Good afternoon, everyone, and welcome to Northern Data Group's Q3 2024 earnings call. Thank you all for joining us today. My name is Jose Cano, Vice President Investor Relations of Northern Data Group. I joined Northern Data in September, following Jens-Philipp's move to a different role within the company. On the call today, we're joined by our CEO and founder, arose Aroosh Thillainathan; alongside our group COO, Rosanne Kincaid-Smith; and Elliot Jordan, our Group's CFO. Before we start, I must remind you that, as always, this presentation and management's comments reflect underlying assumptions and include forward-looking statements. I will not read the full disclaimer, but I will assume it has been read for the purpose of this call. In terms of our agenda today, Aroosh will share with you an overview of Northern Data Group and the key industry trends followed by Rosanne, who will focus on our strategy as well as the main performance drivers and highlights of the first 9 months of 2024. Elliot will run through the financial performance in Q3 and the outlook. And finally, Rosanne will close the presentation with a brief reminder of Northern Data's path forward. After the presentation, as usual, there will be time for Q&A. And with that, I will hand over to Aroosh.

Thank you, Jose, and good afternoon, everyone. For those of you who may not be so familiar with the company, Northern Data Group is a leading provider of AI and high-performance computing solutions. We are powering the next generation of innovation through best-in-class hardware, purpose-built data centers and carbon-neutral energy systems that shape the future today. We are a global company with a European heart, and we work closely with a selected group of partners to optimize our infrastructure and ensure our products meet the increasing demands of the fast-growing AI market. Taiga Cloud is the answer to the exploding generative AI market, which is expected to grow to more than USD 1.3 trillion by 2030. Generative AI tools for speed and automation and their significant benefit to life science and education ensure Gen AI is poised to deliver through societal progress in the long term. Taiga Cloud is Europe's largest generative AI cloud platform. Our solutions are fueled by the very latest NVIDIA technology solutions. They provide true data sovereignty running on carbon-neutral energy. Our Ardent Data Center business is how Northern Data Group is supporting the rapidly expanding demand for data center capacity, driven by both the advancements in cloud technology and generative AI. This need for data center capacity will not only underpin the AI revolution, it will also provide a platform for uniquely sustainable solutions for energy use and supporting technologies like liquid cooling. With Ardent, we are pioneering a new era of high performance and efficiency, which leverages more than 25 years' experience in design and operation. Finally, our heritage Peak Mining business supports the global adoption of digital assets into the wider financial ecosystem. Bitcoin mining is our heritage business. Having started off as a bitcoin miner, we know what it means to scale and make use of efficiencies and above all, capitalize on opportunities that arise from market trends. Northern Data is poised to capitalize on the major trends that are shaping the future of the industry. The advent of GenAI has brought exponential growth to the HPC space, and the speed at which AI is being adopted has increased demand of the infrastructure required to train and support the increasingly large LLMs we continue to see. Further, full access to the compute hardware required to run AI and LLM programs, combined with the limited know-how for maintaining data center infrastructure, have led companies and individuals scrambling for resources to keep up. As an elite partner to NVIDIA, Northern Data is one of just 11 storing AI clouds globally to be selected as a partner for the latest Grace Blackwell technology by NVIDIA. Our data centers are purpose-built to support this technology-specific requirements. For example, the new AI HPC data centers are adopting the latest on-chip liquid-cooled technologies, enabling power densities up to 120 kilowatts per rack. Finally, as these GPUs are significantly more power hungry than the prior generations, adding more pressure to the energy grid, Northern Data has taken steps to offer carbon-neutral solutions to our clients. I'm proud of the work we have done over the last 2 years to stay ahead of the -- these trends and believe we are well positioned to capitalize on the business opportunity ahead of us by delivering for our clients who are on the cutting edge of Gen AI innovation. Let me now hand over to Rosanne to share some additional insights on how we are bringing this to life.

Thank you, Aroosh, and hello, everyone. To capitalize on the opportunity of AI and high-performance compute, we are building on strong foundations in 5 key areas

commercial clarity, thanks to our clear delineation of services across cloud, data centers and mining; industry expertise, thanks to a portfolio that is led by industry and subject matter experts; access to growth capital from successful -- successive capital raises; state-of-the-art technology through the acquisition of the latest GPU and ASIC-based technology; alliances with the right partners to maximize hardware and deployment. With the ongoing support of our major shareholders, as demonstrated by the capital raise in July and our close supply partnerships, we are empowered to make investments and deliver on our technology-first strategy. And importantly, we have put in place the corporate structure to allow for sustainable growth across our group, deploying assets and expertise to capitalize on new opportunities. Our path to solidifying our position as a leader in AI and HPC solutions is clear, and we are executing against each of these areas to realize that ambition. I am genuinely excited about future of Northern Data Group. Now turning to our operational highlights so far this year. I'm pleased to say we've had a strong 9 months. We have successfully built Europe's first and largest dedicated Genesis AI cloud. Our platform provides access to the latest NVIDIA technology running on carbon neutral energy. At the end of the quarter, we have circa 19,000 H100 Tensor Core GPUs with customers already onboarded or in the process of being onboarded. The delivery of GPUs has been consistent, thanks to our strong partnerships with our suppliers, and our deployment has been strong over the course of the year. But it is important to distinguish between the delivery and deployment of GPUs and the proportion of GPUs being billed in any quarter. Particularly when onboarding large customers across many thousands of GPUs, there is almost always significant testing and tailored configuration required before billing will commence, and we are proud to be a provider that can offer its customers bespoke solutions, which will in turn foster long-term commercial relationship. We have been focused on building the foundations of our business, staying ahead of market and initiating strategic partnerships that will augment our capabilities. We purchased the next generation NVIDIA H200 Tensor Core GPUs, which has been delivered to our partner sites in the U.K., and we are now preparing to deploy this next-generation technology. As Aroosh noted, we were selected to be one of the first sovereign cloud service providers to offer NVIDIA's Grace Blackwell technology. We also inked partnerships with VAST Data storage and Supermicro, a well-known premium technology leader. Since the official launch of our AI Accelerator, we have received applications from start-ups, representing 21 countries across 5 continents, spanning pre-seed to series B+. In September, our review panel consisting of Northern Data Group leadership and external partners like HP and Supermicro, we met to select the final 5 start-ups to progress with our AI Accelerator. More recently, we extended our partnership with NVIDIA to include NVIDIA's AI Enterprise an end-to-end cloud native software platform that accelerates data science pipelines and streamlines development and deployment of production-grade copilot and other Generative AI applications. Our data center business also expanded its capacity breaking ground on the recently acquired site in Pittsburgh, Pennsylvania and commercial retrofit to make it a cutting-edge, high-density data center environment, which will deliver 20 megawatts of capacity on completion. We are expecting the first 5 megawatts of flagship direct-to-chip liquid cooled capacity to be available in early 2025. This means that our portfolio now has access to over 60 megawatts. Additionally, we have more than 1.2 gigawatts of site acquisition targets. This represents a significant increase of our focus in this area under due diligence or LOI. Our own capacity is complemented by our growing ecosystem of third-party co-location partners. And our mining business has continued to add committed capacity and remains on track to deliver 7.9 EH/s by year-end. Momentum in our cloud business was driven by the further progress of our GPU deployments and the onboarding of customers. As I mentioned earlier, to the end of the quarter, we have taken delivery of or installed circa 19,000 NVIDIA H100 Tensor Core GPUs. Our H100 estate spans 6 different locations. Our H200 will further expand our estate and as part of our commitment to providing best-in-class hardware, we are de-racking our A100 fleet as part of our hardware life cycle management program. This will aid in reinvestment and enables us to continue to offer customers access to only the latest and best cutting-edge hardware. Our structured deployment schedule and our commitment to operational excellence represent undeniable progress towards solidifying our position as a leader in the AI and HPC landscape. And of course, and most excitingly, further investments are in the pipeline. Generative AI is a subset of AI and its focus was on teaching machines to produce original and creative content. Unlike traditional AI, which operates based on predetermined rules, Generative AI has the ability to learn from data and generate content autonomously. This technology leverages complex algorithms and neural networks to understand patterns and produce output that mimic human-like creativity. The significance of this lies in its potential to revolutionize industries across the board from content creation to software development to higher-order use cases like health care and climate change. Generative AI tools are paving the way for greater efficiency, creativity and innovation. Companies are increasingly adopting these tools to streamline their processes, reduce manual effort and unlock new possibilities that were once unimaginable. Our guiding principle at Northern Data is innovation bravery. That means being bold in our decisions and putting our resources behind the technologies that we believe will make the biggest difference. To enable this, we will continue to innovate and provide customers with the tools they need to make the breakthrough, which will change the course of history. In short, this means we are doing much more than deploying GPUs. We are building a customer-focused AI ecosystem. We have the privilege of running the foundational technology of the AI revolution. But importantly, we are also building the services that will allow us to partner with our customers on an enhanced basis to establish what capabilities they need to bring their ideas to life and importantly, enabling them to bring about the positive societal change that we would like to see. The solutions we are developing will form a seamless product platform. AI Platform as a Service, which you will all be familiar with as the scalable GPU-based infrastructure that serves the foundation of our business today, powered by the cutting edge NVIDIA technology and providing true data sovereignty, it runs on carbon-neutral energy. Intelligence as a Service, the NVIDIA best in NVAIE, which offers easy-to-use tools to accelerate the development and deployment of production-grade co-pilots and other Generative AI applications, dynamic enablement solutions, which will provide on-demand capabilities and the human expertise of skilled AI engineers and AI talent solutions to turn innovative ideas into reality. Our cutting-edge colocation offering, ensuring that those customers who grow from using our infrastructure into owning their own can continue to enjoy the superior enablement and services they've become accustomed to in our future-proof colocation environment with continual innovation to better optimize scalable deployment and be flexible to grow with our customers across our global footprint. Northern Data emphasizes the collaboration between human creativity and AI capabilities, ensuring refined and tailored outcomes that drive industries towards remarkable achievements. The final quarter of this year continues to be focused on execution, but also preparation for 2025. In summary, our cloud business is focused on the rapid deployment of our H200 GPUs and the onboarding of more customers into our H100 capacity. In addition, we will roll out the NVIDIA AI Enterprise software. As already mentioned, we are in the process of completing our unique Grace Blackwell technology configuration design with the support of NVIDIA. And this will be a defining moment for us as we enter 2025. In our data center business, we are focused on the substantial upgrades to our data center in Pittsburgh, Pennsylvania, and completing the due diligence on new sites. And of course, our mining business will complete the committed 7.9 EH/s. We really are poised for strong continued commercial growth into 2025 and beyond. And I now have the pleasure to hand over to Elliot, who will talk us through our financial performance.

Thanks, Rosanne, and good afternoon, everyone. Our financial performance in Q3 demonstrates strong execution against our HPC strategy with group revenue of EUR 59 million, up 235% year-on-year. Adjusted EBITDA of EUR 25 million, delivering a 42% margin and an operating cash outflow of EUR 6 million. Our overall position year-to-date is revenue of EUR 115 million, up 109% year-on-year, driven by a 7% year-on-year increase in mining and 530% year-on-year growth from cloud compute, which puts us on track to achieve the full year guidance of EUR 200 million to EUR 240 million in revenue for 2024. Adjusted EBITDA for Q3 was EUR 25 million, a significant increase in profitability compared to the 2 prior quarters, which means we are at approximately EUR 26 million adjusted EBITDA year-to-date. This result, combined with an expected strong Q4 means we are also on track to achieve EUR 50 million to EUR 80 million in adjusted EBITDA for 2024. Looking at revenue in more detail. Cloud revenue stepped up 296% quarter-on-quarter from EUR 12 million to EUR 48 million, representing 80% of group revenue in the quarter. This significant ramp-up in revenue mirrors the step-up in deployment of GPUs over the summer alongside the onboarding of new customers and expanding existing customer relationships to achieve good levels of utilization in Q3. We have further opportunities to increase revenue from this capacity as we continue to onboard more customers during the coming months. In addition, we expect further revenue growth as we move towards 24,000 GPUs deployed in Q4 and further growth into Q1 '25 as utilization rates increase after clients onboard, test and configure their systems with us. Mining revenue stepped back from EUR 14 million in Q2 to EUR 12 million in Q3. Our investment in efficient miners and growth in capacity meant our share of the block increased quarter-on-quarter to 0.767%. However, our hardware uptime reduced quarter-on-quarter to 66%, which means we produced 209.6 coins in Q3, which is 8% below Q2. With an average bitcoin price down quarter-on-quarter to EUR 55,464, peak revenue was EUR 12 million in the quarter, 17% below both the prior year and the prior quarter. It's important to note how the execution of our investment strategy is transforming Northern Data's profitability. We are moving from a business mix historically dominated by mining at approximately 80% of 2023 group revenue to a business mix where cloud is estimated to become the largest business segment reaching around 80% of group revenue in 2025. Our business cloud revenue is recurring, predictable and delivers a high EBITDA margin and in turn, a strong positive operating cash flow. This allows us to achieve an expected payback on GPU infrastructure in 2 to 2.5 years. We are already seeing this transformation in our revenue mix this year with Q3 at 80% from cloud compute, a significant change in just 3 quarters from 26% in the first quarter of the year. With strong momentum behind our GPU deployment and sales pipeline as we enter Q4, we remain on track for our previously stated revenue and EBITDA targets. As a result, we would expect to see revenue to increase sequentially again in Q4 and reach at least EUR 85 million with the deployment date and finalizing client contracts determining the overall revenue position for the quarter. Our current estimate for revenue therefore remains at EUR 200 million to EUR 240 million for 2024. This Q4 revenue position drops through to another quarter of strong adjusted EBITDA margins, and therefore, we continue to expect an adjusted EBITDA for the full year of EUR 50 million to EUR 80 million. Turning to cash. Q3 saw a very active deployment of capital with payment for GPU infrastructure and the final payments for efficiency and capacity investment at our mining sites. Total spend for Q3 on capital projects was approximately EUR 370 million, a total of EUR 970 million year-to-date. The main items of the spend year-to-date were

EUR 790 million into Taiga Cloud infrastructure, primarily GPU acquisition and deployment; EUR 35 million into our Ardent Data Center business, primarily into the retrofit of our Pittsburgh location; and EUR 140 million on mining infrastructure as we grew our efficient mining capacity. Note, we also had around EUR 70 million in receipts year-to-date from hardware sales. So the year-to-date net cash flow from investing is EUR 900 million. For the full year, we continue to remain well funded with an estimated year-end closing cash position of between EUR 100 million to EUR 200 million. In Q4, we are focusing our investments on the retrofit of our Pittsburgh data center location as well as early payments associated with a second U.S. data center. Within Taiga Cloud, we have the final payments for the H200 island which is coming on stream in late Q4. Finally, there will be no additional material CapEx for Peak Mining this year. As Rosanne mentioned earlier in the presentation, we expect to sell our 2,500 A100 GPUs as part of our well-managed hardware life cycle program. This cash inflow in Q4 will aid reinvestment in the latest technology as we move into 2025. As a result, total growth investments for the full year is expected at just over EUR 1 billion, and net investment after asset sales will be just under EUR 1 billion. The successful execution of our strategy is delivering the financial results we expected at our Capital Markets Day this time last year. This includes our expectations for 2025, which remain unchanged. The final deployment and sales cycles across Q4 2024 and into Q1 2025 put us on track to deliver EUR 520 million to EUR 570 million in revenue and EUR 300 million to EUR 350 million in adjusted EBITDA next year. This demonstrates an opportunity for us to annualize on the ramp-up in capacity we have achieved in 2024. As I mentioned earlier, circa 80% of this revenue is expected to come from our AI Cloud Technology segment driving the strong EBITDA margins overall. I look forward to discussing the Q4 and full year results with you in a few months' time. But for now, I'll hand back over to Rosanne.

Thank you, Elliot. Now a quick look at our path forward and the road map for success in '25 and beyond. Our path to solidifying our position as a leader in AI and HPC solutions is quickly being realized. Seamless execution has been our absolute focus in 2024. The delivery of our technology and product road map was paramount to this and inherent in that was the strategic investments in the rapid deployment of our NVIDIA H100 Tensor Core GPUs. Now in quarter 4, our attention has turned to our H200 with 2,000 set to be installed across the remainder of the year. Our success is building, and we are on track to meet our targets for the year. Looking ahead to 2025, you will see us accelerate with NVIDIA Blackwell technology, new innovative software and specialized services. We will scale, capitalize on growing market opportunities and deliver sustainable, predictable, repeatable growth. We will expand our North technology and product estate, establishing our leadership position in the technology market and continue to invest for future growth. In 2026, we will double our ambition for further expansion. We will develop new cutting-edge products and solutions while forging new strategic partnerships, which will enable us to increase our market share. We will realize scale and optimize capability while continuing to deliver value to shareholders. These foundations are the drivers for continued commercial success not just today, but in the future too. We have our North Star, and we are poised to harness the opportunities that the future will bring. To conclude, we are proud of the progress we have made so far this year. We are wholly focused on advancing the technology which creates leading high-performance computing solutions, and we are well positioned to continue to execute on our strategy. We have built strong foundations with diversified and clear core capabilities, commercial clarity backed by industry expertise. Our strong strategic partnerships and smart capital allocation secure our future growth. Our investors' confidence in us is reflected in consecutive successful capital raises in 2023 and in 2024, securing as Elliot mentioned, nearly EUR 1 billion in debt and equity funding. As such, we are in a strong position financially and primed for delivery. We've executed rigorously against our road map in 2024, and we have begun to see the results, and we expect to see very much more. We are well positioned to accelerate into 2025, and we are very excited for the next 12 months and beyond.

Thank you, Rosanne. This concludes the presentation, and we can now open the lines for Q&A.

[Operator Instructions] Our first question comes from Lucas Pipes with B. Riley Securities.

This is Nick Giles asking questions on behalf of Lucas. Guys, congratulations on the progress so far. You mentioned some acquisitions in the pipeline. And my first question is wondering if you could give us a flavor of kind of size and geography of the ones in the pipeline?

I think we've been -- thank you for joining the call. I think we've been transparent that we have been very -- partnering very closely with NVIDIA on Grace Blackwell technology. And there is also a significant amount of more capacity from a data center perspective that we've been looking at. And those will be the focus of our investment and all of the related infrastructure to continue to deliver to the market the innovative and direct-to-chip liquid cooling technology and advanced architecture that we've been focused on. So that will be the focus of our ongoing investments.

I appreciate that. Should we think about these opportunities, is there a desire to expand further in the U.S.? And if so, would these be brownfield in nature, similar to Pittsburgh or are you looking at more greenfield opportunities?

So we will expand our footprint. And I think we've also been transparent about that we're a global company although our heart is European and that does mean expansion into North America more aggressively perhaps than what we have indicated in the past. And the data center acquisitions that we have looked at, that strategy remains in place. We prioritized retrofitting. So buying what do you refer to as brownfield, which is buying existing data centers and retrofitting them for our unique designs.

Understood. Maybe just one more before I turn it over. Sorry if I missed this, but for the H100s, what's the split today between delivered versus deployed?

So when we say they are delivered, we've taken receipt of them, they're currently in design, and they're getting ready for deployment. So there are currently 12,000 -- just over 12,000 that are actively deployed and the remaining GPUs are currently in design or being built according to customer specifications.

Our next question is from Michael Roost with Baader-Helvea.

It's Mike from Baader-Helvea. So actually, I've got 3 questions. I'll put them forward all in a row, and then you can sort of push them through to whoever can answer them. So the first question, I hate to sort of bang on about the same thing because you mentioned that in the presentation quite a few times. Obviously, your guidance for the year. There is obviously a huge element of back-end loading into the guidance, i.e., that Q4 is obviously the big quarter in terms of revenue generation and also on the EBITDA side. From today's perspective, what is the risk in terms of that guidance? The second question for me is on the cash, Elliot, I think you mentioned, for example, on the cash position that you're expecting cash position around between EUR 100 million and EUR 200 million. Can you just give me a bit of an idea in terms of what are the positives or negatives affecting that number there? And then finally, actually, just more of a general question. I know you guys -- although you're crypto miners, you're not necessarily focused on the price of cryptocurrencies, but what is your general view of the next let say, 6 to 12 months in terms of cryptocurrencies?

Mike, thanks for those questions. I will dive in. It's Elliot here. So in terms of Q4, we're very confident that the EUR 200 million to EUR 240 million is achievable this year. As Rosanne just answered in the previous question, of the 18,000, 19,000 H100 GPUs that we have taken receipt of, only about 12,000 or so were generating revenue across Q3. And so that leaves us with an opportunity to start to monetize the other 6,000 or so plus we've got a few more another island of H100s coming on stream across Q4. So monetizing those into the quarter sees another significant step-up in revenue from the cloud compute business. So we'll be going from EUR 47 million up to EUR 75 million sort of levels, I guess, plus depending on the full sell-through rate. I guess the key risk to delivering these numbers is around client onboarding and testing. As Rosanne said earlier on in the presentation, we are becoming a lot more bespoke in tailoring our offering for the client requirements. And what we've learned during those processes is to get it right for our clients, we are taking our time to test and configure and test again and deliver high-quality infrastructure and services to the clients. So will take a little bit longer to do that and get it right, and then the revenue flows from that. So the key risk to delivering within the range, of course, is how long that testing might take. But the longer we take on the testing, the better we did it, and therefore, revenue for next year is more secure because we don't have to move from 1 client to the next. In terms of profitability, you can see that once the cloud revenue starts to drive the bulk of the P&L, it falls through very strongly. Pretty much all of the profitability that we delivered in Q3 came from the cloud business. The mining business was sort of breakeven in those sort of levels. So if you sort of take that same EBITDA margin on cloud and apply it to this Q4 expectation, you can see that we'll pretty easily get into the range of the adjusted EBITDA guidance for the full year. So confident at those levels. In terms of cash, the EUR 100 million to EUR 200 million, as I said, is sort of midpoint probably, if you want to really sort of zero in on your model so around EUR 150 million. That includes positive cash flow now coming through from our operations. So as we monetize the cloud revenue and get us a few receivables in that we had at the end of the quarter, we'll be delivering positive operating cash flow across Q4. And so the net position in terms of investing is probably around EUR 65 million to EUR 70 million in the quarter. And then last but not least, the final aspects of the capital raise, the injection of capital that we announced in July, will be coming through into Q4 to mean that overall, Q4 is a pretty neutral quarter, the same level of cash we've got now. We've got EUR 149 million in the bank at the end of September, so the same level expected at the end of December. I hope that answers the modeling questions. I'll hand over to Aroosh, who is more of an expert on crypto than me.

Mike, yes, to answer the question. So actually on the infrastructure side, right, so the crypto where you can look at the digital asset side and the bitcoin mining side as more on the infrastructure side. You can see that profitabilities are right now stabilizing according to the higher pricing. I think going forward, definitely, as the largest market for this is the U.S., I think the U.S. elections will definitely have an impact on the overall crypto market. One candidate is more bullish on the digital asset space than the other. But as we've seen in the past, when you see on the mining perspective, you have always the basis is we have the phase of high profitability and then the correction. So I think for the next 12 to 14 months, I think the profitability will stay in line. But after that, we have to see how the price development of bitcoin will go if we can keep the profitability what is today.

Our next question is from Milo Bussell with Edison Group.

This is Milo Bussell from Edison. Congratulations on the progress and thanks for the presentation. I've got 3 questions, so maybe we'll take them one by one. So firstly, it would be good -- your results obviously show strong demand within Europe for HPC cloud services. So could you give us an idea of the competitive landscape within the European context? And additionally, how are you finding upselling managed services and software to existing clients?

And so I can take that one, Milo, and thanks for joining the call. And in terms of the competitive landscape, I mean, I think it's fair to say that there is increasing competition. And where we certainly had the edge is that we were first to market, and we also have the largest clusters. So we're able to cater not just to small customers, but to scale customers as sort of demonstrated by the profile of the customers that we have shared with the market. So we expect to see growing competition, but we also expect, to your point, to have those services, which allow us to institutionalize our customers to our product platform rather than just selling GPU capacity. So the start off of our managed services or the enablement solutions that we referenced earlier on in the presentation, will start with intelligence service, which is the tool NVAIE, which will allow customers to develop and deploy their solutions. And then the advancement of offering them expansions to their own teams through AI talented solution and on-demand spin-up, spin-down solutions. So there is a huge demand for those types of solutions. And we are looking to really have those to scale in 2025.

Great. My second question is just on the GPU rollout. So I think previously, you discussed the H100 cluster being fully deployed by Q3 -- the end of Q3. I know it's only a difference of 2,000 or so GPUs, but are you facing any delays or bottlenecks in the deployment, just thinking about the rollout of the H200s? And I guess on the H200, just could you give an indication of the demand from customers?

Yes. So I'll just cover the delays to start off with. And so of course, somebody will stop the 1,034 GPUs that we're missing off of the receipt map. And those 1,034 are in fact in hand now. They did at the end of September get stuck at border control and which is inevitably one of those things that are somewhat outside of our control, but they are in fact in receipt now in October and will be added to the cluster in Portugal. And the H200 demand. And so H200 demand has actually been extremely high. And the H200s and -- technology is significantly more performant for training, and the pipeline for that has been very interesting to see develop and we've got customers who are very much competing for that capacity at the moment.

Great. And just finally, on the Pittsburgh facility. So I mean, on the remaining 15 megawatts, how should we think about the phasing of that in FY '25?

So the remaining capacity -- so actually 5 megawatts early 2025, we expect the next 5 megawatts towards the end of 2025 and then the remaining 10 megawatts early in 2026. Part of that is about the utility delivering the power, and part of that is about us expanding the capacity in line with future proofing innovation. So we know that the next generation of technology will certainly require more cooling facilities and higher power demand. And so we expect the rack setup to change slightly. So we've staggered that ramp up to meet with that demand. As I mentioned earlier on in the presentation, we view Pittsburgh as a flagship for our own innovative technology. And so we've staggered the ramp up accordingly.

Our next question comes from Gerhard Orgonas with Berenberg.

I also have 3 questions, please. First one, could you remind us about the contract structure and the utilization of your GPUs that you currently have in place? What kind of rental contracts do you have? What's the average duration in main clients?

So the average duration of our contracts ranges between 3 to 12 months at the moment. But in our pipeline, we have contracts that range anywhere from 3 months to 24 months. And it very much depends on the configuration that the customer wants, but as it stands now, 3 to 12 months.

Okay. Any major client who has taken up a big part of that? Or is it...

I know you love the question, Gerhard. I know you love it. I mean we have been transparent with the market on a number of our customers. So we've had poolside AI. They were on stage with us at the GTC in March. We've had NUAI on stage with us on multiple occasions. And we also had Zain who is a legal copilot in Europe for legal references. The majority of our larger customers tend to be under NDA because they are developing proprietary software or other technologies so we do not share their names, which is pretty standard with the market, but I hope that that helps.

Okay. Second question. Could you tell me again how much you made in cash from asset sales year-to-date? And how much you expect to get for the A100? Are you planning to sell all A100s by the end of this year?

So year-to-date, we've monetized about EUR 70 million worth of assets that we were able to sell into the market. And the A100s, we obviously don't want to give too much away, but it's somewhere between EUR 20 million to EUR 30 million is likely to be raised in Q4 on the A100s for all 2,500 units.

Okay. My last question is on mining. It has been kind of said what -- are you now fully ramped, if you like, or fully set up for Q4? And what is kind of the Q4 revenue range that you expect?

I'm glad you asked this question because I realized I didn't actually include it as part of the question earlier on around Q4 guidance. So the previous numbers I gave were around the cloud business. On mining, we are expecting to see revenue step up again between Q3 and Q4. I think the key driver of that is our uptime. So as I said in my prepared remarks, we were at 66% uptime, which was a big drop back from where we were across the first half at around 88%. That's because we were bringing our new facilities onstream in Texas, in particular, and we ramped up the use of that technology quite slowly across Q3. So as we go into Q4, I'm expecting to be back in the 80s, high 80s at least for uptime, which will drive a step up in revenue and in terms of coin production. Plus also our share of the overall market is expected to increase a little bit again between Q3 and Q4 as a bit more of our infrastructure comes on stream. As Rosanne said we were at 7.9EH/s by the end of this year. So an increase in the share of the block reward, increased uptime drives a significant step-up in coin production. So revenue from mining should get somewhere between EUR 15 million to EUR 20 million, that's sort of a range probably in Q4 or subject, of course, to bitcoin price, which we, at the moment, are just forecasting flat between Q3 and Q4.

Our next question is from Tim Wunderlich with HAIB.

I think in Q3, if I did the math correctly, AI had a 50% adjusted EBITDA margin. Do you still expect this to move up to 70%, which I think is your operating model. So that would be my first question. And then the second, just a reminder, the difference between adjusted EBITDA and reported EBITDA because you're talking about adjusted EBITDA in Q3. Could you just remind us, I mean do you expect any material difference between the two? And then finally, on bitcoin mining, I mean, you gave us some color on what you expect in terms of revenue going into Q4. And I think if I understood you correctly, you said mining was breakeven in Q3. So with this slightly higher revenues and the higher uptime in Q4, would you expect mining to be profitable in Q4?

Tim, I'll take all 3 of those. Last one first, yes, we would expect the Peak Mining operations to be profitable again in Q4, as you say, driven mainly down to improvements in uptime. And that's it on that really. In terms of the cloud margins, you're right, that is around about the margin that we delivered from the cloud business, 52% there or thereabouts in Q3. That will increase absolutely as we see more of this capacity come on stream and drive more scale out of the infrastructure that we are deploying as we've sort of been alluding to, we've done quite a lot of testing work with our customers. That comes at cost to Northern Data investment in the future. So as we move through Q4, but more importantly, into 2025, and we're humming on all cylinders, we will see the adjusted EBITDA margin from the cloud business move towards the numbers that you talk about. Sort of we back solve contracts to be north of 70% adjusted EBITDA margin, but there's obviously central overhead costs that need to come off that number. So as we go into next year, the guidance is overall group adjusted margins of sort of 55% to 60%, which has delivered group adjusted EBITDA margins. The difference at the moment between adjusted EBITDA and EBITDA is only the charge for share-based payments, which we have forecast at EUR 20 million for the year. That's EUR 5 million per quarter. So EBITDA would be EUR 5 million less at EUR 21 million for Q3 and that's -- yes, that's the only adjustment we're seeing. There might be more bits and pieces, but materially, it's EUR 5 million a quarter from share-based payments.

That's very helpful. And maybe just one follow-up regarding the monetization or utilization of the H100. You said 12,000 were monetized in Q3 on average. And what is your best guess, what should we model in Q4?

So I think you could get closer to the 20,000 mark across Q4. As Rosanne said, we have in hand all of the units now, and we're getting good at deploying. So maybe not the full 20,000, but certainly 18,000, 19,000 is probably the number that you should be putting into your models. And then the only variable of that, as I said earlier in the remarks, is sell-through rate really clients onboarded. So that number is an estimate at the moment to deliver the overall revenue numbers that we gave you.

Our next question is from Kingsley Crane with Canaccord Genuity.

Just one for me. So your relationship with NVIDIA is clearly a major asset, and it's encouraging to hear about your plans for Blackwell next year. Just curious to hear your thoughts on other players potentially becoming more relevant in the market as GPU providers? How agnostic is your architecture to support potential advancements? And then could that help lower costs, potentially improve your cost structure?

Kingsley, thank you for joining. We've been transparent, I think, right throughout that we look at other technology and we test that technology. We're currently testing AMD's 300X with our partner HP, and I think that it's inevitable that there will be shifts in the market and new technology will come to bear. We're also seeing consolidation with some of the announcements that Intel has made. And so there will be challenges in the market. I think that our relationship with NVIDIA has meant that we have had access to this very premium technology. We design alongside them. They support all of our initiatives. And that, as you say, has been very, very valuable. But of course, there is a broader market and all of our peers and the sort of broader comps that you see in the U.S. also look at other technology. And as a company with innovation bravery as is part of our ethos, and looking at that technology and seeing how it can contribute to the overall AI ecosystem will be a natural part of our evolution. And I think in terms of your question on cost, I mean, the hardware that we see in the market more generally and the associated infrastructure tends to be reasonably comparative in price, and there's anything between 10% and 20% between that. So it will really depend, I guess, on how we configure that or the customer requirements and whether or not the associated architecture is a nonnegotiable like it is with NVIDIA because that will allow us to be flexible on price.

Thank you. And that concludes the Q&A session. I will now hand back to Jose for any closing remarks.

Thank you so much. I've seen there's a couple of questions on the webcast, which I will address directly as we're running out of time. I just wanted to thank everyone for joining us today. And again, if you have any further questions, please don't hesitate to reach out directly to the IR team at Northern Data. Thank you so much.