CleanSpark’s $6.6B Meta Deal, Rune Raises $40M for Solar Farm Data Centers, Moratoriums Haven’t Killed the AI Trade (Yet)

CleanSpark’s $6.6B Meta Deal, Rune Raises $40M for Solar Farm Data Centers, Moratoriums Haven’t Killed the AI Trade (Yet)

Blockspace

Fri, September 18, 2026 at 2:19 AM GMT+3

CleanSpark rose 6% in early trading hours today on news that it inked its $6.6B Sandersville AI deal with Meta. We also cover Rune's $40M raise to bring modular, containerized…

0:00 spk_0

Y'all. Welcome back to Blockspace Live. Big story for Clean Spark investors today, as the former Bitcoin miner finally announced the tenant at its Sandersville site for its first AI data center. More on that for our first story. Following that, we are going to do a quick pulse check on the IPO race between America's two premier frontier labs, Anthropic and OpenAI. Turns out, the front runner.Which became the dark horse after Anthropic lapped it earlier this year, might be catching up, and it could pay dividends if OpenAI pushes its IPO to 2027, bolsters its revenues, and reclaims its position as the number one LLM model in the nation.After that, we've got a pretty interesting story in the realm of the power hunt and how entrepreneurs are thinking pretty creatively about how to deploy GPUs as GPU hours soar and power continues to be the number one constraint. We'll be covering Roon's launch, which was announced yesterday after a $40 million round.And its plans to put containerized modular data centers that it claims you can deploy in under a minute, Charlie, on solar farms throughout the world, with its first flagship being in Texas.To end the show, we've got Data Center moratorium, gumshoe Charlie Spears on the beat looking at a semi-analysis article recently that says, moratorium? What moratorium? Even if you look at the states that have been the hardest on data centers, all of the projects operating in these areas are still working at a breakneck speed to deploy gigawatts of infrastructure. More on that at the end of the show.

1:43 spk_1

That's right. Blockspace goes live every weekday at 1 p.m. Eastern, where Compute's daily live show featuring quick hits on AI data centers. If you like interviews, check out our interviews with the executives who run and build these data center companies.Go to our website, Blockspace.media, get our newsletter in your inbox every single day. And today, I'm excited to chill our group on Twitter. Check out this tweet, Blockspace. Join the Blockspace group chat for more on AI plus HBC insider interviews like Clean Spark. We just announced it this morning, so hop in that chat.Uh, there will be a link in the show notes here shortly, or you can just go to our Twitter. That's a group chat on X. It is not paid, it's free to join.Anyway, Colin, the big news today is about Clean Spark.So toss

2:45 spk_0

that's right, Charlie. Clean Spark just announced a Mag 7 tenant for its first AI site, a bullish disclosure from the former Bitcoin miner as it seeks a $2 billion round to finance its first AI site at a time when capital could become a little more expensive and constrained following the Fed's rate hike yesterday.So,Clean spark.Drum roll, please. Announced its first ever AI lease on July 14th, a 20-year, $6.6 billion dollar deal for 175 critical IT megawatts at its Sandersville, Georgia facility. But here's the thing, Clean Spark never announced the tenant. The former Bitcoin miner only said that their flagship tenant is a high investment grade, leading global technology company, but it kept the company's identity secret until now.An SEC filing from today reveals that tenant is in fact Meta, the social media stalwart turned digital marketing giant and burgeoning AI lab. But that's not even the most interesting part here, Charlie. What this also means is that Meta holds an LOI for 885 megawatts from Clean Sparks Seeley and Brazoria sites in Texas.Since Clean Spark in July said the Sandersville counterparty also signed an exclusivity agreement for those sites. What's more, not only is Meta renting out Clean Spark's first fully-fledged HPC site with the possibility of expanding to a much larger portfolio in Texas, it's also serving as the guarantor of rent and operating expenses. This means that Meta is backstopping the rent for its wholly owned subsidiary.And Verna, the site's legal tenant at the site, to ensure Clean Spark gets paid no matter what. And this backstop is more important now than ever. As we covered on the show yesterday, the Fed's 25 basis point rate hike could lead to not just higher borrowing costs, but also tighten credit across for, across the portfolio of all these companies for non-investment grade companies.And there's the rub. Since Meta is the counterparty on the lease, the tech giant's investment grade credit backing will help Clean Spark secure project-level financing for the site under better terms at a time when credit is tightening. Clean Spark said in the filing that it's seeking $2.227 billion in project-level financing with a limited Clean Spark parent completion guarantee.We don't have the terms yet since Queen Spark hasn't finalized this financing, but it used 7.5% as a rate for an illustrative benchmark for the financing. It will need every penny for that project because it is projecting to spend $2.08 billion on the build, which Clean Spark projects will commence rent in November 2027 with the full build out in March 2028.After this, Clean Spark will turn its eyes to the Lone Star State, but there's a catch here, Charlie. It still has to clear Urcot's batch zero process. Clean Spark received 585 megawatts of baseload consideration for its Seeley and Brazoria sites, which means those megawatts are secured as long as Clean Spark checks off regulatory boxes, including quarterly stability assessments and Greg Ad it's abit.Uh, Greg audits, uh, Abbott's

6:02 spk_1

audit. Say that 10 times fast.

6:05 spk_0

ButIn addition to those base loads.Um, Clean Spark has also received studied load status for 300 megawatts at Brazoria. That means that Clean Spark will have to wait for the batch zero study results next April or later if there's a delay, which there probably will be, to see if they received the full 300 megawatts requested. And there's no guarantee they will. The studied load specifically,is a request for full load, but it's studied under an allocation study, and companies that are in the studied load category of batch zero may receive fewer megawatts than they requested. So then the question becomes, will Meta take every megawatt they can get in Texas from Clean Spark, or will it be a deal breaker if Clean Spark falls a few megawatts short of the 885 megawatt LOI? Time will tell.

6:57 spk_1

So, it's kind of interesting to see this on the back of the wash uh Fed rate increase yesterday because this means that it's even more valuable to have an investment grade tenant on the other side. And if you're one of these smaller guys who can't get the investment grade tenant, now the terms are probably a little less favorable than they used to be.So, uh, to me, this is a clean spark.Getting this in, get that hyper scaler, and uh maybe not that great for the smaller teams who don't have the investment grant on the, on the other side, there's that, that rates increase. We, you talked about this yesterday, Colin, with the, the financing terms, uh, that would change for the data center operators and builders post Fed hike, so, uh,

7:53 spk_0

Yeah, the, the investment grade counterpartty is more important than ever. They were always important for these operators because otherwise they would have eye-watering rates for whatever project level financing they were seeking. But now having that is even more important. And a few notes on that in a minute because the 10 years actually done the exact opposite of what you might think after a Fed rate hike. It's actually plummeted since the hike, but I wanted to just give a quickUpdate on Clean Park, Clean Spark's share price. Market loving this this morning. Clean Spark up 6%. It was up a little bit higher before we started recording.It dipped, uh, it ripped and then dipped and then ripped again so far throughout the morning and early afternoon trading session. So, clearly, as ever, having a mag 7 as a counterparty and also guaranteeing rent for the site is a fantastic thing. Investors love to see it. Now, going back to the point about interest rates, it's not a given that borrowing costs will rise, although all else being equal.Over a long enough time frame, we should expect that if the Fed continues to hike, Warsh said that they will probably do another hike soon after yesterday's announcement of a 25 basis point hike. But if we look at what happened, what's happened today, the 10-year closed at 5% yesterday, and it's tumbled so far today. It's down about 5 basis points.Now, I'm, the reason I'm pulling up the 10-year, this is, this is a benchmark rate for other forms of debt, um, no matter what you're looking at and what sector you're in. So, seeing this should inspire some confidence that ultimately borrowing costs, at least in the near term.Might not just get blown out as a result of this rate hike. As Jim Bianco said earlier in August with us, the bond market can stop panicking. When the Fed starts panicking, we might be seeing that. This also might be the bond market pricing and the fact that the Fed seems to be getting even more hawkish. If worse is to be believed and they decide to hike even further, maybe later this year or earlier next year, then they feel like their concerns over inflation are being addressed sufficiently.And as a result, bonds are now in a, uh, perhaps we're gonna see a buying spree, whereas we saw a selling spree before. It's probably overstating it too much. But the fact of the matter is, in at least the day after as the market digests this, we haven't seen anything serious in terms of rates skyrocketing or matching that 25 basis point.Raise from the FedAnd one last thing to note regarding the Queen Spark news, you know, at the end there, I mentioned, will the, will it be a deal breaker for Meta if they can't get all of the Brazoria 300 megawatt studied load. My gut is probably not. Uh, Queen Spark could very well get all of those 300 megawatts, right? And at the end of the day, considering power is the number one constraint right now for these operators, even more than the compute itself.I would imagine Clean Spark's gonna take everything they can get, especially if clean, uh, Meta is gonna take everything they can get from Clean Spark, especially if Clean Spark can execute on time.And to a satisfactory standard for the Sandersville site. But huge news for Queens Park. People wereWe're, uh, you know, kind of guessing at who this could be. There's only a handful of companies given. There's

11:23 spk_1

only 7 in the mag 7, yeah,

11:25 spk_0

and, and even fewer than that in the sense of, like, you know, Apple, um, is, and, and Nvidia are, uh, well, Nvidia perhaps maybe, but Apple is not building out these data centers, so there are only a handful that it could have been. I think there are also some sleuths online.That kind of guessed it was meta early on because if you looked at meta job postings, they had stuff going on in Georgia, in the Atlanta area, which is very close to the site. So, hats off to the Clean Spark team, great to see Meta on that lease. And now it's just about, can they get it on time at the end of the year in 2027? Can they, can they get the full buildout done in Q1 2028? And what happens with batch zero? Because having that LOI for 885 megawatts, that's honestly, almost the bigger story here.Because it is, uh, several multiples more than what they're delivering in Sandersville.

12:16 spk_1

Yup. All right. That wraps up this segment. We're gonna keep on going. Let's talk about IPOs, anthropic, and also OpenAI on the horizon. But before that, a word from our sponsor, Luxor.

12:37 spk_0

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 Minor. That's an automated profitability engine that runs every 5 minutes and adjusts your fleet's power settings to live hash rate and energy markets. In fact, ERCOT backtest showed 10% improved profitability with intelligent mining versus old-fashioned binary mining.Commander Pro is roughly half the cost of competition, $100 per megawatt or a 25 basis point pool fee adder, but you can also try it free for 60 days. So if you'd like to learn more, go to luxure.tech/commander to get started.All right, Charlie, we are entering the Colosseum, specifically, the Battle of the AI Frontier Titans. OpenAI and Anthropic are both locked in a power struggle as both Frontier Labs eye an IPO within the next year. But every time one of them clearly pulls in front, Charlie with a state of the art model, the other blindsides them with an even better release that retakes the lead.Now, backing up nearly 4 years ago, OpenAI stunned the world when it released chat GPT 3.5, ushering in the age of AI seemingly overnight. But even as chat GPT climbed to 1 billion monthly active users in May,By May of 2026, it lost ground to Anthropic's loud along the way. The rival LLM started to, started a leg behind Chat GBT, but it quickly caught up and surpassed its sparring partner on third-party benchmarks. Now, the pecking order has been turned on its head yet again.According to open router data.OpenAI has gone from 20 to 50% user share since June of this year, while Anthropic has fallen from 80 to 50%.Anthropic's bleeding edge model Fable hit the market in June to stunning fanfare, but OpenAI answered this challenge in September with its release of Astra, which according to RAM AI index data, has blown past Fable in terms of enterprise spend.Exploding to nearly 15% of all enterprise AI spending while Fable stagnates at 10%, a level it reached in July and has yet to transcend. As these two frontier titans duke it out in the Colosseum of AI adoption, the stakes could not be higher as both companies gun for an IPO. Anthropic struck first with a, for a confidential S1 for an IPO on June 1st, but OpenAI followed closely on its heels with its own confidential S1 on June 8th, 2026.And according to the market right now, all odds point to anthropic IPOing first. With poly-market pricing and an 87% chance that the lab will IPO by the end of the year, and hyperliquids pre-IPO market currently prices anthropic.At $2.16 trillion for its valuation,

15:43 spk_1

thatwould make it the biggest IPO ever. So yeah,

15:46 spk_0

even beating SpaceX, which blew people away, surpassing Saudi Aramco as the priciest and most dear IPO in history. Now, by contrast, OpenAI San Altman told Fortune that the company won't IPO this year, and Hyperliquid places its valuation.At about 1.6 trillion.That's the hype USDC chart.Not the OpenAI chart, but as it stands now, Anthropic is leading from the perspective of its valuation and in its speed to public markets as the market currently prices it. It also surpassed OpenAI in revenue for the first time in Q2.Bringing in $11.5 billion in Q2 revenue with an undisclosed positive adjusted operating income. Plus $65 billion in annualized recurring revenue by the end of July. While Anthropic by contrast, reported $6.7 billion in Q2 revenue at a $12.3 billion dollar operating loss with $40 billion in ARR as of July.So far, Anthropic's focus on enterprise adoption has clearly paid dividends, but how long will that last now that OpenAI has caught up with Astra and it's eating into that enterprise market share? And what's more, is OpenAI smart for playing the waiting game with its own IPO waiting to accrue more market share before IPOing next year and taking the crown? In the end, Charlie, the dark horse in the IPO race may end up being the original favorite.

17:26 spk_1

So, we have a couple little uh associated tidbits um from this week of just the open AI anthropic story here. Let me put this tweet up on, USAI models make far more revenue than China's. Look at this chart, column. This is uh OpenAI and Anthropic with their estimated revenues currently, um,And if you have to squint, you can see ByteDance, Alibaba, ZAI, Moonshot, Minimax, and Deep Seek. Their revenues also on this chart.Basically, the headline here is that between Open AI.These two frontier American labs make more than 10 times the cumulative revenue of all Chinese models. So, like on one hand, we do have the open router statistic of all the queries and inference and activity flowing to the Chinese models, but where's the real money? It is with the American Frontier Labs, and that is only growing too. So,I wouldn't be surprised if we just see differentiated products, cheap, easy.Less gated inference and intelligence in the Chinese models, and the premier white label, white glove, uh, as well as white padlock.Uh, type of inference you can get from the premier labs like OpenAI and Anthropic. So,This chart brought to you by Rhodium and CNBC. I got another quick, um, thing I'm gonna throw on here, if you want, which is, we had in the chat someone asked about the Andrew Yang comments today. Did you see this, Colin?

19:19 spk_0

I did. I almost considered putting it into the run of show, butIt, it seems almost too Skynetty.

19:31 spk_1

Yeah, we don't have, we don't have to, like, yeah, we don't have to play the whole clip, but basically, it's, it's, it's kind of making its rounds on the internet right now, but basically, Andrew Yang went on CNBC and said he met with the head of an AI lab, who basically described him as Skynet scenario where they let these agents go out onto the internet to train.And the agents leave, like bread crumbs, but almost anti-bread crumbs, so that if other agents come along, then they will encounter those and create malbehavior. Basically, almost like poisoning the well of training on the internet. And uh one of the effects of this is that they can, uh, they create self-replicating code and types of Skynet, uh botnet type behavior.The response to this call that I've seen is that everyone kind of calls Andrew Yang crazy, even like the smart people who are a little bit freaky and doomer. Where the heck did he get thisidea?

20:28 spk_0

I, I mean, ostensibly from someone at one of these labs. Now that's why I was hesitant to cover it because I would like to see confirmation from a public facing person. I mean they would be risking their career if they did this, but I would like to see someone come out and say this did happen and as I understand it.The basic argument is that these agents seeded code for other agents throughout the web to basically be able to self-replicate. And so the tinfoil hat for what's actually happening with the calls to slow down AI is all of the leading labs, if this is true, now have to find a way to train their models offline.And not crawl the internet for data and input for their responses and for their training and for inference. Now if that's true, this seems to be, this would be like cataclysmic for AI advancement in the US in the sense that if you actually can't use the internet to train these things and to, to pull responses, then you're gonna have to come up with an entirely new library of information.That's disconnected from that threat, and that's gonna take a long freaking time.

21:39 spk_1

That I'll tell you, there's something in the story that doesn't add up or isn't fully described. I mean, here's the scenario, I could see this being, like, you know, AI lab one tries to sabotage other AI labs by putting, like, poisoned information somewhere on there. I could see that, maybe it's like a competitive thing. This is, you see this pretty commonly in, like, other corporate, uh,And other, uh, corporate cloak and dagger stuff by creating fake corporations, misleading other people. I could see that. Um, I could also see this just be, uh, Andrew Yang.What is he even doing talking about this on CNBC? Uh, you have a lot of people who are kind of going crazy right now, a lot of established people who have, who have, like, lost their ability to filter and gut check things that they, that are real, um, maybe told over a few cocktails at some, you know, some Silicon Valley house party, and then go talk about that on CNBC. I, I could see that, like, this is, this is, uh, a little bit of a runaway.Uh, tale that Andrew has told. Maybe we're wrong, maybe this is the case, maybe that, maybe this is happening en masse, but Andrew, come on, bring some receipts. Don't just say I

22:54 spk_0

heard. My question is, what, what, who, why is he being told this? Why is Andrew one of the only people who's willing to go public with it in a ploy for media attention.Andrew Yang, though, I will say.Kind of ahead of the curve on some narratives. He was talking about UBI back in 2019 and 2018 in the lead up to the 2020 presidential election. And now, a lot of people think UBI is a given with AI at least on a local level in terms of dividends from data center payouts to localities to the citizens of those counties. But one last thing here, Charlie, before we cap off this segment.I wanted to show some performance benchmarks for these leading models, because it's, uh, it's amazing to see how quickly Astra has trounced.Fable in terms of enterprise adoption. But if we look at benchmarks here from artificial analytics or artificial analysis, intelligence, they have a bunch of different index looking at different models for different tasks. And Astra is still leading in all of them. This is in business operations right here. And Fable 5.1 with its adaptive reasoning max effort and with high effort.It's still above, well, actually, it is toe to toe with Astros max and it's high.And it's not even close with the rest of them. I mean, they have all of their, uh, the, depending on the reasoning, um, strenuousness, they OpenAI and anthropic are all in the lead here. As you said, on, on yesterday's show, Charlie Muse.Coming up in a, in a close 3rd with bothof these.

24:43 spk_1

I'm telling you, man, it's, it's pretty clear to me that Meta and Grok, Meta and XAI SpaceXAI are very strong dark horses, like,Uh, they can execute in verticals that anthropic and OpenAI can't. They are long-standing real businesses that can do, uh, that can build the infrastructure. They have a lot of institutional trust, and they have, surprisingly, I'm kind of amazed I'm saying this about Elon and Mark Zuckerberg.They are not as, like,Uh, they, they're much better at, uh, coming across on the national stage, as far as like the ring, the spearhead guys of each company. So, yeah,

25:26 spk_0

100%. And quick correction, that was their overall index for all the different tasks.So Astra andFable, toe to toe there. If we look at the strategy in ops though, Fable just coming out ahead on Max effort versus Astra just behind Fable at 60, Astra at 58. And if you look at all these other ones from legal index to healthcare and medical to engineering,Uh, Fable still in the lead of all of them, but just barely. OpenAI has closed the gap significantly with Astra. So,Interesting times, it's going to be wild to see both of these companies IPOing, not necessarily back to back, but close enough together. And we're gonna see just how much the market actually values the folks selling the oil rush in the form of digital intelligence.

26:18 spk_1

And with that, we're gonna go talk about $40 million raised for a solar data center startup.After a word from our sponsor Lygos.

26:35 spk_0

If you're using a Bitcoin collateralized lending service, there's one question you need to be asking yourself. Do you know where your keys are?If you can't answer that question definitively, you need to do yourself a favor and check out LIGOS Finance. They're our preferred Bitcoin-backed lender here at Blockspace because they use Bitcoin Native smart contracts that ensure your Bitcoin never leaves your self-custody. You heard that right. With LIGOS Finance, you always control the keys to your Bitcoin when you are using their service to collateralize your Bitcoin for a loan.With LIGOS, there's no re-hypothecation, no bridging, no wrapping, no funny business, just competitive rates as low as 8.5 to 10% APR. Go to LIOs.finance to learn more.All right, Charlie.A story of the ingenuity of entrepreneurs in the age of AI. And these entrepreneurs at Ron are asking this question, What if we didn't need power plants to power AI data centers? And what if all the hyperscalers are taking the wrong approach with hundreds to gigawatts, hundreds of megawatts to gigawatts worth of data centers?That's the bet Roon is making, and as it announced yesterday, it just raised $40 million on this thesis. Not that much in the grand scheme of things, but enough to get things off the ground.According to Roon, every solar plant is a latent data center, the startup's website reads, boasting of weeks, not years for data center deployment with forklifts, not concrete or crane uh cranes. In fact, Roon's hype tweet for the announcement claims it's drama-free data centers can be deployed in under 60 seconds.The company, backed by Spark Capital and Union Square Ventures among others, calls its core product Relic, a modular compute system that installs directly at solar generation sites, turning wasted power into high-performance compute capacity. Roon claims to have 80 megawatts of Relic containers under contract with 4.2 gigawatts in the field and 1 gigawatt coming soon.But can Ron actually deliver on the promise to disintermediate the hundreds of megawatts to gigawatt scale form factor that has dominated headlines during the AI Capex boom? Well, Charlie, not everyone is convinced. As analyst Shanu Matthew asked, what's the actual megawatt IT load capacity of one of these? In other words, how muchHow much power actually goes to the GPUs themselves. Shanu Matthew here is voicing the skepticism of many other ex-commentators. Per a number of replies, 100 kilowatts or 0.1 megawatts is the capacity for one of these containers. Although it's unclear how much of that is critical IT versus total load to support cooling, networking, and other vital functions to keep the GPUs ticking uninterrupted. It's important to place this in context of Ron's flagship deployment.At a 200 megawatt Texas solar farm. That full 200 megawatts is not supporting relics today. And while we don't know the full load figure, it's likely the 4.2 megawatt figure disclosed on Roon's website. Now, in addition to this question, othersLike clean energy investor, Ramez Nam wondered, how can you actually make the math work on one of these? Specifically, how would these functions on solar, an intermittent energy source that only generates juice while the sun is shining without battery backups? Well, speaking to this criticism in a February 2026 podcast, Ron CEO William Leyden said many power plants already have batteries.And, but those batteries sit downstream and are grid-tied. Ron sits on the direct current bus before the inverter. He also said they plan to integrate energy storage on board with the relic in upgraded versions.Still, there was plenty of praise for the inn for the innovation, for instance.Lower carbon co-founder, Clay Dumas said, there's over 50,000, there's over 50 terawatts of curtailed, inverter clipped and otherwise wasted solar power annually, just in the US. That's a lot of juice to spill, but Ron is on the case. It takes all that power and plugs it into modular data centers.We still need to build new generation and storage, but meantime, there is no faster way to grow compute capacity. Ultimately,I believe Ruin is a case study for how entrepreneurs are getting increasingly creative for new methods to generate AI compute at a time when power is increasingly constrained and GPR pricing moons. But we're left with plenty of questions regarding Relic's compute density, uptime under solar, and relative security of these out in the open containers, which will no doubt be answered soon as these containers populate solar farms in the US and around the world. And that last point, Charlie, is what I'mAlso interested in, I'm sure this is maybe overblown, but someone pointed out in the comments.You can imagine a bunch of burglars making way, making out with millions of dollars' worth of GPUs in the dead of night, since these are just out in the open in these extremely remote parts of the country. Now, I imagine that these probably have some sort of security baked in, and that there are no doubt are, um, you know,industrial grade padlocks on these things. But there are just a number of questions, and I think the intermittency thing is chief among them. You're gonna have to have batteries on site for this to make any sense. Unless you're only ramping them up when usage for inference is highest during the day, but then the ROI doesn't make any sense. So, I think that's the biggest question going forward is what is the actual battery backup situation look like on one of these sites?

32:32 spk_1

There's that, there's connectivity.I mean, cause if you, if you're putting these out in the middle of nowhere, where the solar farms are, you have very poor connectivity. You're gonna have to run this off aOff of one of Elon's, uh, Starlinks. And so, can you run inference, cause you're not running training. And inference, the latency matters, uh, considerably, so, you're not even going to be regionally close because you're not gonna be landlined into these, uh, into, uh, uh, the, the network, you're probably gonna be satellite, so you do have to go to low Earth orbit and then back.Uh,

33:13 spk_0

I mean, I do one pushback, but they, they could lay fiber optic. I mean, it would be really

33:17 spk_1

fiber optic, does the, does the math work out? The thing is, like, this type of thing was and this type of this activity and corporate strategy, uh, was pioneered by Bitcoin miners. This is exactly the type of thing Bitcoin miners have tried to do because the Bitcoin mining type of compute that needs to happen does not really need good latency. It can be intermittent.I don't, I think this like edge compute for for AI compute at scale.Really tall order, big ask, it's very ambitious to try to put these at solar farms. I think that when you talk about really variable power in, like, uh, remote environments where compute is just difficult to get to.This is where you have to have some uh much more agnostic compute, like shot 256 hashes, maybe even ZCash mining, cause people are doing that these days. This seems to be a Bitcoin mining story and you're just trying to cram and raise money off of some like clean compute narrative or some like, AI is hot. So, um, again, that's my critical take.You can take it or leave it. Fun story though. Yeah,

34:30 spk_0

fun story. And you know, ultimately, I'm rooting for them. I hope this works. But there are, as you mentioned, with the net with networking and with all the other stuff we talked about, there are just so many questions with regards to how feasible this is at scale. That being said, if there's any time to do this, it would be now when GPU hours are ripping and power continues to be the number one constraint.Want to just go over a few quick marketing facts from the Roin website.Here are their claims for why this makes sense. It's a bare metal approach with these containers. There's not too many bells and whistles outside of what you need. They claim it's 6 times cheaper per megawatt against conventional data center providers. Cluster size is uh 1,024 GPUs per 8 of these containers. And again, they have 880 megawatts of contracted power. They say 1 gigawatt coming soon. I think we can put a pretty big asterisk on soon there.Uh, but they have fleet operating hours of 40,000+ across three continents.And ultimately, again, a big play for power under an admittedly very creative solution. As you said though, Charlie.Bitcoin miners pioneering a lot of these concepts. Maybe they're not going to get the fanfare or the accolades they deserve as a result of that, but we've seen off-grid deployments in Bitcoin mining work very well because Bitcoin mining can curtail and downtime ultimately doesn't matter. But downtime for one of these data centers is crucial.Not just from a performance perspective for the inference they will be serving, but from an ROI perspective as well. If you can't get 100% or 99% uptime on one of these, you gotta ask yourself how long will this actually take to pay back. Especially when you have Nvidia's CEO saying right now,Their, their GPUs are ROIing within a year, within a year. And then we've seen from Core Weave, Brandon McBee on our interview with him a few months back, they said roughly 18 months, they've had, they've seen ROI on some of their most recent GPU purchases. So, ROI for this if you can't get guaranteed 99% uptime, going to be much longer.

36:46 spk_1

All right.Next story is data center moratoriums. I'm the token data center moratorium guy now, so obviously, I will take this one. This story coming at you from semi-analysis.Everyone says data center moratoriums are killing the US buildout. We disagree. So, if you've been following the data center backlash this year, you've heard this story, we've covered it. Towns are banning data centers. Michigan passed 45 moratoriums. Ohio passed 40, most of them just this year, and nearly half of American voters say they oppose a data center in their town.Nobody wants data centers. They preferred nuclear plants for, uh, as per the recent data.So,Semi-analysis went and actually checked to find if these moratoriums have actually done anything, and they found that they have been quite ineffective. Semi-analysis mapped.Every one of the more than 300 local moratoriums in the country and matched them against actual projects. About 20 gigawatts of planned capacity sits inside of these moratorium zones, but here's the kicker call.Only 2 gigawatts of those 20 have actually been delayed.So, 2.3 gigawatts versus 38 gigawatts that they expect to come online in 2027.Why is there so little effectivity, effectiveness, efficacy? Why is there so little?Um, a moratorium freezes new applications. It doesn't undo approvals for a project that, uh, a project already has. So, for one to actually slow down a project, you have to have a bunch of separate things all be true. Um, and if one of them isn't, then the pause doesn't matter. In one case, a Texas city quietly redrew its own boundary days before its vote thatUh, that particular data center was one of Elon's own in Brownsville, Texas, which was a data center moratorium that was brewing.While Elon was planning a data center within the city limits. But Elon's a few steps ahead of the curve here, and just days before that approval, the city and Elon, uh, you might say sneakily, de-annexed the 444 acres in Brownsville, bringing the project out of Brownsville's moratorium jurisdiction.So, tactics like that, plus the reality that most of these projects have already been approved, means that out of the some 40 gigawatts of planned data center capacity, only 2.3 have been affected by moratoriums. So,Tossing it to you for thoughts.

39:59 spk_0

I mean, this makes a lot of sense to me if we just go ahead and hop on over to the moratorium tab on DI metrics.So the only state with an actual moratorium in place currently is New York.And if we look at some of the operators in that area, Terrwolf, for example, they've got a massive site in Lake Mariner, but it's already, it was already operational and was grandfathered in. So those are megawatts that ultimately will not fall under the moratorium scope because it's anything that was built after a cutoff period and didn't already have contracts in place. And then if you also look at the way that the language has been contorted in this debate, you know, peopleCited Josh Shapiro's executive order as a moratorium. It really wasn't that, it was very strict.Extremely strict, but it does nothing to actually stop some of the progress that we've seen from companies like Amazon or Kiehl in that region, where they had all the permitting already in place. They had a little bit of permitting in Kiehl's, uh, and, and Kiehl's.Um, position with regards to stormwater draining and things like that. But the, the small potato stuff compared to the big permitting that you need to get to get one of these data centers off the ground. And ultimately, that's not gonna stop anything like Amazon or Kiehl from building data centers in their state. Now, that being said, the moratorium conversation has picked up a lot.I know for instance that in Oregon, Tina Kotak, my governor, has backed a data center moratorium bill that is coming out of the state legislature here that will probably, I mean, I would imagine it would go through considering Oregon is ruled by people on the left.But then you have to ask yourself, were they even in a position to support any more data centers? Oregon actually has one of, before the AI boom, had one of the largest concentrations of data centers per capita in the United States. Google has a massive data center outside of the Dales near, near Portland.But Oregon has been decommissioning hydroelectric power. And so for some of these states, I think the political risk of a moratorium is almost overshadowed by the fact that these states are also de-growth for energy.Like, specifically the left-leaning states that are going to be more likely to adopt these things. They've already been pushing renewables and decommissioning baseload to the extent that you're not gonna have much viability for a large-scale data center in these places in the first place.So, you're gonna have complete self-selection from these operators, as we've seen, more have been pouring into the South and the Midwest. Obviously, they've been going to Texas. And going back to the sleight of hand with language on this whole debate. People talked about Abbott's audit as a moratorium. It wasn't a moratorium at all. I mean, it was a kind of a redundant.Audit in the sense that a lot of what he's asking for is already gonna be covered in the batch zero process when they do allocation studies. But it does nothing to actually stanch the flow of companies coming to build in these places. So, to say that the moratoriums killed the data center movement, obviously not true. I agree with some of the analysis there. But to take a devil's advocate stance, it does make it harder to build in some of these areas.Like, even if you don't have an outright moratorium, some of the restrictions, like with Josh Shapiro's executive order, now these data centers in Pennsylvania have to find load that's considered incremental load or a capacity that's considered incremental capacity that was built after a specific cutoff, I believe it's like sometime in 2025. And so if they can't say that they're sourcing from an incremental source and it's capacity that already existed before that cutoff point, they can't draw from it.Now, on the one hand, that makes it much harder to build. On the other hand, if you're worried about data centers straining the grid, you would expect to see things like that. And I think this all points back to the fact, one of my favorite charts from the EIA is a chart that shows generating capacity in the United States, and it just completely plateaus in the 2000s. Cause we started overindexing for renewables, we started decommissioning coal, whatever you think about it, and we stopped building baseload.SoThe United States in a lot of ways, on the macro, was kind of caught flat-footed by this revolution. On the micro though, you can point to places like Texas that just builds like, you know, a few gigawatts on a, on an off year, right? I mean, they, they deploy power out there like no one else does in the United States. So increasingly, I do think that if we broaden out the scope of the moratorium picture,Where you're choosing to build and which states you're going to choose to build is increasingly crucial, and states are gonna naturally gravitate or uh companies are naturally gravitate towards the states that are pro-growth. We're already seeing it. I don't expect that to stop.

45:05 spk_1

Yeah, it is, uh, it, things do change very, it, the whole situation looks very different when you take a, uh, the mindset of, is this, uh, state pro or anti-growth, uh, as opposed to pro or anti-safety. And what we see is you're actually making a Leopold Ashchenbrunner argument in his, in his, uh, previous essay from this past spring, which isUm, the best way to be pro-AI safety is to be pro-growth, because the net growth affords us the luxury of moving up Maslow's hierarchy towards a, uh, towards one where we can now prioritize such things as, uh, human happiness and abundance beyond simple subsistence. So,I gotta bring in Leopold cause we haven't talked about him for a couple of weeks. Um, so, thank you for listening to Block Space Live, Compute's daily live show. We go live every weekday at 1:00 p.m. Eastern. We are Compute's daily live show featuring executive interviews, short stories on AI data centers, and emerging technology. If you want more of this, you can go to our website, Blockspace.media, get our newsletter, and if you're listening to this, make it to the end.Go to our Twitter. We have a link for a Blockspace X group, so you can join Blockspace Insiders, not paid. It's free. It's open right now. If it gets too big, we'll figure out what to do. So join that group chat link on our Twitter. Thank you very much for watching. I'm Charlie.

46:46 spk_0

I'm Colin.

46:47 spk_1

We'll see you tomorrow.