Bitcoin Mining in the Age of AI | with MARA CEO Fred Thiel

1
00:00:00,000 --> 00:00:10,380
This year has been a year where we have been quietly putting in place the pieces that will be very evident to the marketplace in 26 regarding what it means when you own large amounts of power.
2
00:00:10,580 --> 00:00:24,720
You have land and the ability to build a variety of types of data centers and you're able to deploy and operate tier three and tier four data centers for the most sensitive data types for the most critical customers that need to operate with full sovereignty.
3
00:00:24,720 --> 00:00:34,480
Welcome, Fred Teal. Thank you so much for joining the Bitcoin Rails podcast. I feel like this has been a long time coming that we've been planning this.
4
00:00:34,480 --> 00:00:36,880
Thank you. It has. Yeah. Quite a long time.
5
00:00:36,880 --> 00:00:54,340
Quite a long time. And for me, an especially extra long time, because I don't know if many of my regular listeners even know that I've actually been working with Mara for a little over a year now, particularly on the Enduro and Slipstream projects, which Slipstream we talk about quite a bit on the show.
6
00:00:55,220 --> 00:01:01,100
Very influential project on sort of the broader Bitcoin tech world, which is what the show typically covers.
7
00:01:01,680 --> 00:01:07,660
But today we're going to do something a little different and kind of zoom out into the broader mining industry.
8
00:01:08,160 --> 00:01:08,320
Excellent.
9
00:01:08,320 --> 00:01:18,520
Yeah. So kind of to just sort of get started, I figured that we would talk about how Mara became the largest Bitcoin miner in the world.
10
00:01:19,440 --> 00:01:28,000
My understanding, Mara was a public company well before it got into the Bitcoin mining space in, what, 2017? Is that right?
11
00:01:28,560 --> 00:01:34,040
Yeah. So Mara was originally a company that did vanadium mining, which is a mineral.
12
00:01:34,640 --> 00:01:36,900
So they've been a miner for quite some time.
13
00:01:37,820 --> 00:01:41,280
And they didn't do very well at that. That was back in, I think, 2010.
14
00:01:42,140 --> 00:01:46,480
And then they did some real estate and then the company did some oil and gas.
15
00:01:46,480 --> 00:01:49,780
and then eventually was taken over by a group
16
00:01:49,780 --> 00:01:51,580
who turned it into a patent troll,
17
00:01:51,860 --> 00:01:52,940
Marathon Patent Group.
18
00:01:53,820 --> 00:01:55,580
And as a patent troll,
19
00:01:55,920 --> 00:01:57,400
what companies that are patent trolls do
20
00:01:57,400 --> 00:02:00,680
is they acquire typically bankrupt companies' assets.
21
00:02:01,580 --> 00:02:03,980
And one of the patents they acquired
22
00:02:03,980 --> 00:02:07,480
was a patent for, very basic patent,
23
00:02:07,480 --> 00:02:11,000
for using a limited instruction set
24
00:02:11,000 --> 00:02:15,040
for voice to control a portable or mobile device.
25
00:02:15,040 --> 00:02:18,940
which just happened to like tick the box
26
00:02:18,940 --> 00:02:22,600
that Siri is a voice controlled device as is Alexa.
27
00:02:23,060 --> 00:02:25,320
And so Marathon Patent Group sued Apple.
28
00:02:26,240 --> 00:02:29,600
Apple paid a license fee to them.
29
00:02:30,300 --> 00:02:33,540
And Marathon Patent Group then got very excited
30
00:02:33,540 --> 00:02:34,980
and sued Amazon.
31
00:02:35,240 --> 00:02:38,140
And that case is still going on.
32
00:02:38,200 --> 00:02:39,180
I don't know how many years now.
33
00:02:39,540 --> 00:02:39,820
Oh, really?
34
00:02:40,660 --> 00:02:43,540
But so that is still being litigated.
35
00:02:43,540 --> 00:02:51,340
And Marathon partnered with Fortress Investments, which has an arm that litigates patent suits.
36
00:02:52,540 --> 00:02:57,100
And so Fortress is running that process and we'll see if anything ever becomes of it.
37
00:02:57,100 --> 00:03:06,920
But in 2017, a group took control of Marathon Patent Group and decided they want to go into the mining business.
38
00:03:07,820 --> 00:03:13,000
And so they brought on board a friend of mine, a gentleman by the name of America Komodo.
39
00:03:13,540 --> 00:03:21,100
who was a lower mid-market finance guy, had run hedge funds, and was an investor.
40
00:03:22,080 --> 00:03:26,060
And he had to first fix the balance sheet of the company.
41
00:03:26,760 --> 00:03:30,320
It had some notes on it.
42
00:03:30,340 --> 00:03:31,540
It couldn't really raise capital.
43
00:03:31,740 --> 00:03:38,900
It didn't have enough liquid net worth to be able to issue equity on the market and generate equity returns.
44
00:03:39,660 --> 00:03:42,240
And so he took a while to do that.
45
00:03:42,240 --> 00:03:51,640
I was asked to join the board in early 2018 because I was familiar with Bitcoin and crypto and the other board members weren't.
46
00:03:52,920 --> 00:03:58,520
And so we spent some time changing players on the board and getting everything ready to go.
47
00:03:58,520 --> 00:04:04,800
and then lit up the first Bitcoin mine up in Canada
48
00:04:04,800 --> 00:04:09,640
in a building where when we turned on the miners,
49
00:04:10,220 --> 00:04:13,860
there was such a vacuum from all of the fans
50
00:04:13,860 --> 00:04:17,680
that it took the paint off the sides of the walls
51
00:04:17,680 --> 00:04:21,100
and that paint came into the miners and trashed the miners.
52
00:04:22,360 --> 00:04:23,480
And there's a great video.
53
00:04:24,480 --> 00:04:27,500
Somewhere there's a video of this on the internet.
54
00:04:27,500 --> 00:04:35,240
uh from a we did a movie called the road to 23x ash and it's in that movie um and then we had to
55
00:04:35,240 --> 00:04:42,760
kind of reboot start all over again and a long story short um we had some bitcoin mining going
56
00:04:42,760 --> 00:04:51,940
at a uh adjacent to a coal fired plant in hardened montana and then had problems there where there
57
00:04:51,940 --> 00:04:57,400
was a weather event that caused part of that coal plant to literally go offline completely
58
00:04:58,280 --> 00:05:05,300
And so I stepped off the board and into the CEO role in April of 2021 as employee number five.
59
00:05:06,380 --> 00:05:10,080
And we essentially had no hash rate running pretty much.
60
00:05:11,000 --> 00:05:15,020
And we decided it was time to really grow.
61
00:05:15,020 --> 00:05:20,880
We had been able to raise capital through a vehicle called an ATM, which is an at-the-money offering.
62
00:05:21,600 --> 00:05:25,240
At that point, so in 2021, Bitcoin price was on a tear.
63
00:05:25,240 --> 00:05:29,360
everybody was interested in investing in Bitcoin miners, the retail market was going crazy.
64
00:05:30,140 --> 00:05:36,020
And so we were able to raise quite a lot of money. We placed a huge order with Bitmain for
65
00:05:36,020 --> 00:05:41,740
70,000 machines. And then we just had to find some place to plug them in. And the Bitcoin mining
66
00:05:41,740 --> 00:05:45,560
industry really depends on three things. And by the way, the AI industry is the same way.
67
00:05:46,100 --> 00:05:49,840
You have to have capital. When you have capital, you have to get access to compute.
68
00:05:50,280 --> 00:05:52,880
And then you have to have some place to plug them in, which is capacity.
69
00:05:52,880 --> 00:06:21,820
And we were an asset light miner, meaning we were looking for hosting partners where we could plug miners in. And so over the course of 2021, we started deploying some miners. Again, we bought more miners. We did a bond offering, convertible bond offering at the end of 21, raised $870 million and placed a very big order with Bitmain for, at the time, the state of the art S19XPs.
70
00:06:21,820 --> 00:06:26,220
and then deployed more in 22
71
00:06:26,220 --> 00:06:28,800
as the market was kind of crashing.
72
00:06:29,940 --> 00:06:32,600
And by the end of 2023,
73
00:06:32,600 --> 00:06:35,280
we were the largest miner in the world.
74
00:06:35,980 --> 00:06:38,400
And towards the second part of 2023,
75
00:06:38,400 --> 00:06:41,780
we realized we really need to own the infrastructure
76
00:06:41,780 --> 00:06:43,560
we operated and you couldn't be asset light.
77
00:06:44,360 --> 00:06:47,720
Because when you work with a third-party partner,
78
00:06:48,040 --> 00:06:51,400
you're relying on them to operate the machines.
79
00:06:51,400 --> 00:06:54,000
You're relying on them to maximize uptime.
80
00:06:54,160 --> 00:06:55,920
You're relying on them to service the machines.
81
00:06:56,600 --> 00:07:01,520
And what we found was that we were much better at doing that than any of the third-party hosting partners that we had.
82
00:07:02,260 --> 00:07:07,000
And so we were able to acquire, again, go back 2023, the market was horrible.
83
00:07:08,520 --> 00:07:14,560
We were able to acquire 70% of the capacity where we were hosted for less than it would have cost us to build it.
84
00:07:15,340 --> 00:07:15,600
Oh, wow.
85
00:07:15,600 --> 00:07:25,560
So out of curiosity, just going back for a second, during this whole period where you're making the transition into Bitcoin mining, did investors think that you were crazy?
86
00:07:25,980 --> 00:07:31,380
I mean, were there other, were you guys like some of the first public companies to mine Bitcoin?
87
00:07:31,580 --> 00:07:34,180
I mean, what was the environment like at that time?
88
00:07:34,380 --> 00:07:36,880
So basically you had Riot and you had Core Scientific.
89
00:07:37,200 --> 00:07:39,060
They were already public at mining Bitcoin.
90
00:07:39,060 --> 00:07:39,100
They were already public.
91
00:07:40,360 --> 00:07:42,580
I think HUD-8 was already public.
92
00:07:42,580 --> 00:07:45,920
Hive was certainly already public.
93
00:07:46,240 --> 00:07:49,260
When did there start to be public mining Bitcoin companies in general?
94
00:07:49,440 --> 00:07:51,940
This was like a meta at one point, I think I remember.
95
00:07:51,940 --> 00:07:57,300
Yeah, I mean, it was kind of what happened was in 21.
96
00:07:57,640 --> 00:08:03,900
So late 2021 is when most of them tried to either by acquiring a shell.
97
00:08:04,720 --> 00:08:06,180
Riot, I think, actually did an IPO.
98
00:08:06,300 --> 00:08:12,360
I'm not quite sure because they went public much earlier because they were in a different business.
99
00:08:12,360 --> 00:08:16,320
But a lot of these were already public companies that somebody repurposed.
100
00:08:17,180 --> 00:08:19,200
Because they needed like a new revenue.
101
00:08:19,240 --> 00:08:20,380
They needed a new business model.
102
00:08:20,680 --> 00:08:22,300
It was basically what's called an empty shell.
103
00:08:22,420 --> 00:08:24,280
It's sort of what's happening with DATS right now.
104
00:08:24,420 --> 00:08:24,660
Is that?
105
00:08:25,020 --> 00:08:25,320
Yeah.
106
00:08:25,700 --> 00:08:26,100
Okay.
107
00:08:26,380 --> 00:08:26,700
Exactly.
108
00:08:26,940 --> 00:08:27,560
Fair enough.
109
00:08:27,780 --> 00:08:27,960
Okay.
110
00:08:28,360 --> 00:08:30,760
So it was a hot idea at the time.
111
00:08:30,780 --> 00:08:31,560
It was a hot idea.
112
00:08:31,880 --> 00:08:39,440
And, you know, to be fair, at the end of 2021, I think our stock hit a peak of $81 a share.
113
00:08:39,880 --> 00:08:40,360
Oh, wow.
114
00:08:40,360 --> 00:08:59,180
And we had a multi-billion dollar market cap. We weren't a huge hodler. We did buy, we had so much cash on the books in January of 2021 that we actually bought $150 million of Bitcoin in the market because we didn't have our miners deployed. So we had to do something with the money.
115
00:08:59,520 --> 00:09:06,540
At that time, were you seen as having an advantage in Bitcoin mining relative to the other public mining companies or was everyone just ripping?
116
00:09:06,960 --> 00:09:10,380
We were kind of at a disadvantage because we didn't own our infrastructure.
117
00:09:11,000 --> 00:09:15,880
Oh, so Corcyon, Riot, et cetera, were already in the owning power generation phase.
118
00:09:15,980 --> 00:09:18,300
Well, not power generation, but they owned their own data centers.
119
00:09:18,820 --> 00:09:19,200
Okay.
120
00:09:19,420 --> 00:09:20,340
So we were just hosted.
121
00:09:20,840 --> 00:09:21,120
Okay.
122
00:09:21,360 --> 00:09:23,780
So they had PPAs, but they were just...
123
00:09:23,780 --> 00:09:24,060
Exactly.
124
00:09:24,060 --> 00:09:29,240
And so they had lower energy costs, which was an advantage in 23 when the market...
125
00:09:29,240 --> 00:09:31,360
You were extra asset light relative to them.
126
00:09:31,620 --> 00:09:35,320
Now you're extra asset heavy, maybe relative to everyone?
127
00:09:35,320 --> 00:09:36,840
or is everyone kind of getting
128
00:09:36,840 --> 00:09:38,100
asset heavy at this point,
129
00:09:38,160 --> 00:09:39,680
meaning trying to acquire energy?
130
00:09:39,840 --> 00:09:41,380
I think it's all changing.
131
00:09:41,560 --> 00:09:43,300
The ciphers and irons of the world
132
00:09:43,300 --> 00:09:45,340
are going down the AI route very quickly,
133
00:09:45,540 --> 00:09:48,840
which if you just think about the capex spend
134
00:09:48,840 --> 00:09:49,840
is very asset heavy
135
00:09:49,840 --> 00:09:52,680
because all these GPUs cost a lot more than,
136
00:09:52,840 --> 00:09:53,660
you know, it's more than,
137
00:09:54,140 --> 00:09:56,000
it's 10 to 15 times more expensive
138
00:09:56,000 --> 00:09:57,360
to build a data center
139
00:09:57,360 --> 00:10:02,000
and equip it with miners or GPUs
140
00:10:02,000 --> 00:10:03,140
than Bitcoin miners.
141
00:10:04,420 --> 00:10:04,800
You know,
142
00:10:05,320 --> 00:10:14,440
A data center for essentially what's called bare metal, where you're just renting out your GPUs to a hyperscaler versus a Bitcoin mining operation.
143
00:10:15,940 --> 00:10:21,580
You know, on the one case, you're paying maybe a million dollars a megawatt for infrastructure and miners.
144
00:10:22,040 --> 00:10:25,820
In the other case, you're paying 12 to 20 million a megawatt.
145
00:10:25,820 --> 00:10:28,440
And is that that's just because of the info that's in the hole?
146
00:10:28,440 --> 00:10:29,520
Ten times higher return.
147
00:10:29,900 --> 00:10:32,240
It's, you know, AI has a much higher return.
148
00:10:33,140 --> 00:10:36,000
Is it also more expensive to set those kinds of data centers up?
149
00:10:36,420 --> 00:10:43,040
Yeah, that's something that came up in my research when I was sort of thinking about Xion and NARA's recent acquisition of Xion, right?
150
00:10:43,440 --> 00:10:51,520
I don't know if you can share a little bit about the differences between mining Bitcoin practically and being an AI on the data center level.
151
00:10:51,520 --> 00:10:53,640
Let me kind of reframe it.
152
00:10:53,720 --> 00:10:58,140
If you think about what are Bitcoin miners really good at?
153
00:10:58,140 --> 00:11:03,180
They're really good at acquiring land and acquiring power and then building very low-cost infrastructure.
154
00:11:03,820 --> 00:11:05,800
So typically modular using containers.
155
00:11:06,280 --> 00:11:12,160
Some people just use tents to essentially deploy miners at very low cost.
156
00:11:12,380 --> 00:11:14,800
And it's a tier one data center.
157
00:11:15,860 --> 00:11:20,240
Tier four is the highest level of security, capability, etc.
158
00:11:20,380 --> 00:11:22,160
Bitcoin miners are tier one data centers.
159
00:11:22,340 --> 00:11:26,200
Most AI data centers typically want to be tier four, tier three.
160
00:11:27,120 --> 00:11:29,560
And so it's a very different design how you do it.
161
00:11:29,840 --> 00:11:31,880
But Bitcoin miners are really good at acquiring land.
162
00:11:31,980 --> 00:11:34,880
They're great at acquiring power that's inexpensive.
163
00:11:35,620 --> 00:11:38,200
And then they're good at buying ASICs.
164
00:11:38,200 --> 00:11:44,200
And the Bitcoin mining industry went from CPU to GPU to FPGA to ASICs very quickly.
165
00:11:44,540 --> 00:11:47,360
And you're now seeing the AI market do the same thing.
166
00:11:48,480 --> 00:11:54,340
Google, Meta, all these guys are designing their own ASICs or they're buying from Broadcom as opposed to using NVIDIA chips.
167
00:11:54,340 --> 00:11:59,580
So the machines and AI are also going to become rapidly irrelevant.
168
00:11:59,880 --> 00:12:00,560
I mean, they're there.
169
00:12:00,560 --> 00:12:03,440
There's a very rapid technology obsolescence curve.
170
00:12:03,580 --> 00:12:07,280
You know, you had H100s were kind of the thing you were buying three years ago.
171
00:12:07,440 --> 00:12:11,760
Now you're on B300s, I think, is the latest model that's available.
172
00:12:12,560 --> 00:12:24,060
Much faster, uses less energy per calculus, very expensive machines, but they require a huge amount of power infrastructure inside the data center to be able to plug them in.
173
00:12:24,340 --> 00:12:28,900
And so every three to five years, you literally have to rip out and replace everything.
174
00:12:29,300 --> 00:12:30,640
Just like in Bitcoin mining.
175
00:12:30,880 --> 00:12:31,500
It's very simple.
176
00:12:31,640 --> 00:12:32,440
Yeah, it sounds the same.
177
00:12:32,700 --> 00:12:37,940
And so if you think about a data center, it's land and it's power.
178
00:12:38,080 --> 00:12:42,740
And then you choose whether you're going to run Bitcoin mining or AI or HPC.
179
00:12:42,820 --> 00:12:47,020
And in the AI world, you have really two different types of workloads.
180
00:12:47,020 --> 00:12:56,640
You have a workload around training, which are high-intensity workloads running at a constant amount, and you can't shut those off.
181
00:12:56,940 --> 00:12:59,280
It has to operate, can't curtail it.
182
00:13:01,020 --> 00:13:08,720
Inference workloads, which is really when you ask a question and get an answer, those are short-burst workloads.
183
00:13:09,780 --> 00:13:15,160
And those will have a varied amount of power demand in the data center.
184
00:13:15,160 --> 00:13:23,280
If you think about Bitcoin mining, you're running along doing your block calculus, and then the grid operator may say, hey, you need to shut down, right?
185
00:13:23,500 --> 00:13:26,100
And you curtail, and you have curtail, and then you come back on.
186
00:13:26,260 --> 00:13:28,920
It's really hard for AI data centers to do that part.
187
00:13:29,280 --> 00:13:34,300
But if you marry Bitcoin mining and AI in the same data center, you get some flexibility.
188
00:13:35,200 --> 00:13:43,680
And it's very important because in the U.S. grid today, there's an estimated demand of about 40 gigawatts of power plus for AI.
189
00:13:44,460 --> 00:13:54,840
Duke University did a study where they found that there is 70 gigawatts of power available today that AI operators could use if they were flexible 2% of the time.
190
00:13:55,560 --> 00:13:56,580
Only 2%?
191
00:13:56,940 --> 00:13:57,440
That's it.
192
00:13:58,020 --> 00:14:01,240
The problem is then you have to look at, okay, where is that location?
193
00:14:01,600 --> 00:14:02,640
Where that power is available?
194
00:14:03,300 --> 00:14:05,280
Does that power have ample access to internet?
195
00:14:05,540 --> 00:14:06,600
Is there water for cooling?
196
00:14:06,720 --> 00:14:07,260
Things like that.
197
00:14:07,260 --> 00:14:13,060
And this is a key distinction, I believe, with AI in general, is that it's less geographically flexible than Bitcoin mining.
198
00:14:13,520 --> 00:14:14,040
Historically.
199
00:14:14,400 --> 00:14:15,580
Historically, but that's changing.
200
00:14:16,320 --> 00:14:20,880
The fact that you can put data centers in space these days kind of mixes things up.
201
00:14:20,940 --> 00:14:32,240
The fact that Starlink has an amazing network of satellites is starting to make it viable to do inference, not training, but inference in pretty remote locations.
202
00:14:32,240 --> 00:14:50,680
And I think what you'll find over time is the world of inference will look very much like Bitcoin mining, where you have people operating data centers, some of which are near prem, meaning near where the customer is or the user, metro level, and then remote, large scale.
203
00:14:51,000 --> 00:14:54,420
And then there's an orchestration layer that balances between those.
204
00:14:55,160 --> 00:14:56,320
And so very similar to Bitcoin.
205
00:14:56,500 --> 00:14:59,320
We have a pool that we control, which is our orchestration layer.
206
00:14:59,320 --> 00:15:29,300
Right. Mara's pool essentially assembles transactions into blocks and then directs our miners in 16 data centers on four continents to all go solve those blocks. Right. And so in the world of AI, you'll find something similar over time develop around inference, which will be front end load balancing and orchestration that then is able to route jobs to the data center that has the right level of latency, the right amount of capacity available for that specific job.
207
00:15:29,960 --> 00:15:33,040
And then data centers will have other features to them.
208
00:15:33,220 --> 00:15:47,440
So zooming out a little bit, what was the transition from Bitcoin mining exclusively into the sort of multi-pronged AI mining, selling back to the grid, you know, this optionality strategy that I've heard you talk about on other interviews?
209
00:15:47,920 --> 00:15:49,840
When did that happen?
210
00:15:49,840 --> 00:16:00,740
So when we started looking a couple of years ago, as we were buying infrastructure, buying sites, that we decided we were also going to buy power gen.
211
00:16:01,200 --> 00:16:02,800
So we were going to buy wind farms.
212
00:16:02,900 --> 00:16:15,860
We were going to do power generation off of flare gas and oil fields and build a portfolio of sites where the average cost to mine kept moving downwards as close to near zero as possible.
213
00:16:16,880 --> 00:16:18,500
And so how do you do that?
214
00:16:18,500 --> 00:16:23,160
Well, if you own a wind farm, you don't have to pay anything for the wind to blow.
215
00:16:24,180 --> 00:16:26,500
You just have to keep that wind farm operational.
216
00:16:27,600 --> 00:16:28,860
It's a flat fee, basically.
217
00:16:29,280 --> 00:16:30,540
It's a flat fee, right?
218
00:16:30,700 --> 00:16:33,940
And so whenever the wind blows, it's essentially free energy.
219
00:16:35,480 --> 00:16:40,640
And in flare gas, whenever you're using flare gas, there's a very low cost.
220
00:16:40,740 --> 00:16:44,460
I mean, it's maybe a penny a kilowatt for power.
221
00:16:45,180 --> 00:16:46,860
And so it's very low cost.
222
00:16:46,860 --> 00:16:55,340
If you look at grid connected Bitcoin mining, most the average price a miner is paying to be plugged into the grid is somewhere between four to five cents a kilowatt hour.
223
00:16:56,500 --> 00:17:01,740
I mean, I don't know if this is public information, but what are we paying on average for energy across all miners?
224
00:17:02,600 --> 00:17:04,500
I mean, I think, well, it is.
225
00:17:04,600 --> 00:17:11,720
You can do the math, but it's in the somewhere between three to four and a half cents.
226
00:17:11,780 --> 00:17:14,660
It depends on the time of year and amount of curtailment.
227
00:17:14,680 --> 00:17:15,160
Makes sense.
228
00:17:15,160 --> 00:17:24,560
Um, but, um, so we have a mix of direct behind the meter and then grid attached, um, power
229
00:17:24,560 --> 00:17:26,620
asset or mining assets.
230
00:17:26,940 --> 00:17:31,420
So when you look at a data center, it really is a similar portfolio in a data center.
231
00:17:31,600 --> 00:17:33,880
There are different things you can use the power for.
232
00:17:34,280 --> 00:17:36,760
You can run Bitcoin mining as your base business.
233
00:17:37,080 --> 00:17:41,380
You can choose to host an HPC customer.
234
00:17:41,380 --> 00:17:44,020
and by the way, there are HPC customers
235
00:17:44,020 --> 00:17:46,540
who don't need 500 megawatts or a gigawatt of power.
236
00:17:46,960 --> 00:17:49,020
There are some that love 25 megawatts,
237
00:17:49,360 --> 00:17:50,420
50 megawatts of power.
238
00:17:50,640 --> 00:17:51,920
So you could take a data center
239
00:17:51,920 --> 00:17:56,380
and section off half of it to a smaller HPC customer.
240
00:17:56,680 --> 00:17:57,940
Then you have inference AI
241
00:17:57,940 --> 00:18:00,920
and then you have traditional compute workloads,
242
00:18:01,060 --> 00:18:02,960
private cloud infrastructure.
243
00:18:03,620 --> 00:18:05,920
And when you think of what a typical company
244
00:18:05,920 --> 00:18:07,300
of the future needs
245
00:18:07,300 --> 00:18:09,720
is they need somewhere to host their data.
246
00:18:09,720 --> 00:18:12,360
They need someplace to host their models.
247
00:18:12,800 --> 00:18:14,980
They need infrastructure to run their models.
248
00:18:15,540 --> 00:18:20,560
And then they need traditional communications and IT infrastructure.
249
00:18:21,620 --> 00:18:26,180
Most companies want to run this in a private way, private cloud.
250
00:18:26,920 --> 00:18:31,320
The reason why is it's very expensive to run stuff in the public cloud.
251
00:18:31,680 --> 00:18:39,700
If you were to go talk to an enterprise customer today, they would tell you that, you know, it's most probably 10 times more expensive to run something in the Amazon cloud.
252
00:18:39,720 --> 00:18:43,440
especially around AI than to run it in a private environment.
253
00:18:44,380 --> 00:18:46,500
Simply because they're just taking more fees?
254
00:18:46,720 --> 00:18:49,400
Well, what you have to do is if you're a larger enterprise
255
00:18:49,400 --> 00:18:53,620
and you want assured response times on your models,
256
00:18:53,620 --> 00:18:58,340
you have to request that you get directly allocated systems
257
00:18:58,340 --> 00:18:59,220
and that are dedicated.
258
00:18:59,580 --> 00:19:00,440
There's an upcharge for that.
259
00:19:00,480 --> 00:19:01,400
So it's a premium, basically.
260
00:19:01,800 --> 00:19:02,560
You pay a premium for that.
261
00:19:02,960 --> 00:19:06,080
The other big thing is that 70% of corporate data today
262
00:19:06,080 --> 00:19:07,880
is not in the public cloud
263
00:19:07,880 --> 00:19:13,620
because companies don't want it to be somewhere where it's available and used publicly.
264
00:19:13,780 --> 00:19:19,360
Nobody in their right mind wants to put their personal information into OpenAI's cloud.
265
00:19:20,320 --> 00:19:26,200
Every time you upload your health records into ChatGPT and say, interpret this for me,
266
00:19:26,200 --> 00:19:32,660
you are essentially giving them the right to use that data to inform other decisions.
267
00:19:32,860 --> 00:19:33,540
It's not secure.
268
00:19:33,540 --> 00:19:45,573
And so corporations who are building models and running models to analyze seismic data for oil and gas healthcare data for drug discovery financial data for fraud prevention whatever it might be
269
00:19:46,353 --> 00:19:48,173
they're not going to put that data in the cloud ever.
270
00:19:48,913 --> 00:19:50,533
They're going to run it in their own data centers.
271
00:19:51,053 --> 00:19:54,613
So if they're not putting the data in the cloud, it means they can't train a model that's in the cloud.
272
00:19:55,613 --> 00:19:59,873
And right now the hyperscalers are all focused on the big public cloud and fraud.
273
00:19:59,873 --> 00:20:06,833
But no one is really servicing these sort of enterprise clients who will eventually probably want their own private data.
274
00:20:06,973 --> 00:20:08,613
It is just starting to change.
275
00:20:08,853 --> 00:20:14,213
AWS just announced that they're going to offer private cloud services on-prem.
276
00:20:14,673 --> 00:20:16,013
Google is offering on-prem.
277
00:20:16,013 --> 00:20:28,653
There's a Lockheed deal that was recently announced where Google is essentially doing a fully air-gapped version of their solution on-prem for Lockheed that does not connect to the outside world.
278
00:20:29,473 --> 00:20:29,933
Interesting.
279
00:20:30,173 --> 00:20:33,093
Okay, so even the big hyperscalers are kind of getting hip to this strategy.
280
00:20:33,613 --> 00:20:41,093
Do Bitcoin miners and sort of kind of the more asset-heavy mega miners, if you will, have an advantage in this game?
281
00:20:42,153 --> 00:20:49,833
So if you were to go to JP Morgan and say, hey, I'm a Bitcoin miner.
282
00:20:50,073 --> 00:20:52,133
I want to offer you a private cloud.
283
00:20:52,333 --> 00:20:57,073
They likely will just look at you and say, you have no idea what you're talking about.
284
00:20:57,073 --> 00:21:02,353
And so this was one of the reasons we made the Xion investment.
285
00:21:03,713 --> 00:21:08,453
To be taken seriously in the world where you're offering private cloud services and inference AI.
286
00:21:08,613 --> 00:21:09,653
And let me just back up a second.
287
00:21:09,993 --> 00:21:13,673
If you look at what Bitcoin miners are doing today in AI, they're taking a site.
288
00:21:14,093 --> 00:21:20,853
They're essentially building a powered shell, a building with power, if they're even doing that.
289
00:21:21,153 --> 00:21:23,893
And they're offering up to a hyperscaler as a tenant.
290
00:21:24,073 --> 00:21:25,973
So they're going to the real estate business.
291
00:21:25,973 --> 00:21:32,093
They're doing pretty much a lower grade version of what Digital Realty Trust or Equinix does.
292
00:21:33,273 --> 00:21:36,393
They can go to the next step and they can actually put racks in that building.
293
00:21:37,373 --> 00:21:45,653
And they could even, in the case of some miners, will pay to put GPUs in the building and then rent that capacity out.
294
00:21:47,713 --> 00:21:54,593
And so you're either providing what's called power and paying, which is it's a building with power and network connectivity.
295
00:21:54,593 --> 00:22:01,053
and somebody is a tenant, which a number of the guys are doing today. But none of them are doing
296
00:22:01,053 --> 00:22:08,613
full-blown, let's offer inference AI plus private cloud plus all the services that you need to run
297
00:22:08,613 --> 00:22:16,213
that securely such that a major multinational, a company in Saudi Arabia, a bank, a healthcare
298
00:22:16,213 --> 00:22:23,373
your company would want to tenant with you. And so we decided we needed to have a partner
299
00:22:23,373 --> 00:22:28,473
who would give us credibility with customers because they knew how to do what we were trying
300
00:22:28,473 --> 00:22:34,293
to do, who already had customers, and who had technology that was some of the most advanced
301
00:22:34,293 --> 00:22:42,953
in that category. And we found Xion. Xion is a subsidiary of EDF, which is the largest energy
302
00:22:42,953 --> 00:22:49,533
producer in the world for electricity. They operate 58 nuclear reactors. And Xion is the
303
00:22:49,533 --> 00:22:55,913
operator of their tier three and tier four data centers, which house the nuclear data. And so
304
00:22:55,913 --> 00:23:04,133
ultra secure, ultra sovereign, ultra private. And these data centers cannot go down.
305
00:23:04,973 --> 00:23:10,613
Will part of that deal be some sort of relationship also with EDF and sort of the ability to get
306
00:23:10,613 --> 00:23:16,013
nuclear power? I mean, are you planning on using nuclear power in France also for power generation
307
00:23:16,013 --> 00:23:23,733
or no? The deal with Exion is purely an investment in a tech company. And EDF will retain a certain
308
00:23:23,733 --> 00:23:30,253
percentage of the company. So they have a vested interest in continuing to use Exion as a vendor.
309
00:23:30,953 --> 00:23:36,993
But by us coming in and investing money, Exion will be able to now have the growth capital to be
310
00:23:36,993 --> 00:23:40,913
able to really address a much bigger market, which is a huge opportunity for them. So we're
311
00:23:40,913 --> 00:23:46,313
super excited about that. And that deal will hopefully close in the first quarter of next
312
00:23:46,313 --> 00:23:52,493
year. It's just pending regulatory approval in France. We have been very active in France.
313
00:23:52,593 --> 00:23:58,593
We established a headquarters in France for our Europe and Middle East operations. And I think
314
00:23:58,593 --> 00:24:07,353
what you'll find over the next year is, you know, you have three of the top five energy,
315
00:24:07,513 --> 00:24:12,153
electrical energy producing companies in the world happen to be French. A lot of their assets
316
00:24:12,153 --> 00:24:17,933
are not in France, though. A lot of the assets are in the US. They're in places like Brazil.
317
00:24:17,933 --> 00:24:26,013
They're in places like the Middle East, Africa, and other places. And there is a lot of energy
318
00:24:26,013 --> 00:24:31,493
generation capacity that is not being sold today. Meaning that they may sell at peak,
319
00:24:31,593 --> 00:24:35,633
but they have a lot of excess capacity, which is perfect for Bitcoin mining.
320
00:24:36,873 --> 00:24:41,613
And so I think what you'll find is that similar to what I said four years ago at Mining Disrupt,
321
00:24:41,613 --> 00:24:47,673
which was as a Bitcoin miner, by 2028, you are either going to have to be an energy company or
322
00:24:47,673 --> 00:24:52,193
partner with an energy company, because otherwise the energy companies are going to want to do what
323
00:24:52,193 --> 00:24:59,753
you do, because when you look at national grid scale operations, being able to deploy large
324
00:24:59,753 --> 00:25:06,153
flexible loads to balance your grid is a key strategic advantage. And for example, in France,
325
00:25:07,173 --> 00:25:15,433
nuclear represents 60 plus percent of their energy. The vast majority of the rest is renewable energy.
326
00:25:15,653 --> 00:25:20,973
So they're one of the greenest grids on the planet and they have a decreasing demand for energy.
327
00:25:20,973 --> 00:25:26,633
decreasing population perhaps and they have an increasing amount of generation coming online
328
00:25:26,633 --> 00:25:32,693
and so they really need load balancing and so we believe that europe um will be a very attractive
329
00:25:32,693 --> 00:25:40,633
country for bitcoin miners to operate in um it is just a market with a lot more regulatory complexity
330
00:25:40,633 --> 00:25:47,633
it's not like in the u.s where you just get a ppa with a grid operator it's a very more highly
331
00:25:47,633 --> 00:25:53,473
regulated market. And so you really have to start at the top and work your way down while working
332
00:25:53,473 --> 00:25:58,553
from the bottom up in the marketplace. And we've been spending the better part of this year doing
333
00:25:58,553 --> 00:26:04,013
that. And we have very strong relationships from the top down, the bottom up. From a markets
334
00:26:04,013 --> 00:26:09,833
perspective, I mean, was there, you know, kind of, I mean, I guess, how would you differentiate
335
00:26:09,833 --> 00:26:15,413
between sort of excitement around AI versus Bitcoin mining? Are people sort of equally
336
00:26:15,413 --> 00:26:22,053
excited about both of these futures or, you know, is AI sort of the kind of path where
337
00:26:22,053 --> 00:26:26,493
most of these kind of Bitcoin megaminers are ultimately going to generate most of their
338
00:26:26,493 --> 00:26:26,813
revenue?
339
00:26:26,953 --> 00:26:28,293
Like, how do you see that split?
340
00:26:29,233 --> 00:26:33,713
Well, I think my opinion is different than what's written in the press.
341
00:26:33,713 --> 00:26:42,573
So, OK, let's start with if you read the press, go back to October, September time frame,
342
00:26:42,573 --> 00:26:49,993
You know, Bitcoin price nearing 120, AI stocks going through the roof.
343
00:26:50,113 --> 00:26:56,193
If you said you did iced tea with AI in it, you would get a 10x on your multiple on your stock price.
344
00:26:56,833 --> 00:26:58,433
Huge frothiness in the market.
345
00:26:58,753 --> 00:27:04,493
And I said in our earnings call for Q2 that we believe the market's very frothy.
346
00:27:04,573 --> 00:27:05,753
For both of them, basically.
347
00:27:06,033 --> 00:27:06,513
For both of them.
348
00:27:06,513 --> 00:27:21,333
And, you know, while there is no question that there is a huge demand for AI capacity, enterprises have barely started actually finding a viable use case for it.
349
00:27:21,633 --> 00:27:24,713
It's like a far away, very future looking industry.
350
00:27:24,713 --> 00:27:27,593
Go back to the Internet, late 90s.
351
00:27:28,093 --> 00:27:29,813
You put a product catalog online.
352
00:27:30,853 --> 00:27:31,733
Let's see what happens.
353
00:27:32,113 --> 00:27:33,753
Oh, right.
354
00:27:33,833 --> 00:27:35,753
Well, this isn't really generating a lot of revenue yet.
355
00:27:35,753 --> 00:27:58,433
And then five years later, you had, you know, companies like Amazon, Salesforce, and others who had just taken off with it. So the market's coming, and it's going to be predominantly around inference. It'll be predominantly around private models, meaning not using OpenAI necessarily or Gemini 3 for your corporate model.
356
00:27:58,433 --> 00:28:03,473
because why pay for models that are full of consumer data and information
357
00:28:03,473 --> 00:28:06,273
when you have your proprietary information
358
00:28:06,273 --> 00:28:08,633
and you want to train a model on it and nothing else?
359
00:28:08,973 --> 00:28:12,013
You don't want the hallucinations that all that other data inject
360
00:28:12,013 --> 00:28:16,653
or the complexities that that data may add to your product.
361
00:28:16,773 --> 00:28:22,353
So you're already starting to see now the rise of open source models taking over.
362
00:28:22,953 --> 00:28:25,813
You're already starting to see that in the enterprise market,
363
00:28:25,933 --> 00:28:27,673
Anthropic has a higher market share.
364
00:28:28,433 --> 00:28:36,553
than OpenAI does. And you're starting to see now that people, meaning investors, have realized that
365
00:28:36,553 --> 00:28:43,133
the circular OpenAI economy between NVIDIA and OpenAI, this all invest in you and you buy my
366
00:28:43,133 --> 00:28:50,573
product and then you'll invest in me and I'll buy your product, has some serious challenges.
367
00:28:51,033 --> 00:28:56,593
And, you know, Oracle, when they said they were going to do this $300 billion OpenAI deal,
368
00:28:56,593 --> 00:29:02,873
their stock. I think it made Larry Ellison one of the wealthiest people on the planet. And since
369
00:29:02,873 --> 00:29:06,973
then, the stock price has come down because they're now announcing, well, it's going to take
370
00:29:06,973 --> 00:29:12,953
longer than we thought, all this other stuff. Bitcoin mining, by the same token, no pun intended,
371
00:29:13,493 --> 00:29:18,353
you know, when Bitcoin price was at 120, it was like, oh, that's all this stuff is great, amazing.
372
00:29:18,353 --> 00:29:20,813
Well, Bitcoin price drops 25%.
373
00:29:20,813 --> 00:29:30,273
Just for perspective or context, you know, in 2021, at the peak, Bitcoin was at around 61,000.
374
00:29:30,953 --> 00:29:34,433
At January of 23, it was at 15,000.
375
00:29:34,633 --> 00:29:36,633
That's a 75% price drop.
376
00:29:37,673 --> 00:29:37,693
Right.
377
00:29:37,773 --> 00:29:41,233
So the price drop that we're experiencing right now is basically nothing, relatively speaking.
378
00:29:41,413 --> 00:29:45,513
The volatility of Bitcoin has dropped considerably, even over the last year.
379
00:29:45,513 --> 00:29:48,113
Went from 75% vol to about 50%.
380
00:29:48,353 --> 00:29:51,313
which is just a measure of the risk associated to it.
381
00:29:51,873 --> 00:29:53,953
You had all these corporate buyers coming in,
382
00:29:54,013 --> 00:29:54,993
you had the debts coming in,
383
00:29:54,993 --> 00:29:57,053
you had all this kind of demand
384
00:29:57,053 --> 00:30:00,373
over and above what would be normal demand.
385
00:30:00,953 --> 00:30:04,113
You now have a market where most of the top banks
386
00:30:04,113 --> 00:30:07,533
are going to either custody or allow you to hold Bitcoin.
387
00:30:07,793 --> 00:30:09,353
You have most of the wealth advisors
388
00:30:09,353 --> 00:30:12,313
recommending a two to 4% allocation.
389
00:30:13,313 --> 00:30:17,773
You are soon going to see the SEC hopefully recommend that
390
00:30:17,773 --> 00:30:20,393
or allow that Bitcoin can be held in 401ks.
391
00:30:21,593 --> 00:30:25,873
And, and, and, and, and Bitcoin is becoming part of the financial ecosystem.
392
00:30:27,173 --> 00:30:30,833
But what's happened is whales have started selling.
393
00:30:31,453 --> 00:30:33,493
I mean, think of it this way.
394
00:30:33,513 --> 00:30:36,753
If you paid $10 for Bitcoin and Bitcoin was at 100K
395
00:30:36,753 --> 00:30:41,393
and you had been holding for 10 to 13 years,
396
00:30:41,533 --> 00:30:44,393
wouldn't you maybe want to take some chips off the table?
397
00:30:44,873 --> 00:30:47,773
Maybe you're getting to an age where you're thinking about estate planning.
398
00:30:48,133 --> 00:30:53,013
And, you know, I'm a multi-multi-billionaire, and I'd like to somehow use this.
399
00:30:53,593 --> 00:30:55,013
And so what do they do?
400
00:30:55,073 --> 00:30:58,453
They either sell part of the Bitcoin, they convert it into an ETF.
401
00:30:58,813 --> 00:31:05,353
You can contribute in kind, tax-free or tax advantage, and then you can borrow against it.
402
00:31:05,413 --> 00:31:08,773
So now you're not even selling your Bitcoin, but now you can borrow against it at low rates.
403
00:31:08,773 --> 00:31:11,993
And a big thing here, I think, in this conversation, I love this thesis,
404
00:31:11,993 --> 00:31:15,053
This is sort of like Bitcoin's going through its IPO stage.
405
00:31:15,133 --> 00:31:15,773
I've heard this.
406
00:31:15,913 --> 00:31:25,913
I think you actually probably shared about this before, is that, you know, in the past, there just wasn't enough liquidity in Bitcoin to kind of service these kinds of exits, basically.
407
00:31:26,133 --> 00:31:26,993
But now we have that.
408
00:31:27,333 --> 00:31:27,513
Exactly.
409
00:31:27,653 --> 00:31:30,773
So now you have this huge pool of liquidity in the market.
410
00:31:30,893 --> 00:31:32,993
I mean, Bitcoin is a highly liquid asset.
411
00:31:32,993 --> 00:31:48,713
And so you're starting to see that, you know, kind of IPO stage where the initial investors and cash out and the new investors come in and you get this kind of, hey, you know, I bought it on the IPO day and now the price is down 10, 15 percent.
412
00:31:48,833 --> 00:31:50,373
You get some people that then sell out.
413
00:31:50,533 --> 00:31:54,853
And most of the selling has been people who held Bitcoin for less than 150 days.
414
00:31:54,893 --> 00:32:01,613
So they were buying on the up run the last kind of months before Bitcoin hit a peak.
415
00:32:01,613 --> 00:32:18,333
And so when Bitcoin falls out of favor, like it has now for retail, because they were off chasing the AI bunny, and now the AI bunny is looking a little ugly, so they're chasing other things, or just licking their wounds, you have pressure.
416
00:32:18,333 --> 00:32:36,393
I mean, Saylor, look at, you know, MicroStrategy. I mean, I have a huge respect for Michael. He's a good friend. You know, I think his strategy of buying and holding Bitcoin and sort of creating the first real debt was brilliant.
417
00:32:36,393 --> 00:32:42,393
But over time, you have to keep that machine fed and going.
418
00:32:43,033 --> 00:32:51,033
And when you get a large drop in liquidity in the Bitcoin market, like what's happened now, it gets harder.
419
00:32:52,353 --> 00:32:56,553
And so as you start trading below your MNAV, you all of a sudden become less attractive.
420
00:32:57,013 --> 00:33:02,433
There's only so long, basically, that that strategy can run and outperform Bitcoin, essentially.
421
00:33:02,433 --> 00:33:06,293
And, you know, I don't doubt that Michael will come up with something new.
422
00:33:06,393 --> 00:33:10,673
a new way to do it. Never underestimate Michael Thaler. Absolutely not.
423
00:33:12,473 --> 00:33:16,633
And but I think also at the same time, you know, he attracted lots of copycats.
424
00:33:17,153 --> 00:33:22,953
And while who diluted this entire strategy as well, I think that's a key problem that's
425
00:33:22,953 --> 00:33:28,233
happening right now. Exactly. And, you know, a lot of these latecomers to the market
426
00:33:28,233 --> 00:33:35,113
have suffered Nakamoto, Metaplanet, etc. And to be fair, Metaplanet wasn't a latecomer.
427
00:33:35,113 --> 00:33:38,413
They've been at this in a big way.
428
00:33:38,413 --> 00:33:45,873
And, you know, if you look at our Bitcoin holdings, over 50% of it is mined and the rest is purchased.
429
00:33:46,813 --> 00:33:52,433
But I think the DAT model, there will be one or two that remain.
430
00:33:52,573 --> 00:33:55,033
I think strategy will continue to exist.
431
00:33:55,353 --> 00:34:01,913
Michael will adapt, find a new way to raise more capital or find more unique ways to raise capital and be able to keep doing it.
432
00:34:01,913 --> 00:34:23,113
Bitmind, Tom Lee has done an amazing job on the Ethereum side in doing this, but I don't think there's room for a lot of them. I think 21, we'll see. They have to come up with an operating business, but they have Tether and SoftBank behind them. So who knows? Deep pockets, lots of money.
433
00:34:23,113 --> 00:34:28,853
So, you know, we hold, we're the second largest publicly traded holder of Bitcoin.
434
00:34:29,413 --> 00:34:31,333
And for us, it's an item on our balance sheet.
435
00:34:31,433 --> 00:34:32,513
We generate yield with it.
436
00:34:33,173 --> 00:34:39,513
And it's there to use maybe someday as a growth capital source or maybe not.
437
00:34:39,573 --> 00:34:48,633
But as long as the Bitcoin generates an appreciation that's greater than what we can get by doing other things with our capital, then we'll keep it there.
438
00:34:48,733 --> 00:34:49,673
All right, guys.
439
00:34:49,673 --> 00:34:57,093
I want to take a quick break to thank you for watching and remind you to click the follow button on my profile if you want more Bitcoin technical deep dives like this.
440
00:34:57,373 --> 00:35:00,393
I also want to give a shout out to my partners for making the show possible.
441
00:35:00,813 --> 00:35:08,453
My leading partner, Best in Slot, just released the programmability module for BRC20, making the execution of smart contracts possible for the protocol.
442
00:35:08,873 --> 00:35:11,673
I know that's been long awaited by the BRC20 community.
443
00:35:11,673 --> 00:35:17,113
alongside this release, Best in Slot put out what is arguably the most sophisticated DEX for trading
444
00:35:17,113 --> 00:35:21,533
Bitcoin native assets called CoinCooker, which I also highly recommend checking out if you want to
445
00:35:21,533 --> 00:35:27,813
dive into the BRC20 trading game and get started executing smart contracts on Bitcoin. I'd also
446
00:35:27,813 --> 00:35:32,973
like to thank Citraya, building one of the leading implementations of a BitVN style bridge for
447
00:35:32,973 --> 00:35:38,133
Bitcoin. Citraya is building what I think could become the canonical Bitcoin rollup, enabling
448
00:35:38,133 --> 00:35:43,173
programmability, scaling, really anything is possible with a trust-minimized bridge,
449
00:35:43,253 --> 00:35:47,113
which is the UD value prop of these BitBM-style roll-ups on Bitcoin.
450
00:35:47,593 --> 00:35:52,013
Lastly, I'm thrilled to announce my newest partner, Spark, which is the leading state
451
00:35:52,013 --> 00:35:53,293
chains protocol on Bitcoin.
452
00:35:53,913 --> 00:35:58,153
State chains builds upon what was made possible by the Lightning Network for payments, but
453
00:35:58,153 --> 00:36:02,473
without the liquidity constraints and the UX challenges that often make self-custodial
454
00:36:02,473 --> 00:36:03,653
Lightning impractical.
455
00:36:03,653 --> 00:36:10,733
Spark is also led by one of the leading payments infrastructure teams on the planet, including founders from PayPal, Facebook.
456
00:36:11,353 --> 00:36:17,413
They're incredibly well positioned to turn Bitcoin into a true institutional standard for payment systems around the world.
457
00:36:17,593 --> 00:36:20,253
And I couldn't be more excited to support what they're up to.
458
00:36:20,513 --> 00:36:22,473
All right. On that note, let's get back to the show.
459
00:36:23,393 --> 00:36:28,713
So going back really quickly into this sort of asset light versus asset heavy transition.
460
00:36:28,713 --> 00:36:35,473
that transition occurred for you guys before the big AI revolution was really taking place.
461
00:36:35,533 --> 00:36:39,513
So that was something that was a foregone conclusion for you just with Bitcoin mining.
462
00:36:40,053 --> 00:36:45,173
Can you share a little bit more about that and maybe what the implications might be for
463
00:36:45,173 --> 00:36:52,733
small scale miners or miners who may not have the cap dev to, you know, acquire their own energy
464
00:36:52,733 --> 00:36:56,453
sources? Yeah, I think if you look at where.
465
00:36:56,453 --> 00:37:02,913
so they're kind of three drivers as i said earlier um in the bitcoin business it's capital
466
00:37:02,913 --> 00:37:09,773
compute and capacity so the longest pole in the tent is capacity that's the thing that takes the
467
00:37:09,773 --> 00:37:14,213
longest to get you have to find power and find land once you get that now you need to have money
468
00:37:14,213 --> 00:37:19,413
to buy it the land or at least lease it you need to get a ppa you may have to put a huge deposit
469
00:37:19,413 --> 00:37:25,673
down so you need capital for that and then you need to go buy buy miners back in the last cycle
470
00:37:25,673 --> 00:37:31,893
miners, the market was controlled by Bitmain and they purposely kept pricing high.
471
00:37:32,953 --> 00:37:33,773
How did they do that?
472
00:37:33,933 --> 00:37:35,253
Just by constraining supply.
473
00:37:35,433 --> 00:37:35,713
Okay.
474
00:37:36,033 --> 00:37:36,213
Right.
475
00:37:36,613 --> 00:37:37,553
Miners had lots of money.
476
00:37:37,633 --> 00:37:38,593
They had lots of capacity.
477
00:37:38,893 --> 00:37:39,693
They couldn't buy compute.
478
00:37:40,193 --> 00:37:41,813
And so they purposely constrained supply.
479
00:37:42,433 --> 00:37:45,433
And then in 22, 23, the price just came down.
480
00:37:46,313 --> 00:37:49,093
You saw a lot of miners buy machines in 24.
481
00:37:49,993 --> 00:37:54,673
You know, if you look at the number of miners who are now at or near our scale, you know,
482
00:37:54,673 --> 00:37:57,613
there are four or five guys that are right behind us in size.
483
00:37:59,313 --> 00:38:02,893
And many miners are hurting from this 2024 bulk purchase
484
00:38:02,893 --> 00:38:06,113
that hasn't actually accrued the value they were hoping.
485
00:38:06,373 --> 00:38:06,753
Exactly.
486
00:38:07,473 --> 00:38:09,473
And now, if you look in the marketplace,
487
00:38:11,193 --> 00:38:13,353
you know, global hash rate has continued to grow.
488
00:38:14,013 --> 00:38:16,533
You have people like Tether who are in with Electron
489
00:38:16,533 --> 00:38:20,853
buying every small miner, every miner kind of around the place
490
00:38:20,853 --> 00:38:23,433
that is standing that they can buy.
491
00:38:24,673 --> 00:38:27,373
I think it's going to be a challenge for them to integrate those.
492
00:38:27,573 --> 00:38:29,153
When did they get into mining, Tether?
493
00:38:29,793 --> 00:38:38,133
Well, you know, in the same way they got into the gold business and they're now one of the largest buyers of gold in the world because they have the capital to do it.
494
00:38:41,853 --> 00:38:45,553
Investors love to compare our revenue per employee to Tethers.
495
00:38:45,553 --> 00:38:49,173
It's like, OK, you know, how many billion for 50 employees?
496
00:38:49,173 --> 00:39:25,826
Anyway but if you look at the small miners you know there going to be a consolidation happening right Tether consolidating a number of small miners I think you going to see others do it The challenge is if you going after the hyperscaler market if you want to service Microsoft or you want to service Google you need a lot of power So the small miners aren going to do it These small miners have 25 30 40 50 megawatts They can get into the AI game per se They can get into the HPC game HPC game And they can get into a hyperscaler game
497
00:39:26,906 --> 00:39:28,286
They just don't have the capital.
498
00:39:28,646 --> 00:39:29,606
They don't have the ability.
499
00:39:30,526 --> 00:39:41,286
What they can do potentially is essentially offer up their land and power to somebody who has the capital who would want to build inference AI sites.
500
00:39:41,286 --> 00:39:45,306
because inference is just a different type of animal
501
00:39:45,306 --> 00:39:46,866
and it can be done.
502
00:39:47,506 --> 00:39:49,166
So you think even for small-scale miners,
503
00:39:49,326 --> 00:39:53,306
this optionality element will be important or relevant?
504
00:39:53,626 --> 00:39:55,926
I think you have to look at it
505
00:39:55,926 --> 00:39:57,266
like a real estate developer does
506
00:39:57,266 --> 00:39:58,626
and that's how we did.
507
00:39:58,726 --> 00:40:01,426
We have reorganized ourselves around asset management
508
00:40:01,426 --> 00:40:03,746
and every data center is an asset
509
00:40:03,746 --> 00:40:05,486
and how do you optimize the value of that asset?
510
00:40:05,586 --> 00:40:08,286
There's all sorts of things you can do with that asset, right?
511
00:40:08,366 --> 00:40:10,086
You can turn it into,
512
00:40:10,086 --> 00:40:11,846
You can run it as a Bitcoin mining site.
513
00:40:11,946 --> 00:40:13,586
You can run it as an HPC site.
514
00:40:13,706 --> 00:40:17,186
You could run it as an inference AI or traditional IT workload sites.
515
00:40:17,186 --> 00:40:21,986
It depends on where it's located, price of power, curtailment, et cetera.
516
00:40:22,506 --> 00:40:24,026
There are all sorts of things you can do.
517
00:40:24,606 --> 00:40:27,226
We have also always been a technology-first company.
518
00:40:27,646 --> 00:40:32,386
So we lead with the technology that enables us to do what we do.
519
00:40:33,026 --> 00:40:38,566
And so when we started running inference in Granbury, where we're running air-cooled inference
520
00:40:38,566 --> 00:40:41,606
in containers the same way we run Bitcoin mining.
521
00:40:42,266 --> 00:40:43,346
Actually, in Granberg today,
522
00:40:43,386 --> 00:40:44,826
we have more immersion Bitcoin mining
523
00:40:44,826 --> 00:40:45,826
than we do air-cooled.
524
00:40:47,546 --> 00:40:50,946
But if you can deploy inference AI
525
00:40:50,946 --> 00:40:54,226
like you deploy Bitcoin mining in containers,
526
00:40:54,446 --> 00:40:55,966
it means you can do it really fast.
527
00:40:56,906 --> 00:40:58,646
Because the biggest problem is
528
00:40:58,646 --> 00:41:00,446
if you're going to build a building,
529
00:41:01,006 --> 00:41:02,446
put network, put cooling,
530
00:41:02,446 --> 00:41:04,706
put racks to run training
531
00:41:04,706 --> 00:41:08,306
or inference the way most people do it,
532
00:41:08,566 --> 00:41:09,966
It's hugely expensive.
533
00:41:10,586 --> 00:41:21,066
Just the building, the refrigerator, if you would, is $12 million a megawatt versus $400,000 a megawatt for a bunch of containers that run Bitcoin mining.
534
00:41:21,066 --> 00:41:50,746
And so if you can not have to do that, and you can then focus your capital on buying the right GPUs or the right ASICs, and you have the right software stack above that that lets you monetize every compute cycle and optimize every compute cycle, then essentially a mining site that is a hybrid of mining and AI, you can do arbitrage on every electron.
535
00:41:50,746 --> 00:41:51,646
you're operating.
536
00:41:51,726 --> 00:41:52,166
You can say,
537
00:41:52,266 --> 00:41:53,146
hmm, okay,
538
00:41:53,766 --> 00:41:54,506
at this site,
539
00:41:54,586 --> 00:41:55,926
my power cost is maybe
540
00:41:55,926 --> 00:41:57,666
two cents a kilowatt hour
541
00:41:57,666 --> 00:41:59,226
for these hours of the day.
542
00:41:59,726 --> 00:42:00,266
I'm going to run
543
00:42:00,266 --> 00:42:01,526
Bitcoin mining all day long.
544
00:42:02,106 --> 00:42:03,486
And then when my price
545
00:42:03,486 --> 00:42:04,166
starts changing,
546
00:42:04,266 --> 00:42:04,706
I'm going to run
547
00:42:04,706 --> 00:42:05,706
more inference loads.
548
00:42:06,386 --> 00:42:07,186
You would say, though,
549
00:42:07,226 --> 00:42:09,006
even for small-scale miners,
550
00:42:09,406 --> 00:42:10,206
like acquisition
551
00:42:10,206 --> 00:42:11,446
of power generation
552
00:42:11,446 --> 00:42:12,626
will be important
553
00:42:12,626 --> 00:42:13,866
or maybe a differentiator
554
00:42:13,866 --> 00:42:14,186
between...
555
00:42:14,186 --> 00:42:15,486
It is the only thing
556
00:42:15,486 --> 00:42:16,426
that matters today
557
00:42:16,426 --> 00:42:17,686
is owning power.
558
00:42:19,066 --> 00:42:20,486
Jensen Wong said it
559
00:42:20,746 --> 00:42:32,306
Two or three days ago in an interview, he said, if you look at the pyramid of NVIDIA and AI in the world, you know, the U.S. is really strong on chips and data centers.
560
00:42:32,726 --> 00:42:36,286
What the U.S. sucks at is building a grid and power generation.
561
00:42:36,486 --> 00:42:38,466
And China is eating the U.S.'s lunch.
562
00:42:38,706 --> 00:42:38,986
Really?
563
00:42:39,326 --> 00:42:39,526
Yes.
564
00:42:39,626 --> 00:42:41,466
Why is that? I'm surprised to hear that.
565
00:42:41,466 --> 00:42:47,266
China has grown their national grid with such speed, like they do with many things.
566
00:42:47,366 --> 00:42:49,286
Right. Top down.
567
00:42:49,586 --> 00:42:50,566
It's top down, right?
568
00:42:50,566 --> 00:42:53,246
And in the U.S., it's very messy, right?
569
00:42:53,326 --> 00:43:01,366
And we've had a permitting regime, which has been very difficult to live in with the past administrations.
570
00:43:01,546 --> 00:43:02,126
Is that changing?
571
00:43:02,726 --> 00:43:04,266
They're trying to change it.
572
00:43:04,366 --> 00:43:13,826
You know, the FERC and the DOE, you know, it's difficult because when a new administration comes in, they put new people at the top.
573
00:43:13,826 --> 00:43:19,946
But all the people that actually are responsible for doing the work are career civil servants, as it's technically called.
574
00:43:20,566 --> 00:43:24,006
those people don't change from one administration to another.
575
00:43:24,906 --> 00:43:29,846
And so it's like saying to the people who have been responsible for reviewing your tax returns
576
00:43:29,846 --> 00:43:34,466
and say, well, now we want to be nice to taxpayers as opposed to really find every last penny.
577
00:43:34,606 --> 00:43:36,166
It's like, well, no, that's not my job.
578
00:43:36,306 --> 00:43:37,526
It's like tenured professors.
579
00:43:37,946 --> 00:43:38,126
Yes.
580
00:43:40,846 --> 00:43:41,966
Good example.
581
00:43:41,966 --> 00:43:46,306
And so it's very hard to get that change.
582
00:43:46,306 --> 00:43:54,186
And, you know, an administration is in for four years and hopefully eight years and then rinse, wash, repeat, new administration.
583
00:43:55,226 --> 00:44:06,186
This is where countries like Saudi Arabia, UAE, Singapore have huge advantages in that they have governments who are able to lay out very compelling visions.
584
00:44:06,186 --> 00:44:16,446
They have a lot of capital to deploy and they stick to their guns about what they're doing, which means stuff gets done and they're able to very rapidly grow.
585
00:44:16,446 --> 00:44:40,366
And I think having spent time in Saudi Arabia, as we've been setting up our regional headquarters there, and spent time with a lot of the big players there, I think you are going to see a huge explosion of AI data centers in Saudi Arabia because they have great access to internet speeds, low speed or rather high speed internet, low latency.
586
00:44:40,986 --> 00:44:44,526
They have low cost power and they have very low cost capital.
587
00:44:45,526 --> 00:45:01,886
It's interesting because I have heard that, you know, in the sort of vein of what advantages do miners have or more specifically Mara has, one of the things I heard through the grapevine somewhere was that the ability to get international power partnerships or power contracts is sort of unique.
588
00:45:01,886 --> 00:45:12,886
In the United States, most folks, certainly the PubCo miners and maybe other data center providers, generally speaking, in the U.S., are very domestically focused and not as great about international partnerships.
589
00:45:12,886 --> 00:45:13,006
Absolutely.
590
00:45:13,366 --> 00:45:15,586
Why does Mara have an advantage in that?
591
00:45:16,726 --> 00:45:19,966
Well, I mean, I've spent a good portion of my life living outside the U.S.
592
00:45:20,466 --> 00:45:20,866
Okay.
593
00:45:20,986 --> 00:45:22,106
I was born in France.
594
00:45:22,566 --> 00:45:23,406
Parents were Swedish.
595
00:45:23,766 --> 00:45:24,706
Went to high school in England.
596
00:45:26,046 --> 00:45:28,146
I've lived in many parts of the world.
597
00:45:28,566 --> 00:45:32,986
And I built technology businesses that have sold products and technologies all over the world.
598
00:45:33,246 --> 00:45:35,006
I ran a business.
599
00:45:35,026 --> 00:45:36,226
It's been a focus for Mara, basically.
600
00:45:36,386 --> 00:45:37,766
You've prioritized it in your way.
601
00:45:38,066 --> 00:45:38,766
From day one.
602
00:45:38,766 --> 00:45:44,006
And, you know, in 2023, we announced the deal with our partners in UAE.
603
00:45:44,006 --> 00:46:00,186
We went from signing a contract with our partners to being fully operational, two data centers, total of 250 megawatts of power, pulling 140 kilometers of high tension cables underground, raw land, etc.
604
00:46:00,946 --> 00:46:03,146
Twelve and a half months fully operational.
605
00:46:03,146 --> 00:46:10,106
So it sounds like you would agree or it sounds like part of your strategy from day one has been you have to have international partnerships.
606
00:46:10,426 --> 00:46:17,266
Otherwise, you're just sort of restrained to this United States situation, which is regulatorily more difficult.
607
00:46:17,706 --> 00:46:19,166
And just look at tariffs, right?
608
00:46:19,226 --> 00:46:25,226
If it's going to cost me 30 percent more to bring a miner into the U.S. than plugging it in somewhere else.
609
00:46:27,226 --> 00:46:28,446
That's it, basically.
610
00:46:28,446 --> 00:46:35,786
Right. So, you know, one of our stated goals publicly is that 50 percent of our revenue will come from outside the U.S. by 2028.
611
00:46:35,886 --> 00:46:39,746
We said that, you know, over a year ago and we're very focused on making that happen.
612
00:46:40,046 --> 00:46:46,746
Are there specific countries that you're more bullish on or specific regions that you're more bullish on in terms of creating those kinds of partnerships?
613
00:46:48,566 --> 00:46:50,686
So it is a long term game.
614
00:46:51,946 --> 00:46:56,826
Because you can't just waltz into Saudi Arabia and say, hey, I want to build a data center.
615
00:46:56,826 --> 00:46:57,226
Right.
616
00:46:58,446 --> 00:47:23,126
You know, get in line behind everybody else. So you have to go in in many of these places and first get the nod of approval from upstairs. And then that opens doors and you work your way kind of down through the hierarchy, while at the same time being able to work with a local partner to identify opportunities.
617
00:47:24,066 --> 00:47:30,226
Typically in a lot of countries, they want you to work with a local partner because they want to develop their own infrastructure.
618
00:47:30,566 --> 00:47:32,406
They don't just want you to come in and take.
619
00:47:32,926 --> 00:47:33,566
Makes sense.
620
00:47:33,566 --> 00:47:35,106
They want you to invest in the country.
621
00:47:35,206 --> 00:47:38,446
They want you to help build local knowledge, build jobs, et cetera.
622
00:47:39,066 --> 00:47:40,386
Europe is no different.
623
00:47:40,926 --> 00:47:44,746
It's just Europe has a more complicated regulatory regime.
624
00:47:45,266 --> 00:47:46,906
Did that come up in the Exion deal?
625
00:47:47,046 --> 00:47:49,826
I mean, was there like conversations at the government level?
626
00:47:49,826 --> 00:47:54,066
like we are going to be supportive of France in this situation?
627
00:47:54,306 --> 00:47:54,706
Absolutely.
628
00:47:54,966 --> 00:47:58,706
Again, we set up our international headquarters in Paris.
629
00:47:59,386 --> 00:48:00,666
There's a reason for that.
630
00:48:01,866 --> 00:48:06,926
And it's an important part of our strategy for international growth
631
00:48:06,926 --> 00:48:09,826
in both Europe and the Middle East,
632
00:48:10,006 --> 00:48:13,066
which we think are the two most exciting regions right now.
633
00:48:13,486 --> 00:48:18,206
Realizing that for us, we view Bitcoin mining in Europe
634
00:48:18,206 --> 00:48:20,146
as a load balancing tool for the grid.
635
00:48:21,286 --> 00:48:23,666
We don't view it as a place we can find cheap power.
636
00:48:24,506 --> 00:48:24,906
That makes sense.
637
00:48:24,986 --> 00:48:27,046
Right. We're essentially a service provider to the grid.
638
00:48:27,966 --> 00:48:31,626
However, we view inference AI data centers,
639
00:48:32,106 --> 00:48:35,026
private cloud as a huge business in Europe.
640
00:48:35,486 --> 00:48:39,186
Because, you know, in France, for example,
641
00:48:39,186 --> 00:48:42,606
Microsoft cannot offer sovereign cloud services
642
00:48:42,606 --> 00:48:44,586
to French government companies
643
00:48:44,586 --> 00:48:46,566
because of the U.S. Cloud Act,
644
00:48:46,566 --> 00:48:55,586
which allows U.S. law enforcement to reach into a Microsoft customer's data in France and access it.
645
00:48:55,926 --> 00:48:56,166
Oh, wow.
646
00:48:56,666 --> 00:48:57,466
Same thing with Google.
647
00:48:57,746 --> 00:49:01,886
The only way to avoid that is to have a fully air-gapped solution sitting behind.
648
00:49:02,146 --> 00:49:03,866
You need sovereign enterprise solutions.
649
00:49:04,126 --> 00:49:12,286
So French company operating in France using French technology that the French government has approved.
650
00:49:12,286 --> 00:49:30,606
And obviously, the company that is operating the data centers that run the most critical assets for energy in France, which are the nuclear power plants, which is what Xion does on behalf of EDF, means that their technology has already been approved, vetted.
651
00:49:30,606 --> 00:49:40,186
and, you know, you're able to go in and talk to, you know, virtually any other French company
652
00:49:40,186 --> 00:49:43,406
and then by extension, any other EU company.
653
00:49:44,566 --> 00:49:46,666
And, you know, Exion has some other things that are really exciting.
654
00:49:46,846 --> 00:49:51,406
They are licensed under the French regulators as an issuer of tokenized securities.
655
00:49:51,406 --> 00:49:53,246
They issue tokenized bonds.
656
00:49:53,566 --> 00:49:56,166
They've issued stable coins on behalf of French banks.
657
00:49:56,726 --> 00:50:03,946
They have a whole suite of blockchain technologies that are things we haven't really talked a whole lot about that are really exciting.
658
00:50:03,946 --> 00:50:13,386
I believe they have sort of special licensing and special kind of regulatory permits, essentially, to engage in these behaviors, which is rare and unusual.
659
00:50:13,886 --> 00:50:14,686
Very difficult to get.
660
00:50:15,026 --> 00:50:15,866
Very difficult to get.
661
00:50:16,006 --> 00:50:21,306
So we think we have a strategic advantage in the scope of businesses that we can operate.
662
00:50:21,306 --> 00:50:26,186
and we think Xion is going to be very successful
663
00:50:26,186 --> 00:50:30,726
in the world of private cloud, secure data center technology.
664
00:50:31,986 --> 00:50:33,646
Xion won't operate Bitcoin mining.
665
00:50:33,846 --> 00:50:35,446
That's not the business there at all.
666
00:50:36,586 --> 00:50:38,866
That's something Mara will do separately in Europe.
667
00:50:39,406 --> 00:50:41,686
Do you have a point of view about RWAs generally
668
00:50:41,686 --> 00:50:43,726
or we'll see what happens?
669
00:50:43,726 --> 00:50:44,426
It was interesting.
670
00:50:44,586 --> 00:50:47,786
I was at FII, which is the Future Investment Institute
671
00:50:47,786 --> 00:50:50,946
in Saudi Arabia, and I was sitting four rows
672
00:50:50,946 --> 00:50:55,546
from the round table where Jamie Dimon was answering that question
673
00:50:55,546 --> 00:50:57,466
and Larry Fink was answering that question.
674
00:50:57,766 --> 00:51:00,846
And, you know, they all say RWAs are the big thing.
675
00:51:01,386 --> 00:51:01,406
Really?
676
00:51:02,046 --> 00:51:04,066
Matter of fact, Larry Fink was,
677
00:51:04,666 --> 00:51:07,166
there was an article in The Economist a few weeks ago.
678
00:51:07,826 --> 00:51:11,606
Where, you know, I think it's the next big thing.
679
00:51:11,606 --> 00:51:18,686
And by the way, DTCC was just given a no action letter
680
00:51:18,686 --> 00:51:20,766
to be able to do tokenized securities.
681
00:51:20,946 --> 00:51:31,646
So if you think of how the securities world works today, even though you trade on the NASDAQ, you're actually not trading on the NASDAQ.
682
00:51:31,706 --> 00:51:32,946
You're trading through a broker-dealer.
683
00:51:33,446 --> 00:51:36,606
That broker-dealer then executes the trade on the NASDAQ for you.
684
00:51:37,586 --> 00:51:39,626
You don't get physical shares anymore.
685
00:51:39,946 --> 00:51:42,826
They're held in street names, it's called, by your broker-dealer.
686
00:51:44,506 --> 00:51:50,706
And while technically they're pieces of paper, it's a digital entry in DTCC's ledger.
687
00:51:50,946 --> 00:51:53,206
And it's fully custodial.
688
00:51:53,466 --> 00:51:53,786
Exactly.
689
00:51:54,466 --> 00:52:08,666
And so now if you're going to go issue real world assets or securities in a tokenized form, the same custodian, the same people that manage the shuffling of, okay, you just bought this stock and that person bought it.
690
00:52:08,726 --> 00:52:10,886
Now you have to move it from this ledger place to the other.
691
00:52:10,886 --> 00:52:14,146
The most logical thing is that they become tokenized.
692
00:52:14,286 --> 00:52:20,566
They tokenize all of this process so that it can operate friction-free 24-7.
693
00:52:20,566 --> 00:52:21,466
So much more efficient.
694
00:52:21,966 --> 00:52:28,666
And I think what you're going to see now is the stuff that Bitcoin and crypto has been doing, trading 24-7, 365.
695
00:52:29,106 --> 00:52:42,126
You will soon see the exchanges trading, maybe not 24-7, 365, but most probably 22-5, meaning Monday through Friday for 22 hours a day.
696
00:52:42,386 --> 00:52:46,806
And then they'll take the weekend kind of pause and then you'll get to six days a week.
697
00:52:47,526 --> 00:52:49,406
Kind of like the way the futures exchange works.
698
00:52:49,406 --> 00:52:50,446
Oh, wow. Okay.
699
00:52:50,566 --> 00:52:52,746
And then it'll eventually go to 24-7.
700
00:52:52,966 --> 00:52:56,006
It's all a question of all of the pipes and plumbing have to work.
701
00:52:56,106 --> 00:52:57,386
And that takes time to put in place.
702
00:52:57,926 --> 00:53:04,706
But five years from now, I think people will be, you know, the term banker's hours will be gone.
703
00:53:05,566 --> 00:53:07,126
Nine to five will no longer exist.
704
00:53:07,346 --> 00:53:08,386
It will all be crypto hours.
705
00:53:09,106 --> 00:53:11,066
People will be trading 24-7, 365.
706
00:53:11,806 --> 00:53:14,886
You'll have to have global operations, different time zones.
707
00:53:15,906 --> 00:53:20,346
And you'll see liquidity pools balancing out around the world.
708
00:53:20,346 --> 00:53:42,626
And it's going to be a super exciting time. You'll be able to move money from your bank account onto an exchange instantaneously by using a stable coin of some sort, either the one issued by your bank or one issued by a national government or one issued by a third party provider. That'll go on and off exchanges. Money will move at the speed of light. And I think it's going to be a very exciting time.
709
00:53:42,626 --> 00:53:53,166
And, you know, Bitcoin plays a big part in that as one of the most sought after assets for holding your store of value in turbulent times.
710
00:53:53,386 --> 00:54:00,686
I'm curious, while we're on the topic of geopolitics also, if you could sort of touch a little bit on kind of China and sort of China's role.
711
00:54:00,806 --> 00:54:05,066
And is there a sort of mining war happening between China and the U.S.?
712
00:54:05,066 --> 00:54:07,406
Is that overblown?
713
00:54:07,406 --> 00:54:10,966
I mean, it leaves a mining war going between China and the U.S.
714
00:54:10,966 --> 00:54:17,786
Because if anything, the Chinese government at the highest levels is cracking down a little bit on Bitcoin again.
715
00:54:19,006 --> 00:54:27,566
And that's most likely because, and this is just purely my inference, not that I have real data here.
716
00:54:28,686 --> 00:54:39,326
But more than likely, certain members of the government have used Bitcoin in a way that they shouldn't have.
717
00:54:39,326 --> 00:54:46,646
And as you know, Xi Jinping has been going through a big anti-corruption drive to stamp out corruption.
718
00:54:47,366 --> 00:54:52,146
And, you know, there's a ton of Bitcoin in China, a lot of Chinese holders of Bitcoin.
719
00:54:52,586 --> 00:54:54,186
China has allowed people to own Bitcoin.
720
00:54:54,306 --> 00:54:59,166
They just can't use it to buy goods and services purely as an investment.
721
00:55:01,226 --> 00:55:02,106
Similar to the U.S.
722
00:55:02,306 --> 00:55:02,506
Right.
723
00:55:02,586 --> 00:55:05,626
And I think now you're starting to see a crackdown happen.
724
00:55:05,626 --> 00:55:11,926
That being said, major cities that have all this ample power have opened up to Bitcoin miners.
725
00:55:12,546 --> 00:55:12,646
Oh.
726
00:55:13,126 --> 00:55:19,886
And so there's a particular region in China, whose name I forget for the moment, that has opened up.
727
00:55:19,926 --> 00:55:25,926
And China is now, yet again, the third largest site for Bitcoin mining in the world.
728
00:55:26,266 --> 00:55:27,286
And Bitmain is.
729
00:55:27,886 --> 00:55:30,926
And Bitmain, Canaan, and MicroBT are all there.
730
00:55:31,306 --> 00:55:32,826
What are the implications of that?
731
00:55:32,826 --> 00:55:38,546
I mean, so I know that there's sort of you mentioned open source as sort of a movement that's developing earlier.
732
00:55:39,206 --> 00:55:45,906
I think that there's been a lot of discussion about that, especially amongst kind of the smaller scale kind of pleb solo miners that I know.
733
00:55:46,006 --> 00:55:51,966
There's a lot of concern about kind of the amount of control that Bitmain has over production of ASICs.
734
00:55:52,366 --> 00:55:55,366
Do you see that evolving or changing?
735
00:55:55,806 --> 00:56:00,866
I know Jack Dorsey is working on ASIC miners, open source, firmware, everything.
736
00:56:00,866 --> 00:56:03,126
I mean, how do you see that story evolving?
737
00:56:03,386 --> 00:56:12,546
Well, in 2021, we came up with the idea to co-found a U.S. Bitcoin mining ASIC manufacturer who today is called Oradyne.
738
00:56:13,146 --> 00:56:15,486
So we co-funded that company.
739
00:56:15,626 --> 00:56:26,526
We seed funded it, brought in Mayfield, brought in engineering talent from Palo Alto Networks, which is one of the top ASIC designer teams in the U.S.
740
00:56:26,526 --> 00:56:27,886
So they're making chips, actual chips.
741
00:56:27,926 --> 00:56:28,786
They make ASICs, yeah.
742
00:56:28,786 --> 00:56:29,386
Oh, wow.
743
00:56:29,386 --> 00:56:37,446
So, Oradyne, their latest hydro miner was tested recently and performs just as good as the best bit made.
744
00:56:37,546 --> 00:56:38,746
So, they make full-blown miners.
745
00:56:39,406 --> 00:56:42,226
They also will sell chips if you want to assemble your own miner.
746
00:56:42,486 --> 00:56:44,506
How is their production going right now?
747
00:56:44,646 --> 00:56:45,946
It's continually increasing.
748
00:56:46,206 --> 00:56:46,486
Okay.
749
00:56:46,886 --> 00:56:49,906
And so, we're obviously a very large buyer of their systems.
750
00:56:51,366 --> 00:56:58,886
You know, the great thing about their miners is the, one of the biggest problems with the Chinese miners is they're designed as shoebox miners.
751
00:56:58,886 --> 00:57:00,466
that are not designed for industrial mining,
752
00:57:01,326 --> 00:57:02,926
meaning large-scale,
753
00:57:03,506 --> 00:57:07,146
able to control them,
754
00:57:07,266 --> 00:57:08,906
to be responsive to grid signals,
755
00:57:09,106 --> 00:57:09,926
all that sort of stuff,
756
00:57:10,126 --> 00:57:12,786
which is something we wanted.
757
00:57:12,886 --> 00:57:14,866
We wanted them built directly for immersion,
758
00:57:15,066 --> 00:57:17,686
built directly for large-scale remote,
759
00:57:18,626 --> 00:57:21,026
be able to essentially task them to operate
760
00:57:21,026 --> 00:57:23,146
based on power price, global hash rate,
761
00:57:23,226 --> 00:57:23,946
and Bitcoin price.
762
00:57:24,746 --> 00:57:26,006
And this involves, did Mera,
763
00:57:26,006 --> 00:57:31,746
I don't, again, I don't remember if I'm hearing this correctly, but does Mera also operate its own firmware or?
764
00:57:31,946 --> 00:57:32,126
Yes.
765
00:57:32,266 --> 00:57:32,686
It does.
766
00:57:32,826 --> 00:57:32,926
Yeah.
767
00:57:33,026 --> 00:57:33,306
Okay.
768
00:57:33,606 --> 00:57:33,766
Yeah.
769
00:57:33,846 --> 00:57:39,886
I mean, we today, to be fair, use a portfolio of firmware solutions.
770
00:57:40,466 --> 00:57:42,226
A, depending on the machine we're running.
771
00:57:42,406 --> 00:57:44,786
So in the Orodine machines, we run Orodine's firmware.
772
00:57:45,886 --> 00:57:49,586
We run third-party software in any Chinese miner.
773
00:57:49,726 --> 00:57:50,306
We won't run.
774
00:57:50,666 --> 00:57:55,626
So it's either ours or a derivative product of somebody else's.
775
00:57:55,626 --> 00:57:59,786
But we won't run factory original firmware in any miner we operate today.
776
00:57:59,786 --> 00:58:00,286
Okay.
777
00:58:00,866 --> 00:58:09,806
Reason being, we don't want there being any chance for somebody else to, in any way, take control of our miners.
778
00:58:09,946 --> 00:58:10,166
Huh.
779
00:58:10,446 --> 00:58:23,826
The biggest risk to national security in this country is Chinese-owned, Chinese-operated miners that could all of a sudden turn off or turn on and essentially take down the grid.
780
00:58:23,826 --> 00:58:31,866
bitmain has huge capacity in the u.s today through cango other companies they control
781
00:58:31,866 --> 00:58:37,626
it's probably what 80 percent of miners in the united states are bitmain well are bitmain
782
00:58:37,626 --> 00:58:46,386
manufactured but bitmain the company operates hosting or has their miners hosted on their behalf
783
00:58:46,386 --> 00:58:48,526
by many companies.
784
00:58:49,626 --> 00:58:49,886
Interesting.
785
00:58:50,346 --> 00:58:50,606
Okay.
786
00:58:50,726 --> 00:58:51,826
So there's levels to this.
787
00:58:52,066 --> 00:58:52,266
Right.
788
00:58:52,526 --> 00:58:54,286
And so, you know,
789
00:58:54,326 --> 00:59:04,839
imagine if they took a 250 megawatt Bitcoin mining site and just shut it off and the grid operator had a direct You would take out transformers It would cause all sorts of havoc
790
00:59:05,299 --> 00:59:10,639
So there's no Bitcoin mining war happening with China, but we are we want to make sure we're not vulnerable.
791
00:59:11,319 --> 00:59:11,619
Exactly.
792
00:59:11,899 --> 00:59:19,059
So, you know, definitely nobody should be running original, you know, OEM original software.
793
00:59:19,699 --> 00:59:20,299
Replace it.
794
00:59:20,299 --> 00:59:33,319
Which is challenging because a lot of, I mean, again, especially these small-scale miners who, you know, I'm hopefully going to be reaching today, they're sort of dependent on BitMean firmware at the moment.
795
00:59:33,319 --> 00:59:51,059
You know, I think Jack Dorsey's intention is very much to be able to provide those small scale miners with a solution to their problem, which I think is great because, you know, we're a large industrial miner.
796
00:59:51,499 --> 00:59:54,899
And so our focus is on doing things that are very big scale.
797
00:59:54,899 --> 01:00:04,399
And his company is very good at doing things for, you know, large volume of customers that have small volumes themselves, right?
798
01:00:04,439 --> 01:00:09,679
You look at, you know, Square, you look at all those things, you know, designed for mom and pop stores and boutiques.
799
01:00:09,699 --> 01:00:10,299
This is his mission.
800
01:00:10,479 --> 01:00:11,399
It's his mission, right?
801
01:00:11,759 --> 01:00:14,539
And so I think it's a very honorable mission.
802
01:00:14,839 --> 01:00:16,519
I give him kudos for doing it.
803
01:00:18,759 --> 01:00:23,379
But I think over time, Bitcoin mining is just going to continue to industrialize.
804
01:00:23,379 --> 01:00:25,599
and centralize.
805
01:00:25,859 --> 01:00:27,959
I had hopes a few years ago
806
01:00:27,959 --> 01:00:29,599
that you would see Bitcoin mining
807
01:00:29,599 --> 01:00:31,579
embedded in any electrical system
808
01:00:31,579 --> 01:00:33,779
as a way to use excess electricity.
809
01:00:35,599 --> 01:00:37,479
Unfortunately, I think
810
01:00:37,479 --> 01:00:42,619
it's a little bit like the dream
811
01:00:42,619 --> 01:00:44,979
of building load response,
812
01:00:45,139 --> 01:00:46,139
demand response systems
813
01:00:46,139 --> 01:00:48,419
into everybody's home electrical systems
814
01:00:48,419 --> 01:00:49,459
so that the power utility
815
01:00:49,459 --> 01:00:50,439
could turn on and off your,
816
01:00:50,599 --> 01:00:52,559
regulate your air conditioning
817
01:00:52,559 --> 01:00:59,839
to save power for the grid, these systems all end up resulting in some form of centralized control.
818
01:01:00,939 --> 01:01:09,319
And so I think, unfortunately, as mining gets more difficult, as the capital around it gets
819
01:01:09,319 --> 01:01:15,279
more intense, as difficulty rates keep going up, it's going to continue to centralize
820
01:01:15,279 --> 01:01:21,859
around people who have a vested interest in controlling block space, which is owning enough
821
01:01:21,859 --> 01:01:30,599
blocks out of everyday's additions to the Bitcoin blockchain that they can ensure
822
01:01:30,599 --> 01:01:33,759
the ability to verify transactions.
823
01:01:34,299 --> 01:01:39,779
How risky do you think miner centralization in general is? I mean, how are you concerned about
824
01:01:39,779 --> 01:01:44,579
that? Should we as Bitcoiners be concerned about that right now? And is kind of the miner
825
01:01:44,579 --> 01:01:50,419
centralization, just centralization of hash power going to pose a meaningfully different risk than,
826
01:01:50,419 --> 01:01:52,419
you know, just a bunch of pools, basically.
827
01:01:53,399 --> 01:01:57,079
Well, I mean, you've got a couple of issues.
828
01:01:57,239 --> 01:02:02,719
One is you have a lot of mining happening in the U.S.,
829
01:02:02,719 --> 01:02:07,019
but most probably a third of mining hash rate operating in the U.S. is pointing at China.
830
01:02:08,019 --> 01:02:10,539
Again, Bitmain controls a lot of hash rate.
831
01:02:11,659 --> 01:02:14,399
Antpool has a lot of hash rate in the U.S.
832
01:02:14,519 --> 01:02:17,599
So this is Trump's, and this is one of Trump's initiatives,
833
01:02:17,599 --> 01:02:20,479
is we want more hash pointing towards U.S. pools specifically.
834
01:02:20,479 --> 01:02:20,659
Pools, correct.
835
01:02:21,619 --> 01:02:25,119
And, you know, Mara Pool only mines for Mara,
836
01:02:25,179 --> 01:02:27,499
so we don't have any third-party customers,
837
01:02:27,939 --> 01:02:29,299
but Foundry certainly does.
838
01:02:31,059 --> 01:02:34,019
And, you know, there's a belief amongst the U.S. government members,
839
01:02:34,719 --> 01:02:35,399
the U.S. government,
840
01:02:35,659 --> 01:02:39,639
that mining hash rate that is not pointing at a U.S. pool
841
01:02:39,639 --> 01:02:41,079
is not U.S. mining hash rate.
842
01:02:41,879 --> 01:02:42,319
Fair.
843
01:02:42,859 --> 01:02:43,079
Right?
844
01:02:43,419 --> 01:02:46,279
And, you know, by the same reason the U.S. government
845
01:02:46,279 --> 01:02:54,679
wants to have a Bitcoin strategic reserve as a way to be able to defend Bitcoin, just like the
846
01:02:54,679 --> 01:03:00,199
gold reserve is there to defend the U.S. dollar and the oil petroleum reserve is there to defend
847
01:03:00,199 --> 01:03:07,879
the price of oil, which is a carryover from the old legacy of the oil crisis in the 70s.
848
01:03:09,899 --> 01:03:15,379
The U.S. needs a hash rate reserve. Think of it that way. They need enough hash rate in the U.S.
849
01:03:15,379 --> 01:03:19,599
such that if there were an attempt at a 51% attack on Bitcoin,
850
01:03:20,039 --> 01:03:22,059
the U.S. would have enough hash rate to stop it,
851
01:03:22,139 --> 01:03:27,659
which is why the stated goal is 60% of global hash rate in the U.S.
852
01:03:27,659 --> 01:03:33,519
But again, that is totally for naught if, you know, half of that is pointed at China.
853
01:03:33,699 --> 01:03:34,999
Yeah, it doesn't matter. It becomes irrelevant.
854
01:03:35,379 --> 01:03:41,439
Right. So, you know, I think the Trump administration has done a lot of great things for Bitcoin,
855
01:03:41,739 --> 01:03:43,879
obviously cleared a lot of regulatory paths.
856
01:03:45,379 --> 01:03:51,159
opened it up for trad find some great ways but when you look at the centralization around pools
857
01:03:51,159 --> 01:03:59,459
um i don't foresee a u.s pool doing a 51 attack uh the only other risk you could look at is
858
01:03:59,459 --> 01:04:06,059
censorship and you know any pool of any size has the ability to censor transactions if they wanted
859
01:04:06,059 --> 01:04:12,719
to um but then you know everything we do today is censored anyway you know whether it's social
860
01:04:12,719 --> 01:04:19,679
media, whether it's, you know, whatever you do. It's tricky. Why did Mara decide to run its own
861
01:04:19,679 --> 01:04:23,479
pool out of curiosity? And what are the benefits of running your own pool versus, you know,
862
01:04:23,479 --> 01:04:29,099
your competitors who I think probably are using other pools? Yeah. So the benefit of using a pool,
863
01:04:29,099 --> 01:04:37,179
period, is that you can essentially even out the luck factor, right? So in any given month,
864
01:04:37,179 --> 01:04:40,319
you may be 10% above your statistical payout
865
01:04:40,319 --> 01:04:42,639
or 10% below or somewhere, wherever.
866
01:04:43,119 --> 01:04:46,799
The fewer, the lesser amount of hash rate you have,
867
01:04:47,459 --> 01:04:49,539
the greater your luck variation is.
868
01:04:49,659 --> 01:04:52,059
So that sole miner who has one miner.
869
01:04:53,719 --> 01:04:55,039
One in a million chance.
870
01:04:55,179 --> 01:04:57,539
Has no block, no block, no block, no block.
871
01:04:57,839 --> 01:05:01,079
Maybe they hit a block and they're that unicorn, right?
872
01:05:01,479 --> 01:05:04,279
So it's better for them just to point their hash rate at a pool
873
01:05:04,279 --> 01:05:06,439
where they can get a pro rata fee
874
01:05:06,439 --> 01:05:08,339
of whatever that pool's total winnings are.
875
01:05:09,339 --> 01:05:11,559
But when you get to a certain amount of hash rate,
876
01:05:12,679 --> 01:05:14,339
why do you want to pay somebody else
877
01:05:14,339 --> 01:05:16,739
to level out your earnings?
878
01:05:17,419 --> 01:05:19,959
Why don't you want to keep that 1% or 2% yourself?
879
01:05:20,159 --> 01:05:21,259
Yeah, it makes sense to me.
880
01:05:21,379 --> 01:05:21,559
Right?
881
01:05:22,139 --> 01:05:24,799
And so that was the initial premise for it.
882
01:05:24,859 --> 01:05:27,279
Plus, we believed that
883
01:05:27,279 --> 01:05:29,699
because we were predominantly acid light
884
01:05:29,699 --> 01:05:30,219
in the beginning,
885
01:05:30,659 --> 01:05:33,079
we wanted to ensure that nobody
886
01:05:33,079 --> 01:05:34,579
was stealing hash rate from us.
887
01:05:34,979 --> 01:05:36,359
There's no way to do that when it's
888
01:05:36,439 --> 01:05:38,279
your hash rate's being pointed somewhere else.
889
01:05:38,379 --> 01:05:40,239
Again, it's the risks of custodial.
890
01:05:41,279 --> 01:05:43,639
Essentially, a pool is a custodian of hash.
891
01:05:43,819 --> 01:05:43,979
Yeah.
892
01:05:44,139 --> 01:05:44,459
Yeah.
893
01:05:44,599 --> 01:05:47,739
And so that was another reason.
894
01:05:47,879 --> 01:05:52,019
And then lastly, when you run your own pool,
895
01:05:53,299 --> 01:05:55,559
a third-party pool has to deal with everything
896
01:05:55,559 --> 01:05:58,119
from old S9s to the latest machines.
897
01:05:58,119 --> 01:05:59,899
It has to deal with network latencies
898
01:05:59,899 --> 01:06:02,759
that are really low to really high.
899
01:06:02,999 --> 01:06:04,779
And so it has to balance a lot of stuff out.
900
01:06:04,779 --> 01:06:11,299
and so there's a bit of overhead to running a third-party pool relative to how that pool
901
01:06:11,299 --> 01:06:18,959
performs if you operate your own pool and it's tied directly to the firmware running in your
902
01:06:18,959 --> 01:06:24,699
miners you strip out a lot of latency you strip out a lot of issues and you can gain a slight
903
01:06:24,699 --> 01:06:31,019
little advantage and so i think over you know if you look back historically mara pool has
904
01:06:31,019 --> 01:06:38,099
operated a slightly positive to statistical norm advantage.
905
01:06:38,539 --> 01:06:42,579
And whether that's just luck or whether that is the fact that, you know,
906
01:06:42,619 --> 01:06:48,499
we've been able to extract out some of the fat and latencies,
907
01:06:48,839 --> 01:06:52,499
but it's been an advantage for us and we continue to operate it.
908
01:06:52,659 --> 01:06:56,999
And another kind of more unusual advantage maybe is the ability to run something like Slipstream,
909
01:06:57,419 --> 01:06:59,199
which generates additional fees.
910
01:06:59,559 --> 01:07:00,819
So shout out to Slipstream.
911
01:07:01,019 --> 01:07:10,579
Yeah, and the ability to submit blocks that are, you know, use different templates, et cetera, is really critical.
912
01:07:12,219 --> 01:07:16,919
And we've been very happy with what Slipstream has enabled us to do.
913
01:07:17,299 --> 01:07:20,039
Okay, so the last topic that I wanted to make sure that I brought up with you,
914
01:07:20,039 --> 01:07:25,819
I actually polled a bunch of my, I would say, mining enthusiast friends who run smaller scale miners,
915
01:07:25,819 --> 01:07:31,499
pleb miners, so to speak, solo miners, and a topic that they almost all wanted me to make
916
01:07:31,499 --> 01:07:36,499
sure that I brought up with you is heat reuse. This is very sort of exciting theme, I think,
917
01:07:36,539 --> 01:07:41,299
in that community. Finland, obviously, the operation in Finland, the heat reuse district
918
01:07:41,299 --> 01:07:45,839
heating project for Mera in Finland. I was wondering if you could share a little bit about
919
01:07:45,839 --> 01:07:52,799
that. Sure. So in Northern Europe, it's very common to have district heating systems. This was
920
01:07:52,799 --> 01:07:54,339
something that was developed.
921
01:07:56,299 --> 01:07:57,499
And you're Swedish, you said.
922
01:07:57,739 --> 01:07:58,019
I'm Swedish, yes.
923
01:07:58,199 --> 01:08:01,639
So in Sweden, they've, as long as I've known,
924
01:08:02,239 --> 01:08:04,239
so since the 60s most probably,
925
01:08:04,419 --> 01:08:05,439
have had district heating.
926
01:08:05,599 --> 01:08:07,799
So where essentially you have a power plant
927
01:08:07,799 --> 01:08:08,999
that generates heat, right?
928
01:08:09,099 --> 01:08:12,239
Power plant, oils, water, turns into steam,
929
01:08:12,559 --> 01:08:14,159
uses steam to power a turbine
930
01:08:14,159 --> 01:08:15,459
that generates electricity.
931
01:08:15,459 --> 01:08:16,999
Well, that steam has to be cooled
932
01:08:16,999 --> 01:08:18,039
back into water somehow.
933
01:08:19,319 --> 01:08:21,359
And so what they did was
934
01:08:21,359 --> 01:08:27,939
They started piping it around the city and then superheated steam, essentially.
935
01:08:28,239 --> 01:08:31,099
And now you didn't have to have boilers in buildings.
936
01:08:31,139 --> 01:08:34,119
And so the pollution numbers went down and the air became cleaner.
937
01:08:35,239 --> 01:08:37,519
And so Finland has done this as well.
938
01:08:38,299 --> 01:08:45,819
And in Finland specifically, one of the challenges they had is, you know, they have ample amounts of fuels that are wood pellet based.
939
01:08:45,819 --> 01:08:51,759
And so they had wood pellet plants heating water to put in these systems.
940
01:08:52,399 --> 01:08:56,299
So while wood pellets are better than some other fuel sources, it's not the best.
941
01:08:57,179 --> 01:09:09,159
And so we approached them through a third party where there was an opportunity to essentially replace this wood pellet burning plant with a data center.
942
01:09:09,799 --> 01:09:11,199
So this was your idea.
943
01:09:11,299 --> 01:09:12,259
You brought this to them.
944
01:09:12,259 --> 01:09:37,699
Well, to be fair, it was our Finnish partners who brought the opportunity to us. We had been talking for a long time about, you know, heating water and doing shrimp farming and all sorts of other things, aquaponics. And the problem is, if you want to heat greenhouses, well, just a handful of S19s can heat a whole big greenhouse. So it's not good from a large data center perspective.
945
01:09:37,699 --> 01:09:42,539
Okay, so this is a smaller scale operation. How many megawatts is the Finland situation?
946
01:09:42,839 --> 01:09:43,919
It's, I think, nine megawatts.
947
01:09:43,999 --> 01:09:44,219
Okay.
948
01:09:44,359 --> 01:09:45,899
And it heats 80,000 homes.
949
01:09:46,619 --> 01:09:52,019
Wow. Nine megawatts heats 80,000 homes? That's wild.
950
01:09:52,279 --> 01:10:02,319
Yeah. So our intention is to look at using, if you look at the world of inference AI we were talking about, right?
951
01:10:02,319 --> 01:10:08,559
um so think imagine an office building you build a 50-story office building somewhere
952
01:10:08,559 --> 01:10:16,639
you clad that building in pv meaning essentially solar cells uh you now have solar films that
953
01:10:16,639 --> 01:10:23,319
literally like the film you can put on a car um and now that building generates its own electricity
954
01:10:23,319 --> 01:10:28,659
you put a data center in the basement of that building you use the heat from that data center
955
01:10:28,659 --> 01:10:31,879
to heat the hot water, heat the actual building in the winter.
956
01:10:33,539 --> 01:10:40,279
And now that data center has capacity for private cloud and AI in the building,
957
01:10:40,379 --> 01:10:42,539
which means you don't need high-speed internet access
958
01:10:42,539 --> 01:10:44,719
because you're on a local area network in that building.
959
01:10:46,179 --> 01:10:49,379
This is something commercial real estate developers are looking at
960
01:10:49,379 --> 01:10:51,219
as a differentiator for their buildings.
961
01:10:51,579 --> 01:10:54,679
So people will potentially use heat from AI, from HPC,
962
01:10:54,779 --> 01:10:57,959
from high-performance compute for heating buildings.
963
01:10:57,959 --> 01:10:59,439
This is going to be a...
964
01:10:59,439 --> 01:10:59,779
Absolutely.
965
01:11:00,159 --> 01:11:17,299
And then any excess energy that you are not using for AI purposes, you can then choose to either sell into the local market, which from a regulatory perspective is pretty difficult in most places because most building owners don't want to be regulated utilities.
966
01:11:19,279 --> 01:11:22,739
But you could then just mine Bitcoin with the rest of the energy.
967
01:11:23,079 --> 01:11:25,339
And so now you'll have a little bit of Bitcoin mining.
968
01:11:25,339 --> 01:11:29,239
you'll have some AI, private cloud, batteries, and solar cells.
969
01:11:29,259 --> 01:11:31,979
Is this an argument perhaps for, you know,
970
01:11:32,539 --> 01:11:34,719
is this like a way that people make the argument that,
971
01:11:34,859 --> 01:11:36,599
yes, okay, energy demands are going to go up,
972
01:11:36,639 --> 01:11:39,539
but maybe there will also be areas where energy can be saved,
973
01:11:39,619 --> 01:11:43,059
we'll save on heating, and, you know, we'll be reusing energy, etc.
974
01:11:43,319 --> 01:11:43,979
I mean, is there...
975
01:11:43,979 --> 01:11:45,339
Yeah, so the...
976
01:11:45,339 --> 01:11:50,379
If you go to 50,000 feet and you look at the energy problem today,
977
01:11:50,539 --> 01:11:54,799
it's that the consumer of energy is not near where the energy is being generated.
978
01:11:55,339 --> 01:12:07,099
You have a wind farm in West Texas, or rather you have hundreds of wind farms in West Texas, and you have millions of people in Houston, Dallas, Austin, which is very far away from West Texas.
979
01:12:07,839 --> 01:12:10,679
And so you need to use transmission lines to get the energy to them.
980
01:12:11,339 --> 01:12:13,339
Those transmission lines are bottlenecks.
981
01:12:13,679 --> 01:12:14,919
It's like plumbing in your house.
982
01:12:15,579 --> 01:12:22,679
You can't double the amount of water you put through the plumbing in your house because the pipes just don't have the capacity.
983
01:12:22,679 --> 01:12:30,279
And so you start by moving demand data centers to where the power is being generated.
984
01:12:30,919 --> 01:12:40,319
And that way you can use up all that energy that's not being used, that's not being sold into the market or can't get transported on these constrained transmission lines.
985
01:12:40,799 --> 01:12:49,439
And then you start moving to microgrids and metrogrids where you generate power in a local area and only consume it in the local area.
986
01:12:49,439 --> 01:12:59,179
And you use the grid simply to load balance so that that microgrid or metrogrid can sell energy back into a pool or can borrow when needed.
987
01:13:00,139 --> 01:13:11,799
And so it's much easier to build a microgrid with batteries, solar cells, for example, and operate at a microgrid scale.
988
01:13:11,799 --> 01:13:26,899
So a single house, community, an industrial park, a small city, than it is to go build a big nuclear power plant somewhere or build a big gas plant and then build all this transmission capacity.
989
01:13:27,139 --> 01:13:28,319
Takes 20 years to do that.
990
01:13:28,959 --> 01:13:32,359
But you could build a microgrid in a year or two.
991
01:13:33,019 --> 01:13:39,459
Is it, again, the power companies exclusively that are going to be the main beneficiaries of this entire transition?
992
01:13:39,459 --> 01:13:45,599
Or is there a universe where like Con Edison, you know, is mining Bitcoin and making additional revenue?
993
01:13:46,439 --> 01:13:51,639
So I think it's going to be, you know, different courses for different horses, as they say.
994
01:13:53,459 --> 01:13:56,399
The U.S. has multiple grid operators all with their own.
995
01:13:56,559 --> 01:13:58,339
You know, AirCard is different than PJM, etc.
996
01:13:59,979 --> 01:14:03,759
Different parts of the country have different problems infrastructurally with their grids.
997
01:14:04,439 --> 01:14:09,939
I think one thing you'll see in the very near term is large data center operators will become power utilities.
998
01:14:10,159 --> 01:14:12,319
They're all building their own generation capacity.
999
01:14:13,159 --> 01:14:19,699
PGM has an issue where you have a bunch of data centers who have backup power.
1000
01:14:19,939 --> 01:14:25,819
They noticed a drop in frequency on their grid, which is a way to see if you're about to have a power drop.
1001
01:14:26,459 --> 01:14:29,639
And seven data centers, I believe it was, went on backup power.
1002
01:14:29,639 --> 01:14:36,239
So now all of that capacity, all that demand that was sucking power off the grid went off, which creates a power spike on the grid.
1003
01:14:37,859 --> 01:14:44,039
And so now the grid operators are saying, no, no, no, wait, you can't decide when you go on and off backup power.
1004
01:14:44,619 --> 01:14:46,419
The grid needs to tell you when to do that.
1005
01:14:46,639 --> 01:14:52,739
So now you're getting this power play, which is only going to cause data centers to say, you know what, I'm not even going to be on the grid.
1006
01:14:52,899 --> 01:14:53,279
Thank you.
1007
01:14:53,359 --> 01:14:55,879
I'm going to operate power plant in my garage here.
1008
01:14:56,019 --> 01:14:57,859
I'm going to run my data center.
1009
01:14:57,859 --> 01:15:01,859
And, you know, if you want to buy energy from me, happy to sell it to you.
1010
01:15:02,119 --> 01:15:12,859
But that's actually a really interesting and important point that I hadn't considered was, do you see this whole revolution, this whole thing that we've been sort of talking about this entire interview as decentralizing?
1011
01:15:13,999 --> 01:15:14,379
Absolutely.
1012
01:15:14,379 --> 01:15:15,719
Who owns power, right?
1013
01:15:15,719 --> 01:15:28,639
Absolutely. I'm a huge believer in the fact that, and talk to anybody who really understands data center build out and where are the constraints, and they'll tell you it's energy.
1014
01:15:28,779 --> 01:15:37,639
There are data centers ready to be turned on that just don't have, I mean, the GPUs are plugged in, it's ready to go, they just don't have the power.
1015
01:15:38,199 --> 01:15:40,079
And they're waiting for the power to get turned on.
1016
01:15:40,079 --> 01:15:48,619
So if you have power and you have land, there are data center operators who are calling on you, more than likely.
1017
01:15:49,419 --> 01:15:54,699
And so if you are a power generator and you have land, you're even more attractive.
1018
01:15:55,079 --> 01:16:01,539
And so you may recall we recently announced a deal we're doing with MPLX in West Texas.
1019
01:16:02,379 --> 01:16:04,259
MPLX is a large gas refiner.
1020
01:16:04,879 --> 01:16:08,579
They have a gas pipeline in the Delaware Basin, which happens to be in Texas.
1021
01:16:10,079 --> 01:16:14,939
we are going to build three gas-fired power plants with them,
1022
01:16:15,139 --> 01:16:18,799
starting at 400 megawatts, growing to 1.5 gigawatts,
1023
01:16:18,999 --> 01:16:23,679
and build data centers around those gas-fired plants,
1024
01:16:23,679 --> 01:16:27,739
which will be completely self-sufficient and able to operate.
1025
01:16:27,959 --> 01:16:30,619
And we will be selling power back to the grid.
1026
01:16:31,759 --> 01:16:33,779
Are the hyperscalers getting into this game?
1027
01:16:34,019 --> 01:16:36,659
Or is this something that the miners have an advantage in?
1028
01:16:36,839 --> 01:16:39,239
Well, the hyperscalers are already in the game.
1029
01:16:39,679 --> 01:16:40,019
They are.
1030
01:16:40,079 --> 01:16:41,499
They're buying their own power.
1031
01:16:41,739 --> 01:16:53,119
Look at all these deals about Caruso ordering jet engines from Boom, the hypersonic jet company, to generate electricity.
1032
01:16:53,919 --> 01:16:55,919
Look at G. Vernova.
1033
01:16:56,399 --> 01:16:59,359
Look at their recent backlog of orders for turbines.
1034
01:16:59,599 --> 01:17:00,399
Who are the customers?
1035
01:17:01,599 --> 01:17:03,499
Meta, Google, Amazon.
1036
01:17:03,959 --> 01:17:07,259
They're all getting into the power generation game.
1037
01:17:07,259 --> 01:17:21,599
The problem is it's hugely capital intensive. Right. And so, you know, most Bitcoin miners don't understand the world of power generation. They don't know how to build, run and operate a power plant.
1038
01:17:21,659 --> 01:17:22,979
They're all stuck on PPAs.
1039
01:17:22,979 --> 01:17:31,319
they're mostly all stuck on ppas rides that way cypher iron clean spark etc you know we own our
1040
01:17:31,319 --> 01:17:36,259
own wind farms we generate electricity off of flare and we happen to have a whole team of people
1041
01:17:36,259 --> 01:17:40,919
who have done nothing but build and operate gas-fired power plants you know this year has
1042
01:17:40,919 --> 01:17:47,019
been a year where we have been quietly putting in place the pieces that will be very evident to the
1043
01:17:47,019 --> 01:17:54,819
marketplace in 26 regarding what it means when you own large amounts of power. You have land and
1044
01:17:54,819 --> 01:18:00,959
the ability to build a variety of types of data centers, and you're able to deploy and operate
1045
01:18:00,959 --> 01:18:06,619
tier three and tier four data centers for the most sensitive data types, for the most critical
1046
01:18:06,619 --> 01:18:10,759
customers that need to operate with full sovereignty. And I think what you'll find
1047
01:18:10,759 --> 01:18:16,859
as you look at Mara by the end of 26, it's a company that spans power or what we call
1048
01:18:16,859 --> 01:18:22,459
electrons all the way to actions, meaning insights. And it'll be very exciting.
1049
01:18:22,719 --> 01:18:26,539
Perfect. I think that's an amazing place to stop. Thank you so much, Fred. This was
1050
01:18:26,539 --> 01:18:33,919
really interesting chat and I'm excited to see what Mara does beyond just Exahash. Yeah. So thank you.
1051
01:18:33,919 --> 01:18:34,739
Thank you.