Hut 8 CEO on Strategy to Win Tech Trifecta of AI, Data Centers, and Bitcoin

The underlying thing that drives all those together is energy, and so the first layer of our business is energy. We're building a business, and call it a next-generation energy infrastructure platform at the intersection of power and technology. And so the second layer, in terms of how do we actually utilize that power, is our digital infrastructure layer. Right now we're building data centers and physical infrastructure. Then our third layer is compute, where we own ASICs for Bitcoin mining and we own GPUs on the AI side of the business. And so if we think about our core operating business, it's three layers: power, digital infrastructure, and compute. And then we have a fun fourth layer that's our operating layer, is more on our treasury management layer, which is a large Bitcoin position that we hold that is our strategic reserve.

So we do have power generation, but that's not the long-term thesis of the business. We own power generation when we need to, to power the underlying technologies, and I'll explain a little bit in what scenarios. But ultimately, our belief is as technology evolves and as it grows and advances, so too will its reliance on additional energy and consumption, both in terms of scale overall as an industry, but also in terms of each individual load becoming larger and larger. And so, understanding where can you place those loads on the grid, whether it be front of the meter connected to a transmission line or behind the meter at a power plant or power generation facility directly, that is what we believe is really important, because how does the consumption of power help progress our grid, our energy ecosystem, and humanity? The World Bank came out with a great graph recently, and it showed how GDP per capita in every country and energy consumption per capita were directly correlated. So the more power that you consume per capita, the higher your GDP is. And so instead of being afraid of additional energy consumption, how do we do so in a way that helps grow our energy infrastructure and our grid? And so by bringing technology and power together, we believe that we can unlock something really valuable.

I think the biggest bottleneck for AI today is power. There's chips from Nvidia, and that was a big bottleneck that was being discussed last year. Today is where do you actually place those chips in the data centers that can support it, and that's why companies who have access to power, who have platforms, who have the ability to procure power, have been in the headlines, right? And so for us, any technology that requires power, I think we're going to end up being in the headlines. So that was and continues to be Bitcoin mining, that's AI data centers. And so we think about the business a little differently. We're just building digital infrastructure that it could be a tier zero, tier one data center for ASIC compute that powers Bitcoin mining, or it can be a tier three data center for GPU compute that powers AI workloads.

No comment.

The biggest demand that we've seen from the AI data center — so giving a little bit of background on the evolution of AI data centers and data centers for AI compute — when you think about customers historically, if you were to have a customer that wanted 5, 10, 15, 50 megawatts of demand, that was a really, really big contract. When we're looking at opportunities today, we're talking multi-hundred-megawatt demand profiles going up to a gigawatt at a single campus, and that's because the more GPUs you have at a single location clustered together, the larger you can make that single computer to compute larger and larger LLMs for training for AI. And so the company that can actually support that type of demand profile, that type of investment, tends to focus around the hyperscalers that everyone knows about — the Microsofts, AWSs, Oracles, Amazons, Metas of the world — and also large-scale enterprise. We believe there's also going to be an increase in demand from enterprise consumers as their workloads increase, as their workload scale. And so we don't speculate on rumors that have gone out there to your question earlier, but we did file a permit in Louisiana, which kind of sparked all of these conversations regarding a project that we're working on.

Most definitely. When we think about overall power markets, what we look for is what power markets have the dynamics to be able to support a large-scale load, and where in their transmission systems could we actually place them, right? And so we spent a lot of time with our transmission engineers and our energy team in order to look at what regions do we think are friendly, and then also what regions do we think can actually support this loading during what duration, what time period. So Louisiana, for example, is a really interesting state in our opinion because you see demand going out of tier one markets like Virginia and Chicago and being willing to go to new markets. And Louisiana, earlier this year, the governor in the state just passed a data center tax exemption, and so that shows a willingness to bring in more data center customers and to build properties. In addition, we think that it sits in the middle of a couple of data center markets from a fiber perspective, is a large producer, obviously, of fuel in the United States, and so you have a lot of LNG facilities and fuel generation, so power can be generated there as well. And so we think it's a really interesting market and is one of the many markets that we're looking at as we look at these opportunities.

When we look at our business today, if all the power access we had was our existing facilities, we'd be speaking a lot more about the conversion of those facilities to AI. Why? Because on a net present value perspective of a long-term 15-year lease within a hyperscale tenant, you are able to drive a lot more value from that individual megawatt, and there's a lot of financing that will support the growth of those megawatts as well. And so for us, I think where we're uniquely positioned is we have a lot of confidence in our team and our power development pipeline, where the question we ask ourselves is: we have new sites that can be energized in a short period of time. So instead of cannibalizing our existing cash flow, why don't we drive the demand of AI to those new projects? And at a point where we don't believe we have enough new projects to fulfill that demand, we can look at conversion opportunities as well. Not all sites can be converted, but a lot of sites can, especially as customer profiles and demand profiles on redundancy and power availability changes and evolves over time. And so for us, the Bitcoin mining side of the business is really interesting. When we think about our businesses, we don't think about it as Bitcoin mining versus AI data centers. We think about it as digital infrastructure versus compute. And so digital infrastructure, all we're doing is building different types of data centers with different types of redundancy. I think it was hard for the market and for people to understand that a Bitcoin mining facility was actually a data center. And so we announced this new site that we're building called Vega. It's a 25-megawatt data center. It's built to host ASIC computing chips, which is for Bitcoin mining. The site is really interesting because it's actually behind the meter at a wind farm, but also fully interconnected with the grid as well. And so you have this dynamic where you have this data center in between all these wind turbines, interconnected with the grid. What's really interesting about that site, and I think is going to start changing the way people think about these types of data centers, is we're building it direct liquid cooled. And so similar to the AI data centers, it's not air-cooled anymore. In addition, we had the manufacturer Bitmain build a rack design ASIC chip, and so now when you walk in, it's going to be data center halls, and the actual servers will be like a server just like any other data center. So we're normalizing form factors as well. That data center is going to cool up to 200 kilowatts per rack. To remind everyone, the Blackwells are about 120 to 130 kilowatts per rack. These ASIC chips will be 180 kilowatts per rack, so we're adding buffer as well in our design. And that is where I think you'll start seeing people thinking about data centers as different types of redundancy from tier 0 to 1, 2, 3, rather than what the use case is, because the compute will be what drives the actual use case and the quality of the data center that you need to build from a reliability perspective. And I think there will be less conversation on what is the business, more so what type of data center are you building, who's a counterparty, what's a creditworthy profile, and what are they doing with the compute.