Transcript 0:00 on y'all. Welcome back to Blockspace Live, presented by CleanSpark. Charlie, it's your lucky day, brother. 0:05 It is an oil and gas day, my man, because Cipher said that it is going to build 2.5 gigawatts of pipeline capacity to its sites under construction to accelerate development from 2028 plus to energization in 2027. 0:21 Charlie and I are gonna dig into that story. 0:23 He's got some great background on the oil and gas side of things, and we're gonna have a little spirited debate on which sites are most likely to benefit from that pipeline build-out. 0:32 We're also going to touch on the LLM power outage that took place today. Seems like it's resolved. Charlie and I didn't have any issues, so we're wondering if it's you East Coasters that were most maligned by this. 0:43 Yeah, yeah, you, you, uh, fancy East Coasters over on the Atlantic side, get wrecked. You can't use your LLMs. You can't be productive. 0:50 We're here in the, in the mid- in the Middle of the United States and West Coast, we're still rolling. That's why the stream continued. 0:55 West Coast, best coast, Middle American- It all it did was delay the stream, like, five minutes- [laughs]... so worst case scenario. Yeah. West Coast, best coast, and Middle America majesty there, Charlie. Yeah. Mm-hmm. 1:04 So we're gonna touch on that really quickly before we get into today's news, and then after the Cipher story, we are going to touch on Iron tapping $2.4 billion from the latest financing round for GPUs at its facilities. 1:16 As we covered recently, they've got $25 to $30 billion worth of CapEx planned for the fiscal year 2027, so they're gonna need every drop liq- of liquidity they can get. 1:26 After that, we've got a little Twitter roulette we're gonna play with analysts on whether or not the bottleneck trade is over or it's just getting more difficult to find opportunities. That should be a fun one. 1:38 Then we will wrap up with a quick note on why Inscale could IPO as soon as September according to the information. Blockspace goes live every weekday at 1:00 PM Eastern. 1:48 We are compute's daily live show featuring quick hits on AI data centers. We interview the people making the moves, building the things that power the data centers in our world. 1:59 That's powered land, Neocloud's AI data center factories. If you missed the live stream, that's fine. 2:06 Becomes a podcast anywhere podcasts are found later that day, and you can go to our website to get the expanded universe of Blockspace content, a lot of written content. The website is blockspace.media. 2:18 If you aren't s- signed up for our newsletter, you absolutely have to get signed up. That's newsletter.blockspace.media, or just find it on our website every single day, newsletter straight to your inbox. 2:29 So if you were waiting with bated breath for 1:00 PM Eastern for us to go live and do that five-minute countdown, you found out that we were a few minutes late. You might be wondering, "What's going on?" 2:41 Well, it's actually, uh, unrelated production issues. 2:43 But you would have been right if you, uh, or you would've almost been right if you said, uh, it's probably the Claude Groq ChatGPT, uh, outage issue, which I guess we'll just riff on 'cause Colin, look at this. 2:58 Like, earlier today, um, Claude, Groq, and ChatGPT all down at the same time. 3:09 I think, um, uh, other Blockspace, uh, team member, Will- William Foxley couldn't, uh, turn his brain on this morning 'cause he's on the East Coast. 3:20 We have some suspicions about why you and I weren't hit by this trifecta of outages. Um, and that's because we're not plugged in to USA East One, somewhere in Virginia. 3:33 Yeah, just some context on that, and pretty incredible that the triumvirate of AI were kneeling this morning for most of the United States. 3:43 But from kind of digging into this, it looks like, like you said, Charlie, the, at least for the ChatGPT and Codex failure, an Azure East US networking blip occurred earlier today. 3:55 Azure had a confirmed, quote, "network infrastructure impacting event in East US," end quote, that lasted about 51 minutes. 4:02 OpenAI runs a bul- the bulk of ChatGPT and Codex on Azure, so that's why they were, that's, well, that's why they were down during this time. 4:13 Um, and then that timing and region match the OpenAI incident window and the spike in Azure related reports. Uh, Gemini staying up also kind of fits this. Google runs its own cloud, so it was unaffected. 4:24 So the, like, five Gemini users in the chat are saying, "See, I told you so," after this. 4:29 But what's interesting about this, Charlie, is Colossus also had problems, not at the same time, but around the same time, and so this affected, as you said, both Claude and Groq. 4:41 And I think a lot of folks when this happened were wondering if this would have, was affected by, uh, like, a single data center outage, and I think that we can safely say that that's not the case because n- these, these LLM or these, uh, yeah, these LLMs do not live all in the same data center. 5:01 And you can take it for what it's worth from Groq himself. He said, "Not the same data center. xAI runs Colossus clusters. Claude leases capacity from one of them while OpenAI mainly uses Azure and other providers. 5:12 Today's concurrent model outages are under investigation but appear separate, not one shared facility." It is a crazy coincidence. Kind of a crazy coincidence. It, yeah. 5:20 I don't know if there's something else going on under the substrate here, but... Yeah. Uh, again, yeah, uh, Groq's inference happens in, uh, Memphis at the Colossus locations. 5:31 But I'll also, I mean, I was kind of looking, um, it looks like there's some inference happening at Lan- in Atlanta for Groq and, uh, um, with, uh, Oracle, but it's not clear where that Oracle site is. 5:47 Um, so kind of, you know, I, I think sleuths could figure out where your inference is happening, kinda like it helps to know where your meat comes from, uh, so you can go shake the hand of the rancher- That prepped it. 6:04 Um, it'd be nice to know where your inference happens, uh, because that'll help you use the internet better. 6:10 Kind of like, are your search, are your Google, are your Google search, searches going all the way around the world to China and back? Um, no, those happen domestically. So similar to inference, it's like, well, 6:22 I get my ChatGPT queries and inference, uh, locally and regionally. It goes up the trunk, up the, up I-44 interstate across Missouri to Memphis, Tennessee, and then back to Tulsa, Oklahoma. 6:35 Uh, really technically down to Muskogee, then to, uh, Tulsa, Oklahoma. But, you know, helps to know. There's some, there's some guy who's built a small GPU cluster in his farm 6:47 who's, like, running to his wife right now saying, "See, see, this is why we have this spaghetti monster- [laughs]... of wires in our garage, so that we can avoid- Bro, I literally-... six formulas downtime." 6:58 I literally got... Bro, that was literally the response in more than one group chat this morning when I was like, "Hey, looks like all the L- LLMs are down." And immediately, two people, "Not mine." 7:08 [laughs] I was like, "Okay, you spent $40,000 on some RTX 6000s." [laughs] I mean... Okay, we gotta get on with the news, um, now, 'cause all that's over. So the LLMs are back online. 7:21 You can, you can turn your brain back off and start, stop having to hand write notes that you send via pigeon across the United States. So, um, the news this morning is as follows. Uh, Tyler Page, 7:35 CEO of Cipher, dropped this tweet this morning showing Mario jumping out of a green pipe. And what does that mean, Colin? 7:44 What do you think, uh, Mario jumping out of a green pipe could possibly mean for a data center company who might be, uh, power constrained, depending on the future of Texas interconnection issues? 7:55 What do you think it could be? I think it means that he's playing too much Super Smash Brothers Ultimate, or whichever one- That's right... that's from. 8:02 That's exactly- But yeah, Cipher building 2.5 gigawatts of pipeline capacity for their sites in Texas, Charlie. Yeah. That's an incredible figure. Yeah, here it is. 8:13 So the headline from BlockSpace, pulling from the Cipher PR on Global Newswire this morning. "Cipher Starts Gas Laterals to Support 2.5 Gigawatts of HPC Power Expansion." 8:29 The presser basically says, 8:35 "Cipher Digital, a leading developer, owner, and operator of industrial scale data centers, today announced that it has begun the development of lateral pipelines at multiple sites to deliver natural gas to support bring your own generation capacity." 8:49 There is not much in this press release. There's really not a whole lot. Cipher is up, it looks like 10 to 12% this morning. So up big, outperforming a lot of other PowerShell AI data center companies. 9:03 So, um, this is a very sparse announcement. 9:07 I think the real meat is in the conversation you and I are gonna have explaining a little bit about the world of gas pipelines, the implications, and then us doing some deduction and reasoning to figure out 9:22 some of the details that could be a part of this release. Which Cipher sites? How big? Who is the deal with? On what timelines? Okay, so let me preface by explaining a little bit about this. 9:37 So the press release says specifically they're building laterals. This is different than the large pipeline trunks that, that are built, um, across West Texas and certain parts of the United States. 9:51 These are three to four-foot diameter pipelines that are hundreds and hundreds of miles long, um, crossing giant, empty, you know, flyover, uh, states, um, and in particular, West Texas. 10:06 So this is a, this is a what's called a lateral, which is a smaller line off of this larger trunk. That's specifically what this language means. So this is not building the main line. 10:17 It's saying we're gonna tap into a larger, uh, a tap into a larger pipeline. And because Cipher's sites are largely in West Texas, this is kind of a no-brainer. There are pipelines everywhere, uh, that you could... 10:32 Pipelines and gas gathering facilities that you could pull from. And, uh, what's, like, the, the big number? So they, they announced 2.5 gigawatts across multiple pipelines. 10:44 That's multiple laterals, so not a single pipeline. So this is probably they're building multiple laterals off of these, these larger trunk lines to power more than one Cipher digital site. 10:56 Um, some just general math here. 2.5 gigawatts would, uh, roughly eq- eq- is rough- roughly equivalent to about 500 million cubic feet of gas per day, about half a billion cubic feet, half a BCF. 11:13 Um, that.5 BCF would be roughly 20% of one of these large pipelines. So it's a significant chunk of these, but it's not, like, the entire lion's share of these. 11:29 Um, however,.5 BCF for a lateral is very big for a lateral. You might see 100 to 200 million cubic feet per day on these 16 to 24-inch sty- size laterals. 11:42 So we're, you know, we're talking 24 to 36 inches of a pipe diameter, which we call a lateral pulling off a trunk. So we can see a few of these being built for Cipher digital sites. 11:53 On the gas side, um, how long does it take to build one of these? So it's, it's regulated by a different entity in Texas, where we assume this is probably gonna happen. This is the Railroad Commission of Texas. 12:07 Funny story about why the Railroad Commission of Texas governs upstream and midstream.... gas production in the state, but that's just how it is. 12:15 Um, so the permits for these, you gotta have, uh, egress and ingress, and you gotta get, um, some Army Corps of Engineers studies. 12:24 And then you have, like, physical, uh, components, like steel is the large one, and those have lead times. 12:30 Kinda might become like a tariff story, but again, these are typically shorter lead times than the real constraint, which is probably the power generation. So where are the gas turbines gonna come from? 12:43 If we just imagine that one of the Cipher sites is 10 to 20 miles away from one of these larger trunk- pipeline trunks, then you can expect the cost of building a lateral to run roughly 2 to $5 million per mile of lateral. 13:01 So this is a 50 to $100 million project if you're gonna do all of it in one go. 13:07 So that should show up on the balance sheet of Cipher, depending on who's paying for it, and that can affect some things, depending on whether the pipeline company who bills the trunk pays for it, whether Cipher does it, or whether it's some kind of hybrid arrangement. 13:22 Um, so the short version is obviously build your own pipeline. This is kind of a no-brainer. Cipher themselves will have to contract this out. So 13:32 this is the part where Colin and I can go back and forth on which sites, who is the deal probably with. 13:39 I personally think it's very likely that at at least one or multiple of these sites, Cipher has struck a deal with one of the larger pipeline companies in America, Energy Transfer. 13:53 And we can do this by looking at how Energy Transfer has positioned themselves, some of- one of their new pipelines that they're building across West Texas, and then, um, kind of looking at the sizing of the sites. 14:04 So that's my, my, my expose on this. Colin, feel free to chime in, or I can keep going and give my case for which sites I think might be candidates. Um, I just wanted to... 14:17 Well, uh, if I do- if I pull up this information, then I'm just gonna r- like, run into c- which sites that I think it is. But, um- Yeah... is that, uh, like- Go ahead. You go first. Yeah. Yeah. You, uh... 14:29 Explain to me how you got, uh, the math you did and, um- Okay. So-... where you're coming from... and, and one last note on the cost for this, um, you know, anywhere from, m- uh, 50 to 100. 14:41 I got it on the upper bound from, from Jibity, uh, 400 million. Cipher probably gonna sell the rest of its Bitcoin if it, uh, has to finance some of this out of pocket. Yeah, yeah. 14:51 It's like, it's- The generation, it's-... it's, it's difficult to like, to fully estimate the cost- Right, 'cause you don't know how-... because it's a-... far they have to go, right? Exactly. 14:58 But we do know for the generation itself, you're looking at somewhere between 3 to $4 billion for, for 2.5 gigawatts of, of nat gas generation, which is- Yeah... 15:07 I, and I also wonder, we'll be talking about bottlenecks here in a little bit. 15:12 H- how much has that pricing changed, considering a lot of companies are trying to do the same thing and build behind the meter generation so that they can get expedite sites? 15:21 But I wanted to start here, Charlie, by just showing a visual for one of the, um, pipelines in Texas, the Matador, coming out of the Permian. 15:32 Uh, you- this little, uh, what, what do we call this little, uh, offshoot here, Charlie? That's just... I mean- I don't see your, uh- Just, I, you can't see it, but- I'm assuming it's pr- that's probably a lateral. Yeah. 15:42 Okay. So do you have the- Well, no. No, it's... No, that's part of the main pipeline. Sorry, that's part of the main pipeline trunk. The laterals are not gonna really show up- Right... on this size scale map here. 15:49 So you have this vein just, uh, I say just [laughs] due east of Waha. It's, you know, um, it's hundreds of miles though, or about 200 miles. 15:59 And for context, Waha is, uh, kind of the linchpin for where a lot of the gas in the Permian goes, gets collected, processed, and gathered and distributed. 16:10 So because of that, a lot of gas has a actual s- regional pricing called Waha pricing, where natural gas trades at a specific, uh, price based upon that region. 16:22 Contrast that with NYMEX, which is the standard for, like, gas after it's been deli- uh, at kind of at the end of the line. 16:31 And then you have another regional pricing, which is Henry Hub, uh, for, uh, different oil and gas. 16:36 So Waha is typically, because of the constraints to get from Waha to other parts of the world, um, often trades at a discount. Sorry if I front ran that. No, no, that's great. All right. 16:48 I really appreciate that context 'cause I don't have any background in this. Not to be confused, Waha, with the convenience store chain in the upper Midwest, the inferior Buc-ee's, some might say. Yeah. 16:58 [laughs] So y- as you see, the, this little finger that comes up just east of Waha, sorry for those of you who are on audio, but go check out video if you want, if you're interested in this, comes just out of the panhandle. 17:10 And, um, if we look at s- where Cipher is developing currently, 17:16 this finger literally goes right between its sites in the furthest of West Texas and its, uh, its sites just, uh, due east of the ones that are in the furthest part of West Tex- Texas near Midland. 17:31 So it kinda splits these Midland and Abilene sites. And if we look at the Abilene sites, one of them we got Barbara Lake, which is under construction. That's f- that's, uh, 300 megawatts under construction for the AWS... 17:45 Or sorry, for the Fluid Stack and Google deal. Black Pearl is just to the west of that, uh, right here. It's, it's furthest west site. That's 300 megawatts for AWS. 17:58 And then if we look at what is currently near Barbara Lake in this Abilene, uh, San Angelo area, we have Colcus, which is a 1,000-megawatt project, which is currently in batch zero with grid work underw- underway and- Are you sure it's Colcus? 18:14 I think it might be Colchease or Colcheesy. [laughs] Probably. Yeah, I think that 'cause- That's actually, yeah, the name of, like, a town in the area. Okay. Or, like, an area of, like, geographical formation. So. 18:25 Charlie, uh, continuing reminding me that I know nothing about this, about Texas or Oklahoma nomenclature with regards to these cities. Thank you for the correction, though. So Colchis- Or it could be Alcus, whatever. 18:36 Okay. Whatever, man. Whatever. I'm gonna, I'm gonna call it Colchis. [laughs] Um, Colchis/Colchis, roughly one gigawatt with a target of 2028 energization. And again, this is currently in batch zero. We've covered this. 18:51 Um, and it's, the grid work is underway. Land has been secured. There's also the Makaska site, which is 500 megawatts. Land is secured, also in batch zero, targeting 2028 as well. 19:03 Now, Barber Lake has a 500-megawatt expansion that S- Cipher is targeting for 2030 plus. 19:10 I think that there's a strong case to be made that considering Col- Colchis/Colchis and Makaska are in the batch zero process and w- uh, may be affected by delays accordingly, 19:22 I think there's a strong argument to be made that these sites could be considered, uh, or might be some of the sites that will benefit from this power, especially since they're the ones that Cipher is targeting for the, uh, targeting for 2028 and 2029. 19:39 And part of this release, as Cipher noted, was to expedite projects in the pipeline to 2027 energization. 19:45 So if you're worried that you're not going to get enough power to get a deal done at these before 2028, it would stand to reason that you want to bring that power on site. 19:57 Barber Lake could also be, uh, in this, but that expansion is targeted for 2030 plus. I don't know if they've act- Th- like right now they're, um, they're focusing largely at Barber Lake on the first phase. 20:15 Well, let- Which is to be delivered by January 2027. The initial phase was delivered, um, uh, it w- should be delivered this month. 20:23 So- I mean, l- let me, let me tell you why I think Barber Lake is likely one of these options. So Barber Lake is near, uh, Energy Transfer, again, top three pipeline company in the US. 20:34 Um, they are, they are about to bring their Hugh Brinson pipeline up to, like, full commercial status. It's already kind of in operation, but the whole thing selling commercially to clients at scale. 20:45 Energy Transfer specifically, uh, has been marketing this pipeline to data centers. And so, uh, I would not be surprised if Barber Lake pulled, uh, from this pipeline when it comes online in 2027. 21:02 That's kind of my thesis, and especially because Energy Transfer, uh, has been pretty clear, like, saying that, "We want to do business with data center-type companies," like, and this pipeline is, is, is for that. 21:15 I, I-- This is, like, a very strong candidate to me. 21:18 So I put Barber Lake at, like, near the top of deals, and that's kinda what I was teasing with the episode is, like, I think Energy Transfer probably one of the companies involved here. Uh, so, um, you brought up Waha. 21:33 That's another one. That could be separate deals. It could be deals, again, with Energy Transfer or others 'cause there's a lot of gas gathering facilities, and that would allow you to then, um, 21:45 take advantage of the arbitrage opportunity and effectively, often, negative gas pricing regionally. And- Yeah... yeah. And that's like, that's like the play. That's kind of like the golden goose. 21:57 It's almost like, um, you know how, like, the Bitcoin miners, myself, really, really, like, worked to d- to say there's all this, there's abundant natural gas, and it's all across the United States. A lot of it's wasted. 22:08 Um, that is true. 22:11 The bigger fish and the much, uh, m- easier to ex-- Not easier to, just the much more obvious one is that there's re- negative regional pricing in natural gas at these, like, major collection points, Waha being an obvious one. 22:26 So Cipher's, what, three sites there? Um, that would be an incredible move if you're able to capture basically near zero or perhaps sometimes negative gas price. Now you can set up a whole gas trading desk. 22:43 Um- Yeah, so for the three sites in that area, one of them is the Odessa site, which is fully energized, and they only have Bitcoin mining in there right now. Um, they're probably hunting for a lease. 22:53 The other is Stingray, which is, um, quite small, and that's leased out to AWS. They're targeting rent in the first part of 2027. And the other is Black Pearl, one of their flagships. And they do have expansion, um... 23:11 Actually, no, they don't currently have expansion plans for Black Pearl. I wonder if, 23:16 I wonder if they will consider what you're, you know, if they do end up expanding that site, if they will decide to bring gas on sh- on, onto the site. Um, oh, sorry, I need to get this back up. 23:30 Just a few more notes on this, Charlie, and then we can wrap up. You know, I hear what you're saying about Barber Lake, but I think my reservation is the fact that they're still working on the first phase. 23:41 And so they already have a tenant for this site, and if they're not pl- if they're still working on getting, um, this whole 300 megawatts off the ground, um, they have 244-megawatt phase one, which should be hitting soon, and then a phase two in January 2027 for another 56 megawatts. 24:02 Um, rent's supposed to begin in October 2026. Uh, perhaps they will target that site to try to expedite the expansion for 500 megawatts. 24:12 But, um, one of the things they said in this release was that they plan to bring the gas on as a stopgap and still hope to lean on the Interconnection agreements that they have, and they do have, um, agreements for, uh, Colcus specifically, right? 24:33 So they, they, they do have, uh, power purchasing agreements in place for that site as it stands, and I think they probably, um, I haven't done my homework on McLennan or McKeska as well, but they-- I would assume they have something similar in place for those. 24:50 Uh, so all that being said, I think it comes down to the construction timeline for these, and these are ones that they have specifically targeted for expansion or construction within 2028 and 2029. 25:04 And if you're negotiating deals for these sites and they-- you're basically told by the tenant, "Hey, we can't wait that long," then you're gonna do everything you can to speed it up. 25:14 Whereas I think for Barber Lake, you already have the fluid stack relationship there. There's maybe less execution risk, and you can move a little s- more slowly. The last one I wanted to highlight, 25:26 'cause we haven't talked about it, is, uh, th- uh, these two, McLennan and Milsing. They're not-- They're in East Texas, so they're not close to the Permian obviously, 25:39 but they do-- they would theoretically have pipeline access if we look at the network here. Um, you know, they, they, they could s- they're still close to the Matterhorn route. 25:52 Um, there's also the PHP route that they're somewhat close to. Uh, I have no idea, you know, what the way stations are or what companies are operating in that region. But they could theoretically also tap it for those. 26:07 So I guess to conclude, it could be all of them. [laughs] I wouldn't be surprised if it's all of them. If I'm a pipeline company, uh, I'm-- This is just incredible. Williams, uh, H- uh, headquartered here in Tulsa, 26:22 a big deal with Meta in Ohio. Um, that's been kinda word on the street for a while, and they, uh, think they officially came out to say it, um, not too long ago. Um, also Tulsa, Oklahoma, baby, my hometown. 26:35 I'm broadcasting out of there right now. Home to, you know, a quarter of HQ'd from all, you know, a quarter of all these pipeline companies, uh, that are in the running. 26:45 And if you're not a, a pipeline company and you're not headquartered in Houston or Dallas, you are in Tulsa, or at least you have, uh, your second-largest office here in town. 26:53 So bullish on it, and I'm excited for all of my engineer friends who've had the most boring job ever of designing pigs for pipelines. Go Google that. Um, they can finally get a real job and power the data center economy. 27:07 So enough about that. We're gonna keep on- One-- Sorry, Charlie, one last thing. Um. Um, Colcus does have, uh, an agreement with AEP, or Cipher does have an agreement with AEP for Colcus. 27:16 They don't have anything for McKeska currently. Gotcha. So I just, I just wanted to, um, get that on the record. But okay. All right. Back to you, Charlie. 27:25 We're gonna roll onto Iron and $1.2 billion master financing and security agreement after a word from our sponsor, CleanSpark. [gentle music] We are CleanSpark, 27:36 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. 27:53 We are setting the standard for what's next. Learn more about the intersection of energy and Bitcoin at cleanspark.com. All right, Charlie. 28:09 We are moving on from one hot Bitcoin miner turned AI company to another. We've got Iron up next. Blue Owl leads $2.4 billion Iron GPU financing at 9%. 28:22 Now, this is somewhat of an old story, but we haven't covered it, so we're gonna go over the details here today. 28:30 Blue Owl Capital led a $2.4 billion senior secured financing for Iron supporting NVIDIA GPU purchases for its McKenzie campus in British Columbia, according to a Blue O- Owl Capital press release dated August 28th. 28:42 The financing is divided between a senior secured term loan worth $1.2 billion and a senior secured-- and senior secured notes totaling $1.2 billion. 28:52 Both portions carry 9% interest and mature two and a half years after funding. Bloomberg reported, citing a filing regarding the financing. Pacific Investment Management also participated. 9%, Charlie. That is much 29:09 richer than what Iron enjoyed with its Microsoft-backed financing for GPUs earlier this year. 29:17 The 9% for this is versus 6% for the blended rate that Iron got on $3.6 billion for GPU financing for its Microsoft deployment. Obviously, this makes sense. Microsoft is an investment-grade customer. 29:33 These are not for investment-grade customers, so they don't have the privilege of tapping that lower financing. But 9% still not eye-watering. 29:43 We-- You know, we've seen GPU financing deals in the double digits, so ultimately this isn't that, uh, you know, this isn't- We've s- we've seen a few-... super expensive... 29:52 didn't Core get, like, 5% or 6%, like, a few months ago? Um, Core Scientific, I don't think. Yeah. We, we covered Hut getting, like, 6%, but that- Oh, yes... 30:05 you know, um, the facilities tend to have cheaper rates at, at any rate as well because there's, there's less execution risk and depreciation risk with them versus GPU rollouts. 30:17 Um, but that was I think because they have NVIDIA on that Beacon Point lease, right? And I think, again, that just goes to show you how- 30:25 How stout having an investment grade counterparty is for one of these build-outs, because you can really access cheaper c- cheaper rates for your credit. Yeah. 30:35 A few notes on this, Charlie, before I toss it to you for color. Financing is available through December 31st, 2026. So as Eiron accepts each tranche of GPUs, they will submit a funding request and draw on the loan. 30:52 Each individual borrowing then has a 30-month maturity measured from its own funding date. Uh, and that's important because at a completely static $2.4 billion balance, 9% would equal 216 million of annual interest. 31:06 Actual interest expense should be lower because the $2.4 billion is drawn gradually and principal amortizes after funding. Blue Owl has first priority security interest in the finance GPUs and associated rights. 31:22 If things go south, the lenders can enforce against the actual computing equipment. Eiron itself provides an unconditional guarantee of the subsidiary's payment obligations. 31:32 It set up a subsidiary for this purchase or for this funding. There are two protections for Eiron, though. 31:38 It-- Eiron hasn't pledged its parent level assets as collateral, and the guarantee doesn't impose financial maintenance covenants on Eiron Lim- on Eiron Limited itself. 31:47 So this is GPU, GPU-secured subsidiary debt with a parent payment guarantee, but it's non-recourse. Um, but it's not non-, uh, ah, but it's non-recourse project debt to the actual parent. Um, 32:04 the financing sits with an additional 400 million as well. So within all of this, Eiron took on another 400 million at an unspecified rate to bolster its financing. 32:15 And this is for B300 Blackwell GPUs that it will use in air-cooled facilities in British Columbia and Texas. 32:23 If we look at how much CapEx that will roughly cost, we're talking $2.67 billion of underlying GPU and equipment CapEx. So they've gotten 90% of that with this 2.4 billion. 32:33 But as Eiron said in their most recent fiscal year earnings, customer contracts have included prepayments equal to 45% to 55% of GPU CapEx. 32:42 And as a result, Eiron says debt financing plus prepayments actually exceeds the cost of the GPUs, freeing some capital to help pay for data center construction and other CapEx. 32:53 So Eiron, once again, showing that they've really, really starting to get in a groove with a GPU financing flywheel here. Yeah. 33:01 And maybe just-- I was trying to wrap my head around, like, this 6% that Hut was able to get versus, like, a 9% that Eiron got here. 33:12 And it's basically, I guess that's just the cost of if you're not gonna do business with Microsoft on the other end, um, which is what? Of the $1.2 billion, it's something like $70 million, you know, difference per year. 33:28 Uh, again, it varies on how much they actually draw. So, um, financing, uh, it's al- it's, it's kinda like, you know, our peers 33:40 probably don't pay attention to, like, home interest rates and how much that d-dramatically affects, uh, the long-term, like, total cost of owning a home. 33:51 Similarly, I think myself and others probably don't pay-- have not paid enough, um, enough attention to the specific rates that these, uh, that these, uh, operators are getting to finance their operations. Anyway, 34:05 super interesting. We're gonna keep on going to the bottleneck trade. So what's bottlenecked? Is it over? We'll talk about this after a word from our sponsor, Luxor. 34:21 [firework explodes] This episode of Blockspace Live is brought to you by Luxor's Commander, Bitcoin miner management software for enterprise operations. 34:34 Luxor's Commander gives you real-time fleet monitoring, bulk remote commands across your fleet, and Intelligent Miner. 34:40 That's an automated profitability engine that runs every five minutes and adjusts your fleet's power settings to live hash rate and energy markets. 34:47 In fact, ERCOT backtests showed 10% improved profitability with intelligent mining versus old-fashioned binary mining. 34:54 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. 35:04 So if you'd like to learn more, go to luxor.tech/commander to get started. All right, Charlie, one of the analysts at Centrini Research thinks that the bottleneck trade is over. 35:19 And I think that if you were gonna grill him on this on air, he would basically tell you that the easy pickings are gone, and I think I would largely agree with that with some caveats. 35:28 Tweet here says, "I don't think there are any new bottlenecks left to discover at this point. I think we've already identified them all. 35:35 From here, the trade may simply become a rotation of capital between the bottlenecks we've already found." 35:41 Now, I think that that's actually a pretty reasonable claim because if you look at basically any stock that is as- that is associated with AI or that has business exposure to the sector, 35:55 their stock charts are up and to the right over the last year, over the last two or three years particularly. So in terms of finding new opportunities, I think that that is largely gone. 36:07 The question then becomes, 'cause when we look at supply chain considerations and logistics here in a second, there are a lot of constraints, even still, within all of these different businesses. 36:21 Everything from power to chips to memory to optics and down the line, there are real shortages that we could continue to see out many years. So then the question then becomes 36:35 How do you capitalize on the companies, specifically I'm thinking from the data center side, that already have their supply chain secured? This came up a lot in earnings calls, actually. 36:44 When these companies would talk about expansion, some of the more trenchant analysts would say, "Do you already have your supply chain secured? Is the equipment already secured? Do you have orders in the door? 36:55 Are you going to have to wait for the equipment?" So a few things to just highlight here in terms of pushing back against this. 37:03 This one analyst, Schultz Dugan of Schultz Research said, "Jucan posted this morning that the bottleneck trade is over, that there are no new bottlenecks left to discover, and that from here it becomes a rotation of capital between the ones already found. 37:16 The first part is right. The second part treats bottlenecks like an ore body you can mine out. They aren't. A bottleneck is whatever-- is wherever the queue forms and the queue moves every time you widen something." 37:28 And he specifically mentions here a couple of cases. One of these is Cerebrios, um, saying this is the clearest case. 37:38 Wafer scale means no eight-- um, no high bandwidth memory, no chip on wafer on substrate, no three nanometer. And the company says so explicitly because all three were supply limited. 37:51 It sidesteps the entire standards list and still has a bottleneck. Twenty-five point four billion of signed contracts against a full year revenue guide of eight hundred and ninety million dollars. 38:02 The binding input there is megawatts delivered. Concrete doesn't appear on a semiconductor map. Basically, his thesis here is this. 38:11 What changed isn't the supply of bottlenecks, it's the discovery window, and I think this is a pretty good point. In twenty twenty-four, you had quarters to position in chips on wafer, on substrate. 38:21 Now, a constraint appears in a guidance line on Tuesday night and is priced by Wednesday's open. That makes the trade harder. It doesn't make it over. Great example of this. We just covered Cipher. 38:32 You mentioned their stock's up like what, ten, twelve percent on today's news? Yep. 38:36 Because they just proved, or proved, they just signaled that they are trying to solve for the power bottleneck caused by the hiccups to batch zero, and they were rewarded for it immediately. 38:46 And that to me speaks to what he's saying here about the fact that back in twenty twenty-four and maybe parts of twenty twenty-five, if you identified, "Oh shit, 38:56 this is going to be impacted by the AI CapEx boom," you could get in early and you would be rewarded for it in a few months. Now the time for execution is, is minutes, if not hours, if not minutes. 39:08 They're gonna get numerous- It's, it's, it's just like, it's just like crypto metas, or it's really just like the sort of markets in general, which is markets will, uh, find a way. And I think I, I, I buy this thesis. 39:21 I don't-- I wouldn't-- I don't know if I would call the bottleneck trade being over. It's just matured. Um, it's just matured. It's more sophisticated. It's now a meta. Um, it becomes a little bit harder to execute on. 39:36 But given just the fact that everything is constrained, do you think the market has appro-- has like sufficiently priced those constraints? 39:46 You know, do you think that steel or these multi-layer ceramic capacitors are appropriately priced if we need a hundred x more of them? So, uh, maybe, maybe they are. But, uh, 40:01 to me, I think you can still execute on the bottleneck trade, so maybe the language should be a little softer. Yeah, and to build on that, I think partly probably what, um, 40:15 Jucan was saying is that, to your point, the easy, "Oh, this company makes this critical component. This is going to be in short supply. I'm going to buy it just on that fact alone," that meta is almost certainly gone. 40:30 Now the meta then becomes which operators and companies are actually sophisticated enough to find their ways around the bottleneck and capitalize on this opportunity while everyone else kind of flounders in an environment that is increasingly material constrained. 40:48 Because if you look at a lot of these stocks, fundamentals in some ways be damned for some of them. 40:53 It's just you create this thing, you create chips, you create memory, you create optic, you create-- you build transformers, you're a steel manufacturer down the line, et cetera, et cetera. Same with data centers. 41:05 Companies that don't even have good deal flow last year were getting bid up spectacularly because of the thought that all of this will be constrained, so they're in a position to benefit from it. 41:20 Now it's moved on to, like, Ci- Cipher's a great example. 41:23 Can you find ways around the bottlenecks to where you can actually get online before the competition and continue building your deal flow and then drive cash flow as a result? 41:32 Because I wanted to just point this out really quickly. 41:36 This is Serenity, who is everyone's favorite hated trader on, on AI Twitter saying, "We've got Light, M- MTSI, AAOI all pointing to extreme demand imbalances with lasers, TIA, DSP, et cetera from earnings. 41:54 Optical supply is still far behind the market demand. The expected AI optical supply chains are likely to continue for years." Uh, quotes I believe from the earnings call is, "The entire supply chain is out of stock. 42:06 The supply is still far behind the market demand. The shortage will continue to the next few years." 42:11 Another quote, "Whether it is PCB carrier le- board, laser chip, or even related packaging capacity, as long as the AI supply chain link, uh, link almost all are in short supply. 42:23 When demand visibly extends for years with the supply chain bottleneck, it's difficult not to see photonics echoing the memory super cycle entering twenty twenty-seven." 42:32 Now, I, I wanted to highlight this one specifically to kind of underscore the point that I was just making. Because let's, let's just look at Lumentum Holdings. That's the light ticker here on Google Finance. So, 42:51 looking up in the month, about eight, 8%, six months, 18%, year to date, 118%, year over year, 420%. So it's hard for me to look at that and think, you know, okay, maybe there is a super cycle. 43:09 It goes up another two or 3X. That could obviously happen, but this kind of speaks to, I think, what Jikuna was talking about. This is already up spectacularly year over year. It's already had a 420% increase. 43:24 So I don't really think that you can just say, okay, buy light because, um, you know, optics will be the next bottleneck in AI. 43:33 They clearly already are, and if you didn't get in last year, y- aping in now is probably not the best idea unless you actually know how to analyze what the company's doing, right? And I... 43:43 So I, I think I'd largely agree with the point that Jukhan is making, that specifically it's just not easy anymore, and ultimately, as you said, Charlie, the meta is evolving. Yep. So, 44:01 uh, get good, basically. That's what I hear. Get good from Citrini Research. Um, speaking of getting good, let's get good with our self-custody and lending from our sponsor, Ligos. 44:16 [static sound] This episode of Blockspace Live is brought to you by Ligos Finance. Bitcoin's doing pretty good, y'all, but you know what? 44:29 If you're using a Bitcoin-backed lender and you don't know where your Bitcoin is, doesn't matter where Bitcoin's price is, you're risking losing your bag if that lender plays funny business with your gains and with your Bitcoin. 44:43 You need to check out Ligos Finance. They are our preferred Bitcoin-backed lender here at Blockspace. They use Bitcoin-native smart contracts to make sure you always control the keys to your coin. You heard that right. 44:53 With Ligos Finance, you always have self-custody of the Bitcoin you are pledging as collateral for their Bitcoin-backed loans. 45:00 There's no bridging, no rehypothecation, no wrapping, no funny business, just competitive rates as low as 8.5 to 10% APR. Go to ligos.finance to learn more. All right, take us home, Colin. Enscale, IPO. 45:16 What's going on here? Yeah, this one will be pretty short. More just getting it on people's radar here, 'cause there's not, there's not an S1 to comb through yet, so we don't have really any good information. 45:26 But this is coming at y'all by way of the information from Blockspace. Enscale pitches $103 billion contracted revenue ahead of IPO. 45:36 Enscale says it has $103 billion in contracted revenue after securing a $45 billion compute agreement with Anthropic. 45:44 The Information reported Wednesday that the London-based neo cloud could pursue a US initial public offering as soon as September after showing investors the fresh revenue. 45:54 That total has more than doubled from 51 billion before the Anthropic agreement, helped by additional contracts with other AI native customers. 46:02 Enscale signed commitments have an average duration of 5.7 years, equivalent to about $18 billion annually across the portfolio, per the information. 46:12 Enscale estimated that second quarter revenue exceeded $100 million, up from 37 million in the first quarter. 46:19 The second quarter figure excludes its Anthropic contract, which covers capacity scheduled to begin operating in late 2027. 46:26 And Bloomberg previously reported that Enscale could raise as much as 3 billion through the offering. Goldman Sachs and JP Morgan are working on the potential transaction according to that report. 46:37 Looking at that Anthropic deal specifically, it covers six years for 460 megawatts at Enscale's Monarch campus in Mason County, West Virginia. Country roads take me home, baby. 46:48 The commitment averages 7.5 billion per year and calls for Nvidia's Vera Rubin system to s- enter service in stages beginning late next year. 46:59 According to Enscale, Monarch spans 22,250 acres and is planned to have as much as 1.35 gigawatts of AI compute capacity in phase one. 47:09 Anthropic's allocation equals about 34% of that figure, while Enscale has outlined a longer term power runway exceeding eight gigawatts at the site. 47:17 Which is crazy, because as we covered yesterday, Charlie, with regards to SB Energy, one of the principal members of the Stargate project, 47:29 that site in Ohio that OpenAI and Nvidia are counterparties on is 8 gigawatts of critical IT, 10 gigawatts gross IT, or 10 gigawatts gross power. That's the largest project in the world. 47:42 So Enscale saying here that they have ambitions to make this on the same scale. I'm not sure if that 8 gigawatt is critical IT or if it is gross megawatts or gigawatts. But that being said, 47:56 that's just a huge undertaking, and it's a lot to just say that. You know, don't, uh, don't write checks with your mouth that you can't cash, right? Quick analysis here really quickly with Enscale versus CoreWeave. 48:11 CoreWeave, [inhales] uh, s- reported that it has $104 billion backlog as of June 30th, alongside 2.58 billion of second quarter revenue and about 1.5 gigawatts of active power. 48:24 So that 103 billion that Enscale is saying it has contracted places it near CoreWeave's reported backlog. 48:32 That being said, seems like CoreWeave has more sites operational right now, so that their backlog is not only built on top of stellar revenues that they already have- But they are probably closer to getting some of these sites on- online than Enscale is. 48:50 So not too much more to say on that, just yet another IPO to be watching. Seems like the second half of the year and as we approach Q4 could be a pretty spicy one for IPOs. We have, we have SB Energy forthcoming. 49:06 Enscale could be coming this month, and then we also have Anthropic potentially going public before the year's out. Yeah. And while we were live, um, we had a couple interesting AI analysis I'm gonna spring on you. 49:20 One, Colin, GPT's Astra just dropped. It's the next one. So even though Claude's Anthropic, Anthropic's Claude Fable 5.1 came out earlier this week, fantastic model, absolutely killing it. 49:35 Uh, uh, OpenAI just dropped Astra. Um, maybe that's what took down all of the other models this morning. [laughs] Um, but then actually more, I think, uh, a little more relevant to the stuff we cover. Look at this. 49:49 Bernie Sanders just introduced a bill to permanently ban AI that exceeds human intelligence. Violators would face up to 20 years in prison. Um, so he wants to ban AI super intelligence. 50:02 You can't develop or deploy these things. Pause, they want to pause a- advanced AI development, create a new cabinet-level federal agency, and there's penalties, and then he would also wants to ban this around the world. 50:19 What are we do- what are we doing here? Honestly- So-... what are we doing? The, a, b- well, the thing is, like, even a lot of the tech a- aggressive tech futurists have said this is not a crazy option. 50:32 The, most of their responses are, "We can't stop it." So, uh, we have to then play real politic with, um, AI advancement and regulation. Um, yeah. 50:46 I, I would love to have an AI safety expert, someone that's neutral and understands this because, quite frankly, I'm not equipped to talk about this. B- I don't know what the risk factors are. 51:00 But they do seem to me, if I use what I call the oonga boonga test on them in terms of just intuition, 51:09 it reeks of Skynet fears, Matrix fears, things that are injected into people from the media sphere in terms of kind of catastrophizing the long tail of these things in a kind of fantastical way. 51:25 When the other side of me thinks, if you cap AI intelligence, which it sounds like some of the AI researchers are probably saying that's not even necessarily possible to do, 51:36 what does that do in terms of capping the potential of this technology to solve very real problems that are getting in the way of human flourishing, like energy abundance, curing incurable diseases, et cetera? 51:47 It seems like you're just neut- nerfing this technology in a way that the trade-offs of that outweigh the benefits of doing that. 51:58 And the trade-off here being that the, the AIs don't end up actually solving some of the problems that could really help us as we enter, uh, the next century and beyond, so. 52:06 Well, let me play the, let me play Bernie's advocate here. Um, I do think Skynet fears are pretty warranted. Um, 52:15 on what timeline and to what extent these things would damage the world around us, really difficult to evaluate. 52:22 But I do think everyone feels the gravity that we've, that we're creating things which are v- vastly more powerful than most people r- than, than, like we, we have the ability to contain. Um, 52:36 I, I would say to such a ban, it, the, the, it's really analogous to trying to say, um, we shall not develop nuclear weapons domestically. That may be a noble, uh, that may be a noble goal. However, the reality is 52:53 other people will develop nuclear weapons. 52:55 Ha- this has, this has to be an international cooperation thing, and it's really a game theory, uh, problem where it, the more you, you ban progress like this, the more asymmetric it is, the more asymmetrically benef- benefiting it is for people to then violate that type of progress. 53:18 Um, so I'd also say when in history have we been able to, uh, sufficiently, uh, when is, have we been able to sufficiently stop this type of thing? 53:29 Nuclear is pretty much the only example, and even that's fraught with challenges. 53:34 In fact, I would argue, um, besides, uh, energy security, uh, nuclear-empowered nations are the driver of most of the international conflict today. Every proxy war has nukes behind it. So, uh, 53:49 as, you know, as much as I understand, I, I actually really get this. Um, I think this is a classic example of someone who is more collectivist saying, "I recognize this problem. It's gonna create a lot of social ill. 54:01 How do we stop it? Regulation." 54:04 That's a lot of people's reaction, um, for good reason because regulation does stop things like corporations dumping toxic waste into rivers and, um, you know, destroy, and ruining, you know, the air and, and the environment of all types of things. 54:19 Um, but when it comes to things of natio- international and geopolitical signifi- significance, stuff that is, like, as significant as the Industrial Revolution, possibly more, how do you stop that? I don't know. 54:33 Yeah, and- Um... that's a really good point because, I mean, it's kind of, I know it's a boogeyman to some people, and it's kind of a tired thing, but China's not gonna have any of these safeguards at all. Oh, no. 54:42 Chi- China absolutely doesn't care. Yeah. They don't give a shit. 54:45 China will build their data centers and build nuclear and nukes at, uh, you know, nuclear-powered plants, and they'll just- Dislocate people, they'll imprison people, they'll work them to death, they'll put them in camps, uh, unapologetically. 54:58 And, uh, uh, yeah, because they have, they have people to spare. 55:03 Yeah, see, and that's part of the reason why I have a knee-jerk to this kind of stuff, because so much of this regulation, whether or not is, it ex- is explicitly intended to do this, has the downstream effect of ushering the United States into the role of a client state. 55:17 And, and it, and it w- and, and, like, the end effect is that it will accelerate that trend if we do these things. To use other analogies, the Nuclear Regulatory Commission 55:27 sets up insane red tape for getting a nuke plant off the ground, so much so that the regulation h- h- has ended in ballooning costs for some of these sites that are already really expensive. 55:39 And as a result, we just haven't really built as much. And then you add the environmental movement on that, which the c- counterpart here obviously is the anti-AI data center movement. 55:48 And a lot of these regulations, while maybe well-intended, end up stifling the industries 55:54 that they are regulating in such a way that we lose a very real competitive advantage on the world stage for building these things out. So I don't think this is gonna go anywhere, quite frankly. 56:06 Maybe there, uh, maybe I'm wrong and maybe there's enough steam, but I would also wonder what the legal challenges for this would be if it were passed. Number one, Trump would veto it. Well, yeah. Like outright. 56:17 First of all, you have to- Yeah... first of all, you have to figure out how smart the average human is, and I would say we passed that a long time ago. Yeah. The people out there, they're so dumb. 56:26 We'd have to go back to GPT-3. I mean, you ever played, you ever played Minecraft with some of these smooth brains? You ever been on Facebook lately? Man, we surpassed that long ago. 56:38 We'd have to backdate some of this stuff, so best of luck identifying... The average person has an IQ of, like, 10. Um, so- [laughs] We're gonna... [laughs] I mean... Okay, we gotta, we gotta wrap this up. 56:51 Thank you for listening to Block Space Live. If you want to see us in person, we'll be in Las Vegas at the end of the month at Yotta 2026. 57:01 Join 6,000 industry leaders in Las Vegas, September 28th through 30th, for three days of insights, network, and the partnerships driving the next era of AI infrastructure. 57:11 Use code BlockSpace20, Block Space and number 20, to save 20% on your pass. Colin and I will be there. Come buy us a drink. Maybe we'll buy you a drink. 57:25 Make sure to go to yotta-event.com to find more information. Again, use code BlockSpace20 for 20% off your ticket. Block Space goes live every single weekday at 1:00 PM Eastern. We are Compute.ai's daily live show. 57:40 This live show is brought to you by CleanSpark, NASDAQ-listed ticker CLSK. I'm Charlie. I'm Colin. And we'll see you tomorrow. [dramatic music]