← Blockspace
Blockspace
Nvidia and Friends Defend Open-Weight Models, Galaxy Prices $3.5B Note, TX’s New Rules Claim First Data Center Casualty
00:53:16
Loading summary
Transcript71 lines
- [00:00]
A
What's up, y'? All?
- [00:00]
B
Welcome back to Blockspace live, presented by CleanSpark.
- [00:04]
A
It is Friday.
- [00:05]
B
You can feel the pull of the weekend. But first we've got some news to chew through today. We're going to start off with our Hash Rate index update. And following that for our lead story, Nvidia Meta, Microsoft, a number of other companies sign an open letter in defense of open weight models. The debate about whether or not the frontier models are going to get regulatory protected might be over. Charlie. Following that, we've got a note on Galaxy Senior secured note. The pricing for that just came out today. So we're going to be going through the specifics after that. A data center is pulling out of Texas after it could not comply with new directives from its governor. Texas one of the best place for data centers. But the rules are changing and if you do not change with them, you will not build in the state. And to close today, we are going to be looking at some metrics from DI metrics, data center infrastructure metrics, rental prices, stock prices. Ooh, that one looks kind of ugly. Moratoriums and more at the end of the show. That's right.
- [01:10]
A
Blockspace goes live weekdays at 1pm Eastern, featuring quick hits on AI, data centers, markets and emerging technology. If you like the live stream but you miss it, you can catch it on anywhere.
- [01:24]
B
You get podcasts the day of the
- [01:26]
A
show shortly after the show wraps up. If you want more than just the live stream, head to our website at Blockspace Media. There you can find all of our written content, even more content that we produce alongside the live stream. Strongly recommend you head there and leave us a review. If you haven't already, give us five out of five stars on Spotify or however many stars Apple does. I don't use that platform. This show is brought to you by CleanSpark. Nasdaq listed ticker CLSK. More on CleanSpark later onto the show. So Colin Jensen Huang, the CEO of Nvidia, just tweeted for the first time and I've colored me surprised that on the like AI platform that you're supposed to talk about AI on. Jensen hadn't ever tweeted, but he did and his first tweet was a big joint letter. Really bucking a lot of the narrative over the past two or three months. We'll go to that after we go to Hash Rate Index Update. So I'll let you take this from here.
- [02:41]
B
Yeah, not too much going on. It seems like we're in this perpetual limbo. I guess that's kind of redundant but we're in a limbo. The metrics honestly have not changed very much since the last time we've done this update and that's kind of been the case for the last, I mean this whole year honestly Charlie, Besides some low points with regards to hash price, but hash price is sitting at $31.84 per PETA hash per day. This is your hash rate index update brought to you by Luxor's hash Rate index network. Hash rate sitting at 925. Recovered a little bit from our last update. Not too much though. This is following a negative 5% difficulty adjustment. The estimated difficulty adjustment with 91% of the current epoch cleared is just about half a percentage point, about 0.6% and there's not really much else going on. Charlie. We're in the middle of 4 CP season in Texas so we, we would expect hash rate to stagnate throughout the rest of the summer. In fact we've basically gotten that. Besides a few relief rallies. It's just a descending wedge on hash rate, man. If you look at where we are on the seven day average for the last year, it's just going straight down, my man. And we've said multiple times on this stream it wouldn't shock either of us if hash rate in fact contracts this year, which is an incredibly rare thing. I want to say that there may only be one precedent for this in Bitcoin mining's history. I could be wrong. There may be no precedent but I want to hedge myself there by saying this might have happened once before but I think that maybe we're at a zeta hash at year end but it's really not looking good.
- [04:30]
A
Yeah, I'll actually point out that we don't look at this. The Luxor's live hash rate forward curve. Pretty interesting here. Hash rate is trading in contango for next month in August. So you know what that means you could maybe act on that. Probably related to curtailment across the United
- [04:52]
B
States Contango for the coming month but backwardation for the rest of the year accordingly. Which what is that saying or what is it saying? For the 6 month out contract what are we at all the way at
- [05:05]
A
the six month out that brings it to September, December. So hash rate at 29.5 per PETA hash so down 7% from today, which
- [05:17]
B
frankly pricing in very little hash rate growth.
- [05:20]
A
Yeah, very little hash rate growth and very little price recovery.
- [05:24]
B
So we'll have to keep an eye on it. It's slim pickings out there. For the faithful, the remnant, those of us who still have ASICS online.
- [05:37]
A
Yeah, the few of you. But let's talk about AI Jensen, as I said, tweeted and I'll just read it verbatim. Quote for my first post, I'm sharing a letter Nvidia signed on why Open Models Matter this is a joint letter featuring Andreessen Horowitz, Box Meta Perplexity, Palantir, Mozilla, Mistral, AI that the redheaded stepchild of European AI, Dell, Microsoft, Hugging Face, Y Combinator and others. And what did the model, what did the letter say? Colin, because I'm going to tag you in here.
- [06:21]
B
There were really four pillars to it as far as I can really, as far as I can tell from reading the letter, and they're kind of the, they're the, you know, the talking points that you would expect to hear. So the first pillar was, you know, we need to have these open weight models for the benefits of business. To quote and this is in the third paragraph. Match the right model to the right job at the right cost, reserving frontier capability for genuine frontier problems and running efficient specialized models everywhere else. That discipline is what will make AI economically sustainable as its use scales into the billions of everyday task. America wins the AI era by diffusing it into the workflows of factories, hospitals, farms, classrooms and main street businesses. I think that's simple enough and it makes a lot of sense. Like doing do you need to be running Fable to ask how to cook your lamb shank?
- [07:24]
A
Probably how good of a lamb shake you want, Colin though.
- [07:27]
B
Yeah, I mean, hey man, I'm just saying Sonnet pretty good job even on low effort. So I that to me makes a lot of sense basically saying look, there are going to be different tasks that are better suited for different models. And for some of the open weight ones where the task isn't necessarily super intensive, you might just want to run one of those instead of burning a bunch of tokens on Anthropic or on or on a Claude or on Chat GPT. The second pillar of the argument is competition. This is something that we brought up on the stream recently when we were covering Kimmy K3. I mean, I was talking about the government's, you know, tentative and perhaps tepid response to Kimmy K3 where Axios had this article saying that maybe the government's going to take actions against it. And one of my big pushbacks was actually no, you don't want that. Because if you just ensconce the frontier models into a regulatory moat Are we going to get any meaningful improvements if it's just the two top dogs basically vying for the top spot rather than having this external pressure from other models and so on the competition front, the open letter says open weights create rivalry and spur innovation. They strengthen competition and competition. Open weights also strengthen competition. And competition is what keeps the gains of AI broadly shared rather than concentrated in a few hands. Sounds kind of communist, Charlie. No, I'm just kidding.
- [09:00]
A
Or sounds pro. Freedom.
- [09:02]
B
Yeah, I mean it's ultimately open market. That's why I joke because it's, it is saying, look, if you really want these frontier models to do amazing things and drive innovation, maybe some pressure from a foreign government, or rather, excuse me, foreign actors is what's necessary. And then the final two benefits are customer control. The idea that you want to make sure that organizations invest in AI. They want to know that they will not become locked into a single provider or lose the knowledge and capabilities they build over time. Open weight models help provide that assurance by allowing organizations to control their own data, evaluate and adapt models to their own needs and deploy them wherever their business requirements demand. Now this is something that you pointed out on the stream recently, Charlie. The idea of IP control and it's one that Alex Karp of Palantir had this tirade on. I was at cnbc, it was cnbc
- [10:07]
A
and this is why I brought, this is why I played it. I brought it up at the time because Palantir is a big, often, you know, boogeyman, especially among our generation as the rebellion against flock cameras and the like and Palantir and surveillance and state affiliated corporations building intelligence and weapons. People don't like that. But Alex went on CNBC on a rant as he's want to do and he basically made this case which is we can't be privatizing and gating and creating not just a walled garden but like a walled guarded, enforced surveilled garden just for these, these frontier models. Because otherwise do you, do you want Dario and Sam Altman to one of, one of them and their teams to win the, the race and become basically the arbiters of what is truth and reality. And so yeah.
- [11:05]
B
Oh, sorry, just to cut in here. And one of his like core arguments was they have businesses coming to Palantir that are saying, and I believe Palantir runs something like an aggregator where you can choose different models. Like they have a like institutional offering or you can plug into them and then you can choose between.
- [11:23]
A
It's like open router for corporations. Right, right.
- [11:26]
B
Thank you. I was struggling to remember open router. So he's making this argument that they have customers coming to them all the time and saying, look, we are going to be using this thing to plug into proprietary data and analyzing proprietary data and workflows within our organization that we don't want competitors to have access to. And what most people don't realize, it's similar to Facebook. You know, when people ask, well, you know, who's, who's the product or what's the product? Well, you're the product, your data. Ultimately that became the most valuable aspect of Facebook and Meta's universe is the fact that they have all this proprietary data and individualized data on all of its users that they could then sell, package into marketing things, et cetera. Same same thing here. If you're a business and you're using AI to flesh out a product or you're sharing proprietary information or intellectual property, the model is going to train on that later on. It sucks up everything that you're putting into it and it's used using that to iterate and to improve itself. So the idea here being if you have an open weight model that you can run in a closed environment on your own infrastructure, you don't have any risk of leakage in terms of IP and proprietary data. Now for a consumer, for an average person, that probably doesn't matter as much. You're not thinking about that, you know, you're not worried that, you know, whatever competitive edge you have for getting your, you know, for canning your tomatoes or pickling your cucumbers or your competitive edge for home renovation, like that's ultimately not going to really matter to you unless you're, you know, just like crazy neurodivergent or something. But for businesses, any of that getting out could be an existential threat to your business. And I think this is actually probably in terms of next to competition in terms of the clear benefits of open weight models. This to me for the business sector is probably the key selling point in keeping them open. And the last point that I won't spend too much time on is security. The letter says here, quote, to be sure, open weight, open weights carry real and distinct risks. Once released, the weights are beyond the original developers control and modified versions are difficult to trace or reverse. But the right response to this risk is not to prohibit open weights. In a world where cybersecurity attackers use advanced AI, defenders need access to models with comparable capabilities so they can detect, simulate and respond to emerging threats. Open models broaden defense capabilities, increase transparency and allow vulnerabilities to be discovered and remediated across many teams.
- [13:59]
A
Yeah, I have a number of thoughts. Let's see if I can elucidate them. One, it is notable among the signatories of this letter promoting open weights. Meta Perplexity Palantir. Let's see Microsoft. All of these companies are developing their own proprietary models they want to charge for. And so this is a little bit, this, this challenges the idea. You might call the fat. The FAT model thesis similar to this FAT protocol or FAT App thesis. Where does the value accrue to the model itself? Because in an open weight system the value may not really accrue as much to the, to the model itself. It certainly does need GPUs to run. So this is, you know, obviously Jensen Pro. This because you won't see him writing that we need, you won't see him signing a letter that we need diverse GPU manufacturers. But he will be one where he's like I want competition. All of you can compete against each other for my to use my gpu, my computers. But, but at the same time it, you know, meta signing this kind of interesting. Zuckerberg's been a little bit more open source than people give him credit for, especially as it comes to like AR VR if you know the whole Oculus thing and the developer platforms don't pay attention to his privacy violations and you know, the WhatsApp and Facebook, you know, pretend, you know, separation between the two companies. But, but like you, you see in Microsoft as well, like I think it's very notable that these companies are signing this and it has like because there's, it means there's disagreement among the industry of people who are building the models. Maybe it's them feeling like they're left behind that OpenAI and Anthropic have, have gone, have gone accelerated too far ahead and they won't be able to catch up. Maybe they're worried about those companies building some recursive self improvement or achieving that at some point and this is them trying to hamstring. I don't think it's that. I think it's that, that that people are realizing that if any single person wins totally it could and is and is and is protected by an entrenched bureaucracy and government. This could be very, very bad for everyone else. So this also does challenge the AI Capex cycle thesis because if Anthropic and OpenAI are not able to protect their models, are they worth trillions and trillions of dollars? That's kind of the story they've been telling everybody. What does this mean? This could have big implications. However, I'll say this. You know there is the alternative scenario where Anthropic and OpenAI down the road are not worth nearly as much. They're not these 30 to $50 trillion companies that they hope to be. Someone else is going to buy that compute. One thing is for certain, which is we need the compute. So whether or not the capex comes at year 10 in a 20 year least deal with Anthropic as a client on their end, maybe it's not Anthropic, maybe it's Meta or Microsoft buying that compute. So I don't think this jeopardizes the capex cycle thesis at a structure.
- [17:26]
B
Oh no, this is incredibly bullish for it, right?
- [17:29]
A
Very bullish. Yeah. So I was trying to build, I was trying to build a point to saying like you could try to argue that, but I think it's just let there be compute, let there be competition.
- [17:40]
B
And yeah, yeah, I mean per, we've talked about this on the stream, but like Per Jevons paradox, if Kimmy K3 and these open weight models actually do lower the cost of compute, then that's just good for everyone. That's good for consumers. It's good for the Neo clouds in the sense that now they can have more work put in for less, for less input. And going back to the signatories on this letter, you'll notice obviously like you said, none of the model designers are going to sign this. But of course the folks building the data centers and supplying the data centers with the heartbeat that keeps them running is going to do that. It's like imagine if overnight there was a new mining company that had machines that were 30% more efficient than caterpillars or something. Well, of course the company selling the raw materials, Nvidia and the company that owns and runs the mines, Meta, et cetera, are going to be all for that and they're going to want the competition. So there's nothing really that surprising in this, in the sense that of course these companies are going to come out in favor of it. There was the sentiment I got from the replies to Jensen's tweet was, well, that kind of settles the issue. And I think that's the thing to watch out for. And with regards to whether or not the government actually does take regulatory action and tries to ban these models, Jensen and all these other companies coming out in favor of them really does put some political weight behind the issue in favor of keeping them unmolested, right.
- [19:12]
A
So yeah, we had this tweet earlier. I haven't checked up with the market. It's been an hour and a half since this tweet, but since then Nebius White Fiber, Keel, Core Weave, Applied Digital and Iron all down. Now is that just because secular market correction or is this because those are the companies who may have deals with the private models on the other end? Who knows. But interesting to check to see that. Oh, a single letter again signed by the entire industry with the exception of OpenAI and anthropic, but a single tweet from Jensen does do this. I think we're going to go to the next data center company Top of mind, Galaxy Digital. I'll toss this to you Colin, after a word from our sponsor, Luxor.
- [20:07]
B
This episode is brought to you by Luxor's Commander Bitcoin miner management software for enterprise operations. Luxor Commander gives you real time fleet monitoring bulk remote commands across your fleet and Intelligent Miner that's an automated profitability engine that runs every five minutes and tests your fleet's power settings against live hash rated energy markets. Ercot back tests show 10 improved profitability with intelligent mining versus binary mining. Commander Pro is a hundred dollars per megawatt or a 25 basis point pool fee adder, but you can also try it for free for 60 days. So if you'd like to learn more, go to Luxury Tech Forward slash Commander to get started. Alrighty Charlie. So we covered this earlier this week, but we now finally have pricing for the Note. And this is Galaxy digital pricing. 3.507 billion in secured notes for its Texas AI data center. A lot to chew through with regards to the specifics of this note. I'll go through the key details really quickly before I get into some more of the specifics. So details of the Note are as follows. Again, 3.5 billion coupon is 9.875% fixed. Pretty high compared to some of the other Bitcoin miners that we've been covering. This is largely because Core Weave is the tenant for this Helios site. The note is for the second phase build of Helios. Core Weave is not investment grade. So unlike other miners that we've seen like Cipher and Terra Wolf that have investment grade backing. Galaxy unable to lock that lower interest rate. More on that later. Maturity is August 1, 2031 seniority senior secured project level financing. This will not float up to the actual capital stack of Galaxy. It is secured by the data center itself. Interest is semiannual. February 1st and August 1st. First payment, February 1st, 2027. This is the interesting part. Amortization is only 4% per annum of the original principal, semi annual, subject to adjustment. First payment at least 10 months after project completion. So interest will kick in before the principal begins to be paid down. It's priced as of July 23, 2026, expected to close July 28. And the amateurization piece is the thing that I really wanted to flag here. Because what's interesting about this is by the time the note matures, Charlie, based on the math that I ran with Claude, only 90% of this would be paid off. Now, the reason for this, as far as I can divine, is that ultimately Galaxy Digital just wants to get some financing through the door so we can start building this out. The rate's pretty high. It looks like they are banking on hopefully being able to refi at a lower rate in the future. Maybe Corey's credit profile improves, maybe their own credit profile improves and they'll be able to access better terms. But they're going to have to refinance this unless they just pay out, pay down the entire thing when it matures in 2031. So I thought that was pretty notable. It's not necessarily rare. Other notes that we've covered actually have similar bullets, as they call them, with regards to the fact that it'll have to be refinanced at a later date. But paying down only 10% is actually on the more aggressive end with regards to how little of the actual note will be paid down by the time that this is. Is. Has. Has matured. And that means 1.5 billion in interest. I against a note that retires roughly 420 million of principal over the five years that this takes place.
- [24:02]
A
That's wild. I had not caught that as I was skimming through this story. I mean, your point though, Colin, about maybe anticipating Core Weave being upgraded to investment grade at some point? That is kind of the assumption. It's not said out loud that I can tell that a lot of these deals are signed with. With tenants who like, are in who I. You know, a lot of folks anticipate them going public and then the investment grades being assigned to them at a higher rating. And I mean, look at like Core Weave is the. Is. Is in the platinum tier, the only one in the platinum tier of the semianalysis cloud provider rankings as of this past spring. So you. Yeah, hard for me to have much insight to the feasibility of that, but a lot of people are banking on it. And if you wanted to assume that, then this could be the move to make. Yeah.
- [25:03]
B
And a few more notes on this. I think part of the reason why not just Core Weave's investment profile, but part of the reason why the terms weren't the best is because this is technically a green field for phase two. So there's just more risk here. With some of the notes we've seen underwritten for current data centers for these bitcoin miners turned AI plays, they've already had the construction taken part of the way and then they financed the rest of it. And so for this one, it appears to me that the folks underwriting this loan are saying there's clearly more risk here and we want to see cash flows before we can get you down to a better rate. And that's the other part of the refinancing story here. Once this facility is up and running, they already have phase one running at Helios, Core Weaves operating in it. Once phase two is up and running and they have revenue coming through, it'll be easier to get this rate down. So I think on the face, when you look at how much interest they're going to be paying and how little principle and the fact that the note's still going to have 90% roughly left over by the time that it matures, it seems bad on the face. But ultimately this seems to be a clear indication that Galaxy is going to hold out for better financing once this thing's actually online. So this is almost like a stopgap where you can get enough in the door to get it built and then to pay down the rest of that note, you're going to refinance it at a later date. So something to keep an eye on. We'll definitely have an update sometime next week once this note closes with more information.
- [26:30]
A
All right, let's go. Let's stay in Texas for our next segment on Governor Greg Abbott. Before we do that, a word from our sponsor, CleanSpark.
- [26:44]
B
We are CleanSpark, America's Bitcoin miner, a publicly traded company with the largest operating hash rate powered entirely by self operated infrastructure across four states. This is our proof of work. We are setting the standard for what's next. Learn more about the intersection of energy and bitcoin@cleanspark.com
- [27:15]
A
all right, so here's a doozy in Texas, the most data center friendly state. The governor of Texas just posted a memo celebrating the ending of a data center project in East Texas by one Diode Ventures. Greg Abbott posted this on Facebook yesterday. Proposed East Texas data center ends project admitting they fell Short of the standards I have set, data centers must pay their own way, bring their own power, reuse their own water, reduce electricity costs for Texans and protect Texas neighborhoods. Texans come first. Greg Abbott is up for reelection. Is this a. A pragmatic decision or is this him vying for social for. For positive opinion? Let's dig into it. So Diode Ventures notified Henderson county officials Wednesday that his proposed data center near Cedar Creek Lake did not meet Governor Abbott's executive directives and withdrew the proposal. Diode is a Kansas based developer and it was going to be a 12 year project to build consume 5 million gallons of lake water per day by year five. And here's the kicker. Would have sat in the middle of a developed subdivision and pulled from the Cedar Creek Lake which supplies water to the Fort Worth area which is already kind of low on water. Again, a lot of this centers around water. There was huge opposition to this recently. 2,500 signatures demanding cancellation and entirely volunteer led coalition that doesn't seem, we don't see any indication of massive say Chinese propaganda getting their fingers in this. This could have been just kind of like a not very well thought out data center. Because Greg Abbott's specific requirements are. That. Were made clear on June 10 and a June 10 directed to the Public Utilities Commission and ERCOT were expectations that data centers ensure data centers interconnections will result in reduced residential electrical bills. Require data centers to pay for their electrical infrastructure costs and that no residential ratepayer is burned by the costs. Ensure that data centers add to Texas's electrical capacity, not just its demand. Required that they be water efficient like closed looped and not drain local water resources and to reduce impact on local communities, noise setbacks, et cetera. So my take on just looking at the high level items of this deal here is that this data center might have been a little presumptuous, a little bit over their skis on where it was, how it was set up because like 12 years to build this out. Yeah.
- [30:39]
B
That's crazy.
- [30:41]
A
I think this may have been. I don't want to diminish them too much but no, nothing else. These, this, this sounds kind of like a wacky deal that was put together because it doesn't align with the rest of what we're seeing from the. Yeah. What is serious hyperscalers in the industry.
- [30:56]
B
What are you doing? Are you, are you building the freaking steel refining plant to build the, the beams and the infrastructure? I don't understand. Are you?
- [31:05]
A
I don't know. It's just, it's Just indicative of American industrial decline. I'm just kidding.
- [31:10]
B
I mean, it's indicative of something. Maybe they're trying to crack alchemy so they can, you know, turn rocks into steel and silver for this build or something. Because 12 years makes no sense.
- [31:20]
A
Also, that's three presidents from now. Yeah.
- [31:24]
B
By that time, who knows, it could be a communist.
- [31:26]
A
It'll be. I mean, the Zoomers can be presidents by then, so.
- [31:31]
B
You know the other thing too, Charlie, you probably. I think you touched on it. But 5 million gallons of water per day by the time that it's completed. That's crazy. I mean, so they're using evaporative cooling. That seems wrong to me though. That doesn't converge on any of the industry best standards. I don't know. Is this a traditional data center? Is this an actual HPC data center? The whole thing obviously optically looks really bad. And maybe to your original point, is this pragmatic or a political move? Porcino los dos.
- [32:04]
A
Yeah.
- [32:05]
B
Ultimately everyone needs a sacrificial lamb every now and then to show that you're actually doing something to quell the masses fears. You know, the HOI polo is clamoring about data centers sucking up all the water and driving up their energy costs.
- [32:18]
A
Yeah, maybe this one was going to do that.
- [32:21]
B
You know what's funny is this.
- [32:23]
A
Right.
- [32:23]
B
You know, it would be.5 million gallons per day is quite a lot, especially for a regionally parched area in Texas. And you know what I think is kind of hilarious about this is kind of a slap in the face of either political direction. If you're one of the like, Texas is pro business and is always building things, then this seems to negate that. Although I think that the objections are very understandable and they're not meeting the new directives that PUC and ERCOT are putting out in order to protect ratepayers. So clearly Abbott is showing that what he said about amending the rules is sticky and that they have these new directives have teeth. Right. And it's also a slap in direction on the other way, where a lot of people on the left like to point to Texas as being this like hell state of no regulations. But clearly that's not the case. In fact, I think with regards to the data center boom, you'd be really hard pressed to find a state that's doing things better than Texas. Not just with regards to furnishing new capacity, but being progressive about setting up safeguards and new regulations to make sure that the AI boom does not negatively impact citizens.
- [33:37]
A
The State. Yeah, a couple more angles. So on diametrics, there is a, a political risk assignment to different counties across the United States. Political rest, political risk for like data center development, moratoriums like etc. Are they business friendly in general? This county, Henderson county scored a zero currently as of the last updated run. And they just, they basically just through political organization got this thing killed. So what is. Yeah, so what are you baking in the risk when like what is the risk you're baking in when towards execution? Because like again, the whole narrative this past spring and this summer has been you can get your transformers, you can actually get your megawatts. You just have to wait. Are you going to get your permits? And that's the big question. I think this is also relevant kind of on the back of the ERCOT Batch 0 and the new ERCOT changes. So in that June 10 letter that Governor Abbott wrote to Public Utility Council Chairman Gleason and ERCOT CEO order the PUC to make data centers fully fund the electrical service serving them so costs don't hit residential ratepayers. And it required the joint PUC ercot memo by July 17, which we talked about here. And it also includes a hard deadline for the PUC to initiate action reducing residential transmission costs by July 31st. So next week. And so the PUC with the new batch interconnection process approved on June 18, which has by the way, some 438 gigawatts or 90, almost like 90% of all data centers. The batch zero became effective July 11 and the utility submission package is due to ercot by the 24th. So today the Batch 0 applicants find out next month if they're in and the final transmission plan comes in fall of 2027 also, that's when batch one opens. If you're not familiar, kind of these things are approved in batches. So this is why we all wait with bated breath to find out who is in the batch, who gets their stuff approved. So notably Colin, Batch 0 comes around earnings week. So we will be see, we could be seeing some interesting fireworks as some make or break moments, the badge 0 drops. And does it align with these quarterly earnings estimates? Does it, does someone get rugged or whatever. So, you know, interesting to keep your eyes on that. That's all I have for this, Colin, unless you have.
- [36:48]
B
Yeah, I'll just piggyback on your point about regional risk. I think that the moratoriums are one thing, obviously, and if you are trying to build an estate with one of those. And so far New York's the only one that's passed one on the state level. I believe usually most of these are regional, local, within the states. You're clearly not going to get anything done. But what this is showing is the risk extends out to new regulations, right, and new directives and new permitting for these things. And if you're not staying up to date with that, then you're going to fall behind. You know, I would read this less as Texas is becoming hostile towards business and more Texas is putting the guardrails up. That should be there. Especially I think the most important part of the PUC's new rules. The idea that data centers are going to have to front transmission costs and also grid upgrades and maintenance specifically that is huge. And I think that will go a long way towards quelling some of the fears and proving that, you know, you can have your cake and eat it too with this huge capex boom as long as you devise the proper regulations. I'm sure a lot of the companies don't like it, but at the risk of sounding like a socialist or something, what's another few hundred million dollars here, guys, when you're pouring hundreds of billions into this? Obviously it's going to increase the cost of building these data centers, but I don't really know if there's a better way to go about it. The fact of the matter is the data centers can't really, they can't really push back against this. They need power. That's the bottleneck. And so if it's a seller's market for energy, you're going to have to do what the utility needs you to do in terms of funding that transmission and funding the maintenance if you want to get in the door.
- [38:37]
A
So yeah, I view this as Texas wisening up without diving too much into this specific deal. Sounds to me like they made the right decision. So you know, chalk one up for Charlie being pro the anti data center. Folks, let it, let it be known that I actually, you know, my first, my first take is that a kind of a kind of get this one
- [39:01]
B
think that you've been a double agent this entire time.
- [39:03]
A
I'm a triple agent.
- [39:05]
B
I don't even know my Castro hat to wear on the stream.
- [39:09]
A
Yeah, exactly. Hey look, our, our rivers would still be dirty without the epa. That's all I'm saying. So before we break up this show because the co host disagree, we're gonna. We go to the diametrics section to cap off this episode. We are going to hear from our sponsor Lygos.
- [39:40]
B
The stock market is Crashing. But miraculously, Bitcoin is holding up pretty well. But that doesn't matter. If the lending desk you're using is playing funny games with your coins and rehypothecating them. Do yourself a favor. If you don't want to become the next Celsius or Blockfi victim, but you want to use a Bitcoin lending provider, check out Lygos Finance. Lygos Finance is our preferred Bitcoin backed lender here at blockspace. They use 100% bitcoin native smart contracts to make sure that your Bitcoin never leaves your possession. You heard that right? With Lygos, when you're using their services and you're taking out a loan, you always have custody of your Bitcoin and no one else does. With Lygos Finance, there's no wrapping, no bridging, no rehypothecation, no funny business. Just competitive rates as low as 8.5 to 10% APR. Go to Lygos Finance to learn more. All right, Charlie, a little Dimetrics data center infrastructure metrics corner here to end the show. And I wanted to just pull up some data points that I think illustrate some of the things that we've been talking about today or on the show throughout the week or rather kind of speak to some of the trends that we're seeing out in the market. Specifically, people are worried about rental rates, Charlie, with Kimmy coming out with with Meta saying that they are going to kind of have their own Neo Cloud play. Neo Clouds even need to exist anymore. Well, the fact of the matter is rental rates are not coming down, my guy. And B200 rental rates if you look at the three month are absolutely blowing up. See, three months ago they were at about $5.13 per hour. Now they're up to $7.35 per hour. And even the older Nvidia models are also seeing some gains. H2 hundreds up on the last three months. Three months from 3.8 or $3.81 per hour to $4.35 per hour. H100 up from $2.47 to $2.86. And even A1 hundreds. The S9s of the GPU market have gone from. Well, actually, I believe I can't read this. This is not very granular. Yeah, $0.86 to $0.94 per hour. The appetite for compute has not shown any signs of slowing despite what the market is saying.
- [42:16]
A
So I have some other angles on this. Colin, I'm glad you brought GPU rental rates up because these are the contracted rental rates for the GPUs. Let's contrast this with the contracted offtakes like the long term agreements like the corporate four weave meta the iron Microsoft deals using the Diametrics MCP server which we can't like show on here because it doesn't produce a dashboard on the site. I kind of, I use Fable to kind of scope what are the actual values of the offtake agreements and how do they differentiate from the like the rental rates and what's interesting is and maybe this is a way, maybe everybody else knew this but me the there's actually kind of a convergence around basically $2 per GPU hour for especially for the BT B200s across the board. Look at that and that contract that's
- [43:18]
B
the B2 hundreds that's specifically for contracted
- [43:21]
A
set terms over like five attracted set terms over time. So we have like the secondary market of rental renting these and then we have like the long term contract market which is interestingly are all kind of around like the low two hundreds and these are inferred prices based upon diametrics like inferred contracted rates. So again these are like not fully standardized. So we kind of have to do some inference if you will. But if you were to look at the. Let's just take a look at three OpenAI agreements totaled roughly 22.4 billion all in those agreements would imply a price of 2.16 to 2.17 for GPU hour iron Microsoft that 9.7 billion dollar deal 200 megawatts at $2.10 iron Nvidia which is 3.4 billion air cooled Blackwells at 1.94. So like it's interesting that all you know these rental rates are kind of all in lockstep despite them being signed with different service providers. You have a lot of different parties and different terms, different locations too. There is like an actual long term contracted revenue model. I'm curious what drives that. If you gave me a little more time talking to some finance bros and other folks probably figure that out but kind of little interesting tidbit that I was able to pull.
- [44:59]
B
Yeah and it makes sense to me with regards to if you think about locking in revenue certainty you're going to have lower rates than on demand rental rates for that computer. It reminds me of in some cases to use a bitcoin mining analogy here. If you look at forward rates for Luxor's forwards if you're locking in six months you're going to get. I mean if you play it right you might actually end up getting more than spot hash price. But more often than not, you take a haircut because you're getting guaranteed revenue for the however many 1, 2, 3, 4, 5, 6 months of the contracted term. And to me, it makes sense that you'd see something similar here. Are you smoothing out any sort of volatility you have the revenue certainty. And because it's such a large purchase, like anything buying in bulk, it's going to be a little bit less than what you're seeing for spot prices at any given time. And maybe it tells a different story too, for the marketplaces. If you're short compute and you need some, the demand is so large that you're going to be paying a pretty hefty premium for that on demand computation rather than if you had a contract, you know, drawn out over three, four, five years. That was good context, Charlie. And just for reference there, the MCP that Charlie's referencing is the Diametrics mcp. So if you get a diametric subscription, you can use that MCP to pull all of the Diametrics database data into your models directly. So if you want to pour through Diametrics, that's great. There's a lot of data sets to go through, but you can also just pull from all the data that it's showing on the dashboard locally within your LLM model of choice. Really great feature. It's really leveled up our research for the stream.
- [46:43]
A
Yeah, you can get a link to Dimetrix in this video and stream description. Also, we are proud to partner with Diametrics. And just like Diametrics and Nvidia and Jensen Huang, a little bit of an open question how to pronounce this. So let us know what you think.
- [47:04]
B
Yeah. Comment below with whether or not you think DI metrics or DIME metrics makes the most sense.
- [47:09]
A
Exactly. Perhaps die metrics and we could, we could pivot it on a. It's. It's in reference to the chip, die for silicon or whatever. I don't know, Colin. I think that's all the insights I have that I can bring to the table. Unless you've got some other.
- [47:25]
B
I just wanted to pull up one thing here for the moratorium map. The moratorium map has this great time lapse when the data, the data collection starts in January 2023. There were three moratoriums across the entirety of the U.S. play that tape through to today, Charlie. We have 346 and the vast majority of those came this year. So we're. If we go into 2025, you get a few Popping up. But as we leave the year, we're still under a hundred, then they just explode. This year we added like 200 plus so far this year as data center pushback and the CapEx cycle accelerate. So I thought that was kind of interesting. I was going to maybe pull up, I'll just do this just, just to show how many different features there are here. I want to just show really quickly some of the changes in the, the PowerShell landscape and the Neo cloud landscape with regards to pricing because the market has taken a hit, it's bounced a little bit. But if we look at the monthly here for Neo clouds, they're down 17.2%. We have a little bit of a relief rally this week, but as you mentioned Charlie, earlier in the stream, they're starting to dip down again. There's, they're starting to taper. They have been as we've been recording and the powered shells quite frankly are not doing that much better. If we look at the monthly here, they're down roughly 14% and last I checked, the hyperscalers in the index on DI metrics were about, I think 9% down on the month. So are the bears finally. Right, stay tuned, we'll know soon. I do think that earnings season will be very telling because I mean, Google just came out with 82%. What was it? 82% revenue growth for their cloud business in one quarter. It was something insane, it was huge.
- [49:45]
A
But. And yet they're still, and yet they're still not. And yet they're, they're, their cash, their cash flow or has gone negative.
- [49:56]
B
Yeah. For the first time in the company's history. So their cash flow went negative. They had an 82% increase to cloud revenue and their stock is down 9% in the last five days.
- [50:08]
A
Well, on top of, I mean, on top of that, like, I don't know if you've been watching, but their, their latest model, Gemini 3.5 flash, is very underwhelming. All the benchmarks put it like very low. It's the lowest of the top tier models and it's wild because they poured so much into it and I don't really see them hiring any of the superstars of the AI world. It makes me wonder, like, is Google, is this the beginning of the end of the, of Google or is this like the decline for them? The company which has defined my entire life in technology is, I wonder if
- [50:50]
B
you're starting to see that reflecting in the stock price or at least the fear of that. Right. And that cloud revenue just quick correction, 82% quarter over quarter would be nuts. That was year over year. Still super strong growth there. Their AI model Gemini is not really cutting it compared to the two premier frontier models. And they're burning through cash and you're starting to see signals from the street that people don't really like where this story is heading and maybe there's some fatigue with regards to how much buyer enthusiasm there is. All that being said though, I think this earnings season will be crucial for gauging like how much longer do we have in this crazy euphoric market? Because even Google can beat estimates for their cloud business and operating income can triple. But they're still down almost 10% over the last week.
- [51:50]
A
Yeah. So wild times. IBM down 20% in a day. Google trending down. It's like the incumbents, man. Which hey, this is great. You know Revel Industrial revolution upended a lot of things. If so we could. Maybe we'll see that. Maybe we'll see that. You just gotta I. It's, you know, Poly markets are open for who your next, your next tech overlord is going to be. Is it going to be Dario? Is it going to be Mario? Is it going to be Sam? Who knows? On that note, thank you for listening to Blockspace Live comes to your feed every single weekday at 1pm Eastern. We are COMPUTE's Daily show featuring quick hits on AI, data centers, markets and emerging technology. This show is brought to you by CleanSpark. NASDAQ listed ticker CLSK I'm Charlie. I'm Colin and I will see you you on Monday.