Deal. Yeah. Let's do it. Good to meet everybody. Mike Massey. I am the CFO of RUM Group, which is a combination of what used to be known as Rumble, as well as our Quake AI brand that does AI compute as a service. Speaking of the two businesses, we have Rumble Video, which we were known for previously. Folks know it's a video platform, 50 million-plus monthly active users. The history of it is a free speech platform. As a part of that, it had to build all of its own rails. It couldn't use AWS, couldn't use GCP to host its servers, et cetera. It built its own CDN, it built its own data centers, et cetera. As a part of that, we also offered a cloud service. Just last year, we acquired a company called Northern Data, which was a former Bitcoin miner that then became an HPC/AI compute as a service company. We just closed that transaction in mid-June. As a part of that, we took the Rumble Cloud, the data centers, the CDN, the low latency, et cetera. We combined that with the GPU as a service from Northern Data, and we formed the brand of Quake AI. In addition, we still have our existing Rumble video platform, and RUM Group is the holding company. There are two separate businesses that operate autonomously underneath them. There are strategic synergies between them. For example, we've started to explore with the customers of both Rumble Cloud and Northern Data. There is a lot of interest in our video data for purposes of AI training. We've gotten multiple outreach on that from robotics companies that want to be able to use and leverage some of that video data, especially that is spatial, temporal, et cetera. As we look at moving forward, we're really focused on, and the majority of the financial opportunity is really going to come from Quake AI. I'll talk about that in a bit and tell you about what size opportunity and why we're so excited about it. One footnote to the story, very important strategic partner of ours, owns almost 50% of the combined company, is Tether. Tether is the world's largest stablecoin company, extremely profitable, also makes strategic investments. We are extremely well-aligned with them. We are one of their largest investments, and we think this unlocks for us both a network of and access to technology that's a meaningful differentiator for us. Okay. Quake, let me give you the speaker notes version of this. 22,000 existing Hopper generation GPUs running inside of primarily European estate. It has a diverse set of workloads, so it is not just one hyperscaler. We have 50-plus end users and customers going from inference to training to pre-training to QLoRA, if folks know some of those workloads. Most importantly to the story, if there is one thing I want you to remember out of this, we have 250 megawatts of unmonetized grid-connected power in 2027. I will talk to you about why we have that, because the first question that I often get is, "How do you have that?" Because it has been sold out, right? I will tell you about the journey of why we have it and why we are so excited about the opportunity. The net of it is the monetization of that 250 megawatts of 2027 power represents a $3 billion-plus revenue run rate opportunity at today's rates, which are trending upwards, I would say. What is the story? What is the journey for Quake AI? Northern Data was one of the original HPC AI companies. It saw the strategy, it saw where the future was going. It had the strategy, but it did not have the execution. Customers were disappointed. Utilization was not where it needed to be. In large part, we brought in new management and leadership last year that was very focused deeply on the execution of the existing 22,000 GPUs. Utilization went from sub 20% in the middle of last year to now 83% plus consistently for the first half of this year. We also have those same customers that we needed to prove our execution to now wanting more from us. They do not just want the current capacity, they want even more capacity from us. We had to rebuild the foundation of execution credibility and customer credibility over the last year to get us to the point where we could actually talk about monetizing the estate that we had. Our first deal and the largest deal in the company's history was with a company called Together AI. It was a multiyear contract for latest generation GPU on B300s from NVIDIA. That was $270 million, and it is really our first proof point on latest generation technology with an existing customer of ours that show that we can do technology leadership and AI compute as a service at scale. Where we sit today and where our management focus is now is on we have earned the right and we have earned the credibility and the execution to be able to go monetize the 250 megawatts of capacity. I am going to skip forward a bit and get to the gist, the most important thing for you to take away, and this is the estate. This is where we have power. The signature of that 250 megawatts is our facility in Atlanta, Georgia. It is 180 megawatts. The most important things right now in the AI compute as a service space are in place for Atlanta, Georgia. We have our use permits. We have a Georgia Power CES. The substation is already built. The transformers are already in place. This is a site that is ready to be built. We are engaged with multiple hyperscalers. We're trying to choose the right partner for us because this is really the first of many we see in terms of large-scale deals, and we're really excited about where we are in terms of demand versus supply on that. Again, 2027 power, 2027 estate. In addition to Atlanta, we have a smaller facility in Pittsburgh, which we think will be helpful for AI natives. In addition, we have an additional 70 megawatts in Europe, specifically 50 megawatts in Sweden and another 20 megawatts in Norway. As a management and leadership team, we are focused on our monetization and our execution of this 250 megawatts of 2027. With current rates today, and I'll show a little bit of the math around that because I'm sure there's some folks who know how this works from a revenue per megawatt perspective. We will be at a $3 billion-plus revenue run rate opportunity when we execute on this estate. We think this is not meaningfully understood by the market today, we think this is not meaningfully understood by investors today, and we're really excited because we just need to be able to execute this to grow to a new scale and new magnitude. This is saying what I just said, which is effectively, if you look at the other AI compute as a service companies that are out there, the Nebius, CoreWeave, IREN, others who are private, et cetera, who can deliver AI compute as a service. They trade for the most part on forward revenue and things like backlog, et cetera. It's very clear that the market still sees us as a video platform, and they really still see the existing business that we were. I think that there's starting to be a recognition of just what opportunity we have in front of us if we have a $3 billion-plus annual run rate revenue opportunity. One thing I will mention, sorry to go into the backup, but this is important. There are a lot of amazing business opportunities within the world of AI compute, and there's an industry that is far outstripping demand versus supply. We, in particular, are an AI compute as a service company, AKA GPU as a service company. We do not just do what we would call power and shell. That is a perfectly valid and amazing business model, and we might partner with some of those folks in the future to deliver our services in the end. You can see here the monetization of power and shell is typically anywhere between $1.5 million and $2 million per megawatt per year, contracted over a long period of time. As we look at the run rates for AI compute as a service, our estate today is monetized at roughly $6 million to $7 million per year per megawatt. Latest generation Blackwell, we've seen trade at more like $11 million per year per megawatt. Future generation Vera Rubin, which is what the 250 megawatts will likely be deployed with, we expect to operate at a premium to that. I think you all have probably seen and heard that the industry is certainly firming in terms of price. I've heard numbers thrown around by Elon and others that are significantly higher than this. But nonetheless, we are seeing a really exciting pricing environment, both for our existing estate as well as for our opportunities of the future with 250 megawatts. So in summary, we think we are a meaningful opportunity. We think we have a clear execution ahead of us. We think it is a simple story. It is 250 megawatts. There is a marquee site in Atlanta. There is a growth opportunity where clearly demand outstrips supply. And we think that the market is hopefully going to be catching up with and understanding just how well we are going to be able to execute this and ultimately monetize to a level of revenue growth that is meaningful. So that is our story. And with that, I will invite up Kingsley if you want to have a quick chat. Sure. Yeah. Well, yeah, I will walk up. Yeah. We talked about this 250 megawatts that you have that remains unmonetized. Could you just help us think through the path to monetization there, maybe the CapEx requirements, either total or on a per megawatt basis, just the ability to acquire GPUs? What type of GPU? Where would they come from? Yeah. Then just, I guess, the operational challenges of bringing that all in line. It is a 2027 opportunity, but perhaps that could be a multi-year runway. Yeah. Great questions. First, to start out with some of the operational and execution complexities, you mentioned the CapEx intensity. I think we have talked about, and there is certainly a robust market out there for opportunities to fund the debt as a part of some of these data center build-outs when you have long-term customer-committed contracts with take-or-pay terms. I think that is, as I talked about, targeting the right types of customers for us, especially with the Atlanta facility. I think that is really where we are focused is on financeability and choosing the right partner, not just for this data center, but more broadly. I think with that, we should be able to finance the majority of what we need in terms of the CapEx build-out overall. But in terms of size and scale, to your question, we are seeing the data center portion trade anywhere from $8 million-$12 million per megawatt in general in terms of CapEx. As you start to think about the next generation GPUs, which you asked the question of what we are targeting, in general, we are targeting Vera Rubin architecture. You start to think about the rack, I think SemiAnalysis has had that anywhere from $7 million-$8 million per rack. Just for context on something of 180 megawatts, I think we have talked about this publicly, you could start to think about that as anywhere from 40,000-50,000 GPUs. Okay. Yeah. The second part of your question is: how do we think about execution as a management leadership team, and what are we most worried about? You asked a question about GPU supply. To be honest, we are not as concerned about that. I think we are really well strategically aligned with key partners who want to see us be successful. I think that we have been aligned with them for a really long time, including NVIDIA and others. We look at this as a management and leadership team of, we would not embark and start the project really until we had assurances of the type of supply allocation that is necessary both for us and for the customer of those GPUs. What do I worry about if I do not worry about the GPU supply? It is the boring things that have become really the gates to the industry, right? It's access to steel, things like UPS backup systems, things like chillers, et cetera. When you go into a project like this, the first thing you ask the question on is, do I have the availability of skilled labor that's necessary? Which is why we love being in the Atlanta market, because it's a very robust market, both for data centers as well as for labor and talent. Then, do I have the long lead equipment items that are really going to be necessary to deliver the schedule? The best news is we've got the hardest long lead items, like the transformers already on site, the substation's already built, we've got generators, et cetera. The next wave is those, again, boring things like steel availability, et cetera. You mentioned partners, and NVIDIA's a strong partner. You have a lot of NVIDIA chips on your hands. What about AMD in that mix? Is that something that you're looking at? How does that factor into the cost profile? Then on the partners, can't talk about partners without talking about Tether, who's a big investor. In the past they've also been able to throw their weight around and garner some more GPUs. Do you think that having them in your corner could help you acquire more GPUs? Is that relevant at all? Yeah. I think, to answer the second first, I think Tether is a meaningful strategic partner for us. I wouldn't say throw their weight around, but I think people knowing that Tether is behind us and they're a major equity investor is always a tailwind in every one of our conversations. I think that people understand the level of capital, and people understand the amount of reach that Tether has across borders, et cetera. We've been really happy with them as a strategic partner, and we think that's going to be really valuable moving forward. What was the first part of the question? Sorry. Just thinking about diversifying GPU. Oh, AMD. Sorry. Or AMD that is sitting there. First, we are going to be customer led, first of all. We will look at what customers we want to work with and who we want to work with. In terms of the way we think about our NVIDIA partnership today, number one, they provide amazing technical support. I think things like the DGX Factory, I do not think it is fully appreciated by the market, the extent of value that that is created for folks like us in giving us execution assurance. I think that they provide excellent technical support. In general, our customer feedback so far is that in spite of the fact that prices of the sticker may be higher, the amount of tokens per dollar and tokens per watt that they are able to generate with NVIDIA products as opposed to others is significantly higher just with the software optimizations. For us, we are really happy with where we are with NVIDIA right now. If our customers are happy, we are happy. If there is a point in the future where our customers start saying that they are not seeing the value and they see value in other alternative architectures, we will always be open to that. You have a huge 22,000 H100s, H200s, and spot pricing for that class of chips has only increased over the past six months. You also have a base of Blackwells, I think, that you have signed out to Tether AI, or that is all accounted for. You have mentioned that you are targeting Vera Rubin. I guess just as an operator, what are you expecting for commoditization or pricing trends across the stack there over time? Maybe just candid thoughts on what the useful life of a GPU could be, because it seems like it really could be five to six years. Yeah. I think even that is probably low. I think it is very clear, and I have been in technology industry for a long time, some of it on the product side, some of it on the finance side. What I have found is engineers will always find optimizations the more time that you give them. I think what they have found is there was a time and there was that trough of disillusionment on the Hopper generation where prices just sank, right? Because everybody said, "Oh, there is going to be obsolescence. There is three years. You have got Blackwell generation coming on board." Well, the smart engineers said, "Hey, I can run my latest generation recursive inference model on Blackwell. But hey, these Hopper generation, they are fantastic when it comes to retraining. Let me put together a system where I can leverage those Hoppers." I think that is always going to happen whenever you have effectively deployed high CapEx intensive businesses, engineers will always take advantage of the N -1, N -2 generation. I come from the more traditional cloud business. I worked a lot with the hyperscale clouds as they built out what we would call today a CPU cloud. That business is built on the idea of N minus two, N minus three. You do not make money on the N generation. You always make money on N -1, -2, and -3. The trick of it is how do you keep those around for longer and longer? I think five or six years is probably undercalling how useful some of those Hoppers are going to be in aggregate. Right. Yeah, I think that- In terms of comments on pricing and market demand and dynamics, I think it is very clear that demand is outstripping supply in every portion of the business, so we have certainly seen firming of prices. In the case of us, I mentioned it before, we sit at $6 million-$7 million per year per megawatt in terms of our existing Hopper estate. I think a lot of those contracts, and we did this very intentionally because we talked about building customer credibility and execution over the last year. We signed longer term contracts at a time when the market was probably not as hot as it is right now. We are certainly seeing opportunity in the second half of this year and going into next year that those Hopper generations are trading at a lot higher multiples and generally have a lot firmer demand than supply. In terms of next generations like Blackwell and Vera Rubin, I think the market is starting to come around to the idea of it is really about the output. It is not about just the cost of the GPU, so they are starting to see the benefits of, hey, if I can produce 4x the number of tokens on Vera for the same amount of megawatts or kilowatts, it makes sense for me to use Vera Rubin for these kinds of models and these types of tasks. Then I can use the older generation for the other types of tasks that may need to run on the same recursive loop. Right. Yeah, I think to this point, about training and inference, I think one of the surprises about the durability of the GPU is just the size of the inference opportunity, the shape of that ramp. All of the different classes of GPUs are important. I guess thinking about a customer like Together AI, who has an inference cloud, their end customers are probably more concerned about price per token than they are price per GPU hour. How does that play out in your own customer set? Are you chasing these much larger deals with the inference clouds, or do you think that there's an opportunity to price with some smaller customers that are developing apps maybe more directly with Quake AI? Yeah, I think this is an area where we've made a bet as a leadership team that we think is the right bet, and it's somewhat counterintuitive, to be honest. By the way, I'm not criticizing other strategy, this is our strategy. I think others have started to move up the stack. I think that they've talked about developing things like a PaaS layer to be able to do things like model as a service. I think that's valid. I think our customers, and the ones that we've worked with traditionally, feel as though that's trying to sort of disintermediate them from their end user and customer. I think we are going to be much more focused on being an infrastructure company as opposed to being a software company. We're amazing at running hardware. We're amazing at running data centers. I think that's where we're going to focus because that's where we think the most value is, and I want to be very channel friendly. I don't want to try and disintermediate, I don't want to try and take my customers' margin away from them because there's plenty of margin for me to make. We're really going to focus on hardware and infrastructure. I don't see us making meaningful steps up the stack. To be transparent, I talked about some of the challenges that we really fixed as a leadership team over the last year in terms of the execution of the existing Northern Data estate. A lot of that was because of lack of focus, because they were trying to build on software capabilities that their customers were not necessarily asking them for. I want to be excellent at what my customers want, and ultimately, I think that there's plenty of margin and plenty of value that we create because running GPUs from a hardware perspective at scale is really, really difficult and really, really valuable. One of the things, it's become apparent that it's been difficult to get data centers into production over the past year. Yeah. We're seeing some political barriers. We have midterms coming up, depends on the geo, but I guess given the state of your portfolio, how relevant is that to you as we look over the next couple of months? When you have access to the power, are we already far enough along that you don't think that that's really a roadblock or could be? I think it's relevant to everybody. I think, in the case of Atlanta, we're about as far along as you would like to be. We have use permits in place. We have Georgia Power, CES, we have substations already built, we have transformers already done. But in the end, we want to be a part of communities that want us to be there, right? We don't want to have obstructionists, and I think that we've done a really good job, specifically with the town of Maysville in Atlanta, and really getting the word out there and dispelling some of the myths about data centers. We're really happy, and we think we're in a really good place in terms of community relations. But to your question, I think that's the next big stage for all of us data center and all of us AI compute as a service. You have to first think about the communities that you are a part of before you choose the sites that you're going to develop with. We feel like this portfolio is in a really strong position given the relationships that we have. As we start to look at and start to develop that next wave of the portfolio, call it 2028, we're thinking first about how do we have positive community relationships to make sure we don't see schedule delays. Okay. We are close on time. Okay. Would the audience like to ask a question, they can go for it. Put up the comps again. Slide the comps down. Leave it there for a second. If it is resource constraint, are all the comps doing what you do? Yes. Well, some of them. More or less. More or less. Okay. If you are trading at one times EV to revenues, and everyone else is trading at 3.5x supply constraint industry, are people knocking on your door so you can resplit the difference and took stock in the deal, everybody would make money. Obviously, I cannot reveal any non-public information, but what I would say is that people who are familiar both with our execution and our business have a very similar mindset, as this slide would imply. I think that the market, in a broad sense, does not fully recognize that we are not a video platform anymore, and that we did not just make a small pivot. We bought a meaningful business that has a meaningful position in AI compute as a service and- Did you get rid of the other businesses that you painted with that brush? We still do have our video business. We run it separately, so it has no interactions with our Quake AI in a negative or constructive way. Ultimately, the results of Quake AI are going to far outstrip that of the video platform financially. Of your $3 billion in revenue, how much is the video? Sorry, we're late on time, so I will answer your question, I think. Yeah. Thank you so much, Mike, for the time. It's been amazing. Absolutely. All right. Well, I'll tell him