interconnection requests in ERCOT, just for example, just one grid operator in the US last month they reported 450 gigawatts of large load interconnection requests and their peak load historically is 85 gigawatts, which is a massive difference between those two numbers. Obviously there's nowhere near enough capacity to serve even a small fraction of the large load interconnection requests. Now only a small fraction of those requests will be built. But other forecasts but the build out of Data centers around 100 gigawatts by 2030.
- [16:46]
A
So Chris, one of the more interesting pieces in your article, you don't take too many stabs of saying here's how we should fix this, but one of the things you do say is that the first come first serve model needs to change. And you mentioned an auction based system to where you might actually be able to allocate what power assets do exist a little more, less on a first come first serve basis and more on an as needed or critical infrastructure basis. Can you unpack that for our listeners?
- [17:17]
B
Sure. Well, the interconnection process is a series of engineering studies and building upgrades to allow resources to be built. This is the step in the development process for any power project that takes the longest and that has the most uncertainty around cost. So it's the critical step. And there are a lot of lot of sort of data points to point at to indicate that it's the critical bottleneck. And the interconnection queue is the process by which different plans are prioritized. And currently that's a first come, first serve process, first in first out queue where everybody is served in the order that they arrive, regardless of any of the particular features of the project. But it is the case that some projects are much more valuable than others, much more viable than others. And so the argument goes that you should be prioritizing very large projects that are very viable, that are backed by experienced developers, whatever the case, as opposed to the other ones. A great analogy for this that Alex Tabarrock gave, and I wish I had in the article, was that having Cessnas, it's like having Boeing 747s queued behind Cessnas At Denver International Airport, you can, it makes sense why you would start off with the airport being a first in, first out system. But as the traffic grows, there become significant costs to not having prioritization. Now some grids, I think probably most at this point have adopted some kind of fast track process where they are giving priority to certain projects. But they're doing this using an arbitrary, not arbitrary, but an administrative mechanism, a scoring system, a rubric. And that's certainly preferable. But I think the developers have a lot of private information about the viability of their project, how much pushback they're getting behind the scenes, how firm their permits or their paperwork really are. And so I think that letting developers express their own confidence and the value and viability of their project through an auction system might be a better structure.
- [19:27]
D
You do?
- [19:28]
B
Yeah.
- [19:28]
C
We've been covering like the rising pushback from like multiple angles, whether it be, you know, feasibility of the deal to even just the rising local pushback angle, which maybe we'll talk about. But one thing I'm, I was very surprised. I knew that like a lot of these projects had been withdrawn. Withdrawn. And I'm not really sure of this, like breakdown of rationale. But you, I believe you say 72% withdrawal rate, if I'm not mistaken. That seems crazy to me. Is that somebody, I mean, where's the number come from? And like how much is that affecting the, this whole process here?
- [20:05]
A
And also Chris, I have a question to kind of complicate that. How much of that are like so called phantom loads where it's the same developer requesting the same interconnection from different utilities?
- [20:17]
D
Yeah.
- [20:17]
B
So the source for that statistic, I believe is the Lawrence Berkeley National Laboratory report. Called in the queue, or queued up, something like that. Great annual report on looking at the queues at every ISO. And you're right that phantom loads are a big part of this. And there's a feedback loop that gets worse over time where the reason, when the queue has a lot of speculative projects or is just cluttered with a lot of projects, it's very hard for developers to estimate for themselves what their interconnection cost is going to be at any given location. Because there's just so much going on, it's very difficult to model. And then they don't know what other projects are going to do. They don't know which projects are real or not. And of course neither does the system operator. And so because developers have that uncertainty, it pushes them to submit multiple projects at multiple sites, only intending to Develop a fraction of them, but they're fishing for good locations. This further clutters the queue and makes the problem worse. Now, most grids, I think probably all of them, have taken steps to address this by raising the deposits that are required to put in a study request, by creating requirements that you have to prove that you have control of the land, that kind of thing. But it is still a problem.
- [21:40]
C
And this is where you get into what you call the data center trilemma. Us as bitcoin folks, we love the mental model of a trilemma. Explain the data center trilemma to us and any insights you can pull from it.
- [21:55]
B
Yeah, the data center trilemma is this idea that as a data center developer, you want your project to be as big as possible, to come online as fast as possible and have the highest reliability as possible. But in the situation that exists today, you can kind of get two of those, but they all trade off against each other. And one solution to bringing more data centers on more quickly, large data centers, more quickly, is by having them be flexible. And this is something that bitcoin miners have pioneered in ERCOT is having large loads that can talk in real time with the system operator and be told when to turn down to create, to free up space on the grid for other higher value users. And the idea in the article is that you can use, connect and manage for loads, which is that you connect a load to the grid even before all of the transmission upgrades have been completed, under the condition that it'll turn down if ever it begins to overload parts of the grid. And so this is a way that loads can come online more quickly, but it comes, of course, at the cost of reliability. They're going to run less often, but it's a fast way to get large loads on.
- [23:16]
A
And you mentioned in the article that really ballpark downtime for most of these data centers, depending on the market, would only be like 22 hours a year, something like that. No, no, no, go ahead, Chris, go ahead.
- [23:32]
B
Yeah, because the grid is built to serve peak capacity. So the interconnection studies are run looking at a snapshot of the peak grid stress conditions. And then they build transmission elements to be able to serve all load with this new resource coming online in that peak condition. But that stress condition only occurs a few hours a year. And those 22 hours, for example, wouldn't even be cumulative. That's a cumulative number. So your, your downtime at any given time would be just a few hours, probably just a few hours at a time. And Then you could switch to backup power to have a firm load. But showing the grid flexibility.
- [24:19]
A
Yeah, it just makes me wonder, Sorry Charlie, just I'm kind of thinking out loud here. It makes me wonder like what the best path is for some of these companies because adding enough backup for 22 hours, it's not very much time. But the CAPEX requirements are going to be tremendous. You mentioned in the piece that batteries are a little more economical compared to gas turbines. But it makes me wonder if you're a data center developer, do you want to be in a situation of xai with colossus where you just get the gas already so that you can skip the queue and then you can have, you can energize earlier and then you have that as a backup or do you just build the backup and then wait to get plugged in? I'm curious, this isn't really a question for you. Just thinking out loud like what operators will converge on in terms of what makes the most economic sense.
- [25:06]
C
I'll turn it into a question which is it's really, it's huge. Capex to get these behind the meter solutions you basically have to build your own power plant. I mean does that kind of create like a two tiered system where you're basically where only the hugest of players can build out these power plants that everyone else is just stuck waiting in the queue?
- [25:27]
B
Yeah, there are a lot of disadvantages to behind the meter generation. It's certainly much, it's always going to be costlier, probably much costlier than grid power. It's always going to be less reliable too and you have to over build because some part of the year you're going to have scheduled or unscheduled maintenance on, on your generation and that, that forces you to build more capacity than, than you need at any given time. So you get very low utilization of those resources. And from a social perspective, from the grid cost perspective, this is bad too because we have these very nice turbines that we would like to get high utilization out of but it's only being used, it's being sort of captive to this one power plant. Now as a percent of the total project costs for a data center, I don't know that it's all that high. I don't know how much disparity it creates between players. I guess everybody's pretty well capitalized anyway. But certainly from a grid perspective there are big downsides to having that much generation be behind the meter. But I think probably the incentives of the AI industry point in that direction and We've seen there are studies saying a lot of developers are looking at that. I think it'll probably be a little less popular than, than studies show. But there have been big projects and big deals even recently of off grid projects.
- [26:45]
C
My last question to wrap this up, zoom out. If you could look at the landscape of like FERC and state regulatory rules, is there one like regulation or procedural thing that you would change to really unlock and dramatically improve the process for building out this industry domestically?
- [27:10]
B
I'd love to see the procedures in place to allow large loads to be flexible in real time, to explain, to communicate to grid operators what the value of their load is and to allow grid operators to turn down those loads to manage capacity or congestion. Because one of the big problems with the power grid, how it's structured, is that demand is inflexible as a percent of what percent of load is flexible on the power grid at any given moment. It's very small. Almost everybody is on long term agreements. And so that lack of flexibility creates significant volatility. It creates very irrational behavior like people starting a, a dishwasher that's, that's half empty or half full, you know, when the grid's on, on the verge of a brown out, for example. So having more load respond to prices in real time I think will lower, lower costs and cause more rational behavior.
- [28:15]
C
Chris, thank you so much for coming on the show. Really appreciate the piece. I'll probably hit you up next time because this, all the FERC stuff is like a little mystifying and arcane to me. So thank you so much for your time and we'll follow up later.
- [28:29]
B
Thanks for having me guys. Enjoyed it.
- [28:31]
A
Thanks Chris.
- [28:34]
C
Super smart guy. Glad to have him on the show. We have another super smart guy return of Thomas Brazil back. He's in the backstage. We'll get him on up here and we'll talk about both a recap on Sailor and Stretch, but this time focusing on on seda. Before I bring up Thomas, a word from our sponsor, Luxor.
- [29:01]
A
This episode of Blockspace Live is brought to you by Luxor's Commander Bitcoin miner management software for enterprise operations. Luxor's Commander gives you real time fleet monitoring, bulk remote commands across your fleet and intelligent miner that is an automated profitability engine that runs every five minutes and adjusts your power settings to live energy and hash rate markets. ERCOT backtest show 10 improved profitability with intelligent mining over binary mining. Luxor's commander is 100 per megawatt or a 25 basis point pool. Fee adder and you can try it for free for 60 days if you're interested. So to learn more go to Luxor Tech Forward slash Commander.
- [29:39]
C
We're going to go to talk about the Meta Cloud deal which is absolutely nuking the other Neo Clouds crashing down. Average down 5 to 10% per stock today. Before we go to the Meta Cloud announcement, a word from our sponsor, Lygos.
- [30:01]
A
Hedge funds and lending desks are getting liquidated. Is your Bitcoin safe? It's not just Bitcoin. Price drying up, Big lending firms are reeling as Bitcoin treads water under 60k. It's more important than ever to know who actually controls your Bitcoin. If you're lending with a Bitcoin backed company and with Lygos, that is always yourself, do yourself a favor and don't become the next FTX or Celsius victim. Check out Lygos Finance. They are our preferred Bitcoin backed lender here at Block Space. They use Bitcoin native smart contracts to make sure you are always in control of the keys to your kingdom. To learn more and to unlock rates as low as 8.5 to 10% APR, go to Lygos Finance. All right Charlie, let's talk about disruptions in the AI market with a Mag7 company saying they are going to move into cloud computing services.
- [30:57]
C
Yeah, I'm seeing it being called the Deep Seek moment for NEO clouds. So if you're not familiar, it was just over a year ago Deep Seq, the Chinese AI company came out and basically said all these frontier models from OpenAI, Anthropic, et cetera, we can do it for way cheaper. And so that caused an entire industry repricing across all of AI calling. If you can remember, a lot of people called it the top. It was not the top. So what's happening today? Meta basically saying we've got all this compute capacity, we're going to pull the NEO Cloud model, kind of call it Meta cloud. And that is causing a route across all the Neo clouds. Just to show this, which is DI metrics Neo clouds as a sector down 9.1% today. CHR in down 20%. Core weave down 12% net Nebus down 13.8%. It's a bloodbath out there right now. And the headline just choosing the one because this is a Bloomberg exclusive here. Meta is planning a cloud business to sell AI computing power. Basically they're jumping in the fish tank with Amazon Web Services, Microsoft Azure and Google Cloud. To the average person, I think they might think oh, Meta already does that? No, Meta has not yet pulled the trigger on that. This is them pulling the trigger. Meta, which has been rushing to secure expensive data centers and other infrastructure to fuel its own artificial intelligence ambitions, is forming a business to generate revenue from excess computing power sold to outside customers. According to people familiar with the matter, this would also include selling raw computer capacity. Kind of like neocloud businesses like Core Weave.
- [32:58]
A
Yeah, I mean this, this sets Meta up as a direct competitor to these NEO clouds. And the bloodbath we're seeing is, like you said, Charlie, probably is a deep SEQ moment in the sense of investors are not having to think how they actually price these companies and how they value them. When you consider that one of the most heavily capitalized and powerful tech companies in the world is now going to be a direct competitor. And just to give you an idea of the scale of what Meta is building currently for AI data centers, they have 6 GW under development currently and so far in 2025 and year to date 2026 they have spent 92.06 billion in CAPEX and they project they will spend another 105 to 125 billion throughout the rest of the year. And so here's the biggest, here's the obvious question for me, what does this signal about actual compute demand? Because some people are reading this as bearish on demand, but you could also just read it as bullish in the sense that Meta sees an opportunity to sell compute, that they are not going to use themselves in a NEO cloud model and have the optionality to use it if they want to in the future. And I, I probably lean a little bit more towards that side of things because there's, there are parallels to XAI here because this is what they did XAI did with Colossus and leasing out some of that capacity to Anthropic.
- [34:24]
B
Right.
- [34:25]
A
That XAI is not going to use all that for grok. They'll sell some of it to, to Anthropic. They have the optionality there now they have two different business lines in terms of AI compute and how they're monetizing it. Investors really seem to like that. And that's part of the reason why SpaceX ran up so much after its IPO as we covered here is it's less so the space angle, more so the compute angle that is currently making money right now. Charlie, you're muted.
- [34:55]
C
Here's a tweet from Wayne Nelms, I guess at or we had Oren co founder Kush Bavarian Yesterday, Wayne Nelms speaks to your question and comment. Colin Meta has been aggregating supply and historically been serving excess capacity across its various business units. However, reservations are not the same as utilization, which is the underlying concern for all XPU owners. Zuck Entering the space means margin compression for the clouds, which is bearish for the price of compute, but bullish for market stability and legitimacy. This is a natural evolution of the market. What the world is only beginning to understand is that there are no pure buyers or sellers of compute capacity. Everyone is doing both. Expect compute trading desks to continue popping up not just among the clouds, but also across all large enterprise in the next two quarters for the average consumer, Colin, this is good.
- [35:51]
A
This is great news for the average bag holder of a number of high growth tech stocks, this is terrible news.
- [35:59]
C
Yeah, I mean it is a real
- [36:00]
A
shot across the bow and going back to deep seek. The fears for that were overblown mostly I think partly because we don't actually how much can we trust their assertions that the model was trained at such low cost compared to other models. You know, there was a lot of hemming and hawing or the fact that they piggybacked probably over a lot on a lot of early work by OpenAI and Anthropic in terms of getting the model up to snuff and then took it the rest of the way for their version. But it does call into question, honestly a lot of the companies that we cover here in terms of how much is the okay, if the Neo cloud model is under duress from this news, how much is the powered shell model actually going to be worth? Especially if the current tenants at their sites are probably overpaying compared to what will be paid in two or three or four years.
- [36:58]
B
Right.
- [37:00]
C
I'll say this, I'm gonna. You can be the skeptic and I'll be the bull. We can be bull cop, skeptic cop. And I'll say what Kush said yesterday, which is Jayvon's Paradox Computer comes cheaper, demand goes up, price goes back up again. This is really just more of a repricing moment in a. In a longer market trend. That said there I I do want to. I'm not going to play this very much but this is Alexander Karp, CEO of Palantir. It's a 20 minute video where he's on CNBC talking to Aaron and the other gang and Joe and he's getting very heated on the frontier model topic and talking about how just the gating of this and the way it's priced like the access to these models is priced is really dislocating the value of compute, both for businesses who use this and consumers. And so he's very frustrated at this. And I think that the Meta story does play into this. Here's a semianalysis tweet with three circles in a Venn diagram. Meta, Iron and Allbirds all in the center pivot to Neo Cloud. Meta has wearables, Allbirds has wearables. Iron and Allbirds both have an identity crisis, he says. And then Meta and Iron both have shitloads of GPUs. So great meme from semianalysis. It does point to a major trend which is everybody's pivoting neocloud. Here's another tweet from Amit Amit Amit is investing a bullish and a bearish take Bearish take. First, if Meta has excess compute, they're willing to sell to a new cloud business via new cloud business, doesn't that mean we aren't compute constrained? Why would Meta give a deal to coreweave or Iron if they just sell compute themselves? Furthermore, wouldn't they cut capex because Idle Compute as the basis for a new business means they don't need as much compute as they bought, which means capex should come down. That would be bearish for all folks in the semi business. The bullish take is if Meta is building a cloud business, even if they are using Idle computer, which means they aren't compute constrained, they might end up spending more CapEx to compete with GCP, AWS and Azure. Let's go cloud provider. Like, if they realize that selling compute and services on top of a meta cloud is better than just ads, wouldn't they end up having to spend in the same way that Google, Microsoft, Amazon do in order to build out a full cloud business? More capex, which would be good for semiconductors.
- [39:57]
A
Everything's good for Nvidia.
- [39:58]
C
Everything's good for Nvidia.
- [40:00]
D
Yeah.
- [40:01]
A
You know, and so going to the bull thesis here, I think maybe the bearish case could be understated some because just because Meta doesn't have use for all of the capex or for all of the data center space that they're building doesn't mean that the wider market doesn't. And it could just be a capitulation here for Meta to say it's not really worth us worth it for us to try to consume all of this compute internally for our own products and software when we're not building a frontier model. You know, we're not building anything that will ever compete with OpenAI or with Anthropic, and that's where all of the demand is. And so maybe they're looking at this just from a pure okay, well if we have all this compute, then we're just going to rent it out for more money than we can make building our own tooling internally. I think that there's probably a good chance that this ends up paying off for them and is pretty damaging for the wider sector of NEO cloud providers. But and one note on Jevons Paradox maybe before we we close here, and I actually have one thing I want to pull up as well, Charlie. In terms of disruptions in AI and kind of black swan events, Jevons Paradox plays out for a lot of technologies on a long enough time frame. But the railroad capex boom had a bust, right? And the, you know, the, the fiber optic or broadband boom in 2000s had a bust as well. Yes, over a long enough time frame. I believe that that is probably 100% true. But we're still in the stumbling infancy of this technology and we haven't found fully unlock all of the use cases for AI yet. A lot of the use cases, the most revolutionary ones, and the applications are. We're probably nowhere close to figuring those out. And full integration into businesses, all of these things will come with time, if AI really is what everyone who's bullish says it's going to be. But we'll need more tooling, we'll need more R and D on what actually makes sense for business applications. There's going to be a lot of painful lessons learned along the way, I think before we get to that escape velocity where more supply just means more demand. But on that note, I kind of wanted to close with this and I think you might have some thoughts on this as well, Charlie. This came out a few days ago from the Information and I don't have the unpaywalled article, but OpenAI reportedly discovers a new way to cut inference costs in half. So you're really getting two big news items here in terms of compression of the entire income model for running an AI company or NEO cloud. If inference costs can be cut in half, again, like you were saying, great for users, maybe not so great for some of these companies that have been attaching their cart to this horse. Then you have Meta coming out saying we're just going to go NEO cloud with these data centers. That to me seems like two major disruptions of the assumptions for what the business model is currently. And to me it is A reminder that advancements in this space can cause immediate disruptions and immediately reprice some of these stocks. As we've seen today, a lot of the Neo clouds down double digits.
- [43:25]
C
Yeah. My thought is that token pricing should decrease gradually over time, but I think it'll probably also track a little bit with the price of compute. Probably if COMPUTE spikes at the same time that usage spikes, we could see some reflexivity upwards. But on a long timeline this should trade downwards if you zoom out far enough.
- [43:58]
A
Yeah, I mean that makes sense to me. And that's kind of what that one Bull bear tweet of yours was saying is the cost of compute. We should see actual compression of that over time, especially as a big company like Meta gets involved.
- [44:11]
B
Yeah.
- [44:11]
A
If it really is a commodity like people say it is, it's, it's, it's going to see that compression, it'll trend
- [44:17]
C
towards the cost of production on a long, on a long, long timeline. Now I think a lot of this does depend on the geopolitical nature of this. Right now Chinese open models, comparatively open models are so much cheaper. They're 80, 90% cheaper per token cost. For really close, really competitive inference with the near frontier models. If for some reason the east and west does become divided or closed off, that could create similar to the straight of Hormuz scenario where you have local pricing of tokens or local pricing of energy and oil because the Americans can't get that sweet, sweet cheap Chinese infrastructure.
- [45:08]
B
Yeah.
- [45:08]
A
So the, the, the relationship will be flipped here though instead of like Asia was, was get hit very hard by the lack of oil exports because they don't produce as much there in this scenario. The US could be hit by this. I, I do, you know, I'm going to put a tinfoil hat on for a second. I find it very hard to believe that these Chinese models are that much more efficient. I have to imagine there is some government subsidizing going on in the back end for this. Like I just, I find it very difficult to believe that somehow they found a way to do this more efficiently when quite frankly the premier labs for working on this technology have been western and specifically us. But I nothing to, nothing to base that on. I'm just, that's, that's gut feeling here.
- [45:59]
C
Oh well, I mean look, it's for any technology the people are building the first version, the, the premier pristine like tip of the spear version are going to incur all the R and D costs. If you're in the business, if your entire like economy for the past 20 years has been figuring out how to copy the cheapest way the the cheap fair out the way to the cheapest way to copy how the how the Americans are doing things. I think that analogy still applies because it's the set. The first person to build a widget incurs all the R and D cost and the second person to build a widget does not have to incur all the same R and D cost. I think this probably applies to inference. And I'll also say that these models are not as good. They're just really, really close. And on top of that, think back. Like if you were to go back and use older versions of anthropic or OpenAI's models, like if you were to use like GPT4 or was it Sonnet 3 or Sonnet 4 or Opus 4? I forget, you know, way cheaper and they're still really, really good. A lot has happened in the last six to nine months, so keep that in mind. I'll I'll say that I think the Chinese models are probably more
- [47:20]
D
more.
- [47:21]
C
They're actually probably more indicative of what they they themselves have spent on it. I think, I think the pricing for inference from anthropic and OpenAI is so high just because it is skeptical and uncertain and they are doing all the legwork on building these frontier models.
- [47:39]
A
Yeah, I was going to say so part of the price of the inference is also price and payback for training the model. And if we assume that there's piggybacking for things like Deep SEQ and the R and D is already done, you don't have to bake that cost into running the inference.
- [47:53]
C
Yeah. And also like the value of the absolute most frontier model could be an order magnitude more like it might be that more valuable to be six months ahead of everyone else, especially as things accelerate. If we're talking about security and we're talking about remaining competitive again, this is me speculating. This is not me making statements over. These are qualitative statements on like, you know, do I think it's correct or not? But is me talking through it as I've spent a lot more time studying AI in the past three months than I have Bitcoin. So with that, thank you all so much for tuning in to Block Space Live. We go live every weekday at 1pm Eastern with the exception of this Friday, which is a US Holiday. Make sure to like subscribe Drop a Review Check out our newsletter, delivered daily to your inbox newsletter.blockspeaks media.com this show is brought to you by CleanSpark NASDAQ listed ticker CLSK check out Clean spark. I am Charlie.
- [49:01]
A
I'm Colin.
- [49:02]
C
And we'll see you tomorrow.