WEBVTT

1
00:00:00.420 --> 00:00:04.680
Publicly listed Bitcoin miner turned AI company to cover. That will be our lead story.

2
00:00:04.780 --> 00:00:18.110
Ionic Digital went public yesterday through a direct listing, not an IPO, so we're going to be heading off the show with a quick rundown of their financial position and their AI HPC contract with Enscale.

3
00:00:18.260 --> 00:00:30.620
Following that, we have Ryan Lane, the co-CEO of Impery Digital, on to talk about why they pivoted away from a Bitcoin treasury strategy to make strategic investments in AI infrastructure.

4
00:00:31.460 --> 00:00:43.660
Following Ryan, we have Ben Puladian of BE- BEP Research on to talk about Bloom Energy tokens and the memory trade.

5
00:00:43.680 --> 00:01:03.340
And then we will cap off today's live stream with a quick newsflash segment, where we will cover Galaxy Digital's pricing for its senior secured note and some investment bank notes from Needham Research on Core Scientific following the AMD deal we covered yesterday, and Applied Digital.

6
00:01:03.420 --> 00:01:08.140
That's right. Blockspace goes live every single weekday at 1:00 p.m. Eastern.

7
00:01:08.200 --> 00:01:21.690
We are compute's daily live show, featuring quick hits on AI, data centers, emerging technology and markets, which by the way, markets are a little bit down today. Thank you, Korea. We'll get into that in a little bit.

8
00:01:22.520 --> 00:01:35.780
If you like what you hear on the live stream, it turns into a podcast anywhere podcasts are found, and if you like the podcast and the live stream and want more Blockspace content, all of our stuff is on our website, blockspace.media.

9
00:01:36.500 --> 00:01:54.440
Once again, that is blockspace.media. This show is brought to you by CleanSpark, NASDAQ listed ticker CLSK. More on CleanSpark later on in the show. Colin, stonks are down again because of Korea.

10
00:01:54.540 --> 00:02:09.520
I was looking, I think we tweeted earlier on the main account, Blockspace account, that Iron was headed toward the lowest prices in, like, a year or so. Um, dang. Things are- It's a bloodbath Yeah, it's a blood... Yeah.

11
00:02:09.940 --> 00:02:19.220
That's not stopping the deal flow, nor is it stopping a direct listing from Ionic that we actually covered on the show a week or two ago.

12
00:02:19.240 --> 00:02:24.320
And I'm gonna go ahead and get the latest note that we have from Blockspace up here.

13
00:02:24.340 --> 00:02:35.970
This is specifically about an activist investor, Sacom, had taking a 6.9% stake in Ionic Digital as the miner pivots from Bitcoin mining to AI and HPC infrastructure.

14
00:02:36.400 --> 00:02:45.440
But we're not gonna be covering the strategic investment exactly. Just a quick note on it. I wanted to use this opportunity, Charlie, to just give a quick briefing on Ionic Digital.

15
00:02:46.240 --> 00:02:58.420
So again, the direct listing hit yesterday. And Ionic Digital, for those who remember, was formed in 2024 out of the Celsius bankruptcy specifically to manage the Celsius mining assets.

16
00:02:59.030 --> 00:03:08.660
They have since, as many of the Bitcoin miners in our orbit have, basically all of them at this point, pivoted into a digital infrastructure and AI and HPC leasing company.

17
00:03:08.700 --> 00:03:13.360
They have five facilities in Texas, three of which are owned and two of which are leased.

18
00:03:13.460 --> 00:03:26.820
And the big focus for Ionic Digital as they go public is this Ward County site, a 234-megawatt site that is currently leased to Enscale under 126-month triple net lease.

19
00:03:27.980 --> 00:03:38.940
They received the first payment for this site in November 2025, an advanced payment of $45.6 million, and they have monthly fixed leases starting in August 2026.

20
00:03:39.000 --> 00:03:45.120
The total contracted revenue under the current lease stands at $1.95 billion.

21
00:03:45.160 --> 00:03:58.820
The annual fixed rent is for 234 megawatts and is equals $182.5 million per year for revenue after an 18-month ramp-up, with a 3% annual escalator after year five.

22
00:04:00.080 --> 00:04:12.900
In a February 2026 amendment, Enscale contractually is obligated to take an additional 89 megawatts at the same per megawatt rate when available, roughly, uh, second half of 2027, according to Ionic Digital's estimates.

23
00:04:13.549 --> 00:04:21.320
That would push the annual run rate for this deal up to 250.5 million, for a total contracted revenue of 2.6 million.

24
00:04:22.160 --> 00:04:29.280
Enscale has also granted Microsoft an additional option for power at Ward County starting in the second half of 2027.

25
00:04:29.840 --> 00:04:39.600
For its longer term ambitions, Ionic Digital expects to expand Ward County to 700 megawatts of total capacity. It's working through the details with ERCOT currently.

26
00:04:40.960 --> 00:04:53.900
And the CapEx for this would be $40 million for the 234 and 89-megawatt build, and then $64 million for the full 700-megawatt build-out, funded from cash and BTC treasury sales.

27
00:04:55.100 --> 00:05:04.040
So the Ward County site is the jewel of their portfolio. They also have four sites in Midland, Texas. East Stiles, 30 megawatts.

28
00:05:04.100 --> 00:05:16.380
Garden City at 12, Rebel at 50 megawatts, and another in Stiles at 20 megawatts for 112 combined. Their total power portfolio sits at 346 megawatts.

29
00:05:16.980 --> 00:05:28.020
And their mining fleet right now is, has been pared down from roughly 120,000 to 28,100 that will remain actively deployed at Midland.

30
00:05:28.060 --> 00:05:42.140
They are going to scrap 73,400 for, uh, you know, sell them or scrap them, and then they're going to have around 19,000 kept for spare parts. Charlie, any Ionic Digital- Not many, not many takes.

31
00:05:42.180 --> 00:05:47.240
Yet another miner wisely pivoting to AI HPC. Many such cases.

32
00:05:47.340 --> 00:06:03.056
I think my only other comment is, I didn't realize they were in Midland, which if you're in oil and gas, that's the suit-and-tie-wearing, uh, city for the, uh, Permian Basin oil and gas play, versus Odessa, Texas, which is the steel toes and jeans...

33
00:06:03.156 --> 00:06:12.415
uh, service, service guy, uh, uh, it's part of the, the, the Twin Cities, if you will. It's kinda like how Minnesota's got Twin Cities. Texas has them too.

34
00:06:13.286 --> 00:06:25.606
And at some point, the AI HPC guys flying in will outnumber the oilmen. But we- we're gonna wrap up with that. We have our guest, Ryan Lane, sitting in the wings.

35
00:06:25.676 --> 00:06:39.036
We'll bring him up here, and we'll talk about, again, what else? Pivoting from Bitcoin to AI. Before we bring up Ryan, though, a word from our sponsor, CleanSpark. [gentle music] We are CleanSpark,

36
00:06:39.956 --> 00:06:52.436
America's Bitcoin miner, a publicly traded company with the largest operating hash rate. Powered entirely by self-operated infrastructure across four states.

37
00:06:54.056 --> 00:07:11.636
This is our proof of work, and we are setting the standard for what's next. Learn more about the intersection of energy and Bitcoin at CleanSpark.com. All right, Charlie, let's get Ryan up here.

38
00:07:11.996 --> 00:07:24.256
Learn a little more about Emporee Digital. All right. Ryan- Ryan Lane... Lane. Welcome to the show, sir. Charlie, Colin, uh, thank you for having me. Absolutely, man. Really appreciate you taking the time.

39
00:07:24.336 --> 00:07:34.706
Obviously, this fits into our beat really well. Everyone is pivoting to AI and infrastructure in some way. We used to cover mostly Bitcoin. Now we are knee-deep- We're, we're just in AI pod now... in data center topics.

40
00:07:34.706 --> 00:07:41.596
Yeah. Yeah. [laughs] And so y'all's company really fits, I think, the zeitgeist of the current market.

41
00:07:42.076 --> 00:07:59.796
And s- uh, as, as a good place to start, can you give us some background on Emporee Digital, and also why y'all made the decision to sell roughly half of the treasury raise, like 87 mil, for this push into AI infrastructure with your strategic investments?

42
00:08:01.116 --> 00:08:12.935
Yeah, sure. There's a lot of overlap here, too, 'cause we're actually working with, uh, Anthony McKiernan, who was, uh, the guy who negotiated the NScale and, uh, Ionic Digital, uh, triple net lease. So he's Cardinal...

43
00:08:12.995 --> 00:08:25.796
He's the CEO of Cardinal Data Power, which we can talk about later. But, uh, so we, we basically hit the buzz saw as soon as we did our, uh, our DAT. We raised about, uh, 500 million, or exactly 500 million.

44
00:08:25.856 --> 00:08:32.116
We, we bought Bitcoin. Uh, our average price was, like, 117, uh, thousand on the Bitcoin.

45
00:08:33.086 --> 00:08:47.676
And then as soon as we launched the DAT, uh, we announced it about a month later, uh, they started to get weak, uh, and we started trading below NAV, uh, like a lot of the DATs out there, if you've seen, whether it was Bitcoin or, or other cryptos.

46
00:08:47.716 --> 00:08:59.456
And so the thesis on the trade kinda busted, uh, pretty quickly. And so then we had to figure out pretty quickly what we were gonna do, what the best use of that balance sheet was.

47
00:09:00.056 --> 00:09:11.496
And we always knew that we would have to figure out something to do with the balance sheet, because the best way to aggregate Bitcoin is through cash flow. And so we started really out of the gate.

48
00:09:11.556 --> 00:09:23.156
I mean, we're hedge fund guys, been running a hedge fund for almost 20 years. Uh, actually some of our seed money was out of, uh, Tulsa. So yeah. Yeah. Yeah, maybe we could talk about that.

49
00:09:23.256 --> 00:09:32.776
We can, we can talk about that. Yeah. Yeah. Uh, very dear to my heart. Uh, so we had, we started looking at different businesses. We, we couldn't...

50
00:09:33.056 --> 00:09:39.716
We had to figure out things that would be accreted to, uh, our NAV, right? We couldn't dilute shareholders.

51
00:09:40.336 --> 00:09:49.796
And with the Bitcoin trade and the DAT trade, it was, you know, selling stock, uh, above NAV and, and aggregating Bitcoin. Uh, but if that's not available, how do you, how do you do that?

52
00:09:49.816 --> 00:09:59.916
So we started leveraging up a little bit, selling some Bitcoin, buying our stock back, still adding Bitcoin per share, but in the meantime, looking for opportunities. And we looked at a lot of stuff.

53
00:10:00.256 --> 00:10:11.036
Um, we had a ton of deal flow coming through our shop, uh, trying to figure out exactly what the best place to spend money was. And we looked at Bitcoin mining, uh, too.

54
00:10:11.536 --> 00:10:20.876
Uh, I mean, the miners are what we determined effectively short power. And we said, "We don't wanna be short power in this environment. That's a disaster. So how do you get long power?"

55
00:10:21.535 --> 00:10:31.716
Uh, and so that's the quest that we were seeking to, to solve, is how do we get long power, and how do we have edge in getting long power?

56
00:10:31.816 --> 00:10:44.696
Because we just don't wanna be someone out there just going out and buying power land, uh, and not being able to negotiate the agreements, not being able to finance, uh, the data center builds. There's a lot to it.

57
00:10:44.776 --> 00:10:53.856
We're finance guys, so we understood all the struggles. We understand credit. We understand, you know, how you would finance one of these and get the proper IRR.

58
00:10:54.916 --> 00:11:07.556
And that's kinda how we got here, and we looked at a lot of stuff, and we can talk about the specific transactions if you'd like. But, uh, that was our thinking, and I don't see a better IRR now.

59
00:11:07.616 --> 00:11:20.436
If you can do these deals right, uh, then you can get from doing power land deals. There's a lot of value in power, and we can talk about that too, 'cause there's a lot of reasons that there's all this bottlenecking.

60
00:11:21.976 --> 00:11:22.156
Yeah.

61
00:11:22.316 --> 00:11:43.656
So you mentioned looking for a good opportunity or opportunities to park this cash, and the bedrock deals so far for Emporee have been the $65 million deal with the Hunt family for investment in a private entity for a Midwest data center, and also this Cardinal Data Power investment as well that y'all took a, uh, I believe 8% stake in.

62
00:11:44.496 --> 00:11:56.056
And so could you walk us through what made those deals so attractive, and y'all thought, "These are the ones that we need to use as the foundation for this new strategy"? Yeah, sure.

63
00:11:56.796 --> 00:12:08.056
So we looked at a, a lot of the stuff that the, the Hunt family's working on. Their, their tentacles are throughout the entire, um, power industry, so they're seeing a lot of stuff.

64
00:12:08.844 --> 00:12:19.744
Um, and just from their connections through the, through Anthony and, and Cardinal DataPower now, which was the end scale and, and, uh, Ionic deal. So we,

65
00:12:20.984 --> 00:12:28.864
we looked at pretty much the entire pipeline or portfolio that they were looking at, and this Midwest property is pretty rock solid.

66
00:12:28.874 --> 00:12:38.184
They had put an application to get the additional power through a power contract with the local utility like four years earlier. Uh, they had the contract in hand.

67
00:12:38.304 --> 00:12:49.384
All they had to, all they had to do, or now we have to do, is put a bond down for about 30 million bucks to do the transmission distribution upgrades to, and, and the interconnects to, to bring it onto property.

68
00:12:50.164 --> 00:12:59.944
So we've got 150 megawatts locked down. They're currently using that, uh, Bitcoin mining. Uh, not all of it, but they're- they're using a substantial portion of it.

69
00:13:00.724 --> 00:13:14.824
And then we have to expand the substation on property to bring in the other 149 megawatts, and then the tenant will use that as well. So we're buying that property for $230 million. Uh, Empry Digital is 25% of that.

70
00:13:14.964 --> 00:13:27.404
Uh, we are triple net leasing that, which in this space people are loosely using that term. But a true triple net lease is what we negotiated here as a, you know, as a real estate property person would characterize.

71
00:13:28.044 --> 00:13:39.784
So we have... They- we basically l- almost lose access to the, to the building. So it's true, true triple net lease, like a McDonald's, right? They'll operate it, we'll pass through the power, uh, to them.

72
00:13:40.124 --> 00:13:50.344
Uh, they have all the expense of, of building the data center. They'll bring in their GPUs and their racks and they'll build out the entire system. Uh, they've done all the spec'ing for that.

73
00:13:50.384 --> 00:14:01.844
They've, you know, done all the environmental for that. So they are substantially ready to go. And then we just sit back and collect rent, and it just drops to the bottom line. So I...

74
00:14:03.024 --> 00:14:13.384
If you, if you, if it stays on schedule, it looks like this. It's about $441 million, uh, in rent payments, uh, over 15 years. Uh,

75
00:14:14.704 --> 00:14:23.644
it, the, the extra or the additional power kicks in in year three, and then they'll start paying on that, and that's accounted for in the $441 million.

76
00:14:24.244 --> 00:14:39.734
So effectively, we pay $65 million, and we get $441 million in, in lease payments, and at the end of the day, we still have the property with the power. So from a net present value perspective of what is that worth,

77
00:14:40.684 --> 00:14:56.824
depending on, you know, how you, what discount rate you use, what your debt looks like, but if your debt's at eight and a half percent and you're using a 6% discount rate or something like that, you're gonna end up at 200 or 250 million in value for 65 million, right?

78
00:14:56.864 --> 00:15:09.844
So y- and it's gonna pretty quickly generate cash flow. So the NPV on that to me is a no-brainer. Effectively, you're getting something that's almost worth, you know, three to four times of what you're spending.

79
00:15:10.804 --> 00:15:23.444
So sign me up for that. I'll do that stuff all day long, and that's one of those properties where it's turnkey, right? The power's there, and there's not too many of those opportunities left in this country.

80
00:15:23.454 --> 00:15:32.624
You know, it's gonna all shift, and I don't know when we're gonna run out of those opportunities, but it's gonna all shift to behind the meter. But while we can do these, these are easy turnkey transactions.

81
00:15:33.984 --> 00:15:42.764
So, uh, that investment is not the total amount of Bitcoin that you sold. There's probably s- there's a few more dollars I think you guys may be holding onto.

82
00:15:43.324 --> 00:15:55.024
And you m- I believe you've hinted perhaps at some more AI announcements coming. I don't know how much you can talk to that, but, uh, maybe high level, are you trying to deploy this cash quickly? Uh,

83
00:15:56.034 --> 00:16:03.984
uh, what's the next few quarters look like? Should I... I'm, I do love a good AI deal flow announcement. What should I, what should I be looking for?

84
00:16:04.924 --> 00:16:14.254
So the second one we announced was the Cardinal DataPower, and that one's different, and there's, uh, there's a lot of stuff that we're looking at. But that one, that one is the behind the meter.

85
00:16:14.824 --> 00:16:23.584
Cardinal DataPower is the Hunt's, uh, entity, uh, that they will do behind the meter transactions and, and that's in the Permian. And

86
00:16:24.804 --> 00:16:36.064
they have all the connections between the, you know, power providers, uh, from Hunt Power. Uh, I mean, I think Hunt Power built the first interconnect between the grids in New Mexico and Texas.

87
00:16:36.124 --> 00:16:42.874
Like these guys have been in it since the late '90s, uh, and they're in the oil and gas industry. So this property's interesting.

88
00:16:43.664 --> 00:17:00.654
With this investment, they secured, uh, about a gigawatt of recips, mostly 10 megawatt recips. Uh, and we'll see some partners announcing, uh, you know, that, that energy, uh, collaboration. But that,

89
00:17:01.744 --> 00:17:13.264
if you have a gigawatt of power, you have gas lines coming in, you have a group that has the discipline to be able to turn that power on and reliably, 'cause that's why everyone loves the grid right now, it's reliable power.

90
00:17:13.784 --> 00:17:22.294
But if you can do that behind the meter, I think that's the real edge in Cardinal DataPower, is they can bring that to the hyperscalers. And if we have the... And you

91
00:17:23.344 --> 00:17:33.464
guys are very well aware, you can't get recips right now. No. We have, we have a gigawatt of recips from- That's, that's 1,000 reciprocating engines that... These are huge.

92
00:17:33.524 --> 00:17:43.144
Like, I'm not familiar of any, like, other sites that have, I don't know- It's the biggest... a golf wood site. 1,000 10 megawatt reciprocating [laughs] engines. Yeah. Deafening.

93
00:17:43.244 --> 00:17:54.084
Ho- that's ba- that's acres and acres and acres. This is a huge project. Right. Oh, it's a huge project. It's... Yeah. The, the, the group that is providing these, um,

94
00:17:55.524 --> 00:18:02.464
uh, recips will make an announcement, but they're the biggest of the biggest. Gotcha. Uh- I probably know who they are... and they'll make an announcement after they, after they have their earnings.

95
00:18:03.024 --> 00:18:11.724
But it's, it's material for them. So when... We can't talk about it just yet until they get through their earnings period, but it's a massive...

96
00:18:11.794 --> 00:18:19.760
And it's not just 10 megawatt ones, it's also some 2.5, so it's actually over 1,000 And there's, there's reasons that you, you set it up that way.

97
00:18:21.140 --> 00:18:30.440
But I think you'll see more and more of that, and I think these m- OEMs that are manufacturing these recips are gonna be a big players within this space, uh, in the next coming decade.

98
00:18:31.880 --> 00:18:43.100
So Ryan, kind of to piggyback off of Charlie's question right there, the Hunt family being an ally here is a powerful ally to have given all of their experience in oil and gas, right? I mean, that's...

99
00:18:43.140 --> 00:18:47.090
Y- y- you almost couldn't ask for a better partner in building out some of these sites.

100
00:18:47.960 --> 00:18:56.520
W- will your expansion strategy in the future, and again, I know there's a lot that you can't say 'cause compliance will be breathing down your neck. Yep. But

101
00:18:57.580 --> 00:19:07.920
I mean, is, is the play here going forward to scope out more of those off grid opportunities given this connection, given that incredible portfolio of recips that you're building up?

102
00:19:08.760 --> 00:19:18.280
I mean, I could imagine opportunities up in the Bakken, tons of opportunities in, in the Permian. Uh, there's cheap gas there. They have access to the cheapest gas.

103
00:19:18.380 --> 00:19:25.230
I, I could see us taking larger stakes, additional stakes in Cardinal DataPower as Empiri Digital.

104
00:19:26.230 --> 00:19:43.620
I can see them doing multiple of these types of projects throughout the country where they have some sort of edge on being able to run gas lines and have cheap access to gas, and I think their partnership with the OEM that's doing the recips is much bigger than just this Permian property.

105
00:19:43.660 --> 00:19:47.780
So I think there's a lot to do there, and I think that's the edge, right?

106
00:19:48.520 --> 00:20:01.440
You make, you make money you, when you bring value to a project, and you get access to these projects, and you get looks at these projects when you bring value, and I think that's what the Hunts bring, is they, they bring the ability to provide reliable power.

107
00:20:01.920 --> 00:20:12.750
They're not a counterparty who's gonna, who's gonna fall over. There's a lot of people falling over in this space, right? It's not easy to build a data center, and we're not proposing to be able to do that. But

108
00:20:13.780 --> 00:20:21.580
JV with a group that can build a data center, and then you can provide the power. I, I think that's, I think that's where you will see a lot of announcements.

109
00:20:21.680 --> 00:20:31.600
I think that's where you will see, um, a lot of activity from us and the Hunts. M- Go ahead, Charlie. Well, you know, I...

110
00:20:31.760 --> 00:20:47.300
You know, you mentioned, uh, some money out of Tulsa helped get Empiri off the ground, and it kind of brings me to, you know, looking at this, you've got this, you've got this partnership with Hunt, or this deal with Hunt, and that's a...

111
00:20:47.360 --> 00:20:59.880
Now, the data centers basically have to go after oil money because there's not a lot of, [laughs] there's not a lot of these, like, uh, big, huge capitalized entities like the Hunt, like Hunt Petroleum and otherwise. Um,

112
00:21:01.200 --> 00:21:06.420
you know, what do you see as the future of, of, like, types of partnerships between data centers,

113
00:21:07.340 --> 00:21:19.960
behind the meter oil and gas, and, uh, especially 'cause the oil and gas has go- has undergone a decade of very, um, strict ca- capital discipline, learning a lot of lessons from the private equity boom of the early 2010s.

114
00:21:20.660 --> 00:21:32.700
Yep. Um, you know, are, is it time to take the gloves off and, uh, does that money come out of Tulsa [laughs] anymore? So... Yeah, I, I don't know what form it'll take, but I think they're... You're right.

115
00:21:32.760 --> 00:21:43.300
There's a lot of money there, and just the fact that the Hunts are looking at this and they're deep in it, they've put together an entire team to execute this, and they're spending money.

116
00:21:43.400 --> 00:22:00.120
I think that they see the shift, and versus o- oil and gas and fracking, politically, environmentally, this is, this is like, this is like angel dust. [laughs] Kid gloves. Right? You know? [laughs] Yeah.

117
00:22:00.680 --> 00:22:03.330
So I, you know, I, from a,

118
00:22:04.300 --> 00:22:20.260
from a political perspective, y- it's under, it's hitting some headwinds right now, and that's a whole nother discussion of how rational and justified that is because the number of jobs that will be provided, number of electricians and carpenters and c- I mean, it's...

119
00:22:20.530 --> 00:22:26.080
The... And they're all getting paid w- wages that are substantially more than they're used to. This is all really good.

120
00:22:26.920 --> 00:22:36.940
Um, and then of course there's some backlash, but from an environmental perspective versus fracking, this is amazing. So if the Permian and these families get involved in this sort of stuff, I think it's, it's great.

121
00:22:38.380 --> 00:22:42.000
Ryan, one quick question with regards to the future of the treasury strategy.

122
00:22:42.320 --> 00:23:08.020
What kind of synergies does having a Bitcoin treasury, all that Bitcoin on the balance sheet, I believe it's roughly 1,500 at this point, and also having these AI infrastructure investments, like how does that position Empiri as one of the more novel stocks within this ecosystem by having that dual Bitcoin treasury balance sheet and then also having these investments in these bleeding edge AI infrastructure plays?

123
00:23:08.100 --> 00:23:15.420
Yeah. So we're, we're not taking Bitcoin off the table, right? We're believers in Bitcoin. We believed in Bitcoin. We still believe in Bitcoin.

124
00:23:15.460 --> 00:23:23.860
We think it has a place in the world economy going forward, right, with all the debasement of currencies. So

125
00:23:25.120 --> 00:23:31.760
right now, we are looking at this opportunity to generate cash flow, and when you generate cash flow, you have cash on the balance sheet.

126
00:23:31.960 --> 00:23:43.080
And if you, if you wanna store your cash on the balance sheet, Bitcoin makes a lot of sense. And from the very beginning, we always looked at businesses that generated cash flow so we could aggregate more Bitcoin.

127
00:23:43.820 --> 00:23:46.840
Right now is not the time to aggregate Bitcoin.

128
00:23:46.880 --> 00:24:00.100
Right now is the time to deploy it into these strategic investments, and as they start to spin off cash, uh, I could see us going back into a mode of, of accumulating at least some and probably a substantial amount of Bitcoin back on the balance sheet.

129
00:24:00.940 --> 00:24:03.540
But as far as the Bitcoin goes right now,

130
00:24:05.020 --> 00:24:21.460
I think that we probably continue to bring it down as we identify investments to make, and if we can find more stuff like the Midwest property, and we can deploy some additional cash into, into Cardinal as they're securing additional recips and, you know, specific use cases, then I think that makes a lot of sense.

131
00:24:23.040 --> 00:24:33.220
We'll be keeping an eye on additional deals coming down the pipe, particularly in the oil and gas sector. Ryan, thank you so much for joining, man. Really appreciate it.

132
00:24:33.300 --> 00:24:43.100
We'll have to tag you back in once there's another news item. Colin, Charlie, appreciate it. Thank you. Yeah. Whenever- Thanks, Ryan... you're in Tulsa, come by. Hit me up. [laughs] Absolutely. Cheers. See you around.

133
00:24:46.660 --> 00:25:00.900
Love Ryan of Harry Digital. Smart, timely pivot. We're gonna keep on going. We have in the wings Ben Paladian with BEP Research. We'll bring up and we'll talk compute, maybe some Bloom Energy,

134
00:25:01.940 --> 00:25:07.060
and, uh, excited to have that conversation. But before that, a word from our sponsor, Luxur.

135
00:25:07.720 --> 00:25:19.440
[lightning] This episode of Block Space Live is brought to you by Luxur's Commander, Bitcoin miner management software for enterprise operations.

136
00:25:19.960 --> 00:25:25.540
Luxur's Commander gives you real-time fleet monitoring, bulk remote commands across your fleet, and Intelligent Miner.

137
00:25:26.000 --> 00:25:33.320
That's an automated profitability engine that runs every five minutes and adjusts your fleet's power settings to live energy and hash rate markets.

138
00:25:33.820 --> 00:25:40.440
Our cop back test showed 10% improved profitability with intelligent mining versus old-fashioned binary mining.

139
00:25:40.460 --> 00:25:56.620
Commander Pro is $100 per megawatt or a 25 basis point pool fee adder, roughly half the cost of competition, but you can also try it free for 60 days. If you'd like to learn more, go to luxur.tech/commander. All right.

140
00:25:57.300 --> 00:26:09.100
We've got Ben in the audience. We're gonna bring him on up here. Ben- Hi, guys. Welcome. Welcome to the show. Thanks for hopping in. Good morning, or good afternoon, wherever you guys are.

141
00:26:09.160 --> 00:26:17.800
We're, we're both East, West, and Mid-Coast, so it's time, times, times, uh, very different. Time is an illusion. Yeah. [laughs] Okay.

142
00:26:18.240 --> 00:26:28.920
So, uh, you popped up on my radar as we started really focusing on covering Compute lately. Um, I'm thinking maybe we talk memory, Bloom Energy, maybe, uh, follow up on your piece, Tokens Are The New Oil.

143
00:26:29.000 --> 00:26:43.270
But right now this-- Right now, memory stocks are dragging down the Korean stock market, and therefore kinda, like, causing, causing a lot of existential questioning among some of the weaker-willed AI trade people. Um,

144
00:26:44.600 --> 00:26:50.770
what are your thoughts on the current state of the market, SK Hynix, memory pulling back?

145
00:26:52.120 --> 00:27:02.400
Yeah, so the, the whole idea behind memory and this AI trade is that the idea of having more memory is-- it's like having, like, a bigger brain, right?

146
00:27:02.460 --> 00:27:10.000
So the more context your AI model has, the more data points, hypothetically it will make a better decision, right?

147
00:27:10.160 --> 00:27:22.200
And as we-- we're training bigger models and we're generating more data, you need more ways to store it, uh, your weights and the information quickly to ret-retrieve it.

148
00:27:22.260 --> 00:27:35.040
So that's why NVIDIA with Hynix, you know, many years ago, uh, came up with this idea of high bandwidth memory. So basically taking the memory that we have, like, in our computers and stacking them like lasagna pieces

149
00:27:36.180 --> 00:27:45.520
higher and close to the GPU. And without-- You can't have memory without-- You can't have AI without memory.

150
00:27:45.740 --> 00:27:57.629
So it- it's like almost like peanut butter and jelly when, when it comes to that stuff, so they all come together. Uh, the challenge is, is from, like, a market perspective, uh, memory companies

151
00:27:59.120 --> 00:28:14.530
have, have always been treated like a commodity because it's, like, a fungible type of asset. And you had three players, uh, Micron, Samsung, and SK Hynix, and I guess CXMT out of China, right? Uh, they just IPO'd.

152
00:28:15.320 --> 00:28:28.560
And there's a, there's a world market for that, and there's, there's pricing, and most of the time, consumer electronics companies like Apple would, would buy their memory for the next iPhone or laptops or Dell or whatever.

153
00:28:29.260 --> 00:28:41.540
It-- You never had, uh, those challenges. And I think when there was, like, a big computing cycle, like, I guess PCs or the, the iPhone in 2007, 2008, yeah, memory prices shot up.

154
00:28:41.620 --> 00:28:56.920
And a lot of these, uh, companies were wary of, uh, spending cap-capital on expanding their fabs because they know when there's a downturn in demand, they're gonna be left with all this extra capacity, and then there's no, no one to utilize it.

155
00:28:56.980 --> 00:29:10.440
And now you have this killer app of AI, which is sucking up all the memory, and it's you're not just using a few pieces, you're using gigabytes and terabytes stacked in these data centers. And it's not only NVIDIA.

156
00:29:10.660 --> 00:29:22.280
Every hyperscaler is trying to secure memories, uh, signing, you know, long-term agreements that go into, into 2030, and there's, like, a huge, uh, memory crunch.

157
00:29:22.340 --> 00:29:35.120
And I wrote this piece, uh, last December called The Memory Wars. Uh, we haven't even got into definitive high-resolution world models with video and audio.

158
00:29:35.180 --> 00:29:39.640
Like, right now, we're still playing, uh, agentically with text and code.

159
00:29:40.140 --> 00:29:53.400
Like, the next level is even crazier, and that's why you need to have these improvements in the compute and the memory to, to bring in more, uh, vivid interpretations of AI which haven't been there yet.

160
00:29:54.360 --> 00:30:10.120
On the market dynamics, I, I have this feeling that, you know, nothing against, uh, crypto and things like that, but the Bitcoin market is sort of, uh, soured in the excitement and the altcoins. And I think

161
00:30:11.020 --> 00:30:23.600
a lot of investors or traders were looking for a way to play a new theme. And obviously, memory is something very exciting, and these companies were dogs for many years.

162
00:30:23.710 --> 00:30:32.740
If you look at the charts, they don't really go anywhere, and they, they traded on a factor of book value of what the assets of the company were worth. And now-[clears throat]

163
00:30:32.950 --> 00:30:48.010
Companies like Micron or Hynix are making, you know, 20 years of earnings in like in a few years that they, they never made. And everyone's going crazy and trading them, and it's, it, it's pretty wild.

164
00:30:48.390 --> 00:31:07.090
Uh, it's not for the faint of heart, and you had things go parabolic [clears throat] in the summer into June, or which was last month, and now you're facing these huge drawdowns, and margin cascades, and leverage, and hyper liquid, and like all this other stuff, which fundamentally is it,

165
00:31:08.150 --> 00:31:20.350
it's, it's scary and it's not easy to, to withstand. I mean, I have a small allocation of my portfolio to that, but nothing as big because I can't handle the, these gyrations, but memory is always important.

166
00:31:21.230 --> 00:31:31.110
So it sounds like the memory trade... Like, the memory bottleneck is still definitely on, and I'm trying... You know, I'm, I'm familiar with the volatility. We've been covering Bitcoin for many years.

167
00:31:31.630 --> 00:31:41.329
Um, but it sounds like the, the structural bottleneck is still on. Um, talk me through this. I'm actually kind of curious, especially in light of the CX and T listing.

168
00:31:41.460 --> 00:31:50.370
It seems like a lot of the investors are trying to move down this high bandwidth stack to look at DRAM as, like, a precursor component. But am I...

169
00:31:50.510 --> 00:32:04.970
Like, where among the, like, memory trades, where are the greatest bottleneck? I think really what it comes down to, even it's not even memory, it's packaging. Advanced packaging is...

170
00:32:05.650 --> 00:32:17.610
The idea is you have to stack these pieces of memory together, and this, this miniature, like, li- co- connector wire has to go through all of them exactly.

171
00:32:17.750 --> 00:32:21.390
They were trying to go to 16 high, and I think they got stopped at 14 because it's...

172
00:32:21.870 --> 00:32:32.470
You have to stack all of them exactly perfectly down to the micrometer to get it to work, and you end up wasting a lot of, uh, wafers because, uh, a lot...

173
00:32:32.850 --> 00:32:35.220
You, your yield isn't there because it's so hard to, like, put together.

174
00:32:35.390 --> 00:32:47.850
So getting that automated and not wasting as much and making sure it works is kind of, like, the real bottleneck, and I think that's why you saw Nvidia invest, I think, 1 point...

175
00:32:47.930 --> 00:32:58.850
Or not invest, but forward purchase $1.5 billion in Amkor packaging in Arizona. Uh, you have a lot of these test companies like Lam and Veeco Instruments.

176
00:32:59.150 --> 00:33:12.030
Uh, like, you, you, you need to make sure this stuff ac- You can make it and it's automated, but not ev- not everything comes out 100% right, and you end up throwing away a lot, and that's, that's a big issue, the waste with the yield.

177
00:33:12.600 --> 00:33:23.030
Mm. I think that's, that's, like, the real bottleneck, in my opinion. Ben, qui- a quick question about a temperature check from you and where you think we are in this current cycle. So as...

178
00:33:23.070 --> 00:33:36.810
You know, we've been touching on SK Hynix, Micron, et cetera. SK Hynix down, like, 25% since its date, since its debut on the Nasdaq. Micron down, like, 33% over the last month. What's your take on this price action?

179
00:33:36.890 --> 00:33:43.610
Do you see this as a natural de-leveraging event, considering these stocks were really hot for the first half of 2026?

180
00:33:44.030 --> 00:33:48.170
Or do you think something has structurally changed in the way that the market is pricing these things?

181
00:33:48.610 --> 00:33:58.690
I think, I think any time you have parabolic moves, it, you know, it, it, it shoots too high on the upside and it shoots down on the up downside.

182
00:33:58.850 --> 00:34:12.330
I think moves like this really test your conviction in the company and what they're doing, and if you really believe in the management, and the roadmap, and where you see AI or the technology that's headed.

183
00:34:12.490 --> 00:34:28.610
I mean, I've lived through many drawdowns from 2016 from being an Nvidia in- investor, from, you know, the Bitcoin meltdown in 2018, uh, to 2022. The huge... Every day it was going down and, and a lot of people

184
00:34:29.510 --> 00:34:39.830
want, uh, great returns, but as an investor, you, you get paid for the volatility, right? That's how you earn your, earn, earn your keep.

185
00:34:39.990 --> 00:34:55.370
Uh, th- these things just ran up way too fast, too much leverage, too many leverage products. I saw traders in Korea who, who don't even know what's going on, were getting 500% leverage. Uh, that's just not normal.

186
00:34:55.430 --> 00:34:56.530
That's not investing.

187
00:34:57.150 --> 00:35:14.990
I'm not gonna quote Warren Buffett and say the market is a huge casino, but some of these products that have been created with leverage are just enabling people to gamble and lose, and I'm always against, uh, leverage products because it, they just, they never work out in the end.

188
00:35:16.810 --> 00:35:24.450
Okay, I wanna pivot now from memory to Bloom Energy, which you've been talking about recently. [laughs] I think we, we did a... And this will...

189
00:35:24.510 --> 00:35:42.890
Colin here will actually be able to di- uh, probably tap on this a bit too, 'cause we just started paying attention after Leopold bought, and then, and then we had the, the Bloom Energy FUD, and lately you have been on a tear saying that it sounds like the FUD is overblown.

190
00:35:43.630 --> 00:35:55.190
Um, I believe you tweeted, uh, um, the CEO's response to the Hunter Brooke report recently, which was pretty blunt. Quote, "We have enough scandium... There's enough scandium to power the planet."

191
00:35:55.390 --> 00:36:04.190
Um, this is the CEO of, uh, of, uh, Bloom. "We're not dependent on China. Everything else is proprietary." So we have...

192
00:36:04.230 --> 00:36:15.920
He's just, you know, uh, saying that all these concerns are over- uh, are not, uh, not worthwhile, but we don't... I don't have any details or answers. Ben, enlighten me. You know, what do I make of this?

193
00:36:16.610 --> 00:36:27.250
I mean, I ran a manufacturing company, a commercial LED lighting business. Like, I wouldn't have my investors and customers come up to me and say, "Where are you sourcing your LEDs from? Where are your PCBs from?"

194
00:36:27.870 --> 00:36:36.430
Like, are your customers buying the product? You know, Brookfield is their largest customer, right? They went from five billion to 25 billion, right? They're underwriting the risk.

195
00:36:36.810 --> 00:36:48.190
They're buying- These Bloom boxes on the behalf of hyperscalers and, uh, neo clouds, and they're basically leasing it to them, and they're making the power and making the delta.

196
00:36:48.590 --> 00:37:00.050
So if Brookfield, one of the world's largest asset managers and infrastructure owners, is comfortable with Bloom Energy, I think everyone else should be. I mean, I don't know what the big deal is.

197
00:37:00.090 --> 00:37:12.750
Like, why you have such a hard-on to see where he's getting scandium from. Like, who cares? Like, it's just one part of the, the product. Like, the product works. I know developers that like it. It's quiet.

198
00:37:13.430 --> 00:37:24.190
Like, you, you don't have to wait four years for a turbine from, uh, GE Vernova. It's at parity with cost, and the whole time is how fast can you energize a product for the first token?

199
00:37:24.730 --> 00:37:30.910
'Cause as a developer, you're borrowing at a high interest rate, and you need to turn over the project to the customer.

200
00:37:31.110 --> 00:37:41.930
So I don't know-- And if they're delivering on time and the product is fungible, if some- there's a delay somewhere else, you can just move the Bloom boxes somewhere else. I mean, it, it's a simple solution.

201
00:37:42.370 --> 00:37:53.670
Yes, the CEO's been doing this for twenty-five years. Yes, he was on sixty minutes and, like, t- twenty-five years ago, and he held up this box and one day it's gonna power a home and whatever.

202
00:37:53.750 --> 00:37:59.370
But I, I just don't see, uh, the smoke and mirrors that everyone thinks they see in this company.

203
00:37:59.630 --> 00:38:13.790
I think they're just executing and doing well, and it's, it's much needed to keep this, uh, AI race ahead of China because we're under-electrified as, as a country. I think the Hunter Brooke report didn't...

204
00:38:13.850 --> 00:38:22.470
Like, Hunter Brooke didn't do itself any favors. I mean, the, the head of Hunter Brooke more or less, I believe it was on CNBC or Bloomberg,

205
00:38:23.590 --> 00:38:34.530
compared them to the Theranos of this cycle, which to me is a really unfair characterization from the fact that, like, Theranos literally had a product that didn't work, and that's just not true here.

206
00:38:35.030 --> 00:38:55.420
But I do think that it was somewhat fair to push back on the CEO's contention that they, contention that they don't source any scandium from China, which is just fundamentally not possible if they're going to capitalize on their goals of having multiple gigawatts of these energy cells rolled out, 'cause China has ninety percent of the refining capacity.

207
00:38:55.450 --> 00:39:03.210
So I guess just to ask to a, a more pointed question here, what, what do you make of his insistence that they don't do that?

208
00:39:03.750 --> 00:39:13.810
Um, do you see that as having any supply chain risk given that now China has instituted strict- stricter ca- um, import or export controls on these critical minerals?

209
00:39:14.370 --> 00:39:29.080
And also, what do you make of the contention that the supply of refined scandium every year won't necessarily be enough even for Bloom to reach these like three gigawatt targets that some people on the street are calling for, for their energy cells?

210
00:39:29.150 --> 00:39:41.650
'Cause I think those are the key points of the Hunter Brooke report. That's correct. I mean, he did make those points, and I did speak with, uh, Sammy, uh, l- on the phone. He's a great guy. He, he's a big Knicks fan,

211
00:39:42.590 --> 00:39:50.399
happy for them winning the championship. And, and it-- And Hunter Brooke does great inve- investigative work. Uh, I think probably what they came up with is true.

212
00:39:50.470 --> 00:40:02.750
Maybe all the scandium does come from China, and it's sold through intermediaries in Hong Kong and Thailand or Vietnam, and they're just buying from those. Like, and technically you could say it's not from China.

213
00:40:02.830 --> 00:40:17.010
But, like, uh, Apple went out of its way and says, "Oh, we wanna buy memory from," uh, was it YMTC or I, I... For, for us to sell iPhones in China, and Micron was going crazy like, "Why are you buying Chinese memory?"

214
00:40:17.090 --> 00:40:28.310
Like, why does it c- why does it matter if you buy Chinese memory if you're selling Chinese, uh, made iPhones in China? Like, what's the big deal? Like, these are all just supply chain issues.

215
00:40:28.540 --> 00:40:44.490
And I think I was speaking with some other investors. There's some mines coming online in Canada or Australia, I believe, that would dovetail into the timeline of the ramp of the scandium from Bloom Energy. So maybe

216
00:40:45.690 --> 00:40:54.089
we're investing in some, uh, other mines around the world that we don't know about, but they do exist, and I, I've, I've been told about those companies. But for me,

217
00:40:55.350 --> 00:41:08.610
I'm not gonna go down to the critical mineral of some company's product which they're basically getting twenty-five billion dollars in orders for, that they're, that they're, they're scaling and they have, you know, Brook- Brookfield basically

218
00:41:09.630 --> 00:41:20.590
underwriting them to the wazoo to make sure everything is right because once you deploy these, the, the useful life has to be at least five to ten years, right? It has to work and has to do everything for these guys.

219
00:41:20.650 --> 00:41:34.870
So like, what, like, why are you always questioning all these things? Like what's... Like, like China get soybeans from America. We're gonna question how many soybeans they buy? Like it just... Th- these things, uh,

220
00:41:35.770 --> 00:41:37.490
are, are good questions, but I think

221
00:41:38.390 --> 00:41:50.750
there's other companies [chuckles] on the market today that are doing quest- more questionable things that deserve more media attention than a mineral in this product that this company is selling. Okay.

222
00:41:50.790 --> 00:42:02.530
I wanna, uh, o- one more question on, uh, generation before I, we go to, to, uh, open models. Um, so time to power is like the focus for everything.

223
00:42:02.770 --> 00:42:15.790
Neo clouds, compute, everybody along the stack, it just wants to shorten time to power. Let's assume Bloom Energy, uh, largely can deliver, I think, up to twenty-five mega, uh, gigawatts.

224
00:42:16.630 --> 00:42:28.930
Like, what is the implication in your view for generation overall? 'Cause right now, like you have to, you have to sell your mother to get a t- a gas turbine.

225
00:42:28.970 --> 00:42:40.940
Like, and those take for, those, you know, eighteen, twenty-four, five months, five years to build. What are the implications for the overall time to power thesis if a company like Bloom Energy can deliver?

226
00:42:42.934 --> 00:42:57.714
We're in a race against China. They're super electrified. They have more electricity per capita. They're building nuclear plants at scale. There's less... Uh, in America, we have NIMBYism, like, "Don't build this here.

227
00:42:58.094 --> 00:43:07.474
Oh my God, this is noisy. Oh, the water's gonna get messed" Like, everyone's complaining. In China, it's like a iron fist. Like, "We're putting this data center in your backyard. Tough luck."

228
00:43:08.194 --> 00:43:11.074
Or this, like, power generation. They're just moving ahead.

229
00:43:11.914 --> 00:43:30.334
Obviously, we can't do that here, but at least a company like Bloom could take one of the biggest natural resources that we have, which is natural gas, that we have so much of that we're exporting now around the world in a liquefied na- nature, and taking that and converting that to electricity in a quiet, sustainable manner.

230
00:43:30.934 --> 00:43:41.034
I mean, like, this is like the, the killer app that we, we've been waiting for, and hopefully they can scale this so much and we can use it for homes and commercial buildings.

231
00:43:41.594 --> 00:43:54.314
Natural gas everything because I don't think as a nation we can get enough nuclear online and solar, uh, in the scale because, uh, because we're gonna have s- so much, uh, electricity needs, you know?

232
00:43:55.014 --> 00:44:04.314
Power is everything, and I'm gonna keep, keep on saying that. Like, power equals compute equals revenues. You have all those things, you win. If you don't have one of them, you're a loser.

233
00:44:04.714 --> 00:44:15.634
I mean, that's really what it comes down to. Okay, let's pivot to open models because you had, um, some good insights. You had this, uh, X article.

234
00:44:15.694 --> 00:44:22.994
You've written about it, and this was, mind you, at the beginning of July. Um, the open source war is over. You've tweeted a bit about it.

235
00:44:23.094 --> 00:44:29.634
Um, quote, "Contrary to the obituary wo- obituary wi- uh, writers, the open source AI isn't dead."

236
00:44:29.674 --> 00:44:40.024
Since then, Ben, Jensen Huang made his first tweet co-signed by, like, the entire AI ecosystem except for Anthropic on open models.

237
00:44:40.614 --> 00:44:56.424
What's changed since this article because you have massive industry sign-on to a similar thesis here? Yeah, J- Jensen and I are good friends. We hang out a lot at Denny's, and we- [laughs]... we talk about AI [laughs]

238
00:44:57.094 --> 00:45:08.934
with, uh, the grand slam. Uh- [laughs] Uh, a- actually, I heard he likes chicken fried steak. But no, I've, I've studied Jensen for, you know, over a decade.

239
00:45:09.274 --> 00:45:16.854
I've, I've known about him for over 20 years when I used to work in investment banking. I mean, I have friends that have worked in Nvidia for a long time.

240
00:45:16.954 --> 00:45:34.254
And you look at the parallels of, of technology and the curves, and really what we're seeing is a, a lab like Anthropic is just getting too big and too strong, and they're becoming like a juggernaut. And it's...

241
00:45:34.614 --> 00:45:49.694
They have great products, but the way they're approaching it is they wanna basically decimate the whole ecosystem and create some sort of regulatory capture where they are control everything, even down to the silicon.

242
00:45:49.734 --> 00:45:59.624
They... I think they even went to some of the Korean memory makers like, "Hey, can you give us some materials? Like, we wanna make our own, like, fab and GPUs, like, 'cause we can do it with Claude." Like, like,

243
00:46:00.654 --> 00:46:11.934
don't try-- Like, in business, for business to work, you need to let everyone to kind of make a little bit of money to survive, not just try to put everyone out of business because you're the best.

244
00:46:12.834 --> 00:46:24.514
I, I get that, and they have a great model and everything's good. But the whole point of open source is, is you can't be stuck with Coke or Pepsi for the rest of your life. And then

245
00:46:25.534 --> 00:46:42.274
this Coke, because they make it so well, is like, they say it's, like, $500 a can. And then a similar type of Coke in China or, like, RC Cola is, like, 5 cents a can, and it does-- it gives you that same,

246
00:46:43.394 --> 00:46:52.284
uh, bubbly feeling, and it does most of the same type of work, so, like, why are you paying so much for it, right? I think I said that. Okay. Yeah, you, you used the exact Coke and Pepsi analogy. [laughs] Yeah.

247
00:46:52.294 --> 00:47:07.014
Well, it's funny, Ben, because at the time when, like, you made this analogy and Jensen had put out the letter signed by everybody, um, OpenAI and Anthropic had signed. OpenAI kinda signed late it seems.

248
00:47:07.074 --> 00:47:15.754
Do you think this was a, them realizing they have to join forces with the rest of the industry against their main competitor, Anthropic? Like, you know... I don't know.

249
00:47:15.794 --> 00:47:22.654
H- does, does this change this duopoly Coke, Pepsi thing? Yeah, there was, there, there, this... There's a huge Game of Thrones game theory behind all this stuff. There...

250
00:47:22.734 --> 00:47:33.294
Anthropic as a company has hedged its silicon between Google and then Amazon and then using AMD and then some Nvidia GPUs. They're kind of all over the place.

251
00:47:33.814 --> 00:47:44.634
They're not-- They don't hold alliance to, like, one company, as you can see, and, uh, they're becoming a threat to those companies that they're also being a part of.

252
00:47:44.694 --> 00:47:52.714
I think they're also, like, a threat to Google to some extent. So, uh, OpenAI is somewhat beholden to its investors.

253
00:47:52.794 --> 00:48:03.454
Nvidia is a big investor in OpenAI, and their whole compute stack is co-designed with Nvidia and OpenAI, and they need that to work.

254
00:48:03.574 --> 00:48:16.074
The whole, the whole purpose of frontier versus open model is that there's always gonna be a market for closed models and frontier, uh, because they're better and they're more secure.

255
00:48:16.094 --> 00:48:26.294
For you to run an open model, it actually might even cost you more because you need to set up the hardware, the security, the parameters, and, and do all that stuff yourself.

256
00:48:26.334 --> 00:48:36.914
So the world, the way I see it, is, is a mixture of, you know, frontier models which are really good, like OpenAI or Claude, that are gonna do

257
00:48:38.134 --> 00:48:55.410
most of the, the heavy lifting, and then the grunt work is gonna be done by the open source models w- which could be, come from China, GLM or Kimi, or we have, uh, you know- Oh, new open source models in America from, like, I think Reflection, Reflection is a company, and then- Reflection AI, yeah...

258
00:48:55.450 --> 00:49:07.490
and then obviously Nvidia has its own open source model, Nvidia Nemotron, which is frontier, frontier-level class. So if you're a nation, say, like, you're, uh,

259
00:49:08.450 --> 00:49:20.790
a small country in Europe, say you're Italy and you wanna have sovereign AI, how are you gonna set that up? Obviously you're gonna try to buy Nvidia GPUs, but what model are you gonna run on it?

260
00:49:20.810 --> 00:49:30.250
Are you gonna run OpenAI or Anthropic and pay them billions of dollars a year in API fees, or are you gonna try to get some open source model tailored to the Italian language and culture,

261
00:49:31.310 --> 00:49:42.550
open weights and, uh, and, and modify it to, and tune it to what your country needs? And that's probably, that's what Nemotron is. It's for those people that need all that stuff.

262
00:49:42.610 --> 00:49:57.310
But then you could still also use OpenAI as a harness or, or Claude to, to do all that stuff together. So the quest- it's a mix. It's not one or the other, and I don't see open models as a threat. And to, to give you,

263
00:49:58.270 --> 00:50:10.650
uh, better color on that is the whole Hugging Face hacking incident. So you had the OpenAI model break out and hack Hugging Face, and the next thing you find out is, like, there's no

264
00:50:11.830 --> 00:50:26.210
CrowdStrike or Palo Alto Networks, there was no big publicly traded software company that could help defend against this rogue AI attack except GLM 5.2. So they had to fight fire with fire, right?

265
00:50:26.230 --> 00:50:40.710
They needed an open source model to basically take out this closed source threat. So that's why you need open source models and closed models to live together in this ecosystem to create balance and harmony, or else

266
00:50:41.610 --> 00:50:58.499
being in this closed world, we'll have more problems. So, uh, guess last question is then, um, in light of the open source model, uh, letter signatories, kind of a, a renaissance for open models lately,

267
00:50:59.570 --> 00:51:07.370
what does this mean for our main beat, the NeoClouds and the data centers, the compute providers? Um, is this bullish?

268
00:51:07.450 --> 00:51:15.820
Because right now there's a little bit of ex- existential angst here as token pricing goes down, GPU rental rates, depending on the rig, go down.

269
00:51:16.470 --> 00:51:31.970
What do you think this means, uh, in the medium term, say next year, for, uh, demand for compute? I see a token explosion. I don't see demand of compute going down. It's just only gonna increase, and

270
00:51:32.910 --> 00:51:50.450
we haven't even entered the robotics phase of, of AI at scale or edge cases, and there's still so many people that aren't even using Claude or OpenAI to do, uh,

271
00:51:51.650 --> 00:52:00.170
task work through, like, agentic capabilities. They're just using it as, like, a chatbot. Like, "Hey, what's this bug that I saw? Okay, cool. Yeah."

272
00:52:00.270 --> 00:52:14.770
Or, but, like, if you actually use it for real work or the co-work and it's sequentially doing things for you, and there are agents running in parallel and researching, like, I'm doing that stuff, and it's like, wow, this is, this is pretty groundbreaking.

273
00:52:14.850 --> 00:52:26.730
So you have basically this computer using thousands of other computers at the same time doing... and searches and doing everything and coming back and giving you answers.

274
00:52:26.990 --> 00:52:40.750
Uh, and from a NeoCloud perspective, uh, I said on the, the Jack Farley podcast, or you can see the clip, the NeoClouds, not all of them are created equal. You really need to provide value to your customers,

275
00:52:41.770 --> 00:52:55.710
and it's just not bare metal GPUs. You need to be able to serve different models at different speeds and at the same time give them the uptime. And if you even have an open source model, that's fine.

276
00:52:56.030 --> 00:53:01.170
You still have to host it and charge for it. It's not gonna kill your business 'cause you can have a blend of tokens.

277
00:53:01.870 --> 00:53:12.630
But if you don't have, uh, those services, I think, I think Nebius has a bunch of services, it's sort of like an AWS for, for AI w- with that,

278
00:53:13.530 --> 00:53:17.930
then, then you, you'll fall behind if you're just basically just selling raw compute.

279
00:53:18.470 --> 00:53:30.410
It, it, raw compute at some point will be, it's like a tradable commodity as you can see with, like, the Orange charts or the Silicon data charts. It's what are the value added services that you layer on top of it,

280
00:53:31.690 --> 00:53:38.630
and that's basically, like, a software layer that these NeoClouds like CoreWeave or, or Nebius are adding.

281
00:53:38.770 --> 00:53:51.070
Uh, but if you're just a pure Bitcoin miner that's just plugging in GPUs, I think you're gonna have some challenges in the future. And like I, I said in that video clip from, uh, Batman,

282
00:53:52.010 --> 00:54:03.110
when Bane says, like, "They expect one of us in the wreckage, brother." [laughs] And, uh, you're gonna find a few of these, uh, NeoClouds in the wreckage because you needed them to start the fire, right?

283
00:54:03.470 --> 00:54:08.550
Because no one believed in HPC and, uh, this type of compute.

284
00:54:08.890 --> 00:54:19.810
To, and to get to the, the hyperscalers like Google and Amazon and Microsoft to move, you needed these NeoClouds to exist to show, like, look, there's a real business here. You really need these types of GPUs.

285
00:54:20.230 --> 00:54:33.990
There is demand for it, right? And planting those seeds has created this new ecosystem and, and businesses that didn't exist, you know, five years ago. Ben, thank you so much for your time. Thanks for your insights.

286
00:54:34.030 --> 00:54:44.570
We covered a lot of ground. Yeah. So appreciate, appreciate you, uh, riding with us. Um, hope to have you on again sometime soon and, uh, cheers. Thanks, guys. Thank you, Ben. Have a good one. Bye, you too.

287
00:54:46.890 --> 00:55:00.750
That last comment has echoes in Bitcoin mining and- They expect one of us in the wreckage. You know [laughs]. And this gives you- Yeah... you have raw compute, and this gives you- [laughs]... power over me? Yeah.

288
00:55:00.870 --> 00:55:09.798
Um, but his point about the raw compute won't be enough in the future reminds me of Bitcoin mining pools- During the shift in the China mining ban.

289
00:55:09.858 --> 00:55:17.688
It's not a perfect parallel because in this instance, the miners were selling [lip smacks] hash rate to the pools, right?

290
00:55:17.948 --> 00:55:32.938
But the basic idea being it came to a point where margins were so thin in the mining business that if you just had a mining pool, you couldn't compete, and you had to have additional services like financial, ASIC financing, for example, firmware, et cetera, right?

291
00:55:33.438 --> 00:55:36.198
And it got to the point where the pool was the,

292
00:55:37.298 --> 00:55:47.238
was the service that attracted business and pooled business into your company, and then you used that as a vector to sell into other business models and other business lines.

293
00:55:47.558 --> 00:56:02.657
Hey, we're, we're both-- we both spent some time at Luxor. We know this playbook well. On that note, we are gonna do a quick newsflash. Galaxy Core's Applied Digital. But before that, a word from our sponsor, Ligos.

294
00:56:03.278 --> 00:56:21.558
[lasers blasting] Bitcoin-backed lenders have a sordid history because more often than not, these lenders would take your coins and then they would re-hypothecate them and stretch their balance sheet thin.

295
00:56:22.858 --> 00:56:28.658
When it came time to pay you, they didn't have the coins when you tried to liquidate your position and withdraw your funds.

296
00:56:29.378 --> 00:56:38.498
The good news is there are now alternatives in this year, twenty twenty-six, and our favorite alternative in this landscape is Ligos Finance.

297
00:56:38.538 --> 00:56:49.218
They are our preferred lender here at Blockspace because they use one hundred percent Bitcoin native smart contracts to make sure that you always control your keys. That's right. You heard that right.

298
00:56:49.658 --> 00:57:01.238
With Ligos, you don't have to worry about being the next Celsius or BlockFi victim because you always self-custody your keys when you're using their Bitcoin-backed lending services. So do yourself a favor.

299
00:57:01.378 --> 00:57:06.118
If you're looking at a Bitcoin-backed lender or you're currently using one, check out Ligos Finance.

300
00:57:06.418 --> 00:57:17.857
There's no bridging, no wrapping, no re-hypothecation, just competitive rates as low as eight point five to ten percent APR for pure self-custodied Bitcoin-backed lending. Go to ligos.finance to learn more.

301
00:57:19.618 --> 00:57:32.798
All right, Charlie, quick newsflash to end the stream here today, and we will kick it off with Galaxy Digital pricing its three point five billion dollars senior secured note round.

302
00:57:33.498 --> 00:57:44.218
We've covered this a few times on the stream. I just wanted to get this in the newf- newsflash, though, because the pricing finally hit. And the TLDR of the terms for this deal are as follows.

303
00:57:44.838 --> 00:57:51.778
The senior secured notes will be due twenty thirty-one, priced at nine point eight seven five percent for interest.

304
00:57:52.378 --> 00:58:00.378
The aggregate principal is three million five hundred and seven-- or sorry, three billion five hundred and seven million.

305
00:58:00.398 --> 00:58:12.558
They're issued at ninety-nine point five percent of par, so a slight discount for what Galaxy's receiving on the other end for the loan. Again, maturity is August first, twenty thirty-one.

306
00:58:12.678 --> 00:58:18.978
Interest is paid semi-annually, February first and August first, starting February first, twenty twenty-seven.

307
00:58:19.658 --> 00:58:27.158
They amortize semi-annually per the indentured schedule with no payments due until at least ten months after the project completion.

308
00:58:27.558 --> 00:58:35.178
And that project completion is for the phase two of its Helios site, which has CoreWeave as a tenant.

309
00:58:35.798 --> 00:58:44.278
The finance will constr-- the, the con-- uh, finance will help with construction of two buildings and eight dinners-- uh, eight dinners-- [sighs]

310
00:58:44.698 --> 00:58:54.838
eight data center halls, say that ten times fast, totaling four hundred megawatts of utility capacity, but two hundred and sixty megawatts of critical IT.

311
00:58:55.878 --> 00:59:08.068
That's again on a two hundred and sixty acre site in Dickens County, Texas, phase two of the Helios site. This, uh... Galaxy also has provided an uncapped completion guarantee for the notes.

312
00:59:08.558 --> 00:59:20.698
So if the financing falls short, Charlie, they will cover the difference. It's important to understand about this. I'll move on to our next newsflash unless you have any thoughts on this pricing.

313
00:59:21.558 --> 00:59:35.287
Let's go to Core Scientific. What's the play? Yeah, so the investment banks have been having a field day with their targets as they're want to do. This one's coming from Needham and Company, [lip smacks]

314
00:59:35.318 --> 00:59:45.798
where they have raised their price target for Core Scientific from twenty-nine dollars, which they released actually yesterday, to thirty-five dollars as of today.

315
00:59:45.838 --> 00:59:50.198
This comes on the back of the AMD deal that we covered on the stream.

316
00:59:50.218 --> 01:00:06.558
Core Scientific has entered into a hosting deal for AMD and an undisclosed Neocloud for a total consideration of five hundred and thirty megawatts in a fifteen-year deal valued at just over fourteen billion dollars.

317
01:00:06.598 --> 01:00:20.778
Both leases, again, there are two here. One is for three hundred and seventy-seven megawatts to AMD, the other is to a hundred and fift- for, uh, roughly a hundred and fifty megawatts for the undisclosed Neocloud.

318
01:00:21.158 --> 01:00:24.678
[lip smacks] There are three five-year extensions for this deal.

319
01:00:25.218 --> 01:00:42.398
And Needham here says that the lease averages about a hundred and seventy-six million in annual revenue per megawatt, which Needham calculated at about fifteen point three percent weighted yield on cost, while estimating the weighted net operating income margin of nearly ninety-five percent.

320
01:00:42.458 --> 01:00:49.388
For this, the AMD deal net operating income margin is ninety-nine percent, the Neocloud deal eighty-five percent, according to Needham.

321
01:00:50.138 --> 01:01:05.918
And as a result of this deal, Needham has increased its twenty twenty-seven revenue estimate to one point two three four billion, up from nine hundred and seventy-five million, and adjusted EBITDA to nine hundred and ten million, up from seven hundred and twenty-two million.

322
01:01:05.978 --> 01:01:16.208
Its twenty twenty-eight forecast calls for one point seven three three billion of revenue and one point three six eight billion of adjusted EBITDA after the AMD and CoreWeave contracts are fully deployed.

323
01:01:16.858 --> 01:01:29.414
The bear case here, Needham covering its bases, is- Includes construction or power procurement delays, high capital requirements, and colocation revenue arriving too slowly to offset declining mining revenue.

324
01:01:29.454 --> 01:01:44.054
That last one, maybe you'd see it, but the fact of the matter is the mining contingent really didn't have any bearing on their Q2 results. It was vanishingly small compared to their HPC revenue, which was north of,

325
01:01:45.174 --> 01:01:54.574
I believe, 100 million in the Q2 earnings. The construction costs, though, are significant here. We covered this on the stream yesterday.

326
01:01:55.094 --> 01:02:09.214
The estimated cost for the AMD and NeoCloud build will be route- about $11 to $12 million per megawatt, up from their CoreWeave, uh, retrofit in 2024, where construction costs were a lot cheaper.

327
01:02:09.664 --> 01:02:17.474
Because out of all of their peers, Core Scientific was the only one doing this. They-- Again, they pioneered this pivot from Bitcoin mining to AI.

328
01:02:17.514 --> 01:02:32.574
Now, everyone and their grandmother is trying to build one of these data centers, Charlie, as we covered, and the firms that build these things are a small piece of the overall contractor landscape in the United States.

329
01:02:32.964 --> 01:02:45.434
And you have to imagine the bidding wars [chuckles] that go on at these companies, considering there are not as many companies who have specialty in building these data centers as there are demand-- as there is demand for these builds.

330
01:02:47.194 --> 01:03:03.354
Yeah, absolutely. Um, AMD's been on a tear with these, uh, deals. Uh, I- I'll even quote, you know, a reference, uh, that our previous guest had on a different podcast. Ben talking about how AMD, which,

331
01:03:04.954 --> 01:03:21.514
uh, uh, really competes in the CPU market, that as GPUs, uh, are really constrained by memory as the industry switch to m- a- a managing agents, uh, CPUs are much better positioned.

332
01:03:21.594 --> 01:03:35.874
So, uh, perhaps these deals with AMD, uh, could lead to, uh, better long-term positioning as the models and metas for managing AI themselves change.

333
01:03:36.274 --> 01:03:50.194
I think it's very possible that, you know, um, the idea of hosting GPUs, uh, could give way to a, uh, an entire sub-sector of hosting CPUs. So, uh- Yeah, that makes sense to me.

334
01:03:50.634 --> 01:04:04.724
And one last note here before we jump to Applied Digital. Colocation revenue for CoreSite last quarter, that's their HBC segment, was $136.7 million, and their total revenue was $164.2.

335
01:04:04.854 --> 01:04:09.214
So their mining revenue accounted for less than $30 million for their Q2 results.

336
01:04:09.574 --> 01:04:20.034
Going back to Needham's, you know, potential bear cases, or rather the things that could actually not material-- or the things that could impede the materialization of the revenue they forecast.

337
01:04:20.754 --> 01:04:27.154
The fact of the matter is the Bitcoin mining is a, a vanishingly small segment of their total revenue, and it will continue to be so.

338
01:04:27.174 --> 01:04:38.954
So far, they have done a really good job of scaling that back without having any shortfalls in revenue for their Bitcoin mining segment for the entire company. So... All right. Last note from Needham here.

339
01:04:39.654 --> 01:04:43.674
An incredibly bullish call for Applied Digital.

340
01:04:44.214 --> 01:04:59.354
Needham has raised its buy rating to $83 per share for Applied Digital, implying a roughly 215% upside from the current-- or the bank's reference rate of $26.38 per share on July 27th.

341
01:04:59.394 --> 01:05:11.954
They're basing this on a 21X enterprise value to EBITDA multiple applied to their projections for Applied's discounted fiscal 2029 adjusted EBITDA estimate.

342
01:05:12.474 --> 01:05:20.774
The bank forecasts 2.33 billion of adjusted EBITDA that year, down from its previous estimate of 2.37 billion.

343
01:05:20.834 --> 01:05:33.454
Applied is also in advanced discussions, Needham points out, with two existing counterparties to cover 250 megawatts of critical IT load, 100 me- me- megawatts at Polaris Forge and 150 megawatts at Delta Forge.

344
01:05:33.994 --> 01:05:44.754
Their forecast is largely contingent on, or m- or contingent on Applied Digital executing leases for these sites. Should they do so, they believe their targets will hold up.

345
01:05:45.774 --> 01:05:53.714
This was a question analysts asked Applied Digital during the earnings call, I believe, specifically when we should expect a tenant at those sites.

346
01:05:54.414 --> 01:06:03.374
It is worth noting here that Needham is on the sell side for Applied Digital. Um, that's probably why you're seeing such a bullish target.

347
01:06:03.514 --> 01:06:15.154
Nonetheless, they have high hopes for the company, and I think it just underscores, Charlie, as earnings seasons kicks off, we're starting to see deal flow pick back up for these companies.

348
01:06:15.684 --> 01:06:25.254
It's not unusual for companies to hold some of these announcements for earnings season to try to give their stocks a pop and fizzle as they report the numbers.

349
01:06:26.294 --> 01:06:42.694
But as many have pointed out, it's unfortunate, specifically for the Core Scientific team, that they had this truly watershed deal with the little brother of GPU manufacturing for this AI boom in a time when the market's getting absolutely hammered.

350
01:06:42.734 --> 01:06:47.734
If this AMD deal had been announced a year ago, the stock would have absolutely ripped.

351
01:06:48.174 --> 01:06:58.594
But the fact of the matter is it was announced into a market that is being weighed heavily by downward price action and fears that maybe we're getting a little top-heavy in this CapEx cycle. So...

352
01:07:00.254 --> 01:07:14.974
Um, you can be the bear, I'll be the bull. Um, I'm buying here. Uh, not financial advice. On that note, thank you for tuning in to Block Space Live. We do this every weekday, 1:00 PM Eastern.

353
01:07:15.454 --> 01:07:28.564
We are Compute's daily live show, featuring quick hits on AI data centers, emerging technology, and markets. If you like what you hear on the live stream, it becomes a podcast, everywhere podcasts are found.

354
01:07:28.574 --> 01:07:45.974
And if that's not enough for you, you can go to our website, blockspace.media for all Blockspace content. That's blockspace.media. Shout out to our sponsor, CleanSpark, NASDAQ listed ticker CLSK. Thank you for listening.

355
01:07:46.714 --> 01:07:58.154
I'm Charlie. I'm Colin. We'll see you tomorrow. [outro music]