WEBVTT

1
00:00:00.060 --> 00:00:09.760
On y'all. Welcome back to Blockspace Live, presented by CleanSpark for another beautiful day of earnings, Charlie, and we've got a lot to chew through, my man.

2
00:00:10.360 --> 00:00:25.900
We have CoreWeave's Q2 earnings coming up first, and big updates with CoreWeave. ARR has been guided upwards to over 10 billion, Charlie, but that's on the back of a lot of debt.

3
00:00:25.910 --> 00:00:33.840
There's a lot in the balance sheet to comb over, so we will be doing that first. Following that, we've got Nebius's Q2 earnings as well.

4
00:00:34.260 --> 00:00:43.160
And then after Nebius, we have CoreWeave Chief Development Officer, Brandon McBee on to talk about some of the more salient aspects of the Q2.

5
00:00:43.520 --> 00:00:55.900
We're gonna be talking about the fact that they just signed a, a contract for A10s through 2029. Where are the depreciation hawks? We'd love to have a word with you. A100s, you mean, right? Yeah. Did I say A100s? Yeah.

6
00:00:56.180 --> 00:01:01.840
Whatever. I don't even know. Thank you. [laughs] A100s are still in the game, and they will be for some years.

7
00:01:01.880 --> 00:01:09.560
We're gonna have some questions on that, as well as how they think about structuring deals going forward if contracts roll over, and also how they're managing all of this debt that they're accumulating.

8
00:01:10.100 --> 00:01:24.580
After that, we have a note on White Fiber's Q2 as well, and then Charlie will cap off the segment with Wintermute moving towards AI trading. Everyone's pivoting. No one cares about crypto anymore.

9
00:01:24.600 --> 00:01:29.940
But we could have told you that. We pivoted a few months ago. Yeah, we could have told you that.

10
00:01:30.600 --> 00:01:44.830
Uh, Blockspace goes live every weekday at 1:00 PM Eastern, because we are Compute's daily live show covering quick hits on AI, infrastructure, data centers, emerging technology, and markets.

11
00:01:44.960 --> 00:01:48.980
If you miss the live stream, you can catch us on podcasts, anywhere podcasts are found.

12
00:01:49.740 --> 00:02:05.780
And if you like that and you want even more, go to our website, blockspace.media, for all the expanded universe of Blockspace content. This show is brought to you by CleanSpark, NASDAQ listed ticker CLSK.

13
00:02:05.880 --> 00:02:07.760
More on CleanSpark later on the show.

14
00:02:08.460 --> 00:02:23.560
And, um, we're just gonna have to get rid of, of all news stories e- uh, today, Colin, because it's just earnings, earnings, earnings, which means we have to play the earnings update intro. [beeping] That's right.

15
00:02:23.640 --> 00:02:37.170
Coming in hot earnings week. Let me show you what the current week looks like. We're halfway through the second week of earnings here at Blockspace. Monday, we did watch parties for Keel, Bitdeer, Riot.

16
00:02:37.400 --> 00:02:46.940
Tuesday, we had watch party for CoreWeave. Today, Wednesday, we have Brandon McBee, as Colin just said, co-founder of CoreWeave, as well as Chief Development Officer.

17
00:02:47.060 --> 00:02:55.960
Had watch parties for Nebius and White Fiber this morning. Tomorrow, Thursday, we have CEO, uh, and captain of Keel Infrastructure, Ben Gagnon.

18
00:02:56.800 --> 00:03:07.740
And Sam T- Tabar of White Fiber, CEO of White Fiber, was originally scheduled today, pushed to tomorrow. So apologies to those who were expecting him today. He'll be here tomorrow.

19
00:03:07.760 --> 00:03:18.860
And then we'll be doing a watch party for Soluna. Speaking of Soluna, Soluna CFO, Mike Pitchie, uh, on the show Friday. Uh, Soluna's CEO also had to push earlier this week.

20
00:03:19.320 --> 00:03:31.800
He will be joining us very soon, probably next week. So stay tuned. Outside of that, I'm gonna toss it back to Colin. Uh, so put on your thinking caps, 'cause Colin's about to dive deep into

21
00:03:32.740 --> 00:03:42.960
the w- the weaving world of CoreWeave. Dive deep indeed. There's a lot to go through here, Charlie, so hopefully we don't get the bends as we dive into it. [laughs] Yeah.

22
00:03:43.020 --> 00:03:53.200
And, and I'll, I'll occasionally pull up, uh, DI metrics as we go through this because there's some, like, reference points for, like, GPU hour rents and, like, where the stock is and et cetera, et cetera, et cetera.

23
00:03:53.540 --> 00:04:09.080
So... So CoreWeave coming out with their Q2 earnings with a big beat, Charlie. A big beat. So much so that the stock is up 20% over the last five days. I mean, look at that gap up.

24
00:04:09.220 --> 00:04:17.320
This is, this reminds me of what happened with Microsoft and Amazon after their cloud segments just blew everyone's expectations out of the way.

25
00:04:18.120 --> 00:04:31.060
So if you heard about death knells, the bell does not toll for this AI stock sector just yet. Things are still rip-roaring. CoreWeave above $100 at 107 as the time of recording.

26
00:04:32.520 --> 00:04:40.480
The headline here on Blockspace reads, "CoreWeave Q2 Revenue Reaches 2.58 Billion as Backlog Grows to 104 Billion."

27
00:04:40.600 --> 00:04:48.350
That was one of the highlights, the backlog of, of remaining, uh, remaining commitments for their neo-cloud line.

28
00:04:48.800 --> 00:04:59.100
We'll be going over the numbers really quickly before we get into some qualitative assessments of what was revealed on the earnings call and also the Q&A session. So here are the headline numbers for CoreWeave from Q2.

29
00:04:59.110 --> 00:05:16.140
Again, 2.5 sev- $2.575 billion in Q2 revenue. That's up 112% year over year. Cost of revenue ballooned to 879 million, up 181%. N- there's no free lunch.

30
00:05:18.120 --> 00:05:36.060
Technology and infrastructure came in at 1.5 billion, up 125%. Operating loss was 49 million, uh, up from, or, or down from 19 million in Q2 2025. And here's the oth- here's the big one. A- and it's what...

31
00:05:36.100 --> 00:05:38.100
You know, funnily enough, this didn't come up in the call.

32
00:05:38.160 --> 00:05:49.719
I guess people can s- just assume they have it covered, but their interest expense ballooned to 640 million, up 140% from 267 million in the same period last year.

33
00:05:50.180 --> 00:06:08.096
Net loss came in at 626 million, up 116%, and earnings per share was negative $1.14 per share, um, which is a- Up from negative, or down from negative $0.60 per share at the same period last year.

34
00:06:08.736 --> 00:06:15.216
So why the loss despite the revenue surge? Interest expense is the real story here.

35
00:06:15.796 --> 00:06:25.336
At six hundred and forty million against two point five eight billion in revenue, roughly twenty five cents of every dollar of revenue goes to servicing debt currently.

36
00:06:25.416 --> 00:06:32.736
Operating loss was forty-nine million, net loss six hundred and twenty-six million. The gap is almost entirely financing cost.

37
00:06:32.796 --> 00:06:39.956
Cash interest paid in H1 alone of twenty twenty-six was eight hundred and six million, and that almost understates the burden.

38
00:06:40.016 --> 00:06:45.936
The company capitalized another a hundred and seventy-six million of interest into construction during the first half of twenty twenty-six.

39
00:06:46.416 --> 00:06:55.256
So total interest incurred is meaningfully higher than what runs through the income statement. We'll be getting into some questions on that with Brandon once he comes online.

40
00:06:55.736 --> 00:07:12.496
But TLDR, as we all know, as you know, Charlie, and listeners know, CapEx is ripping for the AI industry. It's very expensive to finance these GPUs for CoreWeave, and that growth comes at a clear cost.

41
00:07:13.176 --> 00:07:24.236
That being said, there were some interesting deals, or rather there was an interesting financing deal that CoreWeave struck recently right before earnings hit that we will cover with regards to how they are financing short-term contracts.

42
00:07:24.856 --> 00:07:30.696
It's not all bad. As long as the money keeps flowing, they've structured it in such a way that they will be able to address this.

43
00:07:32.756 --> 00:07:43.796
The Q3 quarterly interest expense was guided to-- was written for a guidance of eight hundred and sixty to nine hundred and forty million. Again, big numbers here.

44
00:07:44.576 --> 00:07:55.136
And, uh, one notable thing, despite the huge rise in revenue, stock-based comp was basically flat for the first half of twenty twenty-six at three hundred and eighteen million.

45
00:07:56.336 --> 00:08:01.856
Now, let's dig into a little more of the balance sheet with regards to this debt and some of the other aspects of CoreWeave's business.

46
00:08:01.876 --> 00:08:13.676
Total assets went from forty-nine point three billion to seventy-seven point one billion. Total liabilities scaled accordingly from forty-six billion to tw- seventy-two billion. Equity is currently five billion.

47
00:08:14.296 --> 00:08:23.976
And this growth sits according to the following buckets: property and equipment net went from thirty point six billion to forty-six point seven billion.

48
00:08:24.016 --> 00:08:33.336
Technology equipment alone in the GPU fleet rose from twenty point nine billion to three point three billion. Construction in progress sits at eleven point nine billion.

49
00:08:35.176 --> 00:08:44.496
Recourse debt went from twenty point seven billion to thirty-one point four billion. Uh, the current-- That's current plus non-current.

50
00:08:45.076 --> 00:08:54.256
Non-recourse debt also grew from six hundred and forty-seven million to three point six six billion. So most of the debt that CoreWeave has on its balance sheet right now is recourse to the parent company.

51
00:08:54.276 --> 00:09:02.356
We've got a question for Brandon on that. Cash and cash equivalents sat at five point five two billion plus one point three eight billion restricted.

52
00:09:03.116 --> 00:09:07.996
So total debt is roughly thirty-five billion against five billion in equity.

53
00:09:08.156 --> 00:09:19.436
H1 financing raised-- In H1 twenty twenty-six, they raised sixteen point seven billion of debt gross, two point nine eight billion from private placements of stock against five point two billion of repayments.

54
00:09:19.996 --> 00:09:30.856
There are also lease liabilities that add to the total liabilities here that are not structured specifically as debt recourse to the parent. Operating lease liabilities totaled sixteen point three billion.

55
00:09:30.896 --> 00:09:41.276
That's current plus non-current. They are not counted in that debt figure, so combined with debt, those fixed obligations are approaching fifty-one billion. That's pretty hefty.

56
00:09:41.316 --> 00:09:57.836
Regarding financing, one of the more interesting bits that came out recently before this earnings call and which was touched on in the analyst section, CoreWeave closed a new debt facility that is meant to finance short-term GPU take or pay contracts.

57
00:09:58.836 --> 00:10:07.656
They closed an oversubscribed two point six billion delayed draw term loan facility known as DDTL five point five on August 10th, twenty twenty-six.

58
00:10:07.716 --> 00:10:18.796
The financing features an approximately five-year maturity and is priced at SOFR plus five and a half percent, so roughly nine point ten percent there for the fi- for the interest rate.

59
00:10:19.295 --> 00:10:23.616
And this fund, uh, the funding is for customer-dedicated GPU and high-performance AI cloud infrastructure.

60
00:10:24.036 --> 00:10:36.956
This is specifically short-term financing that on the call CoreWeave billed as being instrumental to allowing them to offer shorter term contracts, roughly two to three years, versus the kind of five-year standard for their longer-term take or pay.

61
00:10:37.316 --> 00:10:47.416
That's the kind of contract that they're signing with guys like Microsoft. And it's-- This is as far as I can understand, like one of the first financing deals of its kind, Charlie.

62
00:10:47.976 --> 00:11:00.776
And what it ultimately points to is that now lenders are getting more comfortable lending at shorter-- on shorter terms for these contracts that are seen as less sticky and potentially less accretive to the business.

63
00:11:01.256 --> 00:11:05.756
But it opens up the potential for CoreWeave to be much more flexible with how they manage their compute.

64
00:11:06.356 --> 00:11:15.126
Also, I have a question with Brandon on that because right now, their guidance for current revenue and forward revenue, ninety-eight percent comes from take or pay contracts.

65
00:11:15.176 --> 00:11:27.036
The majority of that is in the long-term bucket versus the short term. And analysts, uh, had a lot to say on the call as to or a lot to ask on the call regarding how CoreWeave views its fleet in terms of being flexible.

66
00:11:27.466 --> 00:11:40.536
When those contracts roll over, will they put that into shorter term contracts? Will they put it into inference? Will they put it into spot on demand? And, uh, management basically said, "TLDR, we're looking into it.

67
00:11:40.996 --> 00:11:53.296
The economics of what makes the most sense right now and also the landscape doesn't make it clear what is the best option." They're gonna keep their options open for whatever is maximizing value on those GPUs. Cash flow.

68
00:11:54.196 --> 00:12:01.236
Operating cash flow swung from minus a hundred and ninety million a year ago to plus two three point six six billion.

69
00:12:01.736 --> 00:12:09.096
That was driven by two point five four billion of depreciation and amortization add back and a one point three seven billion of deferred revenue growth.

70
00:12:09.756 --> 00:12:18.656
CapEx was fourteen point one billion in the first six months of the year. There's also another nine point eight billion of capital expenditures incurred but not yet paid.

71
00:12:19.204 --> 00:12:23.204
Sitting in liabilities, that includes OEM-financed additions.

72
00:12:23.244 --> 00:12:33.184
Of the one point four seven billion was, that was reclassified into debt during H1 upon execution of OEM financing arrangements, those are equipment purchases that are converting into debt on the balance sheet after the fact.

73
00:12:34.424 --> 00:12:46.324
Now, the number that everyone is interested in, deal flow. ARR management is now guiding ten point three billion as of this latest quarter, and it is about to ramp up.

74
00:12:46.404 --> 00:13:07.324
As of June 30th, CoreWeave reported one hundred and three point seven billion of unsatisfied remaining performance obligations with a disclosed run-off schedule as such, forty-one percent recognized in the initial twenty-four months following June, so that's through June twenty-twenty eight, thirty-nine percent between twenty-five and forty-eight months, and the remainder between forty-nine and seventy-eight months.

75
00:13:08.884 --> 00:13:16.884
That seventy-eight-month tail shows that contract duration has extended well beyond the two to three-year deals that characterized the early neo cloud market.

76
00:13:17.324 --> 00:13:24.444
Roughly forty-two point five billion is expected to convert to revenue within two years against a current annualized run rate of around ten point three billion.

77
00:13:25.024 --> 00:13:32.444
Deferred revenue is nine point seven billion, up from eight point two billion at year-end, and customers are prepaying at scale, as that indicates.

78
00:13:33.024 --> 00:13:41.524
Speaking of those customers, customer concentration is easing, but it's still on the higher side or, according to some analysts.

79
00:13:41.584 --> 00:13:54.924
Customer A, assumed to be Microsoft, CoreWeave's biggest customer, had seventy-one percent concentration this time last year. Now they have thirty-six. Customer B had, has twenty-six percent. Customer C, ten percent.

80
00:13:55.604 --> 00:14:05.084
And Customer B and C both respectively had under ten percent in Q2 twenty-twenty five. That's still pretty heavy concentration between those three customers.

81
00:14:05.924 --> 00:14:19.924
You're looking at seventy-two percent of the business being within three customers specifically. Last notes tr- here, Charlie, before I toss it to you for some color. We picked a few of the earnings call Q&A sections.

82
00:14:20.344 --> 00:14:29.084
One of them was regarded to the ramp up on their sites, and their CFO, Nitin Agarwal, gave a pretty clear breakdown.

83
00:14:29.463 --> 00:14:37.224
There's four point two gigawatts contracted as of today, plus one point five gigawatts of powered lands expansion options and executed LOIs.

84
00:14:37.884 --> 00:14:46.984
So that's close to six gigawatts, and it's just the middle of twenty-twenty six, Agarwal highlighted. That's against an eight gigawatt target by twenty-thirty.

85
00:14:47.484 --> 00:14:51.264
Maybe we get that eight gigawatt target bumped up since they're so close to six already.

86
00:14:52.104 --> 00:15:01.824
And the fi- one point five gigawatt bucket is disclosed in aggregate only, no site-level breakdown, and that's similar to how we've seen other reporting in this sector.

87
00:15:03.764 --> 00:15:13.664
The big question from a few analysts, and one we're gonna cover here in the interview: Can existing infrastructure serve inference, and how does CoreWeave see the edge versus hyperscaler landscape?

88
00:15:15.404 --> 00:15:22.014
The answer from management addressed whether or not inference's heavier CPU and storage requirements force new generation data center designs.

89
00:15:22.664 --> 00:15:31.014
Um, and Trader, uh, CEO and Trader's answer was philosophical rather than technical. CoreWeave doesn't build specific or separately for training versus inference.

90
00:15:31.344 --> 00:15:45.664
It builds, quote, "AI infrastructure meant to serve the full loop." So CoreWeave basically leaving their options open or, uh, indicating that they are doing such as, uh, with regards to how they approach these builds.

91
00:15:46.204 --> 00:15:50.724
Again, ninety-eight percent of their contracted deal flow is in that take or pay model.

92
00:15:51.084 --> 00:16:03.684
I think one of the big questions going forward is at what point does it make sense or does it ever make sense to move towards a more spot or on-demand model or an inference model and token-based model for monetizing that compute?

93
00:16:06.304 --> 00:16:20.824
So a couple takeaways. One, uh, they sold some, uh, GPUs higher than they bought them, right? Am I right about this? Um, I didn't see anything about that.

94
00:16:20.864 --> 00:16:23.464
Actually, I might be mixing this up with a different earnings call.

95
00:16:23.824 --> 00:16:35.364
Regardless, like, uh, I-- if I'm, if I'm reading the numbers right, like, uh, what CoreWeave's getting, uh, per GPU hour does-- Oh, no, it's the A one hundred. They signed that A one hundred deal for ten years.

96
00:16:35.443 --> 00:16:48.144
Oh, I see what you're saying. Yeah. Yeah. Yeah. So the A one hundred deal pretty- And, and I mean, yeah, and, and what we're gonna see across the earnings today, uh, and really this week so far, is that, uh,

97
00:16:49.284 --> 00:17:00.904
GPU depreciation, the death, uh, what is it? The death of GPU and early depreciation were vastly overstated. And this is what Jensen's point was last week or this week.

98
00:17:00.964 --> 00:17:15.004
This is what, uh, we see here with the CoreWeave news. Um, so we're actually seeing in some cases used GPUs go higher. And, um, let me pull up DI metrics here.

99
00:17:18.284 --> 00:17:28.664
Um, CoreWeave up twenty-one percent today. Wild. And, and this-- I mean, the three earnings today, Naevious, CoreWeave, and whi- White Fiber all up. Year to date,

100
00:17:29.744 --> 00:17:34.504
Naevious up one seventy-four, White Fiber up sixty-six point nine, and CoreWeave up thirty-five point three.

101
00:17:35.004 --> 00:17:47.324
You know, I go on Twitter, Colin, and whenever people like look at, you know, whenever I see analysts, the smart analysts I see, um, commenting on this, they almost all say CoreWeave is the king, CoreWeave is the best, so.

102
00:17:48.144 --> 00:18:04.484
Um, and SemiAnalysis does rank them in the platinum tier, the top tier of their NeoCloud, uh, tier list. So, uh, this does, uh, reinforce that view, so. One hundred percent.

103
00:18:04.644 --> 00:18:18.164
And one point of clarity before we move on to Naevius. For that A one hundred deal, their CFO, Agarwal, said that older generations, quote, "Remain higher or at levels that we've seen about a year ago."

104
00:18:20.088 --> 00:18:31.868
Hedging the language there a little bit, but at the very least, management is saying that they are sticky and that the pricing hasn't moved. Which- And, and that tracks- Which is bullish one way or the other. Yeah.

105
00:18:31.928 --> 00:18:45.908
What is... What are we seeing on, on- I mean, that, yeah, that tracks with what we're seeing, um, on DI metrics just for third-party marketplaces, and I think this is pulled from Oren. I mean, B200s a year ago, $4.50.

106
00:18:46.048 --> 00:18:58.148
B200s today, $6. H100s and 200s, 2 and 3, and now 3 and 4. I mean,

107
00:18:59.688 --> 00:19:10.388
um, yeah, we're seeing, uh, GPUs at, as they say, at or above, uh, where they were a year ago. And these are the, these are the, the, the settled prices on markets.

108
00:19:10.428 --> 00:19:23.368
These are not necessarily the actual locked in, like, delivery prices of GPUs. So, um, A100's holding in there [chuckles] really well. I mean, look at this. Just look at, like, the, the last six months.

109
00:19:24.207 --> 00:19:37.958
Um, A100's still actually up on the past couple months from, I think, 80 some odd cents per GPU hour to now a dollar. This is a six-year-old card. If we're looking, [chuckles] you know, another

110
00:19:38.868 --> 00:19:56.248
five plus years on this card, uh, very significant. All your GPU, all your GPU projections are wrong. So, um, I think we'll, uh, let you touch on Nevius' earnings, and then after that we'll have, uh- Act- Yeah.

111
00:19:56.328 --> 00:20:07.058
Actually, turns out Brandon's actually in the wings right now. Rather than just cram Nevius in- Yeah. [chuckles]... and this is actually a good, uh, jump-off point, 'cause- Yeah... my first question is on A100. Yeah.

112
00:20:07.108 --> 00:20:17.288
So I think we'll go ahead and rip the CleanSpark ad, and then- Yeah... have Brandon on. Yeah. We'll have Brandon on right after a word from our sponsor, CleanSpark. [gentle music] We are CleanSpark,

113
00:20:18.208 --> 00:20:30.668
America's Bitcoin miner, a publicly traded company with the largest operating hash rate, powered entirely by self-operated infrastructure across four states.

114
00:20:32.488 --> 00:20:43.948
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.

115
00:20:47.448 --> 00:20:58.288
All righty, we have CoreWeave CDO, Brandon McBee, in the wings, bringing him up to grill him on the Q2 earnings. Brandon, welcome to the show, and thank you for joining, man. Really appreciate it. Hey, guys.

116
00:20:58.328 --> 00:21:09.748
Thanks for the time today. Appreciate it. Yeah, definitely. Lot to chew through. Huge quarter. Congrats on that. Big beats across the board, and stock price is being, uh, is, is pumping as a result.

117
00:21:10.668 --> 00:21:18.768
We were just talking about A100s. I'd love to start off there, specifically with this A100 contract running through 2029. I kinda have two questions here.

118
00:21:18.788 --> 00:21:26.548
Can you give us any information on the detail for these contracts, whether they're priced statically from what y'all were seeing a year ago, at a discount or a premium?

119
00:21:27.188 --> 00:21:34.028
And also, a follow-up on Jensen Huang arguing that the A100 useful life may run closer to a decade.

120
00:21:34.708 --> 00:21:41.148
Does this contract support that claim, and does CoreWeave now depreciate GPU classes on a longer schedule than it did a year ago?

121
00:21:41.168 --> 00:21:51.508
So kind of a two-part question there of, like, what exactly does this deal look like in terms of the specifics, or can you give us a flavor for that, and how do you treat depreciation accordingly? Yeah, thanks.

122
00:21:51.668 --> 00:22:02.648
So, uh, pricing on the infrastructure, A100 specifically, has really held very well in market over the last 12 to 18 months. Right? We really haven't seen much degradation in pricing at all.

123
00:22:02.708 --> 00:22:17.288
I believe even our last queue, we were commenting how A100 pricing had actually increased a bit, uh, over the, the, the prior 12 months. So we're still in 2025 or even, like, late 2024 pricing for that SKU.

124
00:22:18.088 --> 00:22:29.088
Um, given that it's a 2020 release, and I think you guys are focused on the exact right point of you're contracting out through 2029 now. I mean, it's just a massive narrative violation.

125
00:22:29.848 --> 00:22:45.108
Um, I, you know, we've been hammered with questions around depreciable, uh, term and useful life, et cetera. And we've just been consistent that these GPUs are going to have a longer useful life than the market expects.

126
00:22:45.148 --> 00:22:56.888
And super important to clarify that people aren't choosing an A100 because they can't get access to an H100 or a B200, right?

127
00:22:56.968 --> 00:23:10.668
They're coming in specifically asking for that Ampere or that Hopper or that Blackwell because they're pairing their engineering requirement, their workload, with the piece of infrastructure that's most optimized for it.

128
00:23:10.728 --> 00:23:22.288
And guess what? There's tons of workloads that need A100s. There's tons of workloads that need H200s. So I think Jensen's comment that Ampere could be a, a decade-long SKU is, is entirely correct, right?

129
00:23:22.308 --> 00:23:33.788
I mean, we're only a year off from that right now. And even more supportive to your point on what does the contract look like, these are firm economic take or pay agreements, right?

130
00:23:33.828 --> 00:23:45.908
Like, arguably it's the lowest margin profile type of contract that we sign in the sense of it's not an on-demand contract, it's a term fully committed take or pay agreement.

131
00:23:46.188 --> 00:23:57.948
And to have that out all the way through 2029, uh, just speaks about the amount of not only the demand in the space, but the durability of all different types of SKUs that support that demand.

132
00:23:58.968 --> 00:24:13.028
You know, we, we come from, uh, the Bitcoin mining side of things, and one of my favorite, uh, kind of contrasting variables with regards to AI compute versus Bitcoin mining is there's, there's, there's only one form of Bitcoin mining [chuckles] compute.

133
00:24:13.068 --> 00:24:19.288
You're doing SHA-256, and there's- Yeah... no flexibility with regards to an older generation model and a new generation model.

134
00:24:19.688 --> 00:24:26.878
You're worried about the efficiency, and if the revenue doesn't make sense, you have to shut off those older models. And so what I hear from you is that at this point it's not even about-

135
00:24:28.080 --> 00:24:33.800
It's not even about operators looking for the newer infrastructure, and then they can't get it, so they're settling for the older models.

136
00:24:34.180 --> 00:24:44.000
It's that there are legitimate questions as to what is the right fit for workload and the economics of that actual GPU itself. And- I think that's exactly correct.

137
00:24:44.740 --> 00:24:48.340
Well, and that kind of leads into my next question, 'cause one of the questions on...

138
00:24:48.520 --> 00:24:57.200
A, a few analysts had on the earnings call was re- with, with regards to contract, um, uh, designation and with flexibility for the compute.

139
00:24:57.260 --> 00:25:05.240
S- y'all's CFO said committed revenue, roughly 98% of the total was durable, but your CEO, um, said that there...

140
00:25:05.300 --> 00:25:15.740
Y'all are still kind of thinking through whether or not to, like, roll over old contracts into maybe a spot or on-demand inference compute, whether or not maybe shorter term makes sense.

141
00:25:16.390 --> 00:25:23.380
Uh, the new, uh, f- facility that y'all signed on August 10th is, is, is, um, relevant here with regards to those short-term contracts.

142
00:25:24.120 --> 00:25:33.079
So I'm curious from where you're standing, what are, like, the two or three variables that actually decide this cluster will be used for this type of contract versus another type of contract?

143
00:25:33.460 --> 00:25:37.760
Is it remaining useful life, uh, customer credit, inference demand in that region?

144
00:25:38.260 --> 00:25:49.180
And how does the DDTL 5.5 financing change the default, or just add a financing option for whichever workloads already looked like enterprise deals? Yeah.

145
00:25:49.240 --> 00:26:04.780
And, and I think you're circling around the main point that contributes to what sort of, um, revenue class we allocate into, which is financing, right? So for newer SKUs,

146
00:26:06.040 --> 00:26:14.480
w- we're leaning pretty heavily into looking for financing, um, on those, that CapEx as we bring it online, right?

147
00:26:14.640 --> 00:26:28.130
The best way to finance that CapEx is to have long-term take-or-pay contracts that we drop into these DDTL style facilities. Traditionally, you know, we, we've always done between two and six-year contracts.

148
00:26:28.180 --> 00:26:35.000
It's just the volume weighted average contract duration. It's been, you know, around four and a half to five years.

149
00:26:35.180 --> 00:26:50.320
Um, and that's been the kind of bedrock or foundation of how we've established this financing flywheel of our business. But recently, with, uh, facilities like DDTL 5.5, we've made steps towards how do we

150
00:26:51.460 --> 00:26:58.440
finance shorter duration contracts, without just, like, kind of pairing them up with longer duration contracts instead, 'cause that's what we've done before.

151
00:26:58.500 --> 00:27:15.420
But how do you finance just a basket of two to three-year contracts? And 5.5 was our first facility as a term, term loan B offering, where the duration of the contracts was actually shorter than the amort period.

152
00:27:16.580 --> 00:27:29.020
Uh, to put that in simpler terms, we are asking our investors in that facility to underwrite us renewing contracts or remarketing that infrastructure, right?

153
00:27:29.080 --> 00:27:50.380
They're taking on the risk that we will need to go out there and, um, find incremental consumers of those GPUs to handle the full amortization, and that's a deviation from, um, what we've done in the past, and I would say an, an enhancement in our ability to go finance a broader set of different types of contracts.

154
00:27:51.440 --> 00:28:02.859
Um, great positive for us because we are very eager to expand our support of the enterprise space, and we announced, uh, Caterpillar as a client in this quarterly earnings. I don't...

155
00:28:02.920 --> 00:28:09.140
I, I would imagine not many people had that on their bingo card for us to onboard Caterpillar as a client. We're incredibly excited to support them.

156
00:28:09.600 --> 00:28:24.860
But that's a great example of, um, the types of clients that are more focused on two to three-year agreements, and us establishing our own ability to go out and finance, uh, those contracts discreetly is, uh, incredibly important to us.

157
00:28:25.020 --> 00:28:27.480
So that, that was the main feature of 5.5.

158
00:28:28.540 --> 00:28:49.840
So on that note with moving into some of these enterprise contracts that might have been out of left field for some people, I wanna kind of talk about customer concentration for a second because, uh, cust- your primary customer, customer A, which I believe is Microsoft, but we'll plead the fifth on that one if we need to, went down from 71 to 36%, uh, of revenue for the year.

159
00:28:50.160 --> 00:28:57.420
Fantastic, but the top three customers are still roughly 72% combined. Would you talk about these other enterprise customers?

160
00:28:57.960 --> 00:29:09.800
Um, that trend away from diversificating, uh, diversification away from, like, one hyperscaler into two or three, um, i- i- is that, uh, just what happens while, uh, you see this extended backlog?

161
00:29:09.900 --> 00:29:19.020
And does an investment-grade counterparty like Microsoft get priced differently in your credit facilities in a well-funded but non-rated lab or an enterprise customer? Yeah. Uh, yeah.

162
00:29:19.060 --> 00:29:33.720
And, and, and another way to kind of ask that, i-i-is will there be more focus on those shorter term enterprise customers as evidenced by this financing that y'all just struck, or is this really just opportunistic, whatever comes your way?

163
00:29:33.770 --> 00:29:46.780
Y- you will see us allocating our portfolio very thoughtfully across different types of clients in order to drive that customer diversification. Um, we want enterprise. We want hyperscale cloud.

164
00:29:46.820 --> 00:29:54.280
We want hyperscale clients like Meta. Uh, we want AI labs. I, I think all of these channels are incredibly important to us.

165
00:29:54.460 --> 00:30:02.400
Um, but it's up to us to be able to allocate, uh, effectively across those channels, uh, uh, to, to keep growing them all accordingly.

166
00:30:03.300 --> 00:30:17.580
Um, cost of capital spread, uh, I, I think you've seen us finance investment grade at, like, call it plus 250. You've seen us finance non-investment grade at plus, call it, around 500-ish.

167
00:30:17.920 --> 00:30:21.430
So 250 bip spread of financing between those things. I...

168
00:30:21.500 --> 00:30:31.440
Overall, I think that that is a phenomenal advancement in our, um, cost of capital basis for financing both investment grade and non-investment grade counterparties.

169
00:30:31.980 --> 00:30:37.012
Um, and we're able to do so in a, um- Uh, very efficient manner.

170
00:30:37.152 --> 00:30:50.072
When I say that, like our transactions are now measured in kind of days to weeks to, to run these processes, as opposed to when we were a private co, it would take months to go and, uh, uh, do these financings.

171
00:30:50.852 --> 00:30:58.872
Um, so I'm, I'm very proud of our financing organization, and I consider it a, a, a critical asset and muscle that we built within our business.

172
00:31:00.352 --> 00:31:14.372
On, on the note of financing, I think, you know, to speak plainly, one of the things that made some investors and analysts, um, queasy following the call, although you didn't get any questions about this on the call, is specifically the debt load.

173
00:31:14.472 --> 00:31:26.472
Um, I think it's like 31 billion current, non-current, uh, as of the end of Q- Q2. Um, obviously, uh, you don't get a free lunch. You don't get all of this growth without having something to finance it.

174
00:31:26.852 --> 00:31:34.132
But one of the things that was notable is that non-recourse debt was gr- grew pretty heavily over the quarter. So I just have a question about that.

175
00:31:34.332 --> 00:31:38.852
Um, is the strategic goal to shift the mix towards more non-recourse project financing over time?

176
00:31:38.952 --> 00:31:52.282
Is that even possible, um, since, you know, CoreWeave is specifically running the GPUs and not having powered shells or building the infrastructure itself? Um, or is recourse debt still the default going forward? Sure.

177
00:31:52.292 --> 00:32:09.452
And how do you think about balancing the two? So w- with our most recent investment grade DDTL offering, that was the first time we introduced an offering as a non-recourse facility, right?

178
00:32:09.492 --> 00:32:17.532
There's no parent guarantee on it. Um, we were able to successfully kind of remove that, that covenant or that bar- that burden on the parent.

179
00:32:18.392 --> 00:32:34.192
I think going forward, that'll become status quo for investment grade SPVs or DDTLs that we're doing, where it's no longer necessary, uh, to have CoreWeave parent providing additional credit into that facility.

180
00:32:35.012 --> 00:32:45.192
I would expect that it'll remain on the non-investment grade DDTLs in the near term, but we'll be seeking to remove that in the future. At the end of the day, uh,

181
00:32:46.392 --> 00:33:02.852
we will seek to not have CoreWeave parent credit as a credit enhancer on these SPVs. Um, general debt load overall, like, th- this has been a confusing one to me.

182
00:33:03.132 --> 00:33:18.052
I'm, I'm gonna take two perspectives on it. One, when we're financing in a DDTL structure, we're typically, and outside of this DDTL 5.5, um, that was a, a smaller facility overall.

183
00:33:18.152 --> 00:33:33.292
Um, but typically the entire amortization of the facility takes place within the bounds of the contracts that we're signing, right? But it goes beyond amortization.

184
00:33:33.572 --> 00:33:47.092
It is inclusive of facility operations as well. So think about it logically. Revenue flows into an SPV based on the full amortization, which includes interest and principal. That gets paid down.

185
00:33:47.412 --> 00:33:59.312
The operation of the GPUs gets paid down. That's all the data center costs. And then you have a, I think we've disclosed roughly a 25% contribution margin that kicks up to the parent co afterwards.

186
00:33:59.592 --> 00:34:13.532
That contribution margin is just profit that goes to parent. So that debt is fully amortizes within the SPVs. It's being done at a decreasing cost of capital.

187
00:34:14.552 --> 00:34:27.592
Uh, we, we look at that as a benefit, right? Not, not a burden, 'cause, you know, the, the only other way to finance this stuff would be like massive equity issuances, right? And that's, you know, pretty, um,

188
00:34:29.072 --> 00:34:39.442
challenging for equity holders. Um, we didn't have the benefit of huge free cash flow to finance, uh, everything that we're doing, like our peer set, the, the hyperscalers.

189
00:34:40.132 --> 00:34:57.972
Um, but you can see like even they are dipping into debt now because at the end of the day, we are solving the massive demand of AI and the demands of our clients to deliver on an accelerated timeline and, uh, accessing credit markets responsibly to accelerate that timeline.

190
00:34:58.152 --> 00:35:11.052
I, I think that's our job, to go, to go solve that. The, the other variable that I think is not as appreciated on the debt side is just the incremental level of diligence that the equity investor gets, right?

191
00:35:11.252 --> 00:35:21.272
Like f- a- as an equity investor, you don't get access to our data center agreements. You don't get access to our, um, compute offtake agreements that we're signing.

192
00:35:22.032 --> 00:35:32.952
But within these credit facilities, these guys get everything, right? They go through it in intense detail to ensure that the data center contract is, um, bulletproof.

193
00:35:33.152 --> 00:35:38.152
They make sure that the offtake agreement for the GPUs is bulletproof.

194
00:35:38.632 --> 00:35:51.732
You're getting this extra level of underwriting on just the core of our business, the economics of our business from the credit market that I think is a huge value add for the equity investor.

195
00:35:52.212 --> 00:36:01.952
I, I just don't feel like that that is appreciated, um, enough in the market. Yeah, I mean, I think that's when you see some people flag the debt loads of all...

196
00:36:01.992 --> 00:36:05.392
and it's not just CoreWeave, it's literally every single company in the CapEx cycle. Yeah.

197
00:36:05.532 --> 00:36:13.152
Um, people often ignore the actual contracts that go behind that, and I think that's also ties back into the, like, what, what tenant would you rather have?

198
00:36:13.192 --> 00:36:28.472
You know, I would imagine that if you're having that recourse, uh, that recourse debt to the parent company, having those hyperscaler contracts make lenders and should make investors a little more comfortable with regards to the in- the, uh, to, to the credit profile of those offtakers of the compute.

199
00:36:29.532 --> 00:36:35.312
So appreciate the a- answers there. I've just got a couple more for you, Brandon, and then we'll get you out of here.

200
00:36:35.552 --> 00:36:46.504
Um- One of the things that w- came up in this Q2 was this New Jersey, uh, joint venture, uh, where y'all's stake went from 15 to 35% over in a year. I'm curious, at what point,

201
00:36:47.464 --> 00:36:58.924
uh, does that ownership level stop being kind of like w- um, unconsolidated, uh, VIE on paper and start being kind of a subsidiary of substance for you guys? And or does it- Oh, boy.

202
00:36:59.443 --> 00:37:07.144
[laughs] Um, that, that's something I would need our CFO here to be able to comment on, and I don't think that we're providing too much color there. Okay.

203
00:37:07.364 --> 00:37:20.004
I, I'm, and I'm just curious, so maybe a, um, a, a higher level question with that. You know, the JVs seem to be ramping up recently. I'm just curious about the, uh, the strategic calculus there. I- is this a,

204
00:37:21.124 --> 00:37:33.184
is this a capital allocation, uh, consideration? Is this just, uh, the deals make sense so you're signing them? Just curious as to why we're seeing them move into the picture now- Yeah... more than we have in the past.

205
00:37:33.504 --> 00:37:36.784
Yeah. I, I think it has to do with our data center strategy overall.

206
00:37:36.793 --> 00:37:48.484
Like, historically, we've been very focused on leasing data centers, as it was a efficient way to quickly access the market at scale and to leverage the market's expertise of delivering data centers.

207
00:37:49.204 --> 00:38:06.824
Um, you know, a- as we have gone through, I believe, 51 data centers in active operation today, which is just an insane number to, to have globally, we've developed a extremely robust skill set of, um, uh, developing and delivering data centers and working with our partners.

208
00:38:07.284 --> 00:38:15.544
So I, I think it's entirely logical for us to be doing self-build, and you'll see more self-build on the data center side coming out of us.

209
00:38:15.624 --> 00:38:22.784
But when we do that, um, I, I, I think it's less appropriate for us to do the full financing of that self-build, right?

210
00:38:22.824 --> 00:38:34.444
It makes sense for us to go out to partners to be a part of a JV for development, and to have other parties involved within it.

211
00:38:34.464 --> 00:38:49.524
And so I, I view doing JV partnerships on the build of CoreWeave-owned facilities, um, a pretty logical path for us to be going down, that w- we will expand as it makes economic sense to us.

212
00:38:50.564 --> 00:38:55.894
Yeah, that makes sense to me, kind of easing into a kind of more vertically integrated structure with some- Yeah... of these.

213
00:38:55.904 --> 00:39:08.804
You know, uh, one of the questions that often gets asked, and this isn't a question directly to you, but I think most people who don't swim in these waters, they, they, they can't really wrap their head around why you have all of these different parts of the value chain kind of split between these different companies.

214
00:39:08.844 --> 00:39:18.804
Like, "Well, oh, this company only runs the GPUs, and this company only runs the infrastructure." And most people don't realize that, uh, you're playing to these different companies' strengths.

215
00:39:19.304 --> 00:39:34.044
You're also dividing the capital expenditure burden, um, accordingly in a way that makes it more manageable for everyone involved. Uh, all right. I, I just have, um, I think one more question here, uh, for you.

216
00:39:34.804 --> 00:39:42.264
And let me see if I can get it here on my prep sheet. Uh, oh, this, this one is a margin question, and maybe this is o- again, one for the CFO.

217
00:39:42.764 --> 00:39:57.284
But the 5 to 10 point margin step up for the Q2 signings is pretty significant. Yep. And I'm curious, where is that coming from? Is this coming from better deals because compute demand is so voracious?

218
00:39:57.364 --> 00:40:08.764
Is this coming from CoreWeave software stack? Is it coming from advancements in GPU efficiency, or is it all of the above? Like, how are you squeezing out that extra margin on these- Yeah... new deals?

219
00:40:09.054 --> 00:40:15.164
'Cause it really runs afoul of the, um, some of the commentary of w- w- we should see ultimately...

220
00:40:15.824 --> 00:40:23.864
Or, or rather, you know, I think a long enough timeframe we'll see this, but people, I think, are expecting compute to commoditize much more quickly than we're actually seeing out in the field.

221
00:40:23.884 --> 00:40:36.484
I think it's a mixture of everything that you highlighted, but the ultimate driver here, it's just TAM, right? It, it's demand. And demand for this compute is just continuing to inflect higher.

222
00:40:36.624 --> 00:40:48.624
I feel like I say it every quarter. I'm, like, kind of running out of verbs [laughs] to describe what's, what's happening. Um, but w- what is underpinning that demand is ROI for our clients, right?

223
00:40:48.684 --> 00:40:56.644
Which I appreciate was, you know, a large market question for quite some time, which is like, look at all this investment in AI. Where's the ROI on it?

224
00:40:56.724 --> 00:41:12.884
And I, I think that the m- numbers that are starting to get out there are, like, incredibly supportive of, um, there being massive ROI on AI, uh, for CoreWeave clients and, you know, uh, uh, uh, AI entities across the space.

225
00:41:12.904 --> 00:41:25.644
And so that ROI finds its way back into pricing. Um, I think it all stems from inference, or a- as we like to call it, inference is really the monetization of AI, right? Um,

226
00:41:28.064 --> 00:41:36.024
i- that is truly, like, early innings, and it is accelerating in ways that are almost undescribable.

227
00:41:36.464 --> 00:41:44.764
So that 5 to 10 percentage point contribution margin increase, um, we, uh, we're very happy to report that for Q2 contracts.

228
00:41:44.884 --> 00:41:58.044
Um, we also reported a 25 percentage point just pricing increase in our rate cards across all SKUs, right? That's not a mixture. It's not selecting certain SKUs are getting a 25, uh, percent increase in price.

229
00:41:58.524 --> 00:42:13.804
It's across our entire rate card. Um, and it just all comes back to the fact that demand for AI is stronger than it ever has been, and, uh, uh, those consumers of AI, of AI are driving massive returns off of it.

230
00:42:15.084 --> 00:42:21.384
I'll probably wrap up going back to where we started, which is on just GPU depreciation, going back to the A100s.

231
00:42:21.444 --> 00:42:32.804
Which if I'm looking at it right, um, you guys, your, your A100s from earlier this decade fully depreciate in, or like around 2027.

232
00:42:32.824 --> 00:42:45.728
But the contract you signed runs through '29 or maybe later, so- Like, what does this, uh, mean for, like, revenue-generating assets with zero book value?

233
00:42:45.768 --> 00:42:58.148
And then do you think this applies to new g- new generations of rigs? And, like, what's the... Like, zoom out really quick and talk about GPU depreciation in general again. Yeah. We use six-year depreciation.

234
00:42:58.388 --> 00:43:06.408
I think our peer set is five and a half to six years or so. Um, I, I don't think you're gonna see us changing that anytime soon. But useful life is absolutely skewing beyond six years.

235
00:43:06.448 --> 00:43:19.328
I mean, we're, we're seeing in reality, and I, I think we've been quite vocal about this ever since we IPO'd, that useful life, um, absolutely can extend beyond its six-year, uh, depreciable life that everyone uses today.

236
00:43:20.548 --> 00:43:29.548
The way we look at it, and we're former commodity traders, right? We spent, uh, decades in risk management before starting CoreWeave. But, you know, it- it's just

237
00:43:30.588 --> 00:43:35.128
such attractive optionality that we're building by owning these GPU bases, right?

238
00:43:35.508 --> 00:43:48.948
Because i- if these assets truly are monetizable post their depreciable life, which it certainly looks like they are, it is fantastic margin profiles, right?

239
00:43:48.988 --> 00:43:59.868
To, to your point, you, you, you drop depreciation off these things, which, you know, is the largest part of the COGS on GPUs. Um, they're incredibly attractive to the asset holder at this point.

240
00:43:59.928 --> 00:44:18.948
So I, I think that we're building, um, an, an immensely valuable option set of cloud-connected revenue-generating infrastructure where demand, frankly, is just unsolvable in the near term, um, and e- even the medium term.

241
00:44:19.088 --> 00:44:33.768
Uh, we, we don't see a path to satiating the market's need for compute to serve AI workloads. Brannon, thank you so much for joining.

242
00:44:33.888 --> 00:44:46.628
And I just have to say, from ETH miner to 51 data centers under your belt, uh, just been truly one of the more impressive transformations within the tech sector. So again, hats off to y'all.

243
00:44:46.668 --> 00:44:49.128
Looking forward to what y'all have cooking for the rest of the year.

244
00:44:49.248 --> 00:45:00.688
Really appreciate you coming on, and, uh, would love to, would love to, um, round-trip this, uh, during Q through or end of year, uh, financials next year. So thanks for the time, man. That'd be great. Thanks, guys.

245
00:45:00.728 --> 00:45:12.288
Appreciate it. Thank you, Brannon. I ju- I just want to note something. A ama- like, Bitcoin miners just seething in some ways. Yeah. Like, imagine you buy the old...

246
00:45:12.508 --> 00:45:22.558
You buy these GPUs as an older model, and you're gonna depreciate it on the balance sheet, but you're still going to get to juice it for a few more years after that. Bitcoin miners could never dream of this.

247
00:45:22.588 --> 00:45:35.058
That being said, Charlie, I did just think then, you know, in, in a funny way, the AI pivot actually might help Bitcoin miners with depreciation. With the fact that hash price is, is kind of in a, you know, a...

248
00:45:35.848 --> 00:45:39.708
Is in the dog days right now, but hash rate is actually in a bear market.

249
00:45:40.408 --> 00:45:52.648
So you have a funny scen- uh, a funny situation within which, you know, like, S21s and maybe even S23s, that those generations might be able to stay online longer just because there's so much less competition, so.

250
00:45:52.708 --> 00:45:56.428
Look, we've seen that, we've seen that happen in Bitcoin for a long time.

251
00:45:56.688 --> 00:46:07.388
I will say, yeah, there's a lot of ETH miners who turned off thinking that that was it, um, when in fact, they probably should have tr- quadrupled down after the merge. Um, maybe myself included.

252
00:46:07.888 --> 00:46:22.128
Anyway, uh, we're gonna keep on rolling. Uh, we are now going to hand it back to Colin to do Nebius. Uh, but before we review Nebius, I will read, uh, an ad from our sponsor, Luxur.

253
00:46:22.488 --> 00:46:35.488
[firework launches] This episode is brought to you by Luxur's Commander, Bitcoin miner management software built for enterprise operations.

254
00:46:36.068 --> 00:46:50.228
Commander gives you real-time fleet monitoring, bulk remote commands across your fleet, and Intelligent Miner, which is an automated profitability engine that runs every five minutes, adjusting power settings to live hash rate and energy markets.

255
00:46:51.068 --> 00:47:05.268
ERCOT back tests show over 10% more profitability versus binary mining. Commander Pro is 100 bucks per megawatt or a 25 basis point pool fee adder, which is roughly half the price of competitors.

256
00:47:05.868 --> 00:47:21.808
Comes with a 60-day free trial. Get started at luxur.tech/commander. Colin, back to Nebius. Th- th- this is the, this is the freaking CoreWeave show, man. We're 52 minutes in [laughs] as of- I know.

257
00:47:21.948 --> 00:47:33.928
I mean- I mean- Not... It's a Neo- It's like NeoCloud, uh, madness today. Yeah, it is NeoCloud madness. And honestly, all, all the same, I think. We're spending so much time on CoreWeave. I mean, y- you really...

258
00:47:34.008 --> 00:47:48.648
Probably the most, um... You know, all said and done, one of the most exciting IPOs in recent memory. And so I, I imagine most people kind of want more here rather than less. But we will go ahead and

259
00:47:49.568 --> 00:47:59.368
focus on CoreWeave, uh, ch- their chief rival, we might say, Nebius. I mean, it is kind of a little red versus blue here, although Nebius' color scheme is green.

260
00:47:59.448 --> 00:48:04.848
I mean, in, in, you know, in, in, uh, Brannon's words, like, they don't see a path to servicing all the demand. Like,

261
00:48:06.008 --> 00:48:16.678
at the, at a time when, like, demand is infinite, [laughs] uh, your rival is really, where can you get the power and plug in? Right. So maybe that's where the com- competition lies.

262
00:48:16.728 --> 00:48:27.128
And that's kind of one question that I'd... I kind of wish we asked Brannon with regards- Yeah... to the data center moratoria all across the US, and some of the local pushback. They were asked that on the call.

263
00:48:27.348 --> 00:48:35.348
Gotta get, gotta get it back on then, so. Right. They were asked about it on the call, and I don't really know what else you say other than, "Well, you know, we'll just go to places that don't have those."

264
00:48:35.888 --> 00:48:47.148
But then everyone else is thinking that. Anyway, a, a decent number of headwinds, but if you can execute, I would imagine you are, uh, privileged over those obviously who haven't done as much, so.

265
00:48:47.404 --> 00:49:00.604
All right, let's go ahead and hop on over to Nevius. I will get the Blockspace coverage up here. Nevius Q2 revenue jumps four hundred and fifty-four percent as AI capacity pricing and margins rise.

266
00:49:01.504 --> 00:49:12.924
Again, that's coming at you all from Blockspace. You can go read the recap there, and we will start with the numbers here on the live stream. So revenue at five hundred and eighty-two point three million.

267
00:49:13.044 --> 00:49:27.544
Again, that is up four hundred and fifty-four percent from Q2 2025. Cost of revenue also up accordingly, a hundred and thirty-three point six million.

268
00:49:27.554 --> 00:49:40.944
My prep sheet just threw me for a loop here. Give me one second here y'all. Apologies. Here we go. One hundred and thirty-three point six million operate-- Uh, cost of revenue versus thirty point one million in Q2 2025.

269
00:49:40.984 --> 00:49:47.924
That's up three hundred and forty-four percent. SG&A up a hundred and fifty-five percent to a hundred and seventy-three point nine million.

270
00:49:48.394 --> 00:49:55.204
Depreciation and amortization also up to two hundred and fifty-nine point seven million. That's up two hundred and forty-five percent year over year.

271
00:49:55.804 --> 00:50:05.644
Loss from operations came in at neg-- Uh, came in at a hundred and seventy-five point nine million, up from a hundred and eleven point two million in the same period last year.

272
00:50:05.744 --> 00:50:17.584
Interest expense up two thousand three hundred and eighty-one percent. It was only four point eight million in Q2 2025. Now it's a hundred and nineteen point one million.

273
00:50:17.624 --> 00:50:29.124
Net loss came in at a hundred and ninety point four million, which is way down from a net gain of five hundred and two point five million in the same period last year.

274
00:50:29.144 --> 00:50:40.224
Adjusted EBITDA coming in at two hundred and thirty-six point two million, which is a sight better than the twenty-one million dollar negative EBITDA they posted at the same time last year. Now,

275
00:50:42.424 --> 00:50:58.854
regarding the net loss, this is a comp artifact. This is not deterioration of the company. Q2 2025's almost, or roughly half billion net income was almost entirely from a non-cash gain from revaluing equity securities.

276
00:50:59.444 --> 00:51:08.404
Strip that one off, uh, off, off of both periods, and the underlying business went from a heavy loss-making operation to adjusted EBITDA of two hundred and thirty-six point two million. Um,

277
00:51:09.544 --> 00:51:12.293
so just getting that out, out front.

278
00:51:13.004 --> 00:51:25.584
H1 income was-- An H1 net income was still positive at four hundred and thirty point eight million because H1 2026 also carried a seven hundred and eighty point six million equity reevaluation gain.

279
00:51:26.084 --> 00:51:34.964
So unlike Q2, the Q2 standalone, the six-month GAAP figure is flattered by the same mechanism working in Nevius's favor over the first full half.

280
00:51:35.064 --> 00:51:45.114
So little, you know, change in fair value kinda doing a lot of lifting there. Real story here, adjusted EBITDA has inflected, and the margins are expanding fast.

281
00:51:45.144 --> 00:51:56.964
Adjusted EBITDA margin roughly forty-one percent in Q2, up from a negative twenty percent margin a year ago. Cost of revenues also fell from twenty-nine percent to twenty-three percent of revenue.

282
00:51:57.144 --> 00:52:07.404
SG&A fell from sixty-five percent to thirty percent of revenue. And depreciation and amortization in absolute dollars was up two hundred and forty-five percent.

283
00:52:07.424 --> 00:52:17.584
But that-- But it fell as a share of revenue from seventy-two to forty-five percent. So when you account for the revenue coming in the door, those ratios all decreased over the course of Q2.

284
00:52:19.064 --> 00:52:35.644
Um, cleanest evidence is the scale you're working. Uh, you know, revenue is up and those ratios down as a result. Total assets over the quarter went from twelve point four, um-- Or this is over the last six months.

285
00:52:36.124 --> 00:52:48.684
Twel-total assets, twelve point four billion to twenty-eight billion. Property and equipment alone from five point five billion up to thirteen billion. Cash stands at eight billion, up from three point six eight billion.

286
00:52:49.324 --> 00:53:05.213
That's despite eight point one three billion of CapEx in H1. CapEx in Q2 alone was five point six six billion, 10X increase from the same period last year, which is, uh, pr-pretty phenomenal.

287
00:53:05.384 --> 00:53:16.264
Total debt went from four point one three billion to eight point five five billion. Operating cash flows plus two point two five billion in Q2.

288
00:53:16.584 --> 00:53:30.084
It's up from negative operating cash flows of one hundred and sixty-seven point nine million a year ago. And financing activities unsurprisingly ramped up in Q2, up two point eight six billion. So

289
00:53:31.804 --> 00:53:36.924
gonna go into operations here real quick, Charlie, and then I will shut up, toss it to you for some color.

290
00:53:38.224 --> 00:53:48.924
Nevius forward guidance annualized run rate revenue reached roughly three billion at quarter end, which is up fifty-six percent from Q1's one point nine two billion.

291
00:53:49.324 --> 00:54:02.394
Still tracking toward a seven to nine billion dollar ARR year-end target for 2026. Um, though the quarter's pace doesn't close that gap, they're going to need some acceleration in the second half of the year.

292
00:54:04.104 --> 00:54:17.324
Full year 2026 revenue guidance was reaffirmed at three to three point four billion, and adjusted EBITDA margin guidance held at forty percent. Contracted power target has been raised to five gigawatts by year-end.

293
00:54:17.384 --> 00:54:31.024
That is a bump up from prior guidance. While connected power guidance held at eight hundred to one-- Eight hundred megawatts to one gigawatt. Pretty big gap if you're taking that into account.

294
00:54:31.104 --> 00:54:43.144
Again, it seems like Nevius is saying that they thought-- If people thought that the first half of the year was bullish for them, then they're going to have to say, "Hold my beer for the second half," because they're going to need some serious ramp-up in capacity to meet some of these numbers.

295
00:54:44.344 --> 00:54:58.290
Interesting thing here, Charlie, with regards to how they're approaching financing for a lot of this ramp-up Our boy Rittenhouse Research has been extremely critical of Nbius because they're like, I think their 2030 or

296
00:54:59.500 --> 00:55:11.180
their multi-year gigawatt target is pretty insane considering how much they have outstanding with regards to financing. So far, customer prepayments have been doing most of the work.

297
00:55:11.260 --> 00:55:20.100
Customer prepayments reportedly covered 50 to 70% of associated CapEx, um, depending on the source. Their CEO said 50 to 60% on the call.

298
00:55:20.180 --> 00:55:32.480
The earnings summary separately says $9 billion of prepayments were expected in 2026, and 70% of Q2 contracts included upfront financing. So they are taking the upfront financing.

299
00:55:32.500 --> 00:55:44.240
They'll take a haircut on the initial terms of those compute deals as a result of that upfront ca- uh, upfront prepayments. But this also helps them moderate their debt load a little bit.

300
00:55:44.340 --> 00:55:54.920
So, you know, it's kind of deciding what, uh, you know, what trade-offs you wanna take there. Uh, another really interesting bit from the quarter.

301
00:55:54.960 --> 00:55:59.840
Capac-uh, capacity auction and short duration deals were introduced in Q2.

302
00:55:59.850 --> 00:56:13.420
These agreements are up to six months and priced at $40 to $50 million per megawatt, which is a premi- a big premium from the 20 to $25 million per megawatt long-term rate, so roughly two X what those long-term rates are.

303
00:56:13.480 --> 00:56:23.400
And management explicitly framed these as price discovery rather than a shift in strategy. Take or pay terms apply even to these shorter term deals. Interesting note there for management.

304
00:56:23.820 --> 00:56:40.320
It's like they're trying to like feel or grope around in the dark for like what is the proper price point for this compute, but they're not, they're not saying that they're moving towards these g- uh, shorter term deals or these auction capacity deals, but mo- mostly just trying to use it to feel out what the market's saying.

305
00:56:40.360 --> 00:56:48.180
All right, Charlie, I've got some management commentary lined up, but I'll pass it to you for any color or second takes that you have with regards to what we've combed through so far.

306
00:56:49.480 --> 00:56:57.960
Um, I don't have a whole lot of insight 'cause [chuckles] there's so many earnings that it's difficult for me to dive into one. The on- maybe the only take I have is that, um,

307
00:56:59.200 --> 00:57:12.340
the, their capacity auction, uh, doing some numbers with Claude looks like that, uh, the short-term duration deals, which are like under six months,

308
00:57:13.260 --> 00:57:26.140
are clearing at 40 to 50 million per megawatt versus the 20 to 25 million per megawatt long-term deals. So spot is like, like two X the forward,

309
00:57:27.040 --> 00:57:43.660
um, rate, and so that's pretty steep. That's pretty, that's pretty significant backwardation. So, uh, that means people are just paying a ton for GPUs today, and it's not that the GPUs aren't worth as much in the future.

310
00:57:43.680 --> 00:57:58.330
It's just that they're worth so much if you can turn them on immediately. I mean, look at the Bitdeer, um, deal from w- uh, with, uh, presumably Anthropic in Norway from last week. Volta. Anthropic- Volta.

311
00:57:58.330 --> 00:58:09.309
It's via, through Volta, yeah. Yeah, okay. With Anthropic via Volta, I mean, they're paying through the nose for this. It was like $5 million per megawatt. I for... Anyway, the numbers were, were, were crazy.

312
00:58:09.420 --> 00:58:24.570
So, um, you know, I don't see how anyone doesn't not need compute today. I mean, if the deals are going like, clearing like that for today, ima- you know, in, in six months, I don't see that...

313
00:58:24.580 --> 00:58:34.080
Like, we already know, like, there's an upper limit to how much, how many megawatts can come online, um, in the next six months. So like, and demand will probably only increase or stay steady.

314
00:58:34.160 --> 00:58:41.920
And so, like, to me, these future, these six-month rates are just gonna be in the same situation we're in today in six months, so.

315
00:58:42.380 --> 00:58:53.220
And I also wonder, you know, as CoreWeave, as Brandon McBee was talking about on our call, you know, they signed Caterpillar as an enterprise client. So it's like even companies that you don't think need compute

316
00:58:54.160 --> 00:58:59.930
are renting out compute. I thought Caterpillar provided the power. No. [chuckles] They're also the customer. [chuckles] Yeah.

317
00:58:59.940 --> 00:59:09.019
When he said that I was, when he said that, I was like: Wait, did I, did I miss something wrong? Aren't you Caterpillar's customer? No. [chuckles] It's nothing to do with- Yeah. Was, was I tripping with that? Yeah. Yeah.

318
00:59:09.030 --> 00:59:23.000
But it does make me wonder. Like, the fact of the matter is, if you are a large enough company and you have the cash for it, and you are due diligent- due diligencing how you can use AI to augment your workflows,

319
00:59:24.220 --> 00:59:29.500
that speaks to the demand picture right there. Because it's companies that you're not even thinking about looking for compute.

320
00:59:30.140 --> 00:59:38.580
And how many of those companies are going to be signing long-term contracts like the hyperscalers? I would imagine not too many of them unless they're really certain about what they want and need.

321
00:59:39.070 --> 00:59:47.540
A lot of them are probably in the R&D phase for what exactly is going to make sense for duration, for compute type, et cetera, for size.

322
00:59:48.180 --> 01:00:02.100
And so I wonder if over the next quarter we will see more of these shorter term deals becoming a larger segment of revenue for these neo clouds at larger, at higher margins, and I wonder how much that changes the equation

323
01:00:03.080 --> 01:00:12.009
for g- for people like CoreWeave, and this goes back to the analyst questions for their CEO. It's like, what, w- when these longer term contracts expire, what are you gonna do with those GPUs?

324
01:00:12.440 --> 01:00:25.260
Do you expect to roll them back into the same hyperscaler clients, or is there a world in which you are confident enough in short-term demand that you are going to roll them into a bunch of smaller contracts across multiple different off-takers

325
01:00:26.520 --> 01:00:38.220
in a way in which you can juice more margin for that compute? Yeah, you could probably- All of that just-... build a, build a cash flow, um, like powerhouse and use that to finance even more. So. Right.

326
01:00:38.350 --> 01:00:47.060
And that, and I mean, a good, a good point with regards to free cash flows. Maybe potentially they'll get to a place where they're having to lean less on debt. I would imagine that we're quite some time out from that.

327
01:00:47.140 --> 01:00:55.000
But to me, it's just a, I obviously don't have answers to this, but what it, what it surfaces for me are some pretty-

328
01:00:56.184 --> 01:01:14.344
Reasonable questions with regards to how much is the demand landscape shifting in the sense of not just, okay, obviously there's a lot of demand, but in what form does that demand take, and what does that mean for who captures the most out of the value chain here in the AI stack?

329
01:01:14.384 --> 01:01:25.114
Okay, I'll, I'll s- wrap up here pretty quickly, Charlie, just 'cause the, uh, honestly, the Q&A segment, uh, for Nebius was not that exciting. They had prescribed questions,

330
01:01:26.044 --> 01:01:38.604
pre-screened questions, excuse me, and those were allocated to specific executives. They didn't have a live analyst Q&A, and that kinda tells you something about management, I think.

331
01:01:38.704 --> 01:01:48.304
I mean, the fact of the matter is that's less dynamic. It gives management the, uh, benefit of being able to prepare the questions in advance.

332
01:01:49.024 --> 01:01:54.424
The most egregious example of this within Bitcoin miners, uh, w- was, was Mara.

333
01:01:54.464 --> 01:02:05.324
I mean, Mara, for a few quarterly earnings there for a while, literally just had, like, I think, like, one equity analyst on, and they would just ask questions that were pre-screened by Mara beforehand.

334
01:02:05.454 --> 01:02:15.864
Kind of a tangent, but that being said, the Q&A session was not as useful as some of the other ones. One of the more useful questions, though, came from Ryan Lance of Morgan Stanley.

335
01:02:16.324 --> 01:02:24.654
He was asking specifically about this Vineland, New Jersey site that has been in the news for two reasons, uh, that has been in the news recently.

336
01:02:24.754 --> 01:02:34.744
Number one, there is a kind of regula- regulatory snag that Nebius has hit. Number two, as a way to circumvent that or address it, Nebius is leaning on Bloom Energy.

337
01:02:35.364 --> 01:02:45.444
In fact, they mentioned Bloom Energy multiple times throughout the call, and Bloom Energy's stock ripped as a result. It was kind of seen as a validation of their entire fuel cell thesis.

338
01:02:46.194 --> 01:02:54.244
So the most consequential exchange of this call came about this question. It directly updates a risk flagged with that Vineland site.

339
01:02:55.504 --> 01:03:08.104
Lance asked what an adjourned without a vote hearing means for the ramp with regards to the regulatory tape that they are running into for this site. What management reframed it as procedural.

340
01:03:08.164 --> 01:03:18.284
The hearing concerns approval of an amended site layout plan. It was triggered by Nebius' switch to Bloom Energy fuel cells as the power source. So they said this is not a new issue.

341
01:03:18.324 --> 01:03:21.164
It's just a sign-off on a change already disclosed.

342
01:03:21.604 --> 01:03:32.184
Management called it, quote, "Part of the normal process," end quote, already built into the schedule, and said Nebius is, quote, "Confident that the layout complies with the applicable local, state, and federal laws."

343
01:03:32.954 --> 01:03:38.664
Um, they followed up by saying all contractually required tranches have been delivered on schedule to date.

344
01:03:38.764 --> 01:03:50.604
Construction of the building itself finished earlier this summer, and Bloom fuel cell deployment, quote, "Should be fast," end quote. Both, uh, management explicitly denied any expected timeline impact.

345
01:03:51.464 --> 01:04:00.904
Now, it seems to me like they framed it as a this isn't anything to worry about, but as I understand it, there are still regulatory hurdles to clear before that site can start generating revenue.

346
01:04:00.924 --> 01:04:17.464
So I think that was one of the bigger concerns on the call. [lip smack] Um, another question that was worth flagging is the capacity ramp-up, um, for the gi- for the gigawatts they wanna have come online. Um, [lip smack]

347
01:04:17.544 --> 01:04:25.694
management said almost all of the newly raised five gigawatt contracted power capacity comes online over the, quote, "Next two, three, maybe three point five years," end quote.

348
01:04:26.224 --> 01:04:34.264
Pretty big window, um, hedging their bets there with regards to being, you know, pigeonholed into having, to giving a tighter timeline on that.

349
01:04:34.784 --> 01:04:41.184
Management also disclosed that Nebius already has access secured to, quote, "Hundreds of megawatts," end quote, of behind-the-meter generation today.

350
01:04:41.844 --> 01:04:58.584
This reiterates the strategic partnership with Bloom as an accelerant for their site, even if they can't get immediate on-site generation or on, excuse me, on-site, uh, power draw. Last point with regards to financing,

351
01:04:59.644 --> 01:05:11.864
um, one analyst at Wolfe Research asked, "Given debt market volati- volatility and rising all-in costs, is Nebius still comfortable leaning on debt, or should investors expect more equity and convertible issuance?"

352
01:05:12.364 --> 01:05:26.604
The CFO answered that they are prioritizing, uh, financing as following. Number one, customer prepayments and operating cash flow. Number two, asset-backed debt against forty billion dollars in contracted backlog.

353
01:05:26.644 --> 01:05:37.224
Number three, corporate-level debt, currently, quote, "Almost none," end quote, is outstanding. And four, equity and equity-linked instruments at the market offering potential convertibles, et cetera.

354
01:05:38.464 --> 01:05:50.604
So interesting, they're basically saying the number one financing plum- the n- the number one f- uh, source of financing we will plumb going forward are those customer prepayments and operating cash flows.

355
01:05:51.084 --> 01:06:01.004
Little m- uh, d- divergence from what we've seen from CoreWeave, where they are specifically going after financing some of wha- most of which is recourse to the parent based on the contracts that they have signed.

356
01:06:04.984 --> 01:06:19.304
Uh, you're muted there. And Nebius up thirty-four percent today. Uh, pretty nuts if you think about it. CoreWeave up twenty-two percent. W- White Fiber up twenty percent,

357
01:06:20.424 --> 01:06:37.784
uh, as pictured here on DI Metrics, dimetrics.ai. Go check out that site. Colin, I think we round this out with some White Fiber that I'll take and lead after a word from our sponsor, Ligos.

358
01:06:38.224 --> 01:06:52.574
[static sound] Bitcoin is stuck at sixty-four thousand, and that may be good news to some of you. But I'm gonna tell you one thing.

359
01:06:52.984 --> 01:07:01.944
If you're using a Bitcoin-backed lender that's playing funny games with your coins, it doesn't matter if Bitcoin seems to have reached a local bottom. They can still fumble your bag.

360
01:07:02.776 --> 01:07:10.776
Now, if you're using Lagos Finance, you don't have to worry about that. They are our preferred Bitcoin-backed lender here at Blockspace, and I'm gonna tell you why.

361
01:07:10.796 --> 01:07:17.516
They use Bitcoin native smart contracts to make sure you're always in control of your keys. You heard that right.

362
01:07:17.896 --> 01:07:29.216
When you're using a Bitcoin-backed loan from Lagos Finance, you don't have to worry about your funds being rehypothecated, because you can always have them in self-custody.

363
01:07:30.076 --> 01:07:44.076
With Lagos Finance, there's no rehypothecation, no wrapping, no bridging, no funny business, just competitive rates as low as 8.5 to 10% APR. Go to lagos.finance to learn more. All right.

364
01:07:45.916 --> 01:07:56.616
WhiteFiber, which by the way, we have WhiteFiber CEO Sam Tabar on the show tomorrow. Originally scheduled today, pushed to tomorrow, so put that on your calendars.

365
01:07:57.816 --> 01:08:14.856
WhiteFiber, as reported in Blockspace this morning, WhiteFiber raises, uh, revenue rises 54% as NC1 in North Carolina starts billing contracted capacity. Um, they beat revenue expectations this morning, and why is that?

366
01:08:14.936 --> 01:08:32.336
It's because a customer paid them to leave. So WhiteFiber printed a Q2 revenue of 28.8 million, uh, up 54, 55% with a $15 million net loss. The stock is up big.

367
01:08:32.456 --> 01:08:37.796
Now up, I believe, as I said, 20.4%.

368
01:08:39.176 --> 01:08:52.756
So of that revenue, which is up, uh, it's up because 12.3 of that revenue, 12.3 million of that revenue, is a one-time termination fee from their original anchor customer walking away.

369
01:08:53.956 --> 01:09:01.416
Outside of that customer cloud revenue for the company was roughly 11.5 million, which is down 30% year-on-year.

370
01:09:02.776 --> 01:09:19.896
So, uh, they tech-- If you drop the, the termination fee, they technically missed earnings, which were, uh, 19.7 million estimated. And, um, sans termination fee, uh, was 11.5 million. However,

371
01:09:21.156 --> 01:09:25.476
forward looking for the company is quite bullish.

372
01:09:25.576 --> 01:09:43.816
So NC1 in Madison, North Carolina, began billing customers and the, uh, and they expect the full contracted run rate billing across 40 megawatts of critical IT load later in August.

373
01:09:44.556 --> 01:09:52.656
So it's a, an $865 million 10-year lease with a total backlog crossing a billion dollars for the company.

374
01:09:53.536 --> 01:10:11.336
And then, uh, this August, in their announcement, they announced that they brought in roughly s- $360 million of fresh GPU contracts, including their, uh, the company's first, uh, Nvidia Vera Rubin deployment, which I'll circle back on in a second.

375
01:10:12.616 --> 01:10:26.276
Um, their remaining performance obligations, uh, or RPO of contracted future revenue not delivered, is just right at a billion dollars.

376
01:10:26.536 --> 01:10:39.696
Their multiple on invested capital, um, uh, comes from a parent loan requiring paying back at 1.1x the principal minimum, uh, regardless of how fast.

377
01:10:39.736 --> 01:10:47.436
It's a 10% minimum return on 90 new money and which is why the annualized rate on that is at 50.6%. Um,

378
01:10:49.876 --> 01:11:04.376
they also paid, WhiteFiber also paid $120 million of the convert proceeds for the right to receive 5.9 million of its own shares in 2031. This is basically a prepaid buyback that offsets convert dilution.

379
01:11:05.156 --> 01:11:14.636
So here's some, there's some bullish and some, uh, arguably bearish takes from the earnings. Uh, NC1 in North Carolina, billing has commenced. This is bullish.

380
01:11:14.676 --> 01:11:24.016
And per this morning's release, Sam Tabar says that they expect full contracted run rate billing across the 40 megawatts of contracted critical IT load later this month.

381
01:11:25.076 --> 01:11:40.576
This means in Q3, which we are entering into now, is a major revenue inflection quarter for WhiteFiber. Um, this is also just a blitz of money from like prepaying customers. Uh, one customer, Base10,

382
01:11:41.936 --> 01:11:54.326
165.2 million, buying 1,392 Blackwell 300s. Prime Intellect with 108.2 million for 576 Vera Rubin 200s.

383
01:11:55.256 --> 01:12:06.476
Uh, and, uh, 87.5 million for 576 B300s in Iceland. So that brings a total of 635 million of new contracts signed year to date.

384
01:12:07.136 --> 01:12:26.456
On the Vera Rubins from Prime Intellect, this is the first Vera Rubin deployment for WhiteFiber, which interestingly, um, this is, uh, according to the, the quarterly earnings today, the first disclosed, uh, Rubin customer from a small cap Neocloud.

385
01:12:27.416 --> 01:12:35.976
So the big hyperscalers are getting them, but this is the first like smaller cap company to actually, uh, get this deal from Nvidia.

386
01:12:36.036 --> 01:12:50.956
Nvidia, as we know, kind of plays a bit of kingmaker, both in financing and then who gets their primo hardware. So, uh, my take on this is that getting allocation to brand new Nvidia silicon is like a stamp of approval.

387
01:12:52.516 --> 01:13:09.540
Um, so shout out to them. Um, they have exclusivity with a consortium of lenders on NC1, um, now in diligence and documentations. If it closes, the 50.6-...

388
01:13:09.610 --> 01:13:26.220
percent, uh, annualized bridge, uh, loan, uh, kind of solves itself, and, uh, that capital goes back into the 1.5 gigawatt pipeline. Um, some bearish takes. Uh, organic,

389
01:13:27.500 --> 01:13:29.620
their organic quarter on quarter went backwards.

390
01:13:30.260 --> 01:13:41.559
Um, their monthly GPU revenue de- actually decreased, um, because of the customer termination, but the aggregate revenue, uh, total revenue increased from the, uh, termination fee.

391
01:13:41.660 --> 01:13:51.660
Um, and then I don't know if this is bearish or bullish. They have seven customers compared to the CoreWeave's three, as we had earlier. It's like these don't seem like a lot of customers.

392
01:13:51.700 --> 01:13:59.740
Typically, you want to follow like a Pareto principle or something, but- Well, CoreWeave has more than three, but three are 72% of the business. Exa- well, okay, exactly.

393
01:14:00.280 --> 01:14:10.920
And in this case, the largest of the seven customers for, uh, White Fiber, the largest is 63% of the first half of the year revenue.

394
01:14:11.820 --> 01:14:25.020
So, and then the entire growth of the company in, at NC1 is one tenant, which is pretty par for the course. It's just, it's a kind of highly concentrated, narrow, uh, client base.

395
01:14:25.500 --> 01:14:38.900
And then, uh, the company's not currently very liquid. They have 56 million in cash, 28.6 working capital deficit, um, and they have- 486 million in cash? I think so. [laughs] If I'm looking at...

396
01:14:38.960 --> 01:14:52.440
Yeah, if I've, if I've gotten... Yeah. Um, so, uh, they have 485 million of construct- of, uh, construction in progress, and they had to draw another, uh, 30 million of the parent bridge money, uh, last month.

397
01:14:53.520 --> 01:15:02.360
So, uh, this could be, uh, financing, uh, this, you know, could be the, the financing, uh, crunch.

398
01:15:02.900 --> 01:15:13.400
Um, I can toss it to you, Colin, but I think more interesting to me, 'cause I'm not really that much of a finance guy, kind of something like that, I'm interested in like the GPU and per megawatt comps, which I've got pulled up, which are pretty- Yeah, those are super interesting.

399
01:15:13.440 --> 01:15:22.140
Let's get into those. Okay. This is what's really interesting. Finally to the part of the, the part of the segment I understand better.

400
01:15:23.080 --> 01:15:36.260
Um, Inscale's deal at NC1 is 865 million for 10 years on 40 megawatts of critical IT, which comes out to 2.2 million per megawatt of critical IT per year. Um,

401
01:15:38.380 --> 01:15:53.450
this is pretty well inside the range that we're seeing everyone else, so it reinforces, uh, pretty much industry standard. Um, now, uh, the, the difference is, is that, uh,

402
01:15:54.500 --> 01:16:09.640
Inscale is not technically directly the, uh, investment grade credit. They are backed by, they have Microsoft, but they themselves are kind of like a, um, operating arm of that company. Um,

403
01:16:11.160 --> 01:16:27.760
and then Cerebras, uh, comes in also at about 2.4 million, uh, US dollars per megawatt per critical IT per year. Colin, do you have any thoughts on the per megawatt power rates?

404
01:16:29.360 --> 01:16:46.780
Um, White Fiber is... Uh, my, my, my question, uh, let me see. Let me just read this really quickly. I mean, they're a Neocloud, right? Mm-hmm. I mean, uh, you know,

405
01:16:47.960 --> 01:17:02.679
I, I would like to run some analysis really quickly with regards to like, uh, compar- comps, but that's like $2.2 million per megawatt is kind of almost in the PowerShell range of megawatt- Yeah, it is... which,

406
01:17:04.160 --> 01:17:12.520
you know, part of the bull case for the Neocloud, this is why Iron has, you know, outcompeted most of its former Bitcoin miners.

407
01:17:12.769 --> 01:17:23.780
Um, I mean, they're vertically integrated, so this is part of it too, but the idea being that like owning the GPUs is the most valuable part of the chain currently for anyone selling compute or moving into co-location.

408
01:17:24.520 --> 01:17:27.800
So m- my question would just be, you know, what,

409
01:17:28.760 --> 01:17:40.720
w- h-how do you square $2.2 million per megawatt versus PowerShell getting the same thing for just running the infrastructure and having much less risk, also much less CapEx, right?

410
01:17:41.260 --> 01:17:53.340
And also, what does it say about the value of what this, uh, of this compute with regards to how it is being used, um, at the end of the chain? So I mean, that, that's the first thing that stood out to me.

411
01:17:53.820 --> 01:18:02.060
I would have expected- Okay... it to be a little bit higher, but I need to hold my tongue and actually like do some analysis with regards to like what are we seeing with what Iron's getting?

412
01:18:02.100 --> 01:18:13.120
What are we seeing with what CoreWeave and Nebius is getting in terms of dollars per megawatt? I don't think the deals are as transparent for the bigger Neoclouds, though I, I will say that much.

413
01:18:13.160 --> 01:18:25.220
They don't, they're not disclosing these numbers, I don't think, a- as granularly as White Fiber is, which, you know, they don't really have to because they're the two biggest sharks in the pond. Yeah.

414
01:18:25.880 --> 01:18:29.790
Well, let's look at GPU hours. Um, so, uh,

415
01:18:31.060 --> 01:18:43.000
roughly, so the Prime Intellect deal with, uh, the Vera Rubin works out to about 188,000 per GPU over three hou- over three years, which is seven bucks per GPU hour at full utilization.

416
01:18:43.780 --> 01:19:00.560
And Base10's B300 deal at 119K per GPU, uh, comes out to 4.5 bucks per hour. Uh, so if we were to compare Morgan Stanley

417
01:19:01.480 --> 01:19:14.044
saying the Vera Rubin 200's NVL 72 rack costing at 7.8 million, which is double the, the B- Blackwell then- Um,

418
01:19:15.084 --> 01:19:25.584
576 GPUs equals eight racks equals 62 million of hardware against an 108 million of contracted revenue. So, uh,

419
01:19:27.144 --> 01:19:37.944
this does reveal a bit about the cost of the Vera Rubins, their rental rates, uh, their, their actual contracted rental rates.

420
01:19:38.064 --> 01:19:51.944
And, uh, can at least, like, now we have a great, like, reference point in real contracted, uh, numbers for this particular GPU stack.

421
01:19:52.004 --> 01:20:03.874
I don't have a lot of insight to this at the current moment, but it's great to have, like, a number from Morgan Stanley and then s- and then, um, settle... And then, uh, contracted, uh, on the White Fiber side.

422
01:20:06.024 --> 01:20:15.224
Um, another story that I ha- another little interesting insight from this is, you know, on the GPU depreciation story that we were talking about earlier with Brandon at, from CoreWeave.

423
01:20:16.444 --> 01:20:32.744
Um, White Fiber actually sold some used H- H200s for 26.1 million, and they had a book value of 24.3 million, so they gained money on these H200s, these used H200s.

424
01:20:33.504 --> 01:20:40.444
So they're clearing above their book value. Uh, this contradicts the GPU, uh, depreciation schedule.

425
01:20:40.544 --> 01:20:58.204
Again, the theme this week has been for Neoclouds, GPUs aren't depreciating as much in their useful life, and actual market value is higher than book. Um, so, uh, yeah.

426
01:20:58.584 --> 01:21:12.804
Um, I have one last comment, which is kind of an open question for Sam, who's on tomorrow. Um, but- Before you get to that, uh, we need to fact-check ourself and wipe, wipe some egg off of our face. Okay.

427
01:21:12.924 --> 01:21:14.764
Yeah, I thought that this didn't make sense to me.

428
01:21:14.904 --> 01:21:22.324
So yeah, White Fiber, it, it's weird because some of their language suggests that they're actually running the GPUs, but they're, they're just a powered shell for these deals. Okay.

429
01:21:22.364 --> 01:21:34.664
So 2.2 to $2.4 million per megawatt is, is on, uh, is, is, is kind of in the ballpark of what we've been seeing from most of the- Okay... board. So. Okay, great. Um,

430
01:21:36.024 --> 01:21:53.124
okay, so, uh, question for Sam tomorrow is the 10-Q says that Duke Energy, um, s- the study from Duke Energy, uh, supports up to 200 megawatts gross at, uh, the North Carolina site NC1.

431
01:21:53.144 --> 01:22:05.264
However, their press release this morning says that that site has the potential to scale up to 300 gross megawatts, so leaves 100 megawatt gap. Curious where the extra power comes from.

432
01:22:05.364 --> 01:22:23.264
So that's kind of interesting, uh, disparity between the Duke study and the, uh, uh, uh, their, uh, their earnings this morning. So that's all I've got for, uh, White Fiber.

433
01:22:23.704 --> 01:22:38.084
Um, you can toss any comments on it or- I, I don't really have anything else, just, uh, eager to get Sam on to talk about some of the... I'm curious about that contract that lapsed,

434
01:22:39.004 --> 01:22:47.804
um, or that the, you know, that was pulled. Um, few questions for him tomorrow, but if y'all are interested in learning more, tune in. Uh, we really don't cover White Fiber that much.

435
01:22:47.864 --> 01:22:50.184
We should probably add them- No, we don't. We probably should... to the coverage.

436
01:22:50.764 --> 01:22:59.964
You know, they were spun off from Bit Digital specifically for the, uh, HPC segment, and then Bit Digital, I think, was like being rebranded as an eth strategy play.

437
01:23:00.824 --> 01:23:13.404
I, I don't really know what's [chuckles] going on with that. I would imagine that that's probably, uh, not really top of mind for most people. So I think we can go ahead and move on to the, uh, to the final segment here.

438
01:23:17.444 --> 01:23:29.804
Yeah, which I totally have pulled up. [laughs] Um, I don't have the story pulled up, uh, right now. It's all right.

439
01:23:29.943 --> 01:23:43.144
Just give us the TLDR, 'cause this is an interesting one with regards to how the entire crypto world is reshaping itself in the image of the AI machine god. Yeah. So here's the, the TLDR.

440
01:23:43.324 --> 01:23:56.164
Uh, Wintermute, which if you're in crypto, you know their name, they're the largest market maker. And, uh, Wintermute plans $1 billion AI push into traditial-- into traditional markets. Um, say no more.

441
01:23:56.244 --> 01:24:02.084
I mean, uh, this is one of the most profitable, successful market makers in crypto, uh,

442
01:24:03.264 --> 01:24:15.824
who is realizing that they need to, uh, start trading against the likes of Jane Street and Citadel, and actually playing with the big boys because mar- the Bitcoin market is now a dwindling...

443
01:24:16.224 --> 01:24:29.544
Not just Bitcoin, crypto broadly is a dwindling market cap. So they intend to invest about a billion dollars into high-frequency trading and artificial intelligent data center construction over the next five years.

444
01:24:31.464 --> 01:24:50.464
Uh, founder and chief executive offer, uh, officer Evgeny, uh, Gavril in this, uh, interview for the story said, "We are now going up against firms that have spent decades optimizing their technology and infrastructure for these markets, so obviously the level of investment required is significant."

445
01:24:51.664 --> 01:25:02.184
As per the Bloomberg article, the $1 billion commitment underscores how steep the price of admission into Wall Street's high-octane markets has become. Uh, so,

446
01:25:03.624 --> 01:25:14.754
uh, Alex Garco's quant trading firm, XTX Markets, said last year that it would spend a billion dollars building five data centers in Finland, uh, for machine learning.

447
01:25:14.904 --> 01:25:21.828
So Jane Street also- Absolutely dumping money into data centers to keep pace with computing power demands.

448
01:25:21.948 --> 01:25:35.968
Note, one of those data centers down the street from me, about an hour south of me, uh, in a little town called Okmulgee, I believe. So, um, you know the joke is, Colin,

449
01:25:36.908 --> 01:25:44.788
uh, whenever you've been trading shit coins, uh, you've been trading against Wintermute, and this is more or less kind of confirmed in, in, at different levels.

450
01:25:45.308 --> 01:25:54.148
Now Wintermute gets to, uh, see the other side of the coin, 'cause they'll be trading against Ken Griffin and Jane Street. So. Yeah. That's, yeah.

451
01:25:54.528 --> 01:26:08.518
We'll see what happens when they have to go gloves on against some of the titans of, of capital. Yeah. It's kinda like, you know, Jake Paul fighting Mike Tyson. Yeah, but Jake Paul got paid, so he still got paid a lot.

452
01:26:08.528 --> 01:26:17.628
No, my point though is it's like he was always, like, doing these boxing fights against, like, older boxers, and it's like, okay, like, how about you go up against, like, the current heavyweight champion?

453
01:26:20.028 --> 01:26:25.008
I mean, you could, but, um- You can get your sm- your first conviction...

454
01:26:25.018 --> 01:26:40.158
for a, for a, yeah, for a, for a kid who ma- who got famous on Vine and then transitioned to YouTube with his goofy rapper wannabe brother, I would say going up and even beating George Foreman, not in his prime, beating George Foreman.

455
01:26:40.668 --> 01:26:54.458
Not George Foreman. Mike Tyson? Mike Tyson, my bad. Yeah. I mean [laughs] Yeah. Yeah. Dude used to be on Disney Channel. Yeah. And the kids these days don't understand how great Vine was. TikTok- It wasn't-...

456
01:26:54.458 --> 01:27:10.248
is a shadow of what Vine used to be. Once again, Vine was ahead of its time. Uh, so on that note, that's the show today. Um, I will call everyone's attention to, once again,

457
01:27:12.088 --> 01:27:22.068
earnings week. We're halfway through. Today's Wednesday. Tomorrow, Thursday, we have Ben Gagnon, CEO and captain of Keel. Sam, push from today to tomorrow for WhiteFiber.

458
01:27:22.488 --> 01:27:31.648
Soluna watch party tomorrow as well on our channels, and then Mike Pitchy of Solana, sorry, Soluna, on Friday.

459
01:27:31.728 --> 01:27:40.208
Thank you very much for watching Block Space Live every w- day, every win- every weekday at 1:00 PM Eastern, uh, we are Compute's daily live show.

460
01:27:41.488 --> 01:27:55.907
Shout out to our sponsor, brought to you by CleanSpark, NASDAQ listed, ticker CLSK. I'm Charlie. I'm Colin. We'll see you tomorrow. [outro music]