WEBVTT

1
00:00:00.100 --> 00:00:16.820
[upbeat music] What's going on, y'all? Welcome back to Blockspace Live, brought to you by CleanSpark. To kick off today's show, markets are bleeding. Why, Charlie? Everyone's signing these deals.

2
00:00:16.920 --> 00:00:24.820
Uh, what's, what's the problem? Is... No, someone missed the memo. It's a big CapEx party, and you're not a part of it because prices are dropping.

3
00:00:25.160 --> 00:00:36.120
We've got that at the top of the show, and we will follow that up with a quick note on SpaceX and why, if you're thinking about it as a spaceship company, you're doing it wrong. It's a compute company now.

4
00:00:36.180 --> 00:00:41.000
It is a sto- it is a rocket ship company that identifies as an LLM AI company.

5
00:00:41.420 --> 00:00:54.060
Following that, we've got news on Cipher's new 2031 note priced at 6% for a data center that will be furnishing compute to Amazon Web Services. More on that deal.

6
00:00:54.100 --> 00:01:04.060
We're gonna get into the details, and then follow it up with another deal, Applied Digital signing another hyperscaler lease for an undisclosed tenant, probably the same one at two other facilities.

7
00:01:04.099 --> 00:01:13.360
We will dig into those details as well. And then we will wrap up with a quick note on China pouring an investment to close the gap in the AI arms race with the US.

8
00:01:14.560 --> 00:01:27.880
Blockspace goes live every weekday at 1:00 PM Eastern, featuring quick hits on the latest in AI data centers, some Bitcoin, some tech, and especially new emerging tech and markets.

9
00:01:28.540 --> 00:01:39.130
Make sure to like and hit subscribe, leave us a review if you're listening on a podcast feed, and hit that notification bell if you're watching on YouTube so you get the push notifications.

10
00:01:39.160 --> 00:01:46.140
That's right, this becomes a podcast after the live stream wraps up. You can find it anywhere podcasts are found, Spotify, Apple, et cetera.

11
00:01:47.340 --> 00:01:51.860
Speaking of podcast feeds, if you're listening on CoinDesk, we are leaving very, very soon.

12
00:01:52.380 --> 00:02:03.620
So go to Blockspace's feed on whatever platform you're watch- you're listening or watching on and subscribe to that, because we're leaving CoinDesk very soon.

13
00:02:03.680 --> 00:02:18.100
This show is brought to you by CleanSpark, NASDAQ listed, ticker CLSK. More on CleanSpark later on in the show. And Colin, we are introducing a markets segment. [drum beats] Drum roll.

14
00:02:18.460 --> 00:02:31.130
And part of the gimmick for this market segment is that depending on whether the markets are up or down, we will have either a red or a green background. So I'm introducing and rolling out

15
00:02:32.040 --> 00:02:45.380
the red background because we're kicking it off on a down day in the markets. We are taking a shot of red chili and running straight to the head. [laughs] As, as they say down in New Mexico, it's Christmas,

16
00:02:46.980 --> 00:02:58.500
and whether or not you're getting red or green depends on where the winds are flying. But yeah, everything except for healthcare, financial services, commercial is down today. Yeah.

17
00:02:58.560 --> 00:03:07.420
Yeah, a little bit, little bit of a Christmas mix here, but for our purposes, tech and communication's taking a huge hit. Yeah. Tech down 5%. We don't care about healthcare.

18
00:03:07.480 --> 00:03:21.680
We don't care about consumer defense right now, or real estate. We're a technology, communication, energy, and industrials, uh, show, and technology is, like, 25 to 30% of the market right now. As a sector, down 5%.

19
00:03:22.380 --> 00:03:35.040
And as sub-sectors within technology, semiconductors are leading the plunge down 6%. Consumer electronics, software infrastructure, et cetera, et cetera, et cetera. The whole sector, red.

20
00:03:35.280 --> 00:03:49.320
Let's look at some of the companies that we care about. First of all, HUT8. HUT is down now 7%, plummeted today. It's slightly recovered in the past hour.

21
00:03:49.340 --> 00:04:02.260
But you could-- this is be- gonna be a theme across the tickers we look at this morning. HUT8 down. Uh, what's going on? No particular major news, I believe that is a negative catalyst here.

22
00:04:03.100 --> 00:04:19.620
Um, I think one thing that we might, uh, that, what we might, uh, take from today is that the market may feel that, uh, AI HPC factories might be a little overvalued right now, just on this random Tuesday. Colin,

23
00:04:20.579 --> 00:04:35.710
you got any thoughts? Yeah. I... A few things here. If, if we look at, uh, the earnings season that just wrapped up, so Q1 earnings season is behind us. Uh, I believe 100% of S&P 500 companies have reported.

24
00:04:36.560 --> 00:04:52.420
And if you look at earnings growth year over year for Q1, info technology grew at 55% earnings growth. That is absolutely bonkers, and we've seen that reflected in the stock prices, obviously.

25
00:04:52.580 --> 00:04:56.969
All of these AI and semiconductor adjacent stocks have been absolutely ripping.

26
00:04:57.520 --> 00:05:07.800
So to me it seems perfectly reasonable, but that at a time when the market is displaying some weakness, these things would sell off pretty hard because they have been bid up very hard to the upside.

27
00:05:08.380 --> 00:05:21.240
And it seems like what we're getting right now is a continuation of last week's market woes. The NASDAQ lost, like, 4% last week as the market cratered on Thursday and Friday. S&P was also down the week.

28
00:05:21.540 --> 00:05:30.740
The Dow was the, uh, one of, the only of the big three indexes that didn't lose as much. I believe it was down just a few, a few bips, like, like 20 or something like that.

29
00:05:31.820 --> 00:05:40.440
But when you factor in the meteoric growth we've seen, and their revenue growth was 32% year over year, I mean, those are just gangbuster numbers.

30
00:05:40.500 --> 00:05:52.360
So it makes sense that they would be reverting down to maybe a more sane level considering valuations and PE ratios and all of those other financial metrics have just been absolutely surging as of late.

31
00:05:52.370 --> 00:06:05.500
The other thing I wanted to share, Charlie, 'cause I think it's really relevant to this conversation as we look at a warish Fed, who many believe will be more hawkish than his predecessor-The Dixie, the dollar index is rising.

32
00:06:05.680 --> 00:06:16.780
Now, it hasn't moved a crazy amount, but if we look over the last week, it's up from about 99 to 100. And a lot of traders and a lot of analysts are starting to look at rate hike odds.

33
00:06:16.800 --> 00:06:22.500
The CME is now pricing the likelihood of a, of a rate hike by the end of the year at, I believe, like 75%.

34
00:06:23.190 --> 00:06:40.940
And they didn't do it on the most recent one, but if we look forward to, uh, uh, the July, September, October, and December Fed meetings, you can see that the, uh, odds of a hike of at least 25 basis points, in some cases 50 basis points down the road, are starting to rise.

35
00:06:41.000 --> 00:06:49.320
A lot of people have, uh, th- they're worried about inflation fears given the price on energy, specifically oil, as a result of the Iran conflict.

36
00:06:49.700 --> 00:07:07.960
We haven't seen any inflation yet, but that might be coming in Q3 or Q4 when some of these, uh, you know, uh, inventories expire that were based on the lower oil prices, and we start seeing inventories refreshed later in the latter part of the year on new contracts forward-looking for,

37
00:07:08.980 --> 00:07:19.350
uh, a- anything and everything. That, and that also includes in a big part food, right? Because fertilizers are downstream of oil production, and, uh, farmers are complaining that fertilizer costs are skyrocketing.

38
00:07:19.400 --> 00:07:26.780
The price of food's likely going to go up, and w- everything's downstream of oil in our modern society. So inflation fears are rearing their heads again.

39
00:07:27.240 --> 00:07:38.180
The dollar is strengthening as a result of, I believe, expectations that the Fed will hike rates at some time at the end of the year. It's qui- it's funny how quickly we went from rate cuts to rate hikes.

40
00:07:38.260 --> 00:07:48.680
So to me, this is completely rational if you look at how gangbusters these stocks have been going, and you look at potentially a strengthening dollar throughout the rest of the year. Other assets obviously feeling it.

41
00:07:48.740 --> 00:08:04.850
Bitcoin cratered to 60K last week. Gold and silver are getting hammered right now. Gold's almost down to 4,000 an ounce. Silver is roughly at 63 right now. Silver was at 120 in Q1. So yeah, um- It's coming red.

42
00:08:04.850 --> 00:08:19.440
It's not just a secular thing within these AI stocks, right? Yeah. The dollar being behind this, uh, it's like the Scooby-Doo mask, you know, you pull off the, [laughs] the, the mask of the, the villain the whole time.

43
00:08:19.500 --> 00:08:35.820
It was actually the dollar the whole time. [laughs] So, uh... Okay, let's breeze through this. Um, uh, Iron down 9, 10% on the day. Uh, also caught a little bit of a bid in the past hour. Um,

44
00:08:36.740 --> 00:08:50.670
you know, if we go back, we look at, say, the one-year or five-year, look at this. I mean, this, this is the chart that the crypto nerds always wanted, that 10, four, 10 to 40 X plus the volatility all along the way. So

45
00:08:51.740 --> 00:09:03.350
volatile, but volatile to the upside is what these stocks have shown. We also have Keel down 4% on the day. It initially jumped, I think it was up 10%,

46
00:09:04.280 --> 00:09:15.520
uh, but als- that's also, uh, uh, currently down 4% on the day. Who knows? Like, by the end of this stream, we could be like Moses holding up the staff so the Israelites can win.

47
00:09:15.560 --> 00:09:30.340
The longer we stream, the more the markets will recover and go up, so. Yeah, and just for broader context, the S&P is down 1.5% today, and the Nasdaq I would imagine is looking even worse. Oh, wow, that's really odd.

48
00:09:30.400 --> 00:09:41.100
Oh, that's the... That is the exchange. Yeah, Nasdaq's down 2.5%. So like we were saying, this is infecting a lot of sectors.

49
00:09:41.200 --> 00:09:58.930
It's a mixed bowl right now, obviously, with anything that's not AI or semiconductor adjacent is doing okay, but I, I do think that it makes sense that we're getting a, a bit of a reprieve or a reversion after the fact that, I mean, the fact of the matter is, year to date,

50
00:10:00.000 --> 00:10:12.340
m- many of these stocks are up double digits. Yeah. And the run that we've seen so far this year has been crazy. So Nasdaq and S&P and these data center stocks are recovering as we speak.

51
00:10:12.380 --> 00:10:20.480
So maybe we end the markets update on a green background. I was a little optimistic there. [laughs] Okay, go back to red.

52
00:10:20.619 --> 00:10:36.220
[laughs] Before we, uh, go to SpaceX, I'm actually gonna surprise you, Colin, and I'm gonna, uh, I'm going to do a live, uh, update on something which is hitting the timeline right now. Take a look at this.

53
00:10:36.960 --> 00:10:51.320
Claude Mythos just dropped. Ooh. Check this out. Um, under the official product name Fable. So this is basically the Claude Mythos model that we've all been hearing about, but it's a little bit neutered.

54
00:10:51.440 --> 00:10:58.140
It's got some guardrails on it. You can't develop bioweapons on it, or at least theoretically you couldn't develop bioweapons before.

55
00:10:58.240 --> 00:11:09.040
Now you definitely can't develop bioweapons as they are kinda like tweaking and very concerned with safety. So this is Fable, uh, just dropped.

56
00:11:09.140 --> 00:11:21.060
I'm certain that everybody who's monitoring the situation is, uh, racing back to their keyboards. Uh- They're already a dozen prompts deep. Yeah, they're a dozen prompts deep. I'll say this.

57
00:11:21.300 --> 00:11:36.020
Uh, they have a little notification here that says, "Fable is the most capable model and takes 2X the usage of Opus." Oh, surprise, Colin, it's gonna, it's gonna blow through your tokens faster. Surprise. So.

58
00:11:36.080 --> 00:11:48.960
Everyone wants a free lunch, man. Yeah. Everybody wants a free lunch. I will say, yeah, the, the, the loss leader of your frontier models continues to work for Anthropic and OpenAI. So

59
00:11:50.540 --> 00:11:58.540
before we go to the hot IPO of the day, let's hear a word from our sponsor, CleanSpark. We are CleanSpark,

60
00:11:59.440 --> 00:12:15.500
America's Bitcoin miner, a publicly traded company with the largest operating hash rate.Powered entirely by self-operated infrastructure across four states. This is our proof of work.

61
00:12:16.380 --> 00:12:25.200
We are setting the standard for what's next. Learn more about the intersection of energy and Bitcoin at cleanspark.com.

62
00:12:28.760 --> 00:12:39.380
Okay, to SpaceX, which IPOs this Friday, I believe, if my calendar's right. June 12th is the day that it's scheduled for IPO.

63
00:12:39.880 --> 00:12:52.140
And here I am, Bernie Sanders meme saying, "Once again, I am telling you that SpaceX is a data center company, not a rocket ship company." And here is why.

64
00:12:52.680 --> 00:13:12.520
We got-- I've been looking through the SpaceX, what's called roadshow, like, uh, s- uh, slides and everything, and here's kind of the big takeaway I have from that, which is when they show the near-term total addressable market for their company, um, you know, these things are, they, they segment it out.

65
00:13:12.600 --> 00:13:31.080
Um, and they anticipate the total addressable market for the company being twenty-eight point five trillion dollars, of which only-- of the sub-sectors that that twenty-eight point five trillion dollars takes up, only s- uh, space only takes up three hundred and seventy billion dollars of a near-term market.

66
00:13:31.480 --> 00:13:47.840
Now, that would be anticipated to change long term, but space being less than three percent of their entire, like, total addressable market is very surprising because the other two sectors are connectivity and, most bigly, AI.

67
00:13:48.640 --> 00:13:58.160
AI, specifically enterprise applications, SpaceX estimates having a near-term mark-- total addressable market cap of twenty-two point seven trillion dollars.

68
00:13:58.480 --> 00:14:16.170
So that is basically the entire value of the company as they're trying to portray to people who are considering buying into the IPO. I'm gonna share some sections of this, uh, roadshow, uh, document.

69
00:14:16.360 --> 00:14:29.950
And if you look, it's a little bit difficult to see here, but they do kind of show that twenty twenty-six has been a, quote, "momentous year so far," with milestones where they-- it's just AI news, AI news, AI news.

70
00:14:30.060 --> 00:14:38.960
Acquired xAI in February twenty-sixth. Starlink customers, they partner with Courser to advance Grok in April.

71
00:14:39.060 --> 00:14:47.520
They, uh, announced a Starlink agreement with American Airlines and, uh, then they entered into a cloud service agreement with Google.

72
00:14:48.200 --> 00:14:59.440
And, um, all of the SpaceX, like, value is in relating, is basically related to connectivity and AI. I'm gonna jump forward to

73
00:15:01.340 --> 00:15:18.020
their slides on supercomputers and saying that, well, we're the only com-company, only AI company who can launch these computers into space, which we'll get into here in a second. And they also dropped

74
00:15:20.800 --> 00:15:38.120
this video showing the satellite that they will be launching up into space. It's kind of, uh, it's kinda cool to see actual, like, renderings and models of an actual design that they're working on.

75
00:15:38.460 --> 00:15:53.379
S-so this is, like, a, a pilot program for a data center satellite? So this would be, uh, their AI1 satellite, the first generation of the AI satellite, and here's the specs.

76
00:15:54.700 --> 00:15:59.780
They're anticipating about a hundred twenty kilowatts average compute payload, of which it peaks at one hundred and fifty kilowatts

77
00:16:00.720 --> 00:16:16.540
and seventy kilowatts with a little-- with-- and it's interchangeable as to which provider providing the, uh, the compute. Wingspan has got a bunch of big old, um, uh, solar cells on it. Wingspan of seventy meters,

78
00:16:17.900 --> 00:16:31.780
and it has a thermal system to basically radiate the heat. Uh, liquid-cooled radiators. Uh- Whoa, liquid-cooled. Yeah, liquid radiator. Obviously closed loop.

79
00:16:32.679 --> 00:16:45.860
[laughs] That's a- Data centers are gonna suck up all the water from Mars and the Moon too. [laughs] That's gonna be a, that's gonna be a thing. Well, anyway, so this video looks wild.

80
00:16:46.080 --> 00:16:49.920
Um, and, uh, I think a lot of people are...

81
00:16:50.560 --> 00:17:10.220
So, so this, you know, a lot of people are gonna be, like, looking at this particular video, but I think that there's an angle to this, uh, satellite which I haven't seen a lot of other people consider, which is how many satellites do they have to put up into space in order to, like, make a data center roughly equivalent to, like, one of the big ones on ground?

82
00:17:10.560 --> 00:17:24.740
So let's just look at, like, the numbers for what their Colossus expansion is going to be. They're anticipating four hundred-plus megawatts of compute very soon at their Colossus sites.

83
00:17:25.310 --> 00:17:35.060
And, um, how many satellites would that be, and how many Starship or, and how many, uh, Starships would it take to send those satellites to space? So here's the numbers.

84
00:17:37.240 --> 00:17:49.140
At four hundred thousand kilowatts, so four hundred megawatts of the roughly Colossus current size, at a hundred twenty kilowatts per satellite is about three thousand three hundred and thirty-three satellites,

85
00:17:50.720 --> 00:18:06.960
which roughly equivocates to five thousand seven hundred and fourteen metric tons of things that SpaceX has to get into space. That means it's about thirty to sixty Starship launches.

86
00:18:07.120 --> 00:18:20.355
Starship is the imminent, uh, SpaceX, like, new big rocket with a fairly large payload.So we're looking at, at about 100 tons per launch. This is about 57 Starship launches.

87
00:18:20.756 --> 00:18:35.176
So we're basically 57 giant big-ass rockets that SpaceX has to send up packed with satellites in order to build a data center in space roughly the size of Colossus.

88
00:18:35.216 --> 00:18:41.516
This is back of the napkin math, and I think it's pretty interesting. Um, yeah, and, you know,

89
00:18:42.536 --> 00:18:51.216
a, a caveat here I would say is eventually they're, the form factor is going to change probably to look more like a space station, right? You're gonna have something like the ISP,

90
00:18:52.256 --> 00:19:01.316
um, or I- International Space Station, right? Uh, who, who... I mean, well, so it- Or you can, like, chain them together and do them on the launch pad or- They're talking...

91
00:19:01.396 --> 00:19:15.756
Yeah, they're talking about linking these with, um, [lip smacks] uh, lasers, light basically, kind of like Starlink. So, uh, perhaps that could change. I, I don't really know. Um, Elon, actually, I have this queued up.

92
00:19:16.496 --> 00:19:26.456
Elon actually specifically talks about the latency between satellites and Earth, because you're like, "Oh, well, they're up in space. It's gonna take forever." That was my next question. I mean- Yep...

93
00:19:26.476 --> 00:19:28.136
the, the one of the biggest,

94
00:19:29.256 --> 00:19:42.516
uh, one of the largest points being made as we've covered all of these data center companies, specifically with looking for inference prime real estate, is the fact that these data centers need to be on fiber, and they need to be close to a metropolitan area to be the most effective.

95
00:19:43.276 --> 00:19:50.496
So my questions that I don't have answers to, are these gonna be serving inference? Are they going to be training only? How do you solve for that latency issue?

96
00:19:51.616 --> 00:19:59.406
Colin, I'm glad you asked because Elon has already kind of addressed it. It's, it's no problem. We've already solved it. Yeah. [laughs] Exactly. We've already solved it.

97
00:19:59.406 --> 00:20:08.596
There's, there's a lot of space up there, and so even when you're talking thousands or even, you know, up to a million satellites- Yeah... you got plenty of room to move around up there. Yeah, space is really big.

98
00:20:09.006 --> 00:20:15.376
[laughs] So it's not like, it's not like space is gonna get crowded. Um, space is, is enormous. Like if you zoom in close to a satellite, it looks big.

99
00:20:15.756 --> 00:20:25.725
But if you actually look at it relative, relative to Earth, these satellites are so tiny you can, you can't even see them. So they're, they're very, very tiny compared to Earth.

100
00:20:25.736 --> 00:20:35.366
And I mean, we have 10- about 10,000 Starlinks in orbit right now. Yeah. We've got a pretty good idea of how to operate just really large constellations and do it safely now, right?

101
00:20:35.546 --> 00:20:44.256
We, we are the only operator that has any experience at that scale. Uh, it's, it's a great thing that we, you know, we have this background so we know how tightly we can pack the satellites and, and, and fly them safely.

102
00:20:44.296 --> 00:20:50.156
That's, that's a, a number one goal when, when we look at the constellation. We're gonna be building a lot of satellites, and we're gonna be building them-

103
00:20:51.116 --> 00:21:04.676
Okay, so that wasn't the exact clip I thought, but I think just before that, Elon says that the latency of light to low Earth orbit satellites is roughly two to three milliseconds, which if you're a gamer, that ping is negligible.

104
00:21:04.716 --> 00:21:08.856
That's basically like playing on LAN. So I think there's like some addition...

105
00:21:08.896 --> 00:21:18.696
And it- it's really just because they're low Earth orbit, and I think if you wanna like consider latency and speed, look at how like good Starlink is and imagine it's gonna be even better. So,

106
00:21:20.336 --> 00:21:33.676
um- I mean, it's pretty incredible, and we, we covered the SpaceX IPO a week or two ago when, when it was priced. They're looking to r- uh, raise enough to be valued at 1.75 bil- uh, trillion dollars.

107
00:21:34.596 --> 00:21:43.616
[lip smacks] I was actually talking to my family about this, with my dad over the weekend, and he just can't understand it. He can't wrap his head around why the valuation would be that, be that much.

108
00:21:43.696 --> 00:21:50.556
My cousin was the same way. He, he's convinced that SpaceX is gonna tank the market, the IPO because it's so overpriced.

109
00:21:50.596 --> 00:21:59.896
There are fears because Fidelity dropped their retail investor threshold for this IPO from half a million to $2,000.

110
00:22:00.496 --> 00:22:06.516
And the fear there obviously, right, is that the re- retail will be exit liquidity for all the insiders and early investors.

111
00:22:07.656 --> 00:22:18.646
But I, I took the steel man here, not because I think that 17- 1.75 trillion is probably the right price right now. I mean, Nvidia is, what, 5.7, 5.5 trillion.

112
00:22:19.476 --> 00:22:34.536
So when you, when you factor that in and you think about like what's the ceiling on this when you compare it to a ca- a, a massive cash cow and cash flowing company like Nvidia, how much further can we move in the current market environment given, given SpaceX's current cash flows?

113
00:22:35.216 --> 00:22:43.736
But the steel man against that is no one else is doing what this company is doing. Even if you look at the s- They have a moat, they have a moat like nobody has a moat ever. Right.

114
00:22:43.916 --> 00:22:52.085
And the moat is just they've spent the past 30 years building the only- Building-... competent private commercial space company that can do this.

115
00:22:52.456 --> 00:23:01.896
Yeah, building the new state-of-the-art for rockets and going back to, you know, the smallest segment of its, uh, uh, of its addressable market, that 3% with like, you know, actual rockets.

116
00:23:02.856 --> 00:23:09.096
You know, the benefit in that for them is what you were just saying, Charlie. It's the delivery vehicle for all the other things they want to do.

117
00:23:09.176 --> 00:23:19.116
But even if we just look at that, the fact that they licked reusable boosters and all of these other things that, uh, NASA's top scientists weren't able to do in the public sector,

118
00:23:20.016 --> 00:23:30.356
people are gonna price this thing based on the fact that the technology is not even bleeding edge. It's beyond the bleeding edge of what most people think we can even do right now.

119
00:23:30.416 --> 00:23:44.436
So when, and when you factor in, when you factor in Starlink, which is the most incredible internet I've ever used, and you think that Starlink basically ensures that there is nowhere on the world at, uh, nowhere in the world at this point

120
00:23:45.796 --> 00:23:49.516
where there should be a dead spot for Wi-Fi as long as you have a Starlink hookup.

121
00:23:50.656 --> 00:24:02.676
The, the, the advancements that a company like SpaceX is bringing to the market are absolutely incredible, and I think a lot of the hope for this company too is riding on this idea of if they can actually put data centers in space that are viable.

122
00:24:02.736 --> 00:24:12.526
You know, to, to be really corny with it, the star is probably the limit in, in, in this case. But, you know, just going back to the original IPO pricing,

123
00:24:13.616 --> 00:24:26.220
uh, who, who's to say whether or not that's going toBe considered absolutely ridiculous, and everyone who's investing in it now will be exit liquidity for the insiders, or if this thing will actually run up a little bit after it, it goes public.

124
00:24:26.820 --> 00:24:33.420
Based on a lot of the chatter I've seen online, I think most people are kind of at a... That there's a coin toss moment here for a lot of people.

125
00:24:33.760 --> 00:24:38.400
You have the Tesla fanboys and the bulls who think that, like, you know, you don't bet against him.

126
00:24:38.440 --> 00:24:49.540
Then you have other people making very rational analysis saying 1.75 trillion is way too much for a company out of IPO like this, and we should expect to see a drop-off after it comes to market on Friday. Yeah.

127
00:24:50.220 --> 00:25:09.140
Because Tesla has this moat, you can basically take the assumption that whatever the total addressable market for AI in space, space in general, any kind of future value, uh, value to be found in space, Tesla will take the entirety of the market.

128
00:25:09.520 --> 00:25:23.540
So a bet on Tesla is about as good of a bet on that thesis as you can possibly make. So if you think space is a 30 to $100 trillion bet in the future, at, you know, Tesla's the only bet you can make.

129
00:25:23.720 --> 00:25:31.520
I will also take a moment to, to step back and think about the w- the statements we're making here, which is, um,

130
00:25:32.900 --> 00:25:43.300
we, uh, we're like, well, of course I guess we're gonna build a data center in space, and we can st- we can shoot up hundreds and hundreds of rockets featuring tu- tens and hundreds of tons of payload every year.

131
00:25:44.260 --> 00:25:56.920
And, um, to tell that to my father when he was a kid would be absolutely insane. Yeah. And then, and, and now my concern, now our concern, our present concern is, and it might be overvalued. [laughs] Right.

132
00:25:57.290 --> 00:26:10.910
So, like- Right... so, yeah, like, so, you know, the, uh, uh, with this, like, very zoomed out approach, th- you know, uh, that, that's, like, the decision we're faced with. And, um, it, this scene...

133
00:26:11.200 --> 00:26:21.220
To me, Colin, this seems absolutely like the, the challenge is timing the market. It's not whether you- Yeah. That, that makes sense to me. Yeah. It's absolutely not whether you want a piece of SpaceX.

134
00:26:21.460 --> 00:26:31.180
You absolutely want a piece of SpaceX. The question is, when is the best time to buy it? Right. And are you gonna get rinsed after it goes public? One last thing before we wrap up this segment.

135
00:26:31.760 --> 00:26:42.800
If they do end up licking data centers in space earlier than anyone else, you just took the number one resource constraint off the table for SpaceX and X AI, which is energy.

136
00:26:43.420 --> 00:26:51.660
You're, they're, they're gonna be up in orbit soaking up sun 24/7, and at that point, it's just does the CapEx...

137
00:26:51.700 --> 00:26:58.450
Or sorry, does the OpEx make sense to get those satellites into space, and does it make sense to furnish the solar panels?

138
00:26:58.500 --> 00:27:07.190
If you can get 100% utilization, then the, the solar panels are going to, you know, ROI themselves extremely quickly. Yeah. You're basically the- Almost...

139
00:27:07.190 --> 00:27:15.660
the guy who's interviewing Elon in this, in this, uh, SpaceX clip has a shirt that says, "It's always sunny in space." That's the thesis.

140
00:27:15.860 --> 00:27:24.060
So you're unconstrained for permitting how big of a data center you can be, and you're unconstrained with energy availability.

141
00:27:24.100 --> 00:27:29.300
Now you just have to figure out how to integrate- Now you have to figure out how to make them work. Yeah. And that's the last thing I'll say.

142
00:27:29.380 --> 00:27:42.820
I'd love to have a pro and a con guest on this because the pros, we just talked about it, unlimited energy from the sun. Once you have the solar panels up there, you're good. Uh, you're in the vacuum of space.

143
00:27:42.980 --> 00:27:44.120
Cooling is much easier.

144
00:27:44.320 --> 00:27:55.340
And you're in space, and you have a lot of it, so you don't have to worry about land acquisition, and you don't have to worry about pissing off a bunch of locals for erecting a data center in their backyard.

145
00:27:55.350 --> 00:28:05.520
The cons obviously being, what does servicing one of these look like? You know, astronaut technicians. Robots, dude. Right? Robots eventually. It's all robots. But how, how far are we from that, right?

146
00:28:05.640 --> 00:28:16.840
I mean, I know, you know, Elon thinks that we're gonna get Westworld in five years, and that, you know, optimists can already replace most factory workers, and all of these other, you know, high techno future ideas.

147
00:28:17.440 --> 00:28:26.940
But this still doesn't change the calculus that, or, or the, the basic math that servicing one of these is going to be very expensive. You're also up in space where there's a lot of debris flying around.

148
00:28:27.260 --> 00:28:41.520
How often do they get damaged? So you're right, but there's really not. So, um, uh, this ha- you know, this, uh, Kessler Syndrome is a, is a ways away. I'll just, I'll put it like that. Uh,

149
00:28:42.980 --> 00:28:53.840
um, yeah, as far as how they're gonna be serviced, well, Elon's other company, Tesla, the other trillion dollar company, they'll build the robots. I don't know. This is gonna be something we're gonna have to cover.

150
00:28:53.960 --> 00:29:09.340
I anticipate, Colin, I predict, Colin, a year from now we'll be talking about robots about every other day. This is my, this is my prediction. I'm, uh, making it. What's your prediction for when SpaceX flips Nvidia? Ooh.

151
00:29:09.520 --> 00:29:26.200
[laughs] Do we think that's gonna happen? Uh, yes. I'll, I'll make that prediction. I think SpaceX will flip Nvidia, especially as SpaceX- Under, under and over 2028? No, no, no, 2030.

152
00:29:27.720 --> 00:29:38.940
I would say it'll be in 2030, if you had to, like, you know, press me on it, so. Okay. And you know, and, um, but that's not really a big multiple from the, [laughs] where the IPO is right now. So- Right...

153
00:29:38.970 --> 00:29:54.659
this is not me saying buy the IPO. This is me just saying I'm bullish SpaceX and their execution capabilities. Uh, would love to be able to buy SpaceX at a reasonable price, so. Yeah, maybe half a, half a trill. Yeah.

154
00:29:55.180 --> 00:30:09.600
[laughs] Ba- basically, like, um, two great financial crisis backstops- Yeah... you know, is kinda where I'm thinking. 600 billion, what was it? Um, okay. We got to move on, and we're gonna go back to stocks.

155
00:30:09.660 --> 00:30:30.364
If you've been sitting here anxious for us to wait to talk about your data center stocks, we'll get there, but not before a word from our sponsor, Luxur.This episode is brought to you by Luxor's Commander, Bitcoin miner management software for enterprise operations.

156
00:30:31.564 --> 00:30:36.174
Commander gives you real-time fleet monitoring, bulk remote commands across your fleet, and Intelligent Miner.

157
00:30:36.644 --> 00:30:45.344
That's an automated profitability engine that runs every five minutes to adjust your fleet to live power markets and hash rate markets.

158
00:30:45.844 --> 00:30:50.554
ERCOT backtests show over ten percent more profitability with intelligent mining versus binary mining.

159
00:30:51.084 --> 00:31:00.354
Commander Pro is a hundred dollars per megawatt or a twenty-five basis point pool fee adder, roughly half the price of competition, and you can try it for free for 60 days.

160
00:31:00.554 --> 00:31:15.804
If you'd like to get started, go to luxor.tech/commander. All right, we're getting back into our neck of the woods with Cipher Mining. They just announced a price, uh... They just priced their- Cipher Digital, mind you.

161
00:31:15.844 --> 00:31:27.584
Cipher Digital. Cipher Digital. Thank you. I don't want to, I don't want to misidentify any of these companies. They identify as an AI company, or rather- And, and, and- Or a power plant, uh, broker-dealer oligarch.

162
00:31:27.844 --> 00:31:42.483
And as they should, because as we will follow up on this, they just inked their third AI tenant deal with Amazon Web Services, and they are pricing, uh, an eight hundred and ten million dollar senior secured note to fund this data center build-out at Stingray.

163
00:31:42.884 --> 00:31:54.814
So the pricing for this note, it's due twenty thirty-one, and it's at six percent. Now, we covered Iron's most recent senior note, believe, uh, two weeks ago or maybe, uh, early last week.

164
00:31:55.404 --> 00:32:05.734
And we've noticed that the credit for these build-outs are getting-- is getting really attractive. In fact, I'm gonna throw this up as I make a note on that really quickly.

165
00:32:06.354 --> 00:32:14.504
If we hop on over to our friend Rittenhouse Research. He just posted the start of CoreWeave credit spreads and, spreads, and they've absolutely plummeted in twenty twenty-six.

166
00:32:14.564 --> 00:32:23.394
They peaked right before the new year at about eight point five. So this is the spread between, uh, the bonds CoreWeave is issuing for its builds versus Treasuries.

167
00:32:23.764 --> 00:32:30.184
Went from eight point five percent spread down to roughly, uh, four point five percent.

168
00:32:30.664 --> 00:32:39.024
Now, what this means is these bonds are getting bid up in the secondary market, and the actual credit that is being offered on them is dropping.

169
00:32:39.064 --> 00:32:51.094
And it also means that CoreWeave has been getting more attractive credit for its latest, uh, notes. And we've seen that across all of these, uh, AI Bitcoin miners, uh, you know, AI-pivoting Bitcoin miners.

170
00:32:51.664 --> 00:33:00.384
Cipher at six percent. I believe Iron's most recent one, um, it was the SOFR rate plus a few percentage points. It was in the ballpark of six to seven percent.

171
00:33:01.224 --> 00:33:19.224
And we've seen the actual interest rate for these, for these notes drop, which is signaling that the market's saying, "There's not as much risk here as we originally believed," or, "Now that we're so far into this CapEx cycle, the risk is diminishing for these notes," and investors are getting more comfortable holding them.

172
00:33:19.624 --> 00:33:21.864
But going back to Cipher, this is, this si...

173
00:33:21.924 --> 00:33:37.184
this, uh, senior secured note offering is coming on the heels of it announcing its latest contract with Amazon Web Services, a two billion dollar contract on a fifteen-year lease and with three five-year extension options.

174
00:33:37.244 --> 00:33:41.744
This is for seventy megawatts critical IT load, and then from that...

175
00:33:41.804 --> 00:33:49.304
So that'll be a hundred gross with all of the supplementary infrastructure that's needed for that, but seventy megawatts specifically of the AI computers.

176
00:33:50.324 --> 00:34:03.644
And looking into the details of this, this is a very attractive deal for Cipher. So going through a laundry list really quickly. Like I said, fifteen-year base term, two billion dollars over that fifteen years.

177
00:34:03.684 --> 00:34:10.304
There are three five-year extension options for a total contracted potential of five point seven billion. It is a triple net lease.

178
00:34:10.704 --> 00:34:19.284
What this means is that Amazon is going to cover the property taxes, the building insurance, and maintenance and repairs of the innards of the data center.

179
00:34:19.344 --> 00:34:30.024
Cipher will still have to manage maintenance of the actual powered shell, the building itself, but Amazon is covering the IT infrastructure, the cooling infrastructure, and the computers themselves, their maintenance and repairs.

180
00:34:30.844 --> 00:34:39.684
Amazon's parent company, not just AWS, fully guarantees all base rent and operating expenses, so this is kinda like the backstops we've seen with Google with Fluid Stack.

181
00:34:40.244 --> 00:34:50.764
So that I would assume people are loving to see on the other side of this who are rating, uh, the credit f- uh, or rating the, uh, you know, the, the credit risk for this build-out.

182
00:34:51.364 --> 00:34:55.704
And it's a hundred percent pre-leased with no termination for convenience after commencement.

183
00:34:56.144 --> 00:35:06.024
This one is huge because some of these service level agreements with these data center companies, these powered shells, come with termination clauses. If things aren't going well, they're going to back out.

184
00:35:06.524 --> 00:35:16.484
Uh, there is no termination for convenience in this SLA. Covered the Stingray bond. One more time, six percent at twen- uh, due twenty thirty-one.

185
00:35:17.004 --> 00:35:22.913
The other interesting thing about this, really quickly, Charlie, before I toss it to you for second takes. The project,

186
00:35:23.874 --> 00:35:33.884
the, uh, amortization schedule for this, uh, requires Cipher to repay a fixed amount of principal each year, uh, rather than requiring Cipher, excuse me, to repay a fixed amount of principal each year.

187
00:35:33.944 --> 00:35:37.744
The amortization schedule is tied to the cash the project generates after completion.

188
00:35:38.284 --> 00:35:46.524
So it starts at twenty-seven million in year one, rises to forty-five million, then fifty-one, fifty-seven, sixty-four, et cetera, et cetera, et cetera.

189
00:35:47.024 --> 00:35:50.464
And this tracks a three percent annual rent escalator that's built into the lease.

190
00:35:51.224 --> 00:36:00.663
This is notable because this kind of takes this bond or the senior secured note for Stingray out of the junk category, and it's looking a lot more like straight up project finance.

191
00:36:01.604 --> 00:36:08.804
And I would imagine just going back to the deals of the term, people are looking on at this as, as a very favorable lease for Cipher.

192
00:36:09.444 --> 00:36:19.364
Charlie, tossing it to you before I wrap up with a few other notes on this versus Cipher's other AI contracts. I don't have a lot of insight to this. Uh,

193
00:36:20.444 --> 00:36:32.882
I think that almost being like an AI landlord is a very compelling model. Triple net lease, put Amazon as your tenant.You know, that's a sweet deal. So I like it.

194
00:36:32.912 --> 00:36:44.992
Yeah, it, it's a super sweet deal, and it's the second one that Cipher has with Amazon. This one will... Rent commencement is start- targeted to begin April or May 2027. Uh, and so I...

195
00:36:45.032 --> 00:36:48.972
It seems like they've probably been laying the groundwork for this one for a while.

196
00:36:49.032 --> 00:36:57.872
That'll be pretty quick to market if they can get there, and hats off to them with that speed to market, assuming that they hit that target.

197
00:36:59.232 --> 00:37:09.212
I just wanna compare this to Cipher's three deals because so far, let's just remember that they were strictly a Bitcoin mining company a year, a little over a year ago.

198
00:37:09.782 --> 00:37:23.912
And since doing, announcing their AI pivot, they now have 11.4 billion in, in contracts for AI services. The first one of these was at Barber Lake, announced November 2025. This is the Fluid Stack

199
00:37:25.072 --> 00:37:40.472
deal backstop by Google. Uh, 1.73 billion total consideration, and which, uh, and, or sorry, that's the bond for that. Um, the, the bond is 1.73 billion, coupon for that 7.125% due 2030.

200
00:37:40.892 --> 00:37:55.732
They also have their Black Pearl build-out for Amazon. This is that 200-megawatt deal with Amazon Web Services, and now they have their Stingray site, which will be, uh, which, which will also be servicing Amazon.

201
00:37:56.352 --> 00:38:04.912
Uh, going back to Black Pearl really quickly. 300 gross megawatts, uh, s- I believe 200 for critical IT.

202
00:38:05.592 --> 00:38:12.392
Uh, Amazon Web Services again is that tenant, 15-year lease at 5.5 billion, and now they have this other one with Amazon as well.

203
00:38:12.472 --> 00:38:19.872
That is looking pretty, pretty sweet, especially if you're a big Cipher bull like our boy Mike Alfred.

204
00:38:19.932 --> 00:38:30.072
You know, getting a company like Amazon to come in for two of these leases, you have to wonder if another one's coming down the pipe. I own some Cipher stock. I'm bullish on it. I like the guys over there.

205
00:38:30.132 --> 00:38:43.932
I like Lee, I like Drew, I like Tyler. Cool team. Cool team and a lot under the hood right now for them. I think that squares it for that, Charlie. All right. We can go ahead on to Apply Digital.

206
00:38:44.052 --> 00:39:01.552
Yeah, we're gonna keep going to Apply Digital's deal. But before that, a word from our sponsor, Ligos. [static sound effect] Hedge funds are getting liquidated. Is your Bitcoin safe?

207
00:39:01.932 --> 00:39:10.102
It's not just Bitcoin's price drying up. Hedge funds and lending desks are going under after the notorious 10/10 and 2/5 liquidations. And who knows?

208
00:39:10.712 --> 00:39:15.172
Might've seen one blow up last week when Bitcoin revisited that February low of 60,000.

209
00:39:15.872 --> 00:39:25.872
Counterparty risk is rampant, so it's more important than ever to know who controls your Bitcoin, and with Ligos, that is always you. They are our preferred lender over here at Blockspace.

210
00:39:25.932 --> 00:39:36.292
They use Bitcoin native smart contracts to make sure that when you're using their lending services, you always control the keys to your Bitcoin and no one else. Don't become the next Celsius or FTX victim.

211
00:39:36.952 --> 00:39:53.392
With Ligos, there's no rehypothecation, no wrapping, no bridging, just competitive rates as low as 10% APR. Go to ligos.finance to learn more. All right. Another Bitcoin mining company turned AI factory.

212
00:39:53.612 --> 00:40:06.092
We've got a note here on Applied Digital. They just announced a 210-megawatt lease with an undisclosed hyperscaler, but they said it is the same hyperscaler that they are hosting at two other facilities.

213
00:40:07.672 --> 00:40:16.981
This puts two- Is that two, is that two facilities that are currently active or that they've got deals lined up for? Uh, deals lined up. Okay. Uh, I, I do not believe that they have...

214
00:40:16.992 --> 00:40:26.222
They might have partially energized some, but I, the- they're not fully, they're not at full capacity yet. Gotcha. So this is, yeah, multiple part deal, multiple sites. 100%.

215
00:40:26.252 --> 00:40:44.072
And now their contracted revenue has been pushed to 36 billion as a result of this. Wow. So a few- That's more than SpaceX. [laughs] Hey, there's a reason Fidelity did that $2,000 threshold. Yeah.

216
00:40:44.152 --> 00:40:58.072
Anyway. But yeah. Anyway, some key notes here before I toss it to you, Charlie. This lease is at Delta Forge 2, its fifth AI factory campus located somewhere in the Southern US.

217
00:40:58.712 --> 00:41:07.432
Again, the tenant is an unnamed US-based investment grade hyperscaler. It's the same tenant as Delta Forge 1 and Polaris Forge 3.

218
00:41:07.452 --> 00:41:16.672
And if we look at the details, 210 megawatts critical IT load, 15-year take or pay, plus renewals. Base term revenue is 5.2 billion.

219
00:41:16.712 --> 00:41:28.872
With all the renewals, the total potential contracted value over 30 years is 12.7 billion, and operations are slated to start Q1 2028. The cooling, proprietary waterless.

220
00:41:29.552 --> 00:41:45.112
Hey, so if you want to quench the fears of the AI data center NIMBYs, we are... Looks like we're moving closer to a form factor for the cooling where you can take water out of the equation entirely. Hallelujah. No water.

221
00:41:45.312 --> 00:41:52.852
What, what are the Facebook groups gonna protest next? They're gonna protest, uh, energy. Yeah. That's... I think that's gonna, that's gonna be the next one.

222
00:41:52.872 --> 00:42:01.212
If anyone sees any spike in their residential power prices, you know, uh, energy price inflation be damned. It... But data centers are gonna be blamed for it.

223
00:42:01.432 --> 00:42:09.852
And I will say, this is actually one of the concerns that I think is totally reasonable for localities, and it's one of the reasons why we've seen utilities come to data centers now.

224
00:42:10.292 --> 00:42:18.692
ERCOT is working on the rules for this, saying, [clicks tongue] "If you're going to demand so much energy from us, you've gotta foot the bill for transmission."

225
00:42:19.372 --> 00:42:23.152
And to me, some of these companies- That makes sense to me... it makes total sense.

226
00:42:23.422 --> 00:42:34.392
It's the same re- way that, you know, as you know well know, Charlie, like oil companies, uh, putting infrastructure money down for roads. They build roads. Yeah. They do easements. There's... Yeah.

227
00:42:34.472 --> 00:42:48.316
Man, big, big oil gotta be really loving this right nowThe target's off their back at least for the next, you know, few years. Well, Microsoft, Microsoft revised their 2030 climate goals, I think.

228
00:42:48.706 --> 00:42:59.796
Or at least they're talking about it. And I'm like... 'Cause they're like, "Well, we just can't feasibly hit, uh, you know, our energy needs without oil." Like yeah, where the heck were you the past two decades? Yeah.

229
00:42:59.816 --> 00:43:10.556
They just need to plant more trees in South America. That's clearly, they just need more carbon credits. That, that's the, that's the key to this solution. Look, I... Look, it never really was about energy.

230
00:43:10.596 --> 00:43:20.716
It was all about narrative and money. Now, let's get down to brass tacks and talk pragmatically about energy so we can have a useful conversation about climate- Yeah. I mean- But enough about that.

231
00:43:20.936 --> 00:43:28.036
We, we, we can't get too distracted. We gotta finish the deal. Yeah, okay. We'll, we'll finish the deal here. Uh, just one more note on this.

232
00:43:28.096 --> 00:43:39.776
Alongside the lease, Applied also closed a Goldman Sachs arranged $350 million revol- revolving loan, plus 200 million accordion option at SOFR rate plus 225 basis points.

233
00:43:39.816 --> 00:43:48.056
It's roughly five, uh, 5.88%, give or take 6% all in, and that will mature May 2029.

234
00:43:48.596 --> 00:44:02.336
Uh, similarly to Cipher, this is pre-development liquidity which lets them move on new campuses before project-level bond financing is in place. Similar playbook with Cipher. So two really interesting deals coming out.

235
00:44:02.876 --> 00:44:18.296
Uh, I, you know, Applied didn't give as much information in their presentation as Cipher did for this deal. I did see our boy Rittenhouse Research say that the, that the terms for this seemed a little less, uh...

236
00:44:18.336 --> 00:44:20.816
Or maybe they omitted some of the, uh...

237
00:44:20.856 --> 00:44:35.406
If, if it were good news, they probably wouldn't omit it, but there were just more omissions in this, and you didn't see the same, uh, you know, kind of beneficial clauses as we saw announced with the Cipher deal, which one of them I would like to return to that for just a second 'cause I forgot to mention this.

238
00:44:35.436 --> 00:44:40.476
One of the most incredible parts of the Cipher deal is that there is a cap on the development cost.

239
00:44:40.816 --> 00:44:50.106
So anything above $10.5 million per megawatt, Amazon will cover the difference for the CapEx for this, which to me just seems, I...

240
00:44:50.106 --> 00:45:01.506
It seems like an incredible provision to have in here, protecting yourself against potential supply-side inflation and ensuring that you're not gonna run out of cash for this

241
00:45:02.436 --> 00:45:12.696
i- if, if the building costs end up eclipsing the, the estimates that you had going into it. But hey, Charlie, we're back in deal flow territory. Back in deal flow.

242
00:45:12.816 --> 00:45:22.016
I'll be thinking, but we've still got deals coming through the door, which is nice. I think deals will continue until morale, um, uh, subsides, I guess. I don't know.

243
00:45:22.636 --> 00:45:36.676
Deals will continue until the Facebook, uh, protest groups get big enough. Speaking of AI, [laughs] as we have been the whole show, [laughs] I don't think we've mentioned a single thing about crypto.

244
00:45:37.316 --> 00:45:48.996
Or rather speaking of, you know, anti-data center stuff- Yeah... and the potential astroturfing from U- US adversaries. Yeah. Charlie, China. China. China.

245
00:45:50.136 --> 00:46:03.176
China is trying to close its gap with the US in terms of the AI race. I'm, I would get this Bloomberg article up, but it's paywalled and I'm not, I'm not- I've got it. I've actually, I've got it.

246
00:46:03.236 --> 00:46:12.456
Can, can you get it up really quick? Yeah, let me look. Are you, are you less poor than me? Do you actually have the full article? Look at that. It's, it's $5 a month and I, I do actually read Bloomberg, so.

247
00:46:13.216 --> 00:46:20.796
Uh, God bless Charlie Spears for doing the Lord's work and getting past Bloomberg's paywall, uh, without using archive.is.

248
00:46:21.156 --> 00:46:28.716
[laughs] So Bloomberg has reported that a, that China is going to pump a state-directed 2 trillion yuan, which sounds like a lot.

249
00:46:28.736 --> 00:46:34.956
It's only $295 billion, though, um, for over the next five- It's only one great financial crisis, uh, salary.

250
00:46:35.016 --> 00:46:42.076
Yeah, that, [laughs] that, that's, you know, that, that's one great boomer financial crisis back when billions used to mean something.

251
00:46:42.516 --> 00:46:54.036
But they're, they're planning on pouring roughly $300 billion over the next five years to build an interconnected national AI compute network. Uh, this, key details for this, the NDRC is drafting the plan.

252
00:46:54.136 --> 00:47:07.116
State telecoms like China Mobile and China Telecom will operate the hubs, and there is a country mandate to source at least 80% of technology hardware for these build-outs from domestic suppliers.

253
00:47:07.656 --> 00:47:16.666
Now, listeners of Block Space will know we reported a few weeks ago that Taiwan authorities were cracking down on smugglers in Taiwan who have been shipping

254
00:47:17.916 --> 00:47:29.936
Nvidia chips and advanced chips to China by way of intermediary countries like Hong Kong, um, other Southeast Asian countries, Vietnam, et cetera. But China doesn't...

255
00:47:29.996 --> 00:47:43.136
You know, at this point they're saying they don't want those. They want to build out their own domestic capabilities, and those domestic capabilities on the hardware side include a number, a few companies.

256
00:47:43.276 --> 00:47:46.376
I believe, though, the premier one is Huawei.

257
00:47:46.956 --> 00:48:06.456
They have a, a, um, H100, um, or kind of a Nvidia equivalent, video GPU equivalent, the 910C, which delivers approximately 60% of an H100's inference performance per DeepSeek's own published assessment, and this is one of the structural challenges for China.

258
00:48:07.236 --> 00:48:13.536
Uh, you know, the rest of the world is going to converge on Nvidia's specs for these AI data centers.

259
00:48:14.036 --> 00:48:30.676
But for China, since they have to push their own chips because of export controls, uh, switching away from Nvidia means rewriting code, abandoning industry-leading infrastructure, and losing access to applications in the CUDA libraries built up over year, over the years by global developers.

260
00:48:31.536 --> 00:48:58.116
Now, I asked Claude to give me any sort of rough estimates or research on how many Nvidia GPUs have actually poured into China-The estimates from Epoch AI or Epoch AI are that roughly between 290,000 and 1.6 million, so there's a big, you know, uh, error margin there, of H100 equivalent GPUs were smuggled to China through the end of 2025.

261
00:48:58.176 --> 00:49:09.255
So the median estimate comes in at 660,000. And, you know, there's a lot of, uh, there-- People theorize that many of these computers were probably used to train DeepSeek.

262
00:49:09.916 --> 00:49:20.146
Um, DeepSeek's also, as I understand from people who understand training AI models better than me, DeepSeek piggybacked on a lot of work that was already done by the other prior models. Yeah.

263
00:49:20.156 --> 00:49:25.196
It's part of the reason why it was able to bootstrap so quickly and part of the efficiency gains. Yeah.

264
00:49:25.436 --> 00:49:43.596
I mean, China, in classic IP theft, uh, fashion, has a history of letting the Americans do the research, do the Western companies, and then taking it and building an inexpensive one based upon all the legwork that the West has done.

265
00:49:43.676 --> 00:49:51.856
And the thing is, Chinese models are very, very good. They're-- they use way fewer tokens. They're way cheaper to use.

266
00:49:51.976 --> 00:50:05.876
Um, uh, you can't do very benign things such as ask about Tiananmen Square, but who needs to know that anyway? And, [chuckles] and- [laughs]... uh, but, but they're, but you know, uh, we can see this.

267
00:50:06.436 --> 00:50:19.716
It's funny 'cause like I think it's, like, a very Chinese thing to do to spend way less money on, uh, actual, uh, AI data center development. Now, I will say this is, like, the public money from, from China, the state.

268
00:50:20.556 --> 00:50:30.856
Uh, this, this does not include the private capital that, say, ByteDance or, um, uh, uh, DeepSeek companies have invested.

269
00:50:31.536 --> 00:50:40.966
Um, but- I'm glad you, you bring that up because if you look at that, that is-- It-- they, they have not gotten anywhere close to the CapEx investment that we've seen in the US. Yeah.

270
00:50:40.996 --> 00:50:52.716
In 2026 alone, Amazon, Microsoft, Meta, and Alphabet, just those four, are projected by Morgan Stanley to spend 630 billion on data center and AI-related investment.

271
00:50:53.136 --> 00:51:02.936
Chinese internet firms, conversely, are projected to spend 70-plus billion, which is only 15% to 20% of what only hyperscalers are spending.

272
00:51:03.046 --> 00:51:12.796
So this isn't even including neo clouds or other smaller players within the AI landscape. Now, is that $70 billion totally accurate? I find it somewhat hard to believe.

273
00:51:13.206 --> 00:51:23.456
I would imagine it's probably a little bit more than that, and that's part of the problem with even measuring the competitive landscape here because there's a huge East-West divide, not just informationally,

274
00:51:24.396 --> 00:51:34.596
but also linguistically. Now, the linguistic thing is not as much of a problem as it used to be, but we-- it's-- we don't really ever know if the figures we're getting out of China are necessarily even accurate.

275
00:51:34.716 --> 00:51:45.916
I wouldn't even be surprised if the Chinese government is lowballing how much they're saying they're going to spend on AI infrastructure, so that when and if they catch up, it makes it look like they can do all this much more efficiently than the West can.

276
00:51:47.385 --> 00:51:57.136
[lip smacks] But when we look at those gaps, so we just had the gap in investment. If we look at the number of data centers estimated to operate in either country. Now, this is 2025 data, so this is outdated.

277
00:51:58.096 --> 00:52:04.336
But the US had an estimated 5,427 data centers in 2025 compared to 449 in China.

278
00:52:04.846 --> 00:52:16.076
And the US accounted for 45% of the 415 terawatt hours consumed by all data centers around the world in 2024, followed by China at 25% and Europe at 15%.

279
00:52:16.456 --> 00:52:28.876
Now that, now that estimate is kind of at odds with that prior one because you're looking at, you know, the fact that there-- the-- one estimate for the number of data centers is saying China has, you know, um,

280
00:52:30.356 --> 00:52:39.546
roughly like a 10th or even less than a 10th of US data centers footprint, but they're 25% of global data center electricity consumption versus the US is 45%.

281
00:52:39.896 --> 00:52:44.036
I would imagine China actually has more than what is being reported here.

282
00:52:44.096 --> 00:52:56.665
Uh, last, last note, i-in terms of the models, Charlie, this is the place where the gap between the two countries is the slimmest in terms of overall performance and overall, um, availability.

283
00:52:57.976 --> 00:53:14.745
China's leading model [lip smacks] um, let me get here. There we go. China's leading model from Zhipu, Zhipu AI's GLM-5 earned an 85 rating on bench LLM. GPT and Claude are at roughly 93 or 94.

284
00:53:15.456 --> 00:53:27.676
So that gap is slimmer than then, than the gap for actual hardware production and data center footprint, and I think that's one of the things to really watch. Like you said, these models are much more token efficient.

285
00:53:28.076 --> 00:53:39.276
People were blown away by DeepSeek when it came out. Everyone was panicking, you know, for a few weeks there, considering how, uh, advanced it seemed, and it just seemed to come out of nowhere for Western audiences.

286
00:53:39.376 --> 00:53:54.676
Yeah. Um, this is a very big story and kind of an existential question, which is, uh, if, if the Chinese models are just, you know, only 10% worse than the frontier models, while being

287
00:53:55.696 --> 00:54:11.676
so much cheaper and, uh, being able to scale like energy and compute and inference so much faster, then, like, what's the value of these as in a frontier model? Like you-- if, if you could buy a car for,

288
00:54:12.596 --> 00:54:26.096
uh, you know, thir-- one-third the cost of a luxury vehicle in the United States, which had twice the gas mileage and was only 10% to 20%, like not as good as the car, you, you'd probably do it, and you'd buy two of them.

289
00:54:26.436 --> 00:54:41.136
So, uh, yeah, I could, I could see this being an existential thing. Um, but maybe frontier models are an unending like edge that you will have, uh, you know, unending technological edge.

290
00:54:41.216 --> 00:54:48.206
Maybe this does, you know, does continue on the trajectory towards nation-state intrigue and, you know, cryptography and stuff like that.

291
00:54:48.236 --> 00:54:56.380
Yeah, I mean, I would imagine that it follows a similarIt follows a similar path with a number of, you know, you can pick any software, right?

292
00:54:56.420 --> 00:55:03.040
In the sense that there are continual improvements, and there's always competition between the, the frontiers within these industries.

293
00:55:03.600 --> 00:55:07.800
Last thing I'll note for this, Charlie, and then we can cap it unless you have anything else.

294
00:55:08.280 --> 00:55:17.800
Going back to existential questions, the scariest part about the AI race is the fact that China owns the bulk of the critical mineral refining process

295
00:55:18.740 --> 00:55:27.360
that is necessary to build not just the, the GPUs and the ASICs themselves, but also the infrastructure to house them.

296
00:55:28.100 --> 00:55:39.320
You know, depending on what you're looking at, China has 60 to 80 to 90% of critical mineral refinery- uh, refining capacity, uh, for the entire globe.

297
00:55:39.900 --> 00:55:52.260
That's everything from silver and copper to tungsten to germanium to all of- to cobalt, to all of these critical minerals that are going to be necessary to power the AI future.

298
00:55:53.650 --> 00:56:03.640
And I wonder how much, I assume it came up quite a bit, or at least it was a focal point. I just wonder what the negotiation table looked like when Trump went to Beijing.

299
00:56:05.200 --> 00:56:12.380
What, what did you do to address that elephant in the room? Because that's not just critical for AI, it's also critical for defense systems.

300
00:56:13.210 --> 00:56:25.880
And that, to me, is one of the biggest black swans lurking in the shadows in terms of the AI race between China and the US. The manufacturing is all over there. We don't have any of that anymore. Rare earth.

301
00:56:26.000 --> 00:56:41.160
There's, there's a lot to unpack here. It's, it's not just, it's not just enough that we've, you know, found, what is it, tons of lithium in Nevada or, like, tons of cobalt and tungsten in, like, California.

302
00:56:41.960 --> 00:56:49.880
Uh- We can't process any of it. We can't process them. Also, you don't want those processed. You thought the NIMBYism was bad for data centers?

303
00:56:50.040 --> 00:57:01.330
Wait until something they actually don't want goes nearby, such as a, like a tungsten processing facility. You know, there's a reason why China has basically

304
00:57:02.980 --> 00:57:17.720
indentured servants living in entire regions of the country whose whole purpose is to do these incredibly, uh, low quality of life, unbelievably dirty- And near zero environmental regulations. That's the other thing.

305
00:57:17.749 --> 00:57:29.550
Near... Yeah. Yeah. It would be, it would be, I would imagine, depending on the state, it would be incredibly onerous to deal with the waste from those facilities in a way- Yeah...

306
00:57:29.580 --> 00:57:39.040
that wouldn't just get the green thumbs calling for your heads. It wouldn't be the green thumbs. It'd be every single person. I'd protest it. I don't want those in near, near my backyard.

307
00:57:39.120 --> 00:57:55.960
I don't want them even in my state. Uh, so more, more on that as it, [chuckles] as it develops, I suppose. Uh, yeah. Banger of a deal day for AI. Um, I'm sorry to say, but, uh, nothing really happened in crypto.

308
00:57:56.080 --> 00:58:07.360
Bitcoin's still at low 60K and Stretch is not yet pegged back to 100. But we'll talk about those some other day. Thank you so much for listening.

309
00:58:07.380 --> 00:58:14.840
If you're listening on Coindesk, make sure to pop on over to the Blockspace feed. The Coindesk feeds are ending very, very soon.

310
00:58:15.640 --> 00:58:29.460
Like and subscribe, hit our newsletter at newsletter.blockspacemedia.com for daily newsletters, and, uh, leave us a review. Let us know what you do and don't like, and, uh, let us know on Twitter.

311
00:58:29.580 --> 00:58:42.400
Our handles are posted below our names here. Otherwise, thank you very much for tuning in. I'm Charlie. I'm Colin. And this is Blockspace Live. [upbeat music]