All right. We're going to keep going here. Next we've got HIVE Digital. Got the CEO, Aydin Kilic. Aydin, good to see you. Great to be here. Thank you, Darren. I guess first topic, you've deliberately taken a capital light approach on the AI cloud side with lease-to-own for GPUs, third-party hosted deployments in some sites, and conversions of existing Bitcoin infrastructure rather than greenfield builds. Little bit of a different approach than other peers. A few questions to start. I guess, what's the strategic logic? I guess, are you explicitly giving up by not owning more of the infrastructure? Yeah. We have a near-term strategy to ramp revenue and EBITDA through our partnership with Bell Canada, largest telecom company in Canada. What that does, it gives us a near-term ramp to bring on 16 MW of critical IT load. Now, it's CapEx light for us. Bell is funding the construction of that data center. It is co-location, but with where we feel the stock is now at a $500 million market cap, really there's a lot of upside for re-rating, and at that point, as we layer on these long-term GPU contracts, our target for the end of the year is $200 million on GPU cloud. We're at $35 million now. We were actually at $20 million, then we just announced a deal for 2 years, $30 million total TCV, $15 million ARR. It went from CAD 20 million to CAD 35 million, now with this pipeline that we have with Bell, it gives us a line of sight to CAD 200 million ARR. We'll provide the market updates on different deals when we sign them. Just to give you example, we have a near-term target, or sorry, I would say a near-term path with completed data center infrastructure to bring online 4,000 GB200s or GB300s. Based on what we're seeing in the market right now for demand at the rates that these GPUs are renting for, I'm talking two-plus, three-plus year contracts, it's about CAD 60 million ARR per cluster of 2,000 GPUs. If we were to bring online 4,000 GB200s in the next, call it six months, that would add CAD 120 million ARR to our CAD 35 million, that would get us to CAD 155 million. That's when we get to that re-rating. This is, again, it's a CapEx-light ramp to near-term EBITDA and revenue growth. We finance the GPUs, by the way, with our OEM, single-digit financing, CAD 1 residual buyout, that's very good as well. We think that once we break through CAD 100 million HPC revenue ARR, that'll be a good catalyst for the stock to re-rate. Yeah. I guess your point about lease-to-own GPU approach, is that a permanent philosophy or is that a bridge until you have enough contracted cash flow to justify owning the actual asset outright, i.e. not the chip but the data center? Yeah. In addition to the ramp that we have with Bell, we have three data centers that we own outright. We have 70 MW in New Brunswick. That's a Tier 1 that we're going to convert to Tier 3. There'll be 50 MW of IT load after the conversion. We have a seven-megawatt site in northern Sweden that you've been to, actually. It's up in Boden. We have a seven-megawatt site in Toronto, right by the airport. These three sites. We own the land and buildings. They're all Tier 1 data centers, and we're undertaking a conversion to Tier 3 so that those can be long-term leased for a co-location. For example, in Toronto, we had a reverse inquiry from the Department of National Defence. We're seeing with the military, they want a lot of small data centers decentralized rather than big mega data centers with hyperscale spots. Of course, New Brunswick at 70 MW, it's a secondary market. It'd be about $130 a kilowatt. Once fitted out, that would be $85 million of ARR. Strategy there, we bought 32 acres of land. We're going through permitting and design. Sign a co-lo contract with a hyperscaler or enterprise. Once you have that, then of course you go get your LTC project financing and you fund that. Yes, in parallel, we have HIVE owned and operated data centers and the near-term growth ramp on EBITDA with the Bell partnership. I guess at what contract duration or utilization rate do leasing GPUs actually make more economic sense than owning them? The best way to think of it is GPUs ROI after cost in about two and a half years at prevailing market rates. If you're signing a fixed contract for three years, you have fully baked in the repayment of those GPUs and some profit, plus the residual value of those GPUs after the three-year term, which we're seeing right now with the amount of demand for Hopper and other late generation. We're forecasting after three years, GPUs would be worth 50%-60% of what you paid for them. If you've completely paid them off after three years, plus their residual value is 50%, 60%, you're coming out with 150% ROI. The question is, how long can you run those contracts? If you do a five-year contract, now you're baking in 2x the value of those GPUs in your fixed contract. The same way that institutions like fixed revenue because they say, "Hey, why don't you go sign a bunch of 15-year hyperscaler co-location leases, add up those cash flows, and now you've got a replay." The ROI on Tier 3 data center build-outs, you've about eight years. You got a much quicker ROI, two and a half years with a GPU cloud. If you're building up strong levers to return on invested capital through five-year contracts, I think that's a sweet spot. Corey's done exceptionally well following that model. The question may become, well, why would any client pay more than the value of GPUs and lock into a five-year contract when they could just buy it themself? Well, because a model builder or an enterprise isn't in the business of building and operating data centers, going through the procurement, putting all that CapEx up front. Instead, it's a free option call on your term of use. Get access to the latest GPUs, run them as you need them, and you're obviously just paying for usage as you go along. How do you think about the mismatch risk between your contract duration and then the financing of the leases? I guess there's a lot of debate around the useful life, and you just hit on this terminal value, what recontracting looks like going forward. It's just, I guess, hard to know what the market will be like in three to four years. Yeah. Well, we have a lot of implicit DNA because we've been buying large clusters of GPUs. I think our first big NVIDIA fleet we bought in 2021. We bought 38,000 GPUs, and those were the A-series cards. 4,000 of those GPUs are still running today as part of our cloud. They're part of our $35 million ARR, and the other 34,000, we were still mining Ethereum when we bought those. After the merge, we were mining altcoins, and we saw a demand 80%, 90% of their MSRP three years after the fact, Darren, and we acted on that. We crystallized it. We sold 34,000 of those cards, realized a nice net gain because we made back about 40% of the value through mining, sold them at 90% of the value. Net, it was a 30% premium on what we bought, took those proceeds, reinvested into H100 and H200. I think once you've been through a cycle and you've actually gone through the transaction, you know the resellers, et cetera. We've also seen, because we started out this business growing from 400, 4,000 GPUs from $1 million to $10 million ARR. We did that on aggregator platforms. We were exposed to daily prices. We have a backlog of three years of data of variable pricing on A-series cards and Hopper. We've seen the resurgence in demand. By the way, even if you are doing one and two-year contracts, yeah, when there's a spike in demand for late-generation compute, because a lot of enterprises and labs, they're really used to the H100 and H200 infrastructure, and they're like, "Hey, this is great for inference or just ongoing," but it's $ per token per second is much more affordable and attractive. Unless you're a foundational model builder, like one of the big ones, you don't necessarily need pure Rubin Ultra. Can we pause for a second? Given you guys are knee-deep in the Canadian market, we always talk about the U.S. market. Demand for all things AI in the Canadian market is what? Red hot? Look, the largest model builder in Canada is partnered with Bell Canada. That partnership is so that Cohere is going to use Bell Canada Sovereign Compute to train their models. Similarly, Bell Canada is partnered with BUZZ High Performance Computing, our AI subsidiary. That partnership is for BUZZ High Performance Computing to build and operate NVIDIA GPU clusters using reference architecture in Bell data centers. You can connect the dots directionally where that partnership may go. Moreover, as mentioned, we had a reverse inquiry from the Department of National Defence for our Toronto site. Really, I think that as Sovereign Compute, like data as an asset class, there are a lot of enterprises that want their data remaining within borders. We actually are the largest sovereign cloud right now with 5,500 GPUs running. I would even say, like in Paraguay, we're seeing a lot of demand. We were at NVIDIA GTC last week. It's interesting now because NVIDIA's evolved where they will actually help you with offtake. If you're building NVIDIA reference architecture and you're in their good books, which we've been a client for years now, we're a certified NVIDIA NCP, NVIDIA cloud partner, and we've deployed and work very closely with their Canadian team. Jensen even said it in his keynote last year. It's about land and powered shell. Well, guess what a Tier 1 data center is? It's powered shell because you've got your substation, you got your medium voltage distribution, your switch gear, your control. All that stuff is there. We have an extra 100 MW PPA in Paraguay. Here's the big news. Last month, Paraguayan government announced a new tariff, a new electrical rate only for AI, 15 years, CAD 0.038. This is massive news because prior to that, the longest PPA in Paraguay was four years. It was actually the ambassador to Paraguay in the U.S. space in D.C., made a case and said, "Guys, if you want to attract U.S. investment, you really need to have long-term stable PPAs." How are you going to do a 15-year colocation hyperscaler contract if you don't have a PPA that's investment grade? We're really excited about that. We stood up 300 megawatts of Tier 3 data center infrastructure last year in Paraguay. You should come down and check it out. Brady, you too. Shout out to the Roth boys. Now with the other big news from last week, we actually just launched our first AI cloud in Paraguay. We took a cluster of NVIDIA GPUs from Montreal. We sent it to Asunción. In the same way that we partner with Bell Canada, largest telecom player in Canada, the same way we partner with Tigo. Tigo is the largest telecom player in Paraguay. In fact, it's owned by Millicom, which is a $10 billion NASDAQ-listed company. They're all throughout Latin America. We visited their Tier 3 data center. The racks were relatively low power density, but it was enough to fit a few clusters of our GPUs. We installed them. We first announced this partnership in January. The nodes were installed and just got commissioned last week. We have a research lab out of Columbia University in New York, award-winning. They were at NeurIPS. They've been published in all sorts of research papers. They're going to be doing the world's first testing of AI in bandwidth, latency, and tokens per second between New York and Asunción. That's the proof of concept. Once we have that, we build in Iguazú, a purpose-built AI factory based on NVIDIA's reference design architecture, ensuring that you have the fiber, we're going to have the redundancy with UPS's, et cetera. All of a sudden, we're the top contender for HPC in Paraguay. You say, "Well, what's Paraguay? What about" It's a country of six million people that owns half of the world's largest hydro dam, which is Itaipu. Competes with the Three Gorges in China. 14 gigawatts, seven gigawatts go to Brazil, seven gigawatts go to Paraguay. Paraguay's only been using four gigawatts. They have a surplus of three gigawatts that in 2027 they're going to renegotiate with Brazil. How are they going to get that fixed demand? Partnering with data center companies. That's why they created the new tariff. I would say that is a stocking stuffer that's not in any of our fiscal projections, but strategically and directionally, you could see where it's going. The last point on that is Marco Rubio actually signed a status of forces agreement between the U.S. and Paraguay in December. This was big news. It was a way for Paraguay to usher in alignment with the U.S. and domestic investment from the U.S. by showing that the strongest ally in Latin America is actually Paraguay for the U.S. Outsized natural resource to population ratio, U.S. friendly, we've got a great track record there. I'd say the outlook's pretty good. In Canada, $200 million on the GPU cloud, $85 million for New Brunswick. That's $285 million ARR for the year, next 12 months, and market cap's $500 million. I'd say that's a pretty good deal. It's a pretty good stocking stuffer. Yeah. Economically, based on everything you just said, I'll put you on the spot. Why does it make sense? You've got another 100, they've changed infrastructure on pricing for power. You potentially have excess. Does it make sense to use that 200 for Bitcoin mining? Or is there ever a scenario where all that goes to some form of AI infrastructure? Right now, for example, in Sweden, we have a 30-megawatt data center, and it's got S21s. Those will reach their end of life probably in the next couple of years. We don't plan to upgrade those and send you ASICs. We just don't plan to. Similarly, for New Brunswick, we've already signaled that we're going through design development for conversion. We have, conversely, in Paraguay, we have brand new S21+ Hydros that still have a three to four-year economic life cycle. We're going to continue to run those, and we have 300 megawatts existing. It's actually another 100-megawatt PPA that we were awarded last year. We've ordered the substation. It's going to arrive in June. We've been doing site grading and the civil layout, putting the pads. That additional 100 megawatts in Paraguay is what we're targeting for perhaps an H2 2027 conversation about dealing with either an enterprise that wants to get off a cloud regionally in Latin America, or, for example, like Oracle, OCI found out, they're using data centers in Brazil for their compute. São Paulo, which is where the Oracle data center is about one and a half-hour flight from Iguazú. You suddenly get into very U.S.-friendly territory, powered directly from a hydro dam built by HIVE. Sounds pretty good to me. Interesting. Okay, back to kind of the here and now. You're targeting $140 million AI cloud. That improves to $225 million if you sign HPC Colo, all from roughly, I think you said $35 million today, is that right? Correct. Yep. All right. It's like an eightfold increase. Yep. Can you kind of walk us through which specific blocks of megawatts and GPU deployments are kind of contractually committed versus contracts you kind of expect to sign in the next two to three quarters to kind of get to that number? Yeah, definitely. I would say the Bell press release that got us from four megawatts of IT capacity to 16 megawatts was really to signal that the road has been paved, and now expect the traffic to flow. What I mean by that is rough math, a 2,000 GPU cluster, if you want to be more specific, a 29 rack deployment of GB200 NVL72 GB 300. For those of you that cover the sector, you'll know what that means. Would generate about $60 million ARR, and that's five megawatts. In 10 megawatts of IT load, you can bring on $120 million of ARR. We're working two deals right now that are each 2,000 GPU clusters for GB200 and GB 300, and we are at the 10 and 20 yard line of getting those deals done. I would say stay tuned for updates. I'm comfortable talking about those because we've already given the revenue targets, which are predicated upon these deals coming to fruition. The term of those contracts and the TCV we'll announce once the deals are announced. I would definitely say over the next one to two quarters is to bring online those two deals, which we're pretty excited about. I guess of that, how much is at your own capacity for megawatts versus kind of third party? This is all ramping with the near-term Bell partnership, and then the strategy would be to do the conversion for Toronto, which again, we own, and Boden, which you've been to, which we own, and of course, New Brunswick. Right now the GPU cloud target of $200 million is the near-term ramp with Bell partnership, and then the $85 million in addition to that is conversion to HPC for New Brunswick. Got it. On that note, I guess with New Brunswick, what are the specific permitting steps remaining and what's the realistic energization date if everything moves on a proper timeframe? Yeah. Permitting, we bought the additional 32 acres. Good thing is that structural-wise, in the town of Grand Falls, which is where we're based, permitting is less than one month. It's more of an exercise in ordering long lead items. I think once we make an announcement that we've ordered long lead items, then we could provide a ready-for-service date to the market. Otherwise, right now we've really just been budgeting what is the best way to build. Our latest design actually contemplates retrofitting one of the existing four buildings that we have. It's a four-building campus at tier 1, retrofitting the fourth building and then building two new buildings. It'd be like a three-phase project. Is the retrofit, would that be a true tier 3, or? Yes, absolutely. What we realized was, because you have the high ceilings, the slab actually is strong enough to support the racks for GB300. The exercise is actually, you'd have to redo the medium voltage transformers because, not to get technical, but the output voltage of the generators is slightly different than the current medium voltage. I would say from the step down from the switchgear onwards, you would be redoing the medium voltage distribution and adding UPSs and adding generators, and that would pretty much take it to tier 3. Well, and chillers, of course. client interest in that site is from who and is it lean sovereign? it doesn't lean sovereign, interestingly enough. We obviously haven't announced who we've been in talks with. Once we have an MOU or LOI signed, we would announce that in due course. I would say that given its proximity to the U.S., it's close to the border of Maine. The stage at which that site is at, we've had diligence done with one hyperscaler client. There's another conversation. Things are moving so fast. We've talked to a different hyperscaler that actually was interested in doing that entire site as GPU cloud. That site could run 25,000 GPUs. When you're presented with something like that, it all of a sudden becomes very compelling. There's a lot of upside optionality on New Brunswick, and we've provided the base case. I would say stay tuned for updates as that saga unfolds. what I can point to is, again, our previous forecast, we were looking at $2.20 per GPU hour, and when we signed our first Blackwell cluster, we rented it for 35% higher than our target. We rented at $2.90 per GPU hour. What we've really tried to do with BUZZ is, underpromise, overdeliver, which our first cluster deal we've done to the tune of 35%, I'd say that's pretty good. We think, we want to make sure, because once you sign a deal for a 15-year co-location, that's it. You can't go back to the well. Do you want to rush in and just sign the first deal that you get because, oh my god, you got to do it now? You realize the value of what you have, you try to realize the maximum upside potential. We're not seeing demand go away anytime soon. If the idea of doing an entire site for 25,000 GPU cloud with a hyperscaler, and operating at that level, kind of like what IREN pulled off with Microsoft, that's like a mega deal. Those deals are absolute game changers. I think that what HIVE has demonstrated consistently is our technical prowess, our ability to execute, hit our targets. We shield the street in 2025. I think that there's going to be a lot happening in the next couple quarters. Sorry, I'm just watching the clock. I guess one of the last ones on that site, would the retrofit building be done first, and what's the cadence of committing capital for CapEx before there's an LOI/executed lease in place? It's a bit of a balance. Typically what you'd want to do is order long lead items. We call OFCI, owner furnished contractor installed. That include, chillers have the longest lead time right now, then you've got generators and UPSs. Before you deploy capital towards that, we really want to see the EBITDA and revenue from our GPU cloud business break through that $100 million mark. Once we've done that, the stock re-rates, the cost of capital to deploy those first investments into New Brunswick makes a lot more sense. Once you've deployed that capital and you've ordered those long lead, all of a sudden you're in a much better bargaining position with a hyperscaler, to do a co-location or to do a large cloud, for that matter. Near-term, the goal for the next couple quarters, break through that $100 million target on HPC cloud. We've got two $60 million deals, should get us to $150 million, we'll blast right through that. I think the same way we saw our stock re-rate in 2025, again, we hit a over $1.5 billion market cap in October. Actually, we were in Chicago, and Roth took us in a great non-deal roadshow in Chicago when we were there. It was October 9th, actually. We met Magnetar. We take all the credit for that. Yeah. we're just out of time now. Aydin, always good to see you. Appreciate the thoughts and look forward to all the great growth going forward. Thanks. Excellent. Thank you