WEBVTT

1
00:00:00.060 --> 00:00:08.520
Colin, y'all. Welcome back to Blockspace Live. For a big week, Charlie. Earnings week is kicking off this week here at Blockspace.

2
00:00:08.560 --> 00:00:14.100
We've got the bulk of the companies in our, in our coverage announcing earnings this week and next week.

3
00:00:14.160 --> 00:00:24.840
So for today's show, we're going to kick off with a schedule of what y'all can expect in terms of live streaming earnings calls, content on the live stream unpacking those earnings calls, and executive interviews throughout the week.

4
00:00:25.420 --> 00:00:37.180
Following that, a quick stock market update because Christmas came early, Charlie, and it is green. It is not quite Christmas in July, Christmas in August. Everything's popping right now after a slump in July.

5
00:00:37.700 --> 00:00:49.390
Are we so back? It looks like it, but we're going to dig into the numbers just to show what the rebound is looking like, including some notes from investment banks B. Riley and Rosenblatt.

6
00:00:50.040 --> 00:01:03.000
Following that, we have our first guest on today, Eric Ellingsen of Fortitude, CFO, on to discuss the company's pivot into Zcash mining and whether or not they have AI and HPC aspirations.

7
00:01:03.460 --> 00:01:19.980
And then we will cap off today's stream by bringing back on investor and analyst Ben Pouledan to talk about why CapEx guidance might be a little wonky, considering no one can truly verify the OEM numbers for how much they're pumping out this year.

8
00:01:21.520 --> 00:01:34.850
That's right. Blockspace goes live every weekday at 1:00 p.m. Eastern. We are a compute's daily live show featuring quick hits on AI data centers, emerging technology, and markets.

9
00:01:35.320 --> 00:01:45.820
This week is earnings week, so we are covering all things earnings in NeoCloud's data centers and powered land. We will get into that in just a second.

10
00:01:45.940 --> 00:01:56.880
If you like the live stream, you can catch it on podcast form after it goes out. On anywhere you get podcasts, leave us a review. Five out of five stars, my opinion.

11
00:01:56.900 --> 00:02:06.580
And if you can't get enough of Blockspace content in audio or video form, we do written stuff. You can find that on our website, blockspace.media.

12
00:02:06.620 --> 00:02:20.980
We publish a lot of articles and a newsletter daily covering everything we cover on the podcast and more. If you like memes, we always include a meme. This show is brought to you by CleanSpark, NASDAQ-listed ticker CLSK.

13
00:02:21.140 --> 00:02:28.340
More on CleanSpark later on in the show. And guess what, Colin? It is time for the inaugural earnings update.

14
00:02:28.380 --> 00:02:45.660
[beeping] Yes, all hands on deck because we are doing the kickoff for every show this week covering earnings, and this week is a doozy.

15
00:02:45.780 --> 00:02:53.660
I think a lot of people don't actually put, like, earnings on their calendar, so they don't know that, you know, it kinda happens like, "Oh, shoot, uh, earnings are today. I should probably pay attention."

16
00:02:54.580 --> 00:03:08.020
We're covering those. We're gonna give you a forward-looking view into earnings. I've, I've got this little, um, uh, calendar forward-looking chart graphic for this week.

17
00:03:08.080 --> 00:03:20.579
So if you're listening, uh, we're covering earnings week two weeks in a row. This is week one. Um, we do Q2 or whatever quarter it is for the, you know, fiscal year. Cover- coverage for AI data centers and NeoClouds.

18
00:03:20.660 --> 00:03:33.920
We have executive interviews. This week, again, kind of a doozy. Today we have Eric Ellingsen of Fortitude Mining. Tomorrow we have Ionic Digital C-suite CEO or CFO.

19
00:03:33.980 --> 00:03:50.890
Tomor- uh, Wednesday, Patrick Fleury, the CFO of TeraWulf, and Asher Gnutt of Hut 8 on Friday. As far as when the earnings calls are, on Tuesday morning we have Cipher and Hut 8.

20
00:03:51.480 --> 00:04:06.120
Wednesday morning, TeraWulf, Riot, and Galaxy kind of all back-to-back. Early trio of earnings that morning. Thursday in the afternoon, CleanSpark and Mara. And on Friday, Core Scientific.

21
00:04:06.140 --> 00:04:09.680
We'll be streaming the majority of these to our channels.

22
00:04:09.720 --> 00:04:22.089
Make sure to like and subscribe and pay attention to both our X and our YouTube, because depending on the stream clash, we may be streaming one place and not the other. So make sure to check out those.

23
00:04:22.120 --> 00:04:37.600
We will be hosting watch parties for these tickers. Colin. Yeah, a lot of overlap, Charlie. And just to drive that point home, for the times on Tuesday, August 4th, Cipher kicks off at 8:00 a.m.

24
00:04:37.620 --> 00:04:55.320
ET, followed by Hut 8 at 8:30. On the 5th, on Wednesday, Wulf at 8:00 a.m., and then Galaxy and Riot conflicting at 8:30 a.m. all times ET. On Thursday, we have CleanSpark's earnings at 4:30 p.m.

25
00:04:55.400 --> 00:05:08.020
ET, right after market close, followed by Mara at 5:00 p.m. ET. And so far, nothing on Friday. So keep an eye out. As Charlie said, on our-- for our X and YouTube, we will be streaming these live.

26
00:05:08.080 --> 00:05:20.320
We will also unpack the earnings every day on the stream. For Tuesday, we will tackle Cipher and Hut 8, come out same day. Wednesday, we will tackle Wulf, Galaxy, and Riot.

27
00:05:20.920 --> 00:05:27.710
Thursday, we will be taking a breather because there's nothing coming out that morning and nothing at market close on Wednesday.

28
00:05:27.720 --> 00:05:31.740
But on Friday, we will be hitting the ground running with CleanSpark and Mara's earnings as well.

29
00:05:31.780 --> 00:05:46.860
And as Charlie said, we have those interviews lined up with Patrick Fleury and Asher Gnutt, so keep an eye out for those throughout the week. Muted. Charlie. And this is just week one.

30
00:05:46.960 --> 00:05:57.180
Uh, we d- we haven't, uh, rolled out, we haven't pushed week two yet, but there's a lot coming out. Week two is also very big, uh, interviews, and live streams, and earnings calls.

31
00:05:57.190 --> 00:06:04.300
So that is to say, it's a particularly exciting earnings week, Colin, because the market's ripping. And-

32
00:06:05.404 --> 00:06:16.084
Where better- It is ripping Where better else to go to see what is happening in the market than DiMetrix or DI Metrix? What's in our stockings this morning, Charlie?

33
00:06:16.144 --> 00:06:31.184
What, what pumps have we, have the little, uh, market elves left us last night? So pretty much everything. Everything is green. Um, it's not quite the rip that we saw on Friday after Leopold got liquidated. Um, but to...

34
00:06:31.274 --> 00:06:38.004
I mean, if we just look at the power shells, Marah, Right, CleanSpark. Let me actually sort these by, um...

35
00:06:39.244 --> 00:06:47.844
Yeah, I mean, Marah, Right, CleanSpark, Core, Wolf, Cipher, Hut, Bitdeer, Hive, Apply Digital, Soluna, Big Digital Energy.

36
00:06:47.904 --> 00:07:04.634
Every single thing is up from anywhere from 2.3%, um, Hive, um, Hut 1.6%, all the way up to Cipher at 9.4% up on the day. Uh, really pretty wild here to see, uh, earnings week,

37
00:07:05.624 --> 00:07:20.214
uh, kick off to such a strong start. If we wanna go down to the Neoclouds, this is where some of the numbers get even crazier. Leading the pack this morning, CoreWeave up 15.7%, Nevius up 14.2%, Iron 7.2%.

38
00:07:21.164 --> 00:07:35.384
So, uh, and then the rest of the sector, you know, DGX, SHAZ, Whi- W- uh, White Fiber, uh, all of these up, um, you know, one to 8%. So strong start to the week.

39
00:07:35.504 --> 00:07:43.204
Uh, maybe, maybe that means that all of the earnings are going to be pretty positive. You know, who knows? So toss it to you.

40
00:07:44.404 --> 00:07:53.964
Yeah, so a huge rebound, and this is a totally unverified report from Kalshi Finance, but I just have to get it up here because I feel like it's notable given the rebound.

41
00:07:54.644 --> 00:07:59.024
According to this tweet, hedge funds just bought tech stocks at the fastest pace since 2022.

42
00:08:00.704 --> 00:08:12.224
So if you're wondering what is behind this rally, you can thank Citadel for buying the book off of Leopold and the ensuing rush of, of liquidity into the system.

43
00:08:12.724 --> 00:08:27.864
I'm not sure how true this is because if you look back into July, there is this Goldman Sachs report that says, f- and this is from Bloomberg via Yahoo Finance, "Goldman says hedge funds sell US tech stocks at record pace."

44
00:08:27.924 --> 00:08:35.724
Now, this is at the end of July when everything was getting absolutely hammered. It's possible that they're rotating back in now that evaluations have become a little less dear.

45
00:08:36.404 --> 00:08:44.394
But needless to say, it's just an absolutely absurd rebound. We covered this on the stream last week when we touched on some of the earnings. Microsoft up

46
00:08:45.384 --> 00:09:00.414
like 15% after it shattered Q2 expectations, versus Meta, which was down about 8 to 9% following its earnings miss. And then you also had Amazon beat on its cloud segment as did Microsoft.

47
00:09:00.504 --> 00:09:10.584
New all-time highs for Amazon, a $3 trillion company as of this morning. I think, what is it? There's five $3 trillion companies, so Amazon joins that pack. Yeah, that's...

48
00:09:11.344 --> 00:09:22.484
If you look at the charts, it's kind of incredible 'cause it's just a cliff, like it gaps all the way up for Amazon and for Microsoft. And we covered some of those Mag Seven earnings last week.

49
00:09:22.524 --> 00:09:32.264
They're not our primary beat, obviously, so we don't put too much attention to those. But needless to say, what we're starting to see... Yeah, I mean, just look at that chart. That's, that's,

50
00:09:33.504 --> 00:09:46.264
that's insane price action for a $3 trillion company. Amazon up 4.5%. I mean, this is a $3 trillion company market cap, three and some odd trillion, uh, h- up big right at the open of the day.

51
00:09:46.504 --> 00:09:52.264
So- And, and the takeaway for me last week, Charlie, was looking at some of these earnings reports from the big dogs.

52
00:09:52.564 --> 00:09:59.724
It's that investors are starting to really ask what the feasible ROI will be for all this CapEx that they're spending.

53
00:10:00.284 --> 00:10:10.124
If you looked at Meta, they had an earnings miss even though they had a revenue beat, but their CapEx guidance was nudged upwards by about five to 10 billion for the full year 2026.

54
00:10:10.644 --> 00:10:25.024
Microsoft, on the other hand, had as your, its Azure cloud business line completely destroy expectations, over 30% increase in revenue, and its CapEx guidance fell, but that was just some accounting shifting.

55
00:10:25.063 --> 00:10:38.424
They basically took it from, um, the capital expense bucket into another operating cost bucket. But we're-- I do think that we have moved on to a new, a, a, an...

56
00:10:38.724 --> 00:10:47.564
We've moved on to a new phase in the market where announcing the deals isn't enough to move the needle. We saw this recently with Core Scientific and CleanSpark.

57
00:10:47.624 --> 00:11:00.044
They had two banner AI deals inked, and the stocks did not really move as aggressively as we saw in year, i- in the months prior and in last year when a lot of these deals were fresh.

58
00:11:01.544 --> 00:11:11.724
I mean, y- you say that, but, uh, and maybe the market was just a little cautious at the time because Korea was driving the market, whatever. Maybe Iran was back on the front of everybody's minds.

59
00:11:11.844 --> 00:11:23.564
Perhaps people are thinking that things are over-leveraged. But at the same time, I mean, look, this morning, again, you said B. Riley lifted CleanSpark to a $26 price target. What is it right now?

60
00:11:23.644 --> 00:11:39.074
It's, uh, that's over, it's, uh, over double, so it's at 14 right now. So it's basically 90% upside target. You have B. Riley also raising the Terawulf price target to 40, so which is 124% upside. Um,

61
00:11:40.824 --> 00:11:50.484
uh- I'm glad you, you mentioned this, Charlie, 'cause it's a good segue into some of these notes coming out before earnings, which we'll go ahead and tackle. Now, caveat here.

62
00:11:50.524 --> 00:11:59.424
Whenever you see these, these investment bank reports, they're almost always on the sell side, so they're going to try to have higher price targets because they have a financial stake here.

63
00:11:59.444 --> 00:12:08.964
Yeah, they're trying to sell you the stock, but maybe they're right. [chuckles] Yeah. Maybe they're, maybe they are right, and if you look at some of these in 2025, they would've been vindicated, right?

64
00:12:09.184 --> 00:12:18.382
Because a lot of those stocks ended up soaring. But we'll start with CleanSpark here really quickly. This is coming to y'all from Blockspace. And per what Charlie just said, B.

65
00:12:18.512 --> 00:12:23.432
Riley raising its price target from nineteen to twenty-six. This was on Friday.

66
00:12:23.452 --> 00:12:32.132
It maintained a buy rating, and the target is an eighty-one point three percent upside from the closing price early Monday of fourteen dollars and thirty-four cents per share.

67
00:12:33.692 --> 00:12:45.512
What's driving this is not the Sandersville deal, though. B. Riley is arriving at this price target based on its assets in Texas, CleanSpark's two data centers in Texas.

68
00:12:46.452 --> 00:13:01.912
The Sealy campus, they've assigned an additional HPC value to the Sealy campus of three point eight nine billion and seven point zero two billion to the Brazoria County campus. Those figures use a fourteen x

69
00:13:02.992 --> 00:13:20.072
multiple to, um, to the, uh, uh, uh, excuse me. Those deals use fourteen times modeled NOI below the multiple assigned to the Sandersville deal that they model in this, uh, research note as well.

70
00:13:20.132 --> 00:13:30.772
And, um, they are discounting, uh, capital expenditure estimates of about two point five nine billion for Sealy and five point four nine-- or four six billion for Brazoria.

71
00:13:30.812 --> 00:13:39.332
They also have this note out on Wolf, which is a little scanter in terms of looking at potential, uh, you know, potential future HPC revenue.

72
00:13:39.912 --> 00:13:46.392
But they did raise their price target from thirty-two to forty and remain, uh, remaining with a buy rating.

73
00:13:47.372 --> 00:13:58.352
Um, the eight percent increase lifted the bank's target by twenty-five percent, and the new target applies about a hundred and twenty-four percent upside from the reference p- uh, rate when we wrote this article.

74
00:13:59.512 --> 00:14:08.692
And this comes on the heels of TerraWolf's July sixth announcement for the twenty-year lease with Anthropic at its justified data center in Hawesville, Kentucky.

75
00:14:08.732 --> 00:14:24.052
That's a nineteen billion dollar deal over about four hundred and one megawatts of critical IT load. There's also a note out on Galaxy, which we don't have up on the site right now. But

76
00:14:25.512 --> 00:14:36.052
the note itself comes from Rosenblatt, and they are maintaining a buy rating, but they are cutting the price target. Price target went from thirty-nine to thirty-five.

77
00:14:36.612 --> 00:14:43.952
Current price for Galaxy is about twenty-one dollars per share, and the price target cut is valuation me-mechanics driven.

78
00:14:44.032 --> 00:14:55.912
Rosenblatt is applying a twenty-three X multiple to a lowered twenty twenty-eight adjusted EBITDA estimate for Galaxy, which is-- it conflicts with a peer average of about eighteen X.

79
00:14:56.672 --> 00:15:10.512
And this is largely driven by the HPC hosting mix from Galaxy. As y'all might recall, they have their CoreWeave deal, but they also have added a second site in McGregor, Texas,

80
00:15:11.492 --> 00:15:22.512
originally specced for twenty-four megawatts, so they can be under that batch cero threshold there in ERCOT. But they have aspirations to scale up to three hundred megawatts at that site.

81
00:15:22.532 --> 00:15:37.432
What was interesting to me, Charlie, with this Rosenblatt note with Galaxy, though, is they basically reset all of their valuation metrics for the company or their estimates, and they didn't really trim them as much as they totally revamped them.

82
00:15:38.112 --> 00:15:47.812
So Q2 2026 revenue, they dropped their estimate from eleven point zero eight billion to seven point three seven billion.

83
00:15:49.752 --> 00:15:57.012
Fiscal year, uh, full year 2026 revenue, they estimated at thirty-one point two five billion, down from forty-nine billion.

84
00:15:57.672 --> 00:16:08.072
For the fiscal year 2027 end, they have the target of thirty-one point four three billion versus sixty-five point eight four billion for, for the, uh, previous estimate as well.

85
00:16:08.092 --> 00:16:20.052
They also adjusted EBITDA down pretty significantly, including flipping EBITDA negative for the full year of 2026, down from seventy-three point one million to minus three hundred thousand.

86
00:16:20.932 --> 00:16:22.712
I would imagine a lot of that is

87
00:16:23.812 --> 00:16:36.232
them taking into account the hammering that the crypto market is having currently, 'cause Galaxy still has a lot of hands, not just within the HPC business, but within the realm of crypto financial services as well.

88
00:16:36.292 --> 00:16:52.392
So... Yep. Um, we are gonna keep on going. We have DeepSeek in Mongolia. Fun little story this morning. Get to talk about, uh, the landscape of is Mongolia the, the Texas of Asia?

89
00:16:53.732 --> 00:17:05.912
Uh, more on that after a word from our sponsor, CleanSpark. [gentle music] We are CleanSpark, America's Bitcoin miner. A publicly traded company with the largest operating hash rate,

90
00:17:07.392 --> 00:17:19.632
powered entirely by self-operated infrastructure across four states. This is our proof of work, and we are setting the standard for what's next.

91
00:17:20.432 --> 00:17:36.192
Learn more about the intersection of energy and Bitcoin at cleanspark.com. You are-- You, you're muted. Thank you, Charlie.

92
00:17:36.252 --> 00:17:42.332
As you said, this one's a fun story. Coming at y'all from Blockspace by way of Bloomberg and the Japanese Times.

93
00:17:42.392 --> 00:17:48.651
I believe the Japanese Times prob-- or Japan Times probably reported it first, and then Bloomberg did supplementary reporting.

94
00:17:48.692 --> 00:18:02.592
But that is DeepSeek targeting one gigawatt of AI data center expansion in Inner Mongolia, a region that old head Bitcoin miners will remember from the heady days of China's mining dominance.

95
00:18:02.632 --> 00:18:14.012
We'll get into as-- We'll get into some of the numbers as to why that's happening. The region is very coal-rich and one of the, uh, electricity generating powerhouses for mainland China.

96
00:18:14.952 --> 00:18:18.852
But this is coming to y'all from Will Foxley here at Blockspace.

97
00:18:19.640 --> 00:18:32.830
Article says here, quote, "DeepSeek is planning roughly one gigawatt of computing capacity in Ulanqab, Inner Mongolia, through a combination of company-owned infrastructure and capacity leased from other operators.

98
00:18:34.030 --> 00:18:47.370
The Hangzhou-based AI startup wants at least part of the capacity operating by the end of 2027," so they must have just gotten started recently, "or early twen-2028, according to reports from Bloomberg's and the Japan Times."

99
00:18:48.640 --> 00:19:06.980
Notably, Charlie, the chip configuration has not been identified, and we covered stories of, or a Bloomberg report last week that ins-insinuated that Moonshot, the company behind Kimi K3, was actually using roughly 20K NVIDIA Hopper

100
00:19:08.160 --> 00:19:25.560
generation chips from Alibaba for the Kimi K3 model, and this ran afoul into a bunch of [tsks] regulations and export controls that the US currently has over leading-generation NVIDIA chips to Chinese companies.

101
00:19:26.260 --> 00:19:28.230
As we covered on the stream, there are a few workarounds.

102
00:19:28.320 --> 00:19:44.740
Chinese companies can rent capacity outside of China for these leading-generation chips, but they're not allowed to have the bleeding-edge NVIDIA chips for training or for inference for their AI in their own warehouses, in their own data centers.

103
00:19:46.180 --> 00:19:50.060
And so the question for this then becomes, what is DeepSeek going to use?

104
00:19:50.150 --> 00:19:56.780
And of course, the reporting was scant here because I don't think this is going to be something that you can just finagle out of sourcing very easily.

105
00:19:57.139 --> 00:20:08.920
[sniffs] But the, uh, the question is whether or not they're going to use Huawei's Ascend line, which is China's leading domestic alternative, or stick with NVIDIA. Um,

106
00:20:09.840 --> 00:20:19.480
DeepSeek has used Huawei in the past, but early reporting indicated that it ran into difficulties training on Huawei Silicon and has continued to lean on NVIDIA systems.

107
00:20:20.000 --> 00:20:26.000
The chip mix, again, for the one gigawatt build-out has not been disclosed, so there's some tension here between regional and domestic.

108
00:20:26.820 --> 00:20:34.440
Charlie, I'll toss it to you before we get into some of the rationale for why Inner Mongolia has been ch-chosen for this gigawatt build-out.

109
00:20:35.360 --> 00:20:47.380
Um, yeah, and I'll chime in on that too, but, um, I think some context is important. So DeepSeek, from which the term DeepSeek moment, the industry reckoning, what, 18 months or so ago.

110
00:20:47.500 --> 00:21:01.740
Um, DeepSeek wa- is vertically integrating. Now, this was a company, a Chinese near-frontier model company, who made its name on being like the efficient model, the, the token-efficient model. Super, super cheap.

111
00:21:02.260 --> 00:21:10.160
You could basically buy your intelligence and inference, uh, just for pennies on the dollar compared to other, uh, providers.

112
00:21:11.140 --> 00:21:23.400
Uh, so and that's because they kind of supposedly trained their model on a shoestring budget, but now they're going full integration, so this is a, a, a very CapEx-heavy decision, as we assume.

113
00:21:23.520 --> 00:21:32.600
Uh, and, uh, it does challenge this narrative that DeepSeek almost kind of created 18 months ago, which is, you don't need massive compute.

114
00:21:32.960 --> 00:21:41.360
And that had caused like a tanking of the industry, NVIDIA and others, for like a week or two, which was quickly rebounded back.

115
00:21:41.400 --> 00:21:54.150
But basically, DeepSeek, the company who once challenged the idea or caused the industry to wonder if the idea that we need large compute and large CapEx, um, is now leaning into this.

116
00:21:54.200 --> 00:22:09.520
So, uh, this is kind of a reversal, an acknowledgment, an admission, uh, through co-corporate strategy and announcements that, uh, yeah, everybody short compute, including the company who everybody thought demonstrated that we didn't need as much compute.

117
00:22:09.920 --> 00:22:19.200
So I'll toss it back to you to talk about Inner Mongolia. I have a map pulled up if you want, uh- Yeah, can you go ahead and toss that up here? Yeah, 'cause this is really-- I think this is helpful for people. Um,

118
00:22:20.120 --> 00:22:34.540
Inner Mongolia is not the country of Mongolia. It's the province of China, like a large state. The province of Inner Mongolia, which is the large province in northern China, which borders Mongolia.

119
00:22:35.979 --> 00:22:53.720
I call it, I think, maybe thinking of it like the Texas of China may be kind of accurate in that it's sparsely populated, it's resource dense, and it has tons of power distribution and power generation-related resources.

120
00:22:54.700 --> 00:22:55.140
To you. Yeah.

121
00:22:55.720 --> 00:23:09.660
Yeah, it's about 350 kilometers northwest of Beijing, so it'll be just close enough, a little bit further out maybe than you'd want, but it'll be close enough to service computing needs for that city center.

122
00:23:09.690 --> 00:23:21.600
[tsks] And it has a naturally cooler climate that makes it beneficial for running these data centers, and it is flush with coal. So much so that there's a crowded field of companies in there.

123
00:23:21.680 --> 00:23:31.120
Alibaba, UCloud, VNet Group, Zhonglian Data have all set up projects there, and now DeepSeek moving in.

124
00:23:32.920 --> 00:23:46.720
[tsks] Um, Bloomberg reports Inner Mongolia as a whole had 28 large/medium computing and data center projects under construction or renovation in H1 2026, totaling an estimated 117.6 billion renminbi in investment.

125
00:23:47.100 --> 00:24:04.860
[sniffs] And a June 25 regional disclosure had already pegged this area that DeepSeek is moving into at 66 major computing projects with 71,000 petaFLOPS of aggregate capacity, over 90% AI related, with DeepSeek by Dance and JD.com workloads already running there.

126
00:24:04.880 --> 00:24:18.060
So the reason why all of these companies are flocking to Inner Mongolia is because it is a electricity-generating powerhouse for China. Total installed ca- power capacity [tsks] is estimated at 240 million kilowatts.

127
00:24:18.100 --> 00:24:35.812
That's 240 gigawatts, which is the largest of any Chinese province. Coal makes up the lion's share of- Of aggregate megawatt hours produced, it's 117 gigawatts of the whole- of the nameplate load.

128
00:24:35.852 --> 00:24:46.252
But in 2025, coal-fired p- powered plants produced 590 billion kilowatt hours versus solar and wind, which generated about half that at 277 billion.

129
00:24:46.272 --> 00:24:55.252
And this despite the fact that solar and wind actually make up a decent chunk, about 120 gigawatts of the nameplate power in the region, but they're intermittent, so you don't get all of that power obviously.

130
00:24:55.292 --> 00:25:00.992
I mean, it's Mongolia. It's the steppes, the windy steppes of Mongolia. Of course there's wind. Yeah.

131
00:25:01.072 --> 00:25:09.762
I mean, so they've got a bunch of wind there, they got a bunch of solar there, but the coal base load is the thing that will really make this region shine for AI data center development.

132
00:25:10.352 --> 00:25:22.252
In fact, Inner Mongolia is the key node in China's West to East Power Transmission Project. In 2025, roughly 40% of its generation was exported to other provinces.

133
00:25:22.792 --> 00:25:31.572
So there is a bumper crop of electrons in this area, and now that will go towards these AI data centers that are setting up shop there.

134
00:25:31.592 --> 00:25:38.332
The coal reserves in the region total roughly 536.5 billion tons, which is nearly a third of China's national total.

135
00:25:38.912 --> 00:25:53.212
So the fuel security in the region is very, very strong, and they're [laughs] not going to run out of any raw materials for producing this. Or sorry, they're not gonna run out of fuel anytime soon for these data centers.

136
00:25:53.272 --> 00:26:02.692
And, you know, as, uh, we- I touched on at the beginning, Charlie, Inner Mongolia was one of the bedrocks of Bitcoin mining when Bitcoin miners still used to be dominant in China.

137
00:26:02.732 --> 00:26:11.252
In fact, that's where most of the miners still are. The ones that are left in the nation are concentrated in this region because it is so flush with power.

138
00:26:11.932 --> 00:26:19.852
So we'll see how long they stay with all of the AI data centers starting to come in and crowd the trade for energy. Yeah. I'll repackage a bit of what you said.

139
00:26:20.032 --> 00:26:34.672
Basically, um, if we want to compare, uh, West Texas, like ERCOT interconnection queues with Inner Mongolia, um, also the way to pronounce that city is Ulanqab. I had to look it up real quick. Thank you. Ulanqab.

140
00:26:35.172 --> 00:26:54.592
Um, the distance from this, uh, base load that DeepSeek is looking at to, say, a major hub like Beijing, um, the distance is roughly the same distance as Abilene, Texas to Dallas, Texas, just for like a little, like, you know, visual, uh, and geographical comparison.

141
00:26:54.632 --> 00:27:05.252
The difference though, between Inner Mongolia and Texas, is that while Texas has huge demand, it, it means large and long interconnection queues.

142
00:27:06.372 --> 00:27:24.472
May not be the same case with Inner Mongolia in that they already have a lot of the installed capacity, just because, for whatever reason, China likes to, as a command economy, build a lot of things, and, uh, in this case, it may actually work out pretty well for them.

143
00:27:25.392 --> 00:27:38.452
Um, I know that some pe- some folks may say, "Oh, this is green computing," because it... You know, there's a lot of installed wind capacity in Inner Mongolia. I don't know if I think that's entirely true.

144
00:27:38.692 --> 00:27:51.542
As you said, Colin, this is gonna be a coal-powered, uh, energy mix. Yeah. The people who fawn over China as some poster child of the Green Revolution are just deluding themselves.

145
00:27:51.582 --> 00:28:01.072
'Cause they'll point to charts that show how much solar and wind they're building, which is true. But if you look at- They've got, they've got, they've got solar, you know, covering mountains and everything, but- Yeah...

146
00:28:01.472 --> 00:28:09.852
it's not connected to anything, and it- Yeah. It's not connected to anything. And also, if you look at how much coal they're building, it dwarfs the green generation that they're rolling out.

147
00:28:09.912 --> 00:28:16.372
Also, China's not doing this out of the goodness of their heart, or I should say Beijing, the CCP are not doing this.

148
00:28:16.392 --> 00:28:26.512
They're doing it because they produce all the world's solar panels, and they subsidize that production. They have a lot left over, so they're just gonna deploy it for their own use cases. So. All right.

149
00:28:26.612 --> 00:28:35.352
We h- we have Eric Ellingsen of Fortitude in the wings. We'll bring him on up here shortly after a word from our sponsor, Luxur.

150
00:28:36.112 --> 00:28:47.932
[whooshing sound] This episode of Blockspace Live is brought to you by Luxur's Commander, Bitcoin miner management software for enterprise operations.

151
00:28:48.531 --> 00:28:53.982
Luxur's Commander gives you real-time fleet monitoring, bulk remote commands across your fleet, and Intelligent Miner.

152
00:28:54.492 --> 00:29:01.432
That's an automated profitability engine that runs every five minutes and tests your fleet's power settings against live energy and hash rate markets.

153
00:29:02.052 --> 00:29:15.792
In fact, ERCOT backtests show 10% improved profitability with intelligent mining versus binary mining. Commander Pro is $100 per megawatt or a 25 basis point pool fee adder, but you can also try it free for 60 days.

154
00:29:15.852 --> 00:29:30.912
If you'd like to learn more, go to luxur.tech/commander to get started. All righty, Charlie, let's get Eric Ellingsen, CFO of Fortitude, on the stage. Eric, welcome to the show, sir. How you doing? I'm doing well.

155
00:29:30.952 --> 00:29:42.102
Doing well. Thanks, Colin, Charlie. Good to spend time with both of you guys. Appreciate you having me on today. Yeah. Appreciate you taking the time, man. It's, uh, been a long time coming. We've been- [laughs]...

156
00:29:42.132 --> 00:29:51.332
covering Fortitude on the, on the website for quite some time after it was spun off from Foundry, started getting its own footing.

157
00:29:51.432 --> 00:30:02.852
So this was a great opportunity for an update on some of the stuff y'all are planning, especially with this merger that y'all are doing with Heart Sciences for a public listing, and I think that's a good place to start.

158
00:30:03.592 --> 00:30:13.822
What's driving this decision, and what is the market appetite right now for a crypto mining equity when most of the sector is pivoting towards AI? I, I think that's probably forced- Oh...

159
00:30:13.872 --> 00:30:22.952
first and foremost on, in investors' minds. Is... And also, if you can give a rough timeline for when y'all think this will be completed. Absolutely. No, again, good to be here.

160
00:30:22.982 --> 00:30:25.192
And, uh, yeah, excited to talk about the transaction.

161
00:30:25.252 --> 00:30:37.372
So not unlike some of our mining brethren before us, uh, we are looking at a reverse takeover transaction with Heart Sciences, ticker symbol HSCS, as a vehicle for us to enter the public markets.

162
00:30:38.082 --> 00:30:49.862
I think, you know, why was kind of a very specific question from you, Colin, on, on why is kind of now the right time relative to appetite for the market considering mining companies.

163
00:30:50.122 --> 00:30:58.342
Um, from our perspective, and, and Barry talks about this often across podcasts and, and Twitter, he likes to zig when others are zagging, right?

164
00:30:58.402 --> 00:31:06.061
And so as folks are leaving a market, we see a gap in the space for us to enter from a mining perspective.

165
00:31:06.182 --> 00:31:12.962
And I think we think about mining a little bit differently than some of the larger folks that have been in the space for an extended period of time.

166
00:31:13.502 --> 00:31:23.722
We're focused primarily on Zcash, and so the mining economics of Zcash look very similar to that of really early days Bitcoin mining. And so when you think about, you know...

167
00:31:23.782 --> 00:31:34.202
And I wanna talk mining holistically across the space as opposed to talking about kind of our financials specifically. But if you take a look, Zcash Info is a great resource by one of our sister companies, Foundry.

168
00:31:34.582 --> 00:31:40.782
But if you take a look at Zcash Info, you could see some of the very specific economics tied to mining Zcash.

169
00:31:41.402 --> 00:31:50.902
And at 80-plus percent margins on any given day for the Z15 Pro, it's a really, it's a really attractive time for us to be mining the Zcash token.

170
00:31:51.602 --> 00:32:03.182
And as you think about kind of the profile of a mining company over the longer term, there's a need for power assets and there's a significant need to purchase ASICs. And appreciate you bringing that up, too, Charlie.

171
00:32:03.202 --> 00:32:06.642
Then there's a need to purchase ASICs and mining equipment as well, right?

172
00:32:06.812 --> 00:32:13.562
And so if you go to the miner tab in particular, you can see the profitability of some of the units kind of directly with the Z15 Pro being at the top of the page.

173
00:32:13.942 --> 00:32:25.112
So as, you know, as we speak, that 10 cent power cost, which he has listed as the electricity rate above, we're still right at close to 80% mining margins.

174
00:32:25.142 --> 00:32:35.182
And of course, we're at something less than 10% today in our power cost, and we've guided in our investor relations deck is bringing our cost to mine a Zcash token down from $70 to $40.

175
00:32:35.242 --> 00:32:45.242
So a little bit of a long-winded answer on the front end, but I think economics are the reason why. And then why, why enter the public markets? Fundamentally, it's, it's, it's a capital to growth story for us.

176
00:32:47.182 --> 00:32:52.402
I'm glad you talked about the economics, 'cause that was actually gonna be my next question.

177
00:32:52.442 --> 00:32:58.342
When you look at a company's decision to go for something like Zcash versus Bitcoin mining, that's the first thing that comes up.

178
00:32:58.402 --> 00:33:08.602
And have, have y'all done numbers on Bitcoin recently in terms of what the margins are similarly for, like, new hardware, just to give y- listeners a p- uh, kind of a, you know, perspective there? Yeah.

179
00:33:08.642 --> 00:33:16.822
Margins in the Bitcoin space are, depending on power cost, anywhere between 10 and 30% kind of contribution or gross mining margins.

180
00:33:16.902 --> 00:33:27.242
And as mentioned, Zcash is in that, kind of depending on your power cost, between 80 and 90% contribution or gross mining margins, depending again on, on the underlying power cost of the business.

181
00:33:27.742 --> 00:33:40.162
And that 80% was based on, uh, 10 cent power, and I believe on one of Fortitude's press releases, y'all are coming in somewhere around 4 cents a kilowatt hour. Am, am I correct about that? You are.

182
00:33:40.242 --> 00:33:48.162
So we're in b- we're in that four to five cent range, um, and that's, that's based on kind of the hub that we're building in real time in Nebraska.

183
00:33:48.222 --> 00:34:18.432
So we were super excited to announce a recent greenfield site, um, that we energized in Grand Island, Nebraska, and we had purchased earlier this year another site in Aurora, Nebraska, and we're working, uh, to kind of build a hub outside of, again, kind of that theme of zigging while others are zagging in terms of being outside of, of Texas in some less crowded markets that are still kind of very open to an interruptible rate, which fundamentally is the thesis behind all, all crypto mining.

184
00:34:19.342 --> 00:34:26.972
C- can you kind of speak to that hunt for power a little bit, Eric, i- in the sense that you said you're kind of looking outside of some of the more crowded markets?

185
00:34:27.542 --> 00:34:41.062
I mean, what, what, what is the, what is the edge that crypto miners are looking for right now to secure power in such a constrained environment with all of these data centers gobbling up every electron they can find?

186
00:34:41.202 --> 00:34:53.562
Yeah. [laughs] Um, the edge and the differentiating factor for crypto versus the kind of reliable, redundant load required in the AI HPC space

187
00:34:54.462 --> 00:35:01.582
is our flexibility and be able to kind of, you know, shut off and, and curtail when the market requires.

188
00:35:01.602 --> 00:35:15.622
And so that continues to be a theme in both Bitcoin and kind of all of, all of crypto mining, including Zcash as well. And so others are looking, again, for that stable, reliable load, and even redundancy tied to it.

189
00:35:15.682 --> 00:35:34.262
So a lot of, again, a lo- a lot of the larger miners that are pivoting in real time are looking to purchase, right, cogen sets and things like that, that are gonna add redundancy to their grid-connected power today, whereby we're, we're, we're comfortable working directly with the utility to, to put in place a curtailment strategy.

190
00:35:34.982 --> 00:35:40.402
I think the other theme is we're working to find and fill kind of smaller pockets of power.

191
00:35:40.762 --> 00:36:00.182
If you ask specifically kind of outside of the ERCOT grid in, in particular, I think there are agrarian economies across the Midwest that have power infrastructure that supports really kind of twice per year use, and that's harvest season and plant season, again, for a lot of the agrarian economy.

192
00:36:00.762 --> 00:36:17.142
That infrastructure is there and primed to be coupled with, you know, all of crypto mining. So, um, I'm curious specifically about this Grand Island site. And, um, you know,

193
00:36:18.082 --> 00:36:25.402
so the curtailment piece seems to have been pretty big about coming into this area and saying, "Hey, we're not going to drive up power costs.

194
00:36:25.442 --> 00:36:36.202
We're going to make sure that we're shutting off when appropriate," and n- kind of dispelling fears of, of these data centers- Yeah... driving up costs. Yeah. I'm curious, um,

195
00:36:37.322 --> 00:36:50.938
at, uh, I'm curious with regards to some of the moratorium pushback that we've seen in certain jurisdictions and localities. How much of that is- impeding or helping with Fortitude's ability to procure, procure power?

196
00:36:51.058 --> 00:37:01.778
Specifically from the standpoint of, it seems like- Um... most of the hire is directed towards the AI data centers, and there might be a case for crypto miners with that curtailment piece to kind of duck that entirely.

197
00:37:02.118 --> 00:37:12.918
So making it easier, say, "Hey, we're not the same as these guys." Does it make it just a little bit harder considering there is more scrutiny on any power use? What have y'all found? Yeah. I think, um,

198
00:37:13.838 --> 00:37:18.058
this is such a cop-out answer, but it's on a case-by-case basis, Colin.

199
00:37:18.178 --> 00:37:28.938
And, and what I mean by kinda case-by-case basis is if people are willing to listen and understand the nature of the load that we're utilizing, like Ryan Schmitz did specifically at Grand Island, right?

200
00:37:29.018 --> 00:37:44.418
He got on, um, you know, on, on the local news to talk about how this is a great thing for the local economy, and an ability for them to bring in additional revenue and utilize power infrastructure that, again, is only utilized on a couple occasions per year.

201
00:37:44.538 --> 00:37:58.298
So the interruptible load and contract that we signed with the folks in Grand Island, we're, we're really excited about working with, with Ryan and team at GIUD and Grand Island to, to, to build that infrastructure, and now energize that particular infrastructure.

202
00:37:58.678 --> 00:38:16.158
So I think if folks are willing to listen to the benefit that we can bring to the grid and the additional revenue that can be generated for that local economy, not only it's us hiring people, but there's a, there's a construction component to what we were doing in addition to paying for power costs that's supporting, you know, Grand Island and the Grand Island community.

203
00:38:16.278 --> 00:38:31.198
So, um, for folks that are willing to listen, you know, we're willing to partner with everybody from, you know, independent power producers and IPPs across, you know, all the way down to, uh, kind of the local co-ops that, that, that, that wanna provide us, you know, smaller loads.

204
00:38:33.138 --> 00:38:41.978
I'll, uh, throw out kind of a wrapping up question, which is obviously it's case by case. Uh, and- Yeah... and you've, again, zigged when others have zagged.

205
00:38:42.598 --> 00:38:57.778
But does Fortitude have any plans to pursue AI HPC co-location at all? Can you speak to this at all? Um, Charlie, we're not, we're not right now. Um, so I think, you know, it's a never say never mentality, right?

206
00:38:57.858 --> 00:39:10.078
I think when we, when we talk about our business, and one of the cool themes and the, one of the reasons I joined, you know, less than, a little less than a year and a half ago now, was this mindset of being a returns maximalist.

207
00:39:10.818 --> 00:39:22.138
And so I think the team at DCG and, and us at Fortitude are always open to ideas, um, and wanna take that kind of return-on-investor capital mindset first.

208
00:39:22.738 --> 00:39:32.638
And if that, over the long term, means that we're gonna explore opportunities, I think the group would be open to it. Awesome. Eric, thank you so much for joining, man.

209
00:39:32.798 --> 00:39:39.778
Really appreciate the time, and, uh, we'll have y'all back on as things start to develop for Fortitude. Appreciate it.

210
00:39:39.858 --> 00:39:49.008
Yeah, exciting times ahead of us too, so excited to come back on and, and keep you guys abreast of everything we're working on. Thanks, Colin. Thanks, thanks, Charlie. Appreciate the time. Thank you, Eric. Cheers.

211
00:39:50.498 --> 00:39:59.278
Love that guy. Love a curve ball, uh, mining strategy. Um, very compelling. Uh, we are gonna keep on rolling. We have Ben in the audience.

212
00:39:59.288 --> 00:40:13.698
We're gonna bring him on, up here in a moment to talk about chips and what it means for AMD's new chip in his letter that he put out this morning. Before that, a word from our sponsor, Ligos.

213
00:40:14.398 --> 00:40:39.658
[static] If you've been in Bitcoin for a while, you've read headlines of another fund being blown up, of crypto custodians losing funds 'cause they were playing funny games with their clients' coins, and lenders just not being there with due diligence and with proper risk management when dealing with other people's money.

214
00:40:40.498 --> 00:40:49.258
The good news is, if you're looking for a Bitcoin-backed lender, you now have options, and our preferred option here at Blockspace is Ligos Finance.

215
00:40:49.798 --> 00:40:59.538
Ligos Finance is a Bitcoin native, smart contract-driven platform where you always have control over your keys when you're lending with them. That's right.

216
00:40:59.918 --> 00:41:08.007
With Ligos Finance, if you're using their Bitcoin-backed lending services, you never relinquish control of your keys during the loan process.

217
00:41:08.478 --> 00:41:21.018
With Ligos, there's no rehypothecation, no wrapping, no bridging, no funny business. Just competitive rates as low as 8.5% to 10% APR on fully custodied, fully self-custodied Bitcoin-backed lending.

218
00:41:21.438 --> 00:41:34.378
Go to ligos.finance to learn more. All right, we got Ben Paladian in the wings. We're gonna bring him up here again. Gonna run it back. Ben, welcome back to the show. Running it back. Welcome. Good.

219
00:41:34.478 --> 00:41:43.658
Glad to be back with you guys. I missed you guys over the weekend, and I'm, I'm happy to be here again. [laughs] Yeah, and you were busy over the weekend. I mean, you went, uh...

220
00:41:43.718 --> 00:41:55.318
You were type- fiercely typing up this, uh, letter to AMD CEO Lisa Su, um, which I wanna focus on. Like, I was reading through this.

221
00:41:55.378 --> 00:42:04.648
Again, had to use the, all the LMS at my disposal to, uh, get, catch up on, uh, chip architecture. But I think I'm down...

222
00:42:04.698 --> 00:42:17.098
And this is, uh, v- very interesting because there's, you say there's about half a trillion, uh, in CapEx, in AI data center spend allocated based upon self-refor- self-reported, unverified vendor benchmarks,

223
00:42:18.918 --> 00:42:31.978
and you have some open questions. So give me the TLDR of, like, your takeaways from the AMD, uh, conference and presentation, and the overall thesis of, like, what you wanna find out. Yeah.

224
00:42:32.018 --> 00:42:36.098
Basically, I'm a, I'm, I'm a pretty well-known Nvidia shareholder.

225
00:42:36.178 --> 00:42:48.988
So I was at the A- AMD AI event a few weeks ago in San Francisco, and then they put on the screen that their CPU's, you know, 20% better than Vera, Nvidia Vera CPU on single-core performance.

226
00:42:49.078 --> 00:42:59.452
And I was like, "Hmm, okay, like- Let's bring the competition back because, uh, you know, back in the '90s and 2000s, it was always Intel versus AMD on the CPU stuff.

227
00:42:59.562 --> 00:43:12.022
Now on the AI data center front, I think it's Nvidia versus AMD on the CPU. So I went through their whole deck to see, like, what they were talking about, and then I couldn't find the chip that they were talking about.

228
00:43:12.042 --> 00:43:22.782
There was no model number, no wattage, no price, and it's not even on their price list. And then this Tom's Hardware also said the same thing, that this, this chip doesn't exist. So I was like, you know what?

229
00:43:23.882 --> 00:43:34.532
Like, this is something important. I, I reached out to a few friends in the industry and I said like, "Am I like, like making sense here? Um, or am I seeing things?"

230
00:43:34.622 --> 00:43:40.942
He's like, "No, you're right, Ben, like this isn't anywhere." So I said, "You know what? They're reporting earnings on Tuesday, let me

231
00:43:42.382 --> 00:43:51.182
use my social media and, and, uh, publish a letter to see if she can, uh, respond and see what it is." I'm, I'm just open to the discussion. I just...

232
00:43:51.702 --> 00:43:56.262
When people put up these slides of all these footnotes, I just want people to back it up with, like, real data.

233
00:43:56.332 --> 00:44:06.361
'Cause when I ran my LED lighting company, we, we submitted everything to, like, a third-party, uh, testing agent, uh, place, and they would measure lumens per watt.

234
00:44:06.402 --> 00:44:20.762
And then this, and then in this particular, uh, case, they're like fudging the, the wattage number in different areas to kind of, uh, make their performance look better or worse and, or changing the metrics, and it's, it's a little weird.

235
00:44:21.262 --> 00:44:34.362
Uh, and in this era, like, yeah, spending half a trillion dollars, you wanna know what the apples to apples comparison are because every, every watt matters in a data center 'cause you're so power constrained.

236
00:44:35.482 --> 00:44:48.282
And, and that gets into, you put out four questions for Doc- CEO Dr. Lisa Su, hopefully maybe to answer or shed light on this tomorrow. AMD earnings call tomorrow, so we'll, maybe we'll get some answers.

237
00:44:48.342 --> 00:44:51.842
But, um, lay this out for me. Um, what's...

238
00:44:51.882 --> 00:45:02.182
You know, dive a little bit deeper into this, um, and I don't know if I'd call it like the mystery chip, but, like, we don't have the hard, we don't have the hard data is, is basic- is basically the point. Um,

239
00:45:03.102 --> 00:45:10.242
what do you think the chip is, uh, but, like, what do we not know about it? Maybe this chip doesn't exist.

240
00:45:10.522 --> 00:45:17.332
So if it doesn't exist, like, tell us, like, what's closest to it or map it out for us so we can kinda understand. Don't

241
00:45:18.702 --> 00:45:29.402
have people go through all these, like, tiny footnotes like in a maze to kinda try to figure out what this product is and, but make the claim that you're 20% better than Nvidia, right?

242
00:45:29.472 --> 00:45:40.682
'Cause Nvidia's also built a heritage of CPUs which people don't know about. Grace Hopper was the first CPU that they worked on, but they didn't really market it m- 'cause the whole,

243
00:45:41.662 --> 00:45:53.962
uh, AI revolution, if you, if you look at all the things that Jensen has been talking about all these years, it's like the CPU is dead. Moore's Law is dead. Long live the GPU. But in reality, you need both to coexist.

244
00:45:54.042 --> 00:46:06.502
You need the CPU to kinda tell the GPU, uh, the GPUs what to do and have that orchestration to work together. And the question is, is like which one is better and which one has higher performance? And,

245
00:46:07.742 --> 00:46:20.322
uh, if you go back, back in time, like, like in, like I said, the late '90s or early 2000s, it was always, like, AMD versus Intel on which CPU is better on performance and megahertz and all those things.

246
00:46:20.342 --> 00:46:34.102
And I think now we're back to that rivalry again between these two companies, uh, which is exciting because I think, uh, rivalries create better performance and you, uh, it serves the customers better in that aspect.

247
00:46:35.282 --> 00:46:45.522
And this gets into kind of like a standard reporting discrepancy. You talk about, like, this, um, inconsistent denominators when reporting chip performance.

248
00:46:45.582 --> 00:46:57.922
And this extends also, I believe, to, uh, Nvidia in that there's some in... Like, it's not... The, the, we just, it's very difficult to compare apples to apples when comparing, uh, the performance of these chips.

249
00:46:58.622 --> 00:47:01.942
Um, I guess what does this mean at the data center level?

250
00:47:02.042 --> 00:47:14.242
Because most of the, our audience is, like, power acquisition, power deals, but that's downstream of this, like, very, very, very different business of, like, chip stats.

251
00:47:14.362 --> 00:47:24.062
Yeah, 'cause they pushed- What does it mean, like, in the data center? They pushed, they pushed 600 watts on throughput and then 400 watts on agents per watt, but there's, like, no wattage for the single core.

252
00:47:24.091 --> 00:47:36.562
So, like, obviously if you're supposedly saying you're using less watts and out- outputting more, then you're better than the other guy. But what if you're actually using more watts but not, and hiding it, right?

253
00:47:36.682 --> 00:47:41.662
And that, I think that, uh, for, like, a data center designer or owner,

254
00:47:42.682 --> 00:47:52.902
the, that leak, that wattage leakage, right, can negatively reflect our pro- affect our profitability because our energy costs go up and they don't have enough power.

255
00:47:52.962 --> 00:48:01.182
I think that's what people really need to understand is we're super power constrained, and every watt matters, and that needs to be accounted for on the specs of your products.

256
00:48:02.882 --> 00:48:13.702
I mean, does, does, does, does this, this have the potential to derail, you know, uh, the expansion plans some of these companies have? I'm thinking about even just modeling out CapEx.

257
00:48:13.802 --> 00:48:20.322
If the wattage isn't totally aligned with what you're expecting to spec for your data center, that could cause some kinks in the road.

258
00:48:20.402 --> 00:48:30.832
Do you think that's a legitimate issue for some of these operators, or do you think- It is because, uh, when you're talking about, like, rack-scale deployments, you have, like, one rack times, like, 1,000 or whatever is in the data center.

259
00:48:30.942 --> 00:48:35.142
So all these things multiply and scale. So the question is, is, like, what is it?

260
00:48:35.362 --> 00:48:45.062
I need to know that number and that performance because I'm designing this data center that's gonna ba- basically be built in a year and a half based on your specs today.

261
00:48:45.582 --> 00:48:50.862
And the, the load and the wattage and the power supply and everything, everything has to kind of match up.

262
00:48:51.182 --> 00:49:00.094
If not- You're putting the, the operator in a bad position 'cause they end up deploying it, and then it's gonna co- end up costing more than what they actually thought it was gonna be.

263
00:49:00.554 --> 00:49:16.134
And then, like, you're negative on your margins. [laughs] So AMD reports tomorrow. Yeah. Um, do you expect them to answer or clarify anything r- regarding this chip? I mean, you've got four questions.

264
00:49:16.174 --> 00:49:28.734
You know, publish the sub scores of the chips, you know, unify denominators, um, maybe tell us what chip or, you know, whatever. Do you expect them to maybe address any of these open questions that you have?

265
00:49:28.794 --> 00:49:44.074
And then if they do, uh, to what degree does that a- add, like, give investors clarity going forward? I'll be listening. Uh, and recently I've, I've, I've had an influx of AMD employees start following me on X right now.

266
00:49:44.314 --> 00:49:52.174
They're all in these different departments coming out of the, the woodwork, so it's, it's definitely making the rounds, I think, around San Jose and Silicon Valley.

267
00:49:52.394 --> 00:50:00.984
It's a good question, and people, when they work hard on these products and these different groups and it, it, it's a lot of data and numbers. Like,

268
00:50:01.954 --> 00:50:15.194
at the end of the day, you want an apples-to-apples comparison on what you're, you're up against, and if you're kind of moving the target or the denominator to make yourself look better, and 20% is a big number for these types of things, it, it kind of discourages you.

269
00:50:15.394 --> 00:50:23.374
So how do we level the playing field for everybody to kind of play nice? Like, I have nothing against Lisa Su or AMD. I love it.

270
00:50:23.494 --> 00:50:32.254
I also had a question for Nvidia to talk about their memory and architecture, what they're doing. Everyone needs to kind of contribute for this to work in this ecosystem because,

271
00:50:33.534 --> 00:50:37.434
uh, the way I look at it is, uh, you know, we're all on the same team.

272
00:50:37.994 --> 00:50:47.474
It's us against China at this point, so we all need to win, and let's make sure the people buying these products are also winning and getting the right product that they were looking for.

273
00:50:47.874 --> 00:51:02.954
I think that's the most important. You bring up the-- You referenced, like, the Intel-AMD CPU, uh, race of before my time. And- Yeah, I'm kind of old, I guess. [laughs] Yeah.

274
00:51:03.014 --> 00:51:14.574
And, uh, and you call this, quote, "the best CPU race in a decade." Um, you know, what are your thoughts on x86 holding the data center crown?

275
00:51:15.214 --> 00:51:28.204
Um, do we see Vera at the beginning of the Arm takeover for, you know, hosting these nodes? Uh, I don't know. Give, give you some color on what we can look forward to for this competition. Yeah, I mean,

276
00:51:29.134 --> 00:51:39.414
the, the whole x86, uh, thing sort of died. I think you can also look at what Apple did with the-- they designed Intel out of their MacBook.

277
00:51:39.494 --> 00:51:46.314
So like every, pretty much every new MacBook has Apple silicon in it with their own cores, and it's, it's all Arm-based.

278
00:51:46.994 --> 00:52:06.694
Nvidia did the same thing, and now with, uh, agentic AI, you have Arm, who-- which came out with its own AI CPU, and then AMD supposedly has three different CPUs, and then Intel, this whole run-up that you see in the stock prices, they're, they're able to charge more for CPUs because there's so much demand in that.

279
00:52:06.734 --> 00:52:20.974
But the question is, is just throwing more CPUs at the, the problem might not be it. The question is, how do you design the CPU to be more, the most performant in an agentic long-horizon task workload for AI?

280
00:52:21.274 --> 00:52:25.614
And to really do that, I think you need to have the roadmap to make it purpose-built.

281
00:52:25.634 --> 00:52:43.174
That's why, not to be conflicted, but I think Vera Rubin was designed for that, and the roadmap that Nvidia has for their CPU products, I think, is well-versed for long-horizon agentic task loads, and I think that's where it's headed with that.

282
00:52:43.234 --> 00:52:54.854
And then they-- I think they shrink that to some extent, and they put it into the, the, the RTX, uh, personal computer that's coming out or f- for, for enterprises and offices.

283
00:52:54.874 --> 00:53:07.564
And I think once you create that, like, scalable solution, you kind of see all that working. And then this is just my, uh, like, hypothesis. Sometimes, uh, you just have to kind of

284
00:53:08.554 --> 00:53:19.214
think outside the box and dream a little bit. I think these CPUs and these Vera CPUs, right now they're orchestrating virtual agents or, like, basically, like, LLMs to do things.

285
00:53:19.834 --> 00:53:32.174
Eventually, these CPUs will be orchestrating physical AI, which is like robots. So the, the agentic, uh, like, the robots are basically running loops for you, say on, in a manufacturing floor.

286
00:53:32.234 --> 00:53:45.734
Like, pick up this widget, drill it in, test it, close it. Like, that's a loop, right? And you need a CPU to orchestrate that. Like, like a master computer with all these robots running in concert as a robot.

287
00:53:45.774 --> 00:53:52.574
How do you-- Who's doing that operation and who's running it? I mean, I, I think that's like another Nvidia type of play.

288
00:53:52.604 --> 00:54:00.634
And that kind of becomes, uh-- A lot of this is focused around memory and where this memory sits on the GPU and the chip. Um,

289
00:54:01.534 --> 00:54:16.514
and this may not be, like, the best closing question, but you talk a little bit about memory power and how this is kind of, uh, obscured, like the LPDDR5X versus, like, an unpublished MRDIMM figure.

290
00:54:17.094 --> 00:54:29.364
Like, th-- I'm still trying to wrap my head around this, um, but the-- a- as, as, like, context becomes king here or, and, and maybe, I don't know if you dis- if you agree with that.

291
00:54:29.734 --> 00:54:36.213
As context often kind of becomes king for these long-horizon tasks, like, memory power becomes very important.

292
00:54:37.194 --> 00:54:45.774
Um, give me some color as to, like, what we could hope to see in the memory, like, stats game, especially with regards to power draw.

293
00:54:46.574 --> 00:54:56.834
Um, and, you know, hopefully, uh, can we expect it to be more standardized and transparent to perhaps incentivize a little more competition? I don't know. What are your thoughts here?

294
00:54:59.194 --> 00:55:11.994
I think, uh, you know, AMD and NVIDIA made opposite memory bets. AMD gets its bandwidth from mRDIMM and modules with buffers that have to be powered and cooled.

295
00:55:12.454 --> 00:55:28.324
NVIDIA uses low latency or, um, uses low power phone lineage memory. So neither company counts a single watt, uh, of their headline power, and that's where they differ the most. So it's kind of like

296
00:55:29.754 --> 00:55:43.254
two different schools of memory thought. I'm not-- I don't know which one is right or which one is better. So it's, that's where it's kind of hard to, to balance out and figure out which one makes the most sense.

297
00:55:43.294 --> 00:55:56.614
You know? One obviously has, you know, buffers in it, and it's cooled, and the other one uses, like, the same types of memory they use in, like, cell phones. So who's right? We'll see with that.

298
00:55:57.694 --> 00:56:07.534
Maybe zooming out for a final take, you know, we're heading into earnings week here. We're gonna cover a lot of the, the power shells, NeoClouds, data center earnings. Um, what...

299
00:56:07.574 --> 00:56:13.974
You know, you probably look a little bit more at the, like, chip level, the actual upstream production.

300
00:56:14.034 --> 00:56:32.454
But I'm curious, like, when you see these types of open questions on AMD, will, will AMD's, uh, uh, an- any, uh, information they do or don't give tomorrow, will that inform any opinions you have about the rest of the actual, like, data center operators and construction companies?

301
00:56:33.914 --> 00:56:36.864
I think demand is still outstripping supply.

302
00:56:37.884 --> 00:56:49.654
The question from a data center operator is to make sure they have the right stack and equipment to eventually, hopefully ope- operate profitably because I think your biggest cost is energy, right?

303
00:56:50.114 --> 00:57:00.154
So I don't see energy costs going down. So if they're scaling up and it's a finite amount of energy, what is your wattage load and how is it all working?

304
00:57:00.194 --> 00:57:11.634
Because you don't want to get overloaded or have devices take more energy than they say they do on their specs or whatever it is to have those issues. And I think that's something they need to mitigate.

305
00:57:12.234 --> 00:57:26.354
And, and in this situation, because there's so much demand, it's like everybody's getting business. I don't see, like, a big slow- Like, it's not like a, it's not like a short report or anything like that.

306
00:57:26.394 --> 00:57:39.014
It's just like a question to just bring to the industry to just level the playing field. So even if there's other startups that claim these sort of things, like, kind of use the same metrics, so

307
00:57:39.914 --> 00:57:50.374
there's an apples to apples comparison on these. Ben, thank you so much for your insights. I hope you get the answers you want tomorrow. [laughs] Come on, Lisa. Tell me what's going on. Yeah.

308
00:57:50.594 --> 00:58:00.124
Lisa-- Tag Lisa in the comments, everybody. Uh, [laughs] Ben, thanks so much for your time. Mayb- We'll, we'll probably run it back soon, so cheers. Let's go. Let's go. Thanks, guys. Have a good day. You too.

309
00:58:00.134 --> 00:58:17.354
Thanks, Ben. You too. All right, bye-bye. Kind of a renaissance, uh, podcast here, covering all things from power shells to Mongolia to AMD 86 and Arm trends. So that will wrap it up for the show today.

310
00:58:17.554 --> 00:58:28.724
I do want to call everybody really quick to look at-- We, uh, have the schedule posted on our Twitter here for the, uh, earnings weeks events we are covering.

311
00:58:28.874 --> 00:58:44.974
Specifically tomorrow morning, tune into our channels for Cipher Digital and HUT8. Then we will have Ionic Digital C-suite person on the pod to chat, uh, about what they are up to.

312
00:58:45.074 --> 00:58:58.134
But make sure to follow all Blockspace content this week. If you cannot get enough of our video live stream and podcast, head to our website for all written content, blockspace.media.

313
00:58:58.174 --> 00:59:13.654
That's not blockspace.com, it's blockspace.media. This show is brought to you by CleanSpark, NASDAQ listed ticker CLSK. I'm Charlie. I'm Colin. And we'll see you tomorrow.

314
00:59:14.014 --> 00:59:21.434
[outro music]