00:00Mara continues to shift its focus from Bitcoin mining to the unprecedented demand for compute
00:07needed to fuel AI. On Monday, the company said it sold one and a half billion dollars of Bitcoin
00:13to fund data center deals like the acquisition of Longridge Energy and Power in the great state
00:18of Ohio. And the CEO of Mara, Fred Thiel, joins us now. Fred, great to have you on the program.
00:24I'm really excited to ask you about this transition because if you're mining Bitcoin,
00:29you essentially need a ton of compute, a ton of power running rigs that hold microchips. And
00:35the same is true for AI. Can you just make a switch one to one from mining Bitcoin to providing
00:43AI?
00:44Well, it's a little bit different infrastructure, but the key gating item in the AI industry today
00:51is availability of power. And this affects NVIDIA. It affects the memory guys. If you don't have
00:57power where you can plug compute in, you don't have sales. And so everybody's chasing the limited
01:03amount of power that's available. We sit on 1.1 gigawatts of energized capacity. And so for an AI
01:09tenant to come in to one of our sites, it's a question of permitting, building the type of
01:14building they need, and then they bring in the compute and plug in. What that does is it dramatically
01:21shortens the lead time for them to get energized. We have the power available today. We have the land.
01:26They can come in and get it permanently, get it built. 18 to 24 months later, they are energized and
01:31turned on. And as you know, by what's in the news about Anthropic and some of these other model
01:36providers, their ability to grow market share and support their users' need for agents and tokens is
01:44totally driven by how much compute they have. And you have an exponential growth in the size of the
01:49compute needed to run the basic model. Plus the agents are using many fold more tokens than they
01:55used before when people were just chatting with AI. And so you just have an insatiable demand
02:00currently. And all this land and power that we sit on is just a very valuable asset. So we looked
02:06at
02:06this and said there was nothing better for us to do with our assets than do this conversion. It was
02:11the
02:11best thing for our shareholders. So Fred, just in terms of the properties that you already have,
02:16you know, leaving aside the acquisition or the further investment, you have essentially data
02:22centers that are set up with computers running ASICs, right, to mine Bitcoin. And you've got
02:27contracts with power providers that are live. Is the only thing you have to do essentially change out,
02:33you know, the chips or the servers in order to make them neocloud data centers?
02:40Not quite that simple. What we need to do is it depends on the who the tenant is. If it's
02:46somebody
02:46who's going to run training, you typically need to build a powered shell. So a building
02:51that's enclosed that has cooling and power infrastructure to support whatever compute
02:55they're going to put in it. If it's inference, as opposed to training, you can do modular,
03:00which is faster. And our sites are today set up for a conversion to modular that's very easy.
03:05If you're going to build a building, you have a construction timeframe that's a little bit longer.
03:09But essentially, the key thing for any of our prospective tenants, whether it's a hyperscaler or
03:14neocloud or whoever, is the power is available today. The land is available today. All they have
03:19to do is essentially sign the lease and start the process of construction.
03:23Fred, do you have anyone that's actually signed leases? Do you have any binding agreements with
03:28any of the hyperscalers?
03:31No, but we have very active engagements across 90% of our portfolio, which is great. We have
03:37a very competitive process going on at some of the sites where people are looking at capacity,
03:43how quickly they can get online. And our expectation, as we've said before, is we'll
03:47announce one or two leases by the end of this year.
03:49I guess, Fred, there is this fear, I mean, on this sort of idea specifically, that you build
03:53something without a lease. And with every CapEx cycle, we've seen this thing repeated, that it is
03:59very clear and real and fundamental that we don't have enough for AI data centers right now. But then
04:04a lot of times something happens that changes the scarcity economics, innovation, efficiency gains,
04:11the AI race, rather the CapEx race finally plays out. I wonder what is the risk that that happens,
04:17that you and some of your peers and, you know, once shoe companies that turn to be AI, well,
04:22what is the risk that something changes in this cycle and means that that scarcity economics isn't
04:27there anymore?
04:29Yeah, great question. A couple of things. For one thing, we don't build in advance of having a tenant
04:34signed. So we're not going to go and allocate capital to building infrastructure until we
04:40have a tenant signed and on the hook, so to say. So and those processes are fairly far advanced.
04:46So we're not going to go spend the CapEx like some of our peers have done in advance of having
04:50a tenant. Because the key thing is, you know, if Google's a tenant, they have a particular spec of
04:55what they want. Microsoft may have a different spec, AWS, Anthropic, whomever. They all have different
05:01specs for what they want. So it's kind of crazy for us to build something and then have them have
05:07to retrofit. So for us, it's all built to suit. So as soon as the tenant signs, then we go
05:12through
05:12the design process and then we build. And, you know, that's the most capital efficient way of
05:18doing it. And then as well, our partnership with Starwood gives us a huge advantage, we believe,
05:24over our peers in that we have a partner who has signed multiple leases with all the top tier
05:29providers. They have built in EPC and construction. They have built and delivered multiple data
05:35centers to these premier partners, and they know how to put them into operation. They know how to
05:40meet timelines. They know how to reduce risk. And then also, you know, they share in the equity
05:46contributions. And our sites are put into each of these joint ventures, if you would, at kind of
05:52predetermined values, which we think is just a very capital efficient way for us to proceed in this
05:56business.
Comments