ALP

Before we begin, I'd like to remind everyone that today's call may contain forward-looking statements within the meaning of the Private Securities Litigation Reform Act of 1995 and applicable securities laws in Canada and the British Virgin Islands. These statements are based on management's current expectations and assumptions and are subject to risks and uncertainties that could cause actual results to differ materially from those expressed or implied, including the risk factors described in the company's annual report on Form 20-F for the fiscal year ended March 31st, 2026, and in the company's other filings with the SEC. The company undertakes no obligation to update these statements except as required by law. During today's call, management may refer to contracted revenue, projected revenue, qualified pipeline, and other company estimates.

These figures are estimates only and are not a guarantee of future results and should be read together with the risk factors in our filings. With that, I'll turn the call over to Brittany Kaiser.

Thank you, John, and thank you all for joining us. This is our first earnings call as Alpha Compute Corp, I want to start by explaining why that matters and what we are actually building here. A year ago, this company had no compute infrastructure, no confidential computing platform, no revenue, and no presence in the AI industry. Today, we own and operate GPU infrastructure in two countries, we have our first enterprise customer live and generating revenue, we have brought more than 120 million users into our ecosystem through the GAMEE acquisition. I'm not saying that to suggest we have arrived. I'm saying it because the pace of that change is the context every investor needs in order to correctly read our fiscal 2026 results, which do, on their own, tell only part of the story. The thesis behind Alpha Compute is straightforward.

Artificial intelligence today runs largely on infrastructure controlled by a small number of hyperscale cloud providers. For regulated industries, healthcare, financial services, government, defense, that arrangement is a real problem. These institutions cannot put their most sensitive data and models onto shared infrastructure they do not fully control and cannot independently verify. Commodity GPU cloud providers do not solve this either, because they compete on price and scale, not on data isolation. Our answer is to own the hardware ourselves and to build privacy into it at the chip level through confidential computing and trusted execution environments, so that even we, the infrastructure operator, cannot see what is running on our own GPUs. That is a different kind of guarantee than a contract or a privacy policy. It is enforced by the code that controls the hardware itself.

We believe that is the layer this industry is missing, and we believe owning it rather than renting it is what lets us build a lasting business on top of it, from confidential inference and training to the applications that run natively on our network. I want to be direct about where we are on our journey. We are an early-stage company. Our fiscal 2026 results reflect a company that spent a year in transition, first taking over a biotech Nasdaq-listed company that was $4 million in unpaid debts and ran out of money. Then through the digital asset phase with Telegram and Toncoin, which leads us to the AI confidential compute pivot due to our first client being Telegram's Cocoon AI LLM. In the final months of the fiscal year ending March 31st, 2026, we began the migration towards the compute business we operate today.

Wes is going to walk through the fiscal year 2026 numbers shortly, I would ask you to review them in context. What you are looking at is closer to a snapshot taken mid-pivot than a picture of an operating business. Almost none of that revenue that we have since, put under contract, appeared in those results. I will walk through our progress over the past several months, which demonstrates our ability to execute. In August of last year, I joined this company as Chief Executive Officer, alongside Enzo Villani as Executive Chairman and CIO, and Yury Mitin as Chief Business Development Officer. Our mandate to build a digital asset treasury firm that was aligned with Telegram and investing in its ecosystem with over 1 billion users backed by their Toncoin.

We always planned to be an ecosystem company supporting what we saw as the best case for global private computing and messaging with a global tech leader. The following month, we closed an initial financing of approximately $71 million and began operating as AlphaTON Capital. That phase gave us capital as an early foothold in a large developer and user community, but it became the bridge to the larger, more important mission, providing AI confidential compute that couldn't be compromised. Starting in the fourth quarter of last year, we started building a GPU-as-a-Service business focused on AI inference and modeling, procuring GPUs, securing data center capacity, and developing the firmware and software stacks needed to run confidential compute workloads. In February of this year, we secured rights to our first large-scale deployment, 504 NVIDIA B200 GPUs under a 24-month infrastructure agreement housed in a hydroelectric-powered data center in Canada.

We call that cluster ALPHA-01. In April, we completed our rebrand to Alpha Compute Corp. and changed our ticker to ALP. This was not a marketing exercise. It reflected a genuine change in what our company does. Since then, we've been organizing the funding and procurement of additional clusters, ALPHA-2, ALPHA-3, and ALPHA-4. We've also hired top engineers and leadership who have led multimillion-dollar units at companies like Microsoft, IBM, the U.S. Air Force, Google Cloud, HPE, Deloitte, U.S. State Department, CoreWeave, and HIVE Technologies. I want to address our digital asset history directly because I know some of you are still asking about it. We are no longer a digital asset treasury company. We never really were only a DAPT. We do not any longer hold a strategic TON treasury, now Gram, and building one is not part of our forward strategy.

The digital asset positions remaining on our balance sheet are legacy holdings tied to the September 2025 financing. They are in the process of being wound down and returned to investors under put rights granted as a part of that transaction. Our capital, our team's focus, our infrastructure are committed to the AI compute business. If there is one thing I want you to take away from this call about who we are today, it is that. On the commercial side, May was a pivotal month. On May 12, we announced our first enterprise offtake agreement, a two-year, $32.2 million contract with a leading frontier AI research laboratory for dedicated use of the ALPHA-01 cluster. That agreement carries $16.1 million in annual contracted revenue and included an upfront payment of $7.5 million, which we have received.

ALPHA-01 has since went fully live for that customer. It's currently successfully processing AI workloads. To put that in perspective, in the first calendar quarter of this year, before that contract, our compute business was generating roughly $30,000 in quarterly revenue. We went from that to $16.1 million in annual contracted revenue within a few months. Essentially all of it sits outside the fiscal year we are reporting today. Later in May, we closed our acquisition of a 60% controlling interest in GAMEE, a gaming and digital rewards platform with 120 million registered users across Telegram and mobile. GAMEE generated approximately $3.5 million in revenue in 2025, and its revenue has continued to grow this year. We did not acquire GAMEE primarily for its current revenue.

We acquired it because it gives our infrastructure a built-in large-scale source of application demand, which is the model I'm about to describe in more detail. For GPU deployments, our leverage strategy for our deployments secures GPU hardware as collateral rather than Alpha Compute's balance sheet. That structure lets us keep scaling our fleet without relying primarily on issuing new equity. Our second cluster, ALPHA-02, 576 NVIDIA B300 GPUs in a hydroelectric-powered facility in Sweden, is under construction and targeted to go live in the third quarter of this calendar year. Once that is online, we will operate more than 1,080 Blackwell generation GPUs across two continents.

Before I hand this to Wes, I want to spend a few minutes on the commercial opportunity behind these numbers because one contract does not make a business. I do not want anyone on this call to think that is all we have. Beyond our first anchor customer, management's assessment of our qualified pipeline, meaning opportunities we consider credible and are actively working across AI research labs, sovereign entities, and enterprise customers currently exceeds $200 million. I want to be careful with that figure. It is a company estimate of qualified opportunities, not signed revenue. Converting pipeline into contracts takes underwriting, negotiation, in some cases, additional financing on our end to fund the underlying hardware. It tells you that our first contract came out of an active commercial process that continues to run, not an isolated event.

One opportunity already in progress is a partnership with Vertical Data, which we expect, subject to closing conditions, to contribute in the range of $43 million in additional financing currently under review with the non-recourse lender tied to our procurement of B300 capacity. We expect this deployment in Q3 2026. I also want to explain how we think about this pipeline strategically because it connects directly to why we built the company the way we did. We describe Alpha Compute as vertically integrated, and we mean that specifically. At the base is the hardware itself, including the ALPHA-01 GPU cluster we lease, which is now fully deployed, and the ALPHA-02 cluster currently under construction, which we fully own and expect to deploy in the near term.

On top of that sits our confidential compute network, the software and firmware layer built with Telegram's Cocoon protocol and our own Shroud developer access point that turns raw hardware into a privacy-guaranteed service. On top of that sits applications, GAMEE today and others over time, that consume our own compute capacity and generate a second stream of revenue on the same underlying assets. We are not dependent on any single layer to make the model work, and growth at the application layer reinforces utilization at the infrastructure layer underneath it. I would also point you to the market context here because it explains both the size of the opportunity and what we are up against. I would encourage anyone on this call to read the company's own market intelligence newsletter published last month for the full picture.

The large hyperscalers are now spending in the range of $650 billion-$725 billion a year on AI infrastructure, up sharply from last year, and each of them has said publicly that they are constrained by power and hardware availability, not by demand. At the same time, the debt markets have started treating AI infrastructure as a real asset class with billions of dollars in investment-grade, non-recourse project financing closing across the sector in just the last few months. That is the same type of tool we use for our own $31.9 million facility at a much smaller scale. That backdrop cuts both ways for a company our size. It supports the demand side of our thesis. Compute is scarce, and enterprises need privacy guarantees that hyperscalers are not built to offer.

But it also means we are competing for hardware allocation, data center capacity, and capital against much larger, better-capitalized operators, some of whom are now financing multibillion-dollar campuses and against a next hardware generation, NVIDIA's Vera Rubin platform, that will eventually reset the competitive baseline again. We do not yet have the scale of a CoreWeave or an IREN, and I'm not going to suggest otherwise on this call. Our strategy is to compete on a specific defensible niche, hardware-enforced confidential compute rather than on scale alone, and to use the same non-recourse financing tools the rest of the sector is using to grow that niche without overloading our own balance sheet. With that context, I will turn it over to Wes to walk through the specifics of our fiscal 2026 results.

Thanks, Brittany, and good morning, everyone. I'll keep this section brief, both because the numbers are relatively straightforward and because, as Brittany said, this filing captures a company in transition rather than the business we operate today. For the fiscal year ended March 31, 2026, we reported a net loss of approximately $38.6 million, compared with a net loss of $6.8 million in the prior fiscal year. That increase reflects the combined effect of our transition away from the legacy immuno-oncology business, our digital asset treasury activity for part of the year, and the initial cost of standing up our GPU infrastructure, all of which are detailed in the notes to our consolidated financial statements. More than half of that loss is made up of one-time charges that will not apply to our compute business on a go-forward basis.

For example, losses on the fair value of digital assets, impairment of the legacy biotech investment in Compedica, legal costs associated with the pivot from the legacy biotech business to digital assets, and ultimately now to compute infrastructure. On the balance sheet as of March 31st, 2026, we had cash and cash equivalents of approximately $700,000 and total current liabilities of approximately $13 million. Based on those end-of-year figures, we have included a going concern disclosure in our financial filing, as substantial doubt exists about our ability to continue operating without additional capital or revenue as of fiscal year-end. Fortunately, the balance sheet as of March 31st does not reflect where the company stands today.

Our filing includes a subsequent events review through July 13th, 2026, as of that date, our cash and cash equivalents were approximately $10.3 million, which reflects, among other things, the $7.5 million upfront payment received from our anchor compute customer, the effect of closing the GAMEE acquisition, and additional cash raised via share sales under our ATM program. Management's plan for addressing the going concern conditions beyond this is described in our filing and is built around converting the contracted revenue from compute contracts into cash, together with closing additional high-margin deals from our compute pipeline as we scale. I also want to clarify a few points on the revenue timing, because I think this is the part that is easiest to misread.

None of our $16.1 million in annual contracted revenue from the anchor compute customer we spoke of is reflected in our March 31 filing, none of GAMEE's revenue is reflected in the fiscal year we just reported either, because our compute contract was signed after fiscal year-end, acquisition of GAMEE was also after fiscal year-end. That sits entirely in our current fiscal year, which began April 1st of this year. Based on those contracts and the acquisition of GAMEE, we are estimating an annualized revenue run-rate in the range of $21 million-$23 million. A significant increase over our prior fiscal year reporting. I do want to underscore that this is a company estimate, not a guarantee of any kind. It is subject to the same execution and market risks that we describe in our filings.

I do think it illustrates the company's growth curve. This is a business that's moving from roughly $30,000 in quarterly revenue to a multimillion-dollar annualized run-rate within just two quarters. With a further $200 million of qualified pipeline of deals behind it, which Brittany described just a moment ago. With that, I'll hand it then back to Brittany to talk about where do we go from here?

Thank you, Wes. I want to be clear about where we are. This is still early. We have one enterprise customer live on ALPHA-01. We have one more cluster under construction. Our addressable market is much larger than our current size, our balance sheet needs to be strengthened. Anyone telling you otherwise is not giving you the full picture, that is not how we intend to run this company. I also want you to see what that early stage looks like from the inside, because I do not think it looks like a company standing still.

In a relatively short period, we have signed our first enterprise contract, brought a cluster online, closed an acquisition that added more than 120 million users to our ecosystem, built a qualified pipeline management believes exceeds $200 million in annual revenue, and put a non-recourse financing structure in place to keep scaling without diluting the shareholders who have stayed with us since the beginning. That is what execution looks like at this stage of a company's life, before the scale, while the thesis is still being proven in the market. The backdrop has not shifted in our favor accidentally. Demand for AI compute continues to outstrip supply, and the hardware generation we have built on, NVIDIA's Blackwell architecture, is the one that frontier research labs and serious enterprise customers are actively seeking.

At the same time, the regulatory environment around data privacy is becoming more demanding, not less, with frameworks including the EU AI Act moving into force this year. Both trends point in the same direction. Compute that carries a verifiable hardware-level privacy guarantee should command a premium over compute that does not. That is the position we are building toward, one contract and one cluster at a time, on the vertically integrated model I described earlier. Here's what I would ask you to watch for over the next two quarters. First, additional enterprise contracts beyond our first customer, including the vertical data opportunity I described earlier, because a second and third signed agreement will tell you more about repeatability than any single deal can. Second, ALPHA-02 coming online in Sweden in the third quarter, which roughly doubles our operating fleet.

Third, the pace at which our contracted revenue converts into reported revenue and cash quarter-over-quarter. Fourth, continued transparency from this management team, because we know that trust with this shareholder base has to be rebuilt through consistent disclosure over time, not through a single call. We did not take this on because it was easy. We took it on because we believe the next generation of AI infrastructure needs to be built differently. With the hardware itself doing the work of protecting the people and institutions that rely on it. We are a small company today with a great deal left to prove. We believe we are building the right thing, and we intend to keep demonstrating that quarter-by-quarter, rather than asking investors to take it on faith. Thank you for your time today and for your continued interest in Alpha Compute.

This concludes our prepared remarks for today. We are now going to take your questions. Please submit your questions at any time using the Q&A component on the webcast interface. Any questions we do not answer live today, we will answer after the fact directly. We will take a moment for the Q&A roster to be compiled. All right. Our first question. The company has changed its name, ticker, and strategic focus twice in under a year. What is the clearest way for investors to describe what Alpha Compute is today and what it is not?

Alpha Compute is an AI infrastructure company focused on providing GPU-as-a-Service to our enterprise, government, and consumer clients. We are no longer a digital asset treasury company. We are not a biotech company. We are focused on fully vertically integrated AI infrastructure that can provide confidential computing and privacy-centric AI to enterprise, government, and individuals around the world. Alpha Compute is focused right now on GPU-as-a-Service in order to build up what is now a $73 million balance sheet and growing. As we roll out additional clusters, having already had ALPHA-01 go live for our first enterprise client, we already have ALPHA-02 cluster, ALPHA-03, and ALPHA-04 in progress. As we build up our balance sheet and continue to close enterprise deals, we are growing hand over fist and are very excited to not just be GPU-as-a-Service in the future.

We believe that investing in real estate, in long-term power contracts, in mineral rights, and in the ability to own our own energy and power generation to be our long-term play. We believe that concentrating on being fully vertically integrated all the way from the land, real estate, mineral rights, and energy, to the data centers, to the hardware, meaning GPUs, servers, networking, storage, cables, and on top of that, the firmware and software services that allow our clients to run their AI workloads, gives us economies of scale and makes us not just more efficient, but more sustainable. We're really excited to be able to vertically integrate as we grow, and you'll start to see higher profitability margins as we attempt that vertical integration step by step.

Right now, we are hardware, firmware, and software, not just through the services that we provide, but through our strategic M&A acquisitions, like our acquisition of GAMEE and its 120 million users. For us, that's very exciting. For our shareholders, it should be even more exciting because our margins get better as we vertically integrate throughout the entire AI stack.

Okay. Our next question. The fiscal 2026 results reflect the company largely in transition. Now that the company is no longer structured as a digital asset treasury and is fully focused on the AI infrastructure opportunity, when can we expect to see reported results which would fully reflect the AI infrastructure business rather than the prior structure?

In order to address this question appropriately, I think it would be helpful to start from the very beginning, when I became CEO of this company last August, nearly one year ago, where I was appointed CEO in order to raise a pipe to create a new business line in the company. I was added to a biotech company that was in debt and had failed to produce revenue, and I raised a $71 million financing and hired in a team to help me implement a digital asset treasury strategy focused on Telegram, one of the world's largest companies by users, with over a billion monthly active users, and fully focused on privacy-centric technology, including Web3, where it has its TON token integrated into the back end of the platform for hundreds of millions of users around the world.

We were focused on this ecosystem specifically because it allowed us to deploy privacy-centric technologies to over a billion people. Our goal was always to be an infrastructure company where we would be able to invest in the ecosystem and get access to privacy technologies and give that access to people and enterprises. A few weeks after we transitioned to being AlphaTON Capital at the end of last September, where we were focused on Telegram and the TON ecosystem, we were given the launch partnership for Telegram's Cocoon AI, which is the confidential computing AI inference network that is integrated throughout the Telegram platform for use for its a billion users. As you may know, Telegram is one of the most popular platforms in the world for AI inference and AI agents.

Enterprises and individuals are using that platform in order to test out all of their new AI architecture. It was very exciting to be given this launch partnership, and we started purchasing GPUs for our balance sheet and to have as testbeds so that we could figure out how to get a confidential computing network live and functioning for Telegram. We were able to undertake that process throughout October and November of last year. At the end of November 2025, we launched, in partnership with Telegram, the Cocoon AI network live on our GPUs. We spent a few more months upgrading that architecture, working on the code for both the firmware and the software layers of that stack so that users on Telegram would be able to take advantage of the Cocoon AI network.

As we started putting this out in press releases, we realized that there was a huge demand for confidential computing, and we started receiving calls and organic leads coming from all over the world, asking if we would be able to do this for them as well. We started working on particular lead generation, and we realized that Telegram would not be our only client, but we would have enterprise and government and even individual AI developer clients if we were to stand up GPUs for other organizations. We started working on our strategy to not just be servicing Telegram for their GPU needs, but to look at other enterprise and government and individual clients where we could stand up bespoke GPU clusters for their AI inference and training.

This was an organic transition, which led to a complete pivot to Alpha Compute Corp, which we renamed from AlphaTON Capital, showing that we are focused completely on AI compute infrastructure and not just on providing that to Telegram. Since then, we have undertaken financing through our ATM. We have unwound our treasury strategy, giving back investors their original cryptocurrency contributions to our treasury, and we have made the full exit of that business public information as of last week. We are now fully focused on AI infrastructure, not just GPU-as-a-Service, but full vertical integration from real estate to energy to data centers to hardware and the firmware and software stacks that make that AI infrastructure accessible. We expect that we will start reporting results more often.

Right now, as a BVI-based foreign private issuer on Nasdaq, we are a semi-annual reporter, meaning that although right now we are presenting our March 31st results, our next reporting would be after the September 30th period, from March 31st to September 30th. It would be shortly thereafter where we would make an official semi-annual filing. We are assessing this and looking to start reporting more often, perhaps quarterly or even some public results monthly as we close our next few enterprise deals.

Okay. Our next question is, what is the expected timeline for deployment of the ALPHA-02 cluster, and what other near-term catalysts should we be on the lookout for?

Our second enterprise-grade cluster, ALPHA-02, which is currently in construction, due to be deployed in Q3 2026 in our 100% hydroelectric data center in Sweden. Currently, as I said, under construction, our deployment and uptime is slated for later in Q3. Other near-term catalysts that you should be on the lookout for are ALPHA-03 and ALPHA-04, our next two enterprise-grade clusters, which are already in process. As we have already disclosed, management believes we have a $200 million annual revenue pipeline that is qualified and in process. As these next contracts get signed and announced, you will see continued momentum from our growth, which doubles every time we close a new one of these deals, approximately. For us, that's very exciting because the more deals we're closing, the more organic leads we have coming in. Our pipeline is growing.

Our interest is growing from our clients, both in the enterprise and government space. We are very excited to also be looking at fuller vertical integration opportunities as I've said, investment in data centers, land, energy rights, and long-term power contracts.

Okay, our next question. How does Alpha Compute's confidential compute offering compare to confidential computing features now offered by hyperscalers and larger Neoclouds? How do customers independently verify the hardware-level privacy claims being made?

Alpha Compute's confidential compute offering is not just GPUs-as-a-service, but also the firmware and software layered on top of our hardware that allows our clients to have fully encrypted AI workloads. What I mean by that is from their systems down to the hardware in our data centers, there is full end-to-end encryption where no third parties can get access to any of our customers' data. Not even we, NVIDIA, Supermicro, or Dell, any of the organizations involved in the hardware, firmware, and software stack, none of us can get access to the data, and that can be mathematically proven through something called an attestation layer, where you can mathematically prove that it is impossible for any third-parties to get access to the data.

There is zero data sharing, and that also is exciting for our clients because it allows them to be fully compliant with all data protection laws across borders. That independent, verifiable, hardware-level privacy is not just a claim, it's a mathematical proof. For us, we use a variety of different attestation layers, and our clients can even choose their own type of attestation layer. It is a third-party layer that allows our clients to have that provability and not just a promise or a contract from us. The regulatory environment around data privacy also is a huge tailwind to our customer demand. We see that regulated industries from financial services and banking to healthcare, to government and defense, all need this type of hardware-level privacy, and their demand for confidential compute is soaring.

As we start to see more data protection and privacy laws go into place around the world, new AI data protection laws, specifically, confidential computing architecture allows these organizations to be automatically compliant, which is incredibly exciting and also helps with a lot of organizations who have said that they would like to implement AI but have not been able to because they cannot get approvals through their compliance teams. Compliance teams, once they understand what confidential computing actually is, are now saying, okay, you can demand confidential computing architecture is a part of your compute power and a part of that stack, and then you're able to use AI safely within an organization that is very sensitive data sets. For us, that really gives us that niche.

It gives us that ability to compete in the space, because we are able to service our clients in a way that a lot of the hyperscalers are not offering. Hyperscalers are often still having their clients' data go into cloud environments where there is data sharing, and even if it is confidential or in a trusted execution environment, there are still third-party services that are involved in third-party data sharing. Because we are giving our clients direct access to our hardware, even at the bare metal level if that's what they want, they can verify that none of their data ever goes through any third-party services at all through the entire process, creating that trustability that we seek to provide to our clients around the world.

Okay, our next question. Please break down the targeted margin profile associated with the contracts being pursued, and what assumptions drive that target?

The margin profile depends a bit on which of the type of contracts we're talking about. So far, that falls into two main buckets where either we're acquiring a lease interest in a cluster of servers like we did in our first deployment, or whether we're actually acquiring the servers, deploying them at a co-location, then similarly finding an offtaker for the compute from those servers. In both cases, target margins tend to be in 20%-25% range, tending to be a little bit higher for acquisitions, but there's a lot of variables that go into it. On the lease side, if we're just talking about bare metal, at times we're able to, with our network, simply find cluster of servers to lease and then re-lease them at a higher margin.

I hesitate to say arbitrage because that discounts that there is execution risk and these are more complicated than a simple arm trade. We can at least lock in at a high level a higher offtake rate than we are paying at the lease rate. That is obviously beneficial. In the case of the acquisition of servers, it's certainly a lot more variables that go into it. A lot of it comes down to how attractive financing, because that's a large cost in the early few years. We're finding our initial deals in the 25%-30% annual return range. I'm actually very excited about that because since a lot of that's driven by debt financing costs, that's something as we scale up and as we grow as a company and have more deployments under our belt, those financing costs will be significantly more attractive.

In fact, we're already seeing that in some potential upcoming deals where we might be getting even more attractive terms than our first deployment that we've disclosed. More variability in acquisition, but certainly more upside. In both cases, 20%-25% is, I would say, conservatively what we're looking for in a margin on these deals.

Our next question. Can you walk us through the revenue generation model at a high level?

Alpha Compute's revenue generation model is based on being a hardware landlord and leasing out that hardware to an end client. To really take you through every part of this, we are really purchasing hardware and finding a home for it to live, which is in a data center. At the moment, we're mostly using co-location facilities where we are signing a long-term contract to do a power offtake, so a long-term power contract and managed services, and renting the space within a data center where we are paying for the hardware build-out. For the racks and the cables in order to house the servers that we send to that location.

The component parts that we are dealing with as a provider is to finance and pay for a long-term build-out of a data center, which is a construction project, and investing in the hardware and the time and the managed services to get that set up, so that when the servers are sent there, that they can be wired up. That includes everything from storage, switches, cables, networking, and the servers themselves, of course. In order to purchase all of that hardware, we've put down a down payment. We've secured those in a non-recourse loan SPV, so that we have lenders usually giving us 70%-80% LTV based off of our 20%-30% deposit for equity on those machines, and then we are deploying them to those data centers.

We're plugging in everything from our equity contribution to the interest rate on our non-recourse loans, to the long-term power contracts, managed services, and data center build-outs. We're putting that all together as the hardware landlord, then we are coming up with a range of leasing prices for our end client so that we can undertake all of this while still being profitable. The way that we usually figure that out is based on an hourly rate for the compute power, depending on what type of GPU chips we are buying. Right now we have B200s and B300s, which are NVIDIA Blackwell generation chips. Those are the chips that are in the servers that we are purchasing and deploying that level of compute power to our end clients.

Our clients are usually negotiating a two-, three-, four-, or five-year contract where they're paying us an hourly rate, guaranteeing 100% uptime. We get paid on a monthly basis for them to use it 100%, even if they are not fully using it at 100%. For us, that means it is a very low-risk, long-term contract, which locks in our hardware for a huge part of the duration of the life of those machines. These machines usually live for about five years. When we can secure five-year contracts, that's obviously preferable because that's considered the entire lifetime of the machines. That's really the way that we put together our compute capacity prices and really think of it in the way that you lease a car or you lease a house. We're the landlord.

We take care of all of the bills and the maintenance, and you're just the one that is using the end asset. That's really the top-line overview of how we undertake that modeling and, in the end, how we make money. What is the moat here and why everyone isn't in this business because it is a very profitable and in the end, lower risk business, closer to real estate, less like software, is that there's a lot of market supply and demand dynamics in this business. Obviously there's changing power costs, there's changing demand on the managed services, there's changing demand on data center space and how many megawatts, or for smaller deployments, even kilowatts are available. Gigawatts for larger hyperscaler deployments.

Getting access to those long-term power contracts and those skilled managed services of the individuals that know how to do the repair and the uptime and the management of the machines, to actually getting allocations of these chips from NVIDIA, buying hardware from big companies like Supermicro or Dell. On top of that, layering our software and firmware services for enterprises that need software and firmware stacks that don't come with their own. That's an extra margin on top of that. All of those products and services have market fluctuations and sometimes are very scarce and hard to get and have long wait times, sometimes months and months ahead of time that you have to place orders in order to make sure that things are being delivered on time for your clients.

This type of dynamic pricing environment means that it does take a very serious complex set of skills, relationships, procurement partnerships in order to put all of these deals together. That's why there's not so many compute companies in the market and why our typical compute capacity contracts are very large and very long-term because our clients really want to lock in these types of deals years in the future because they take a long time to negotiate. Once they close, they are worth a lot of money. As you've seen in our press releases and disclosures and 6-Ks that we've already put out, these deals are usually between $20 million-$100 million deals every single deal. You're looking at between a minimum of $16 million-$20 million of revenue annually as the minimum size of a contract.

I think it's really important to understand that the large scale of these deals and the complexity of these deals does mean that when you do finally close a contract, that is very much worth the time and the investment in getting there. It means that with a handful of deals under our belt, we can be a very large company already.

Okay. Our next question, can you please unpack some of the specifics behind the stated $200 million pipeline? How do you define these pipeline opportunities, and what do you realistically expect for a conversion rate?

To unpack some of the specifics of Alpha Compute's stated $200 million pipeline in annual revenue, I can define a little bit of what those opportunities look like. Most of these are large enterprises. Some of them are frontier research labs, some of them are hyperscalers that are coming to us in order to stand up either their bare metal or their confidential computing deployments. What they are asking us for is to source the hardware, the networking, the storage, to source the data centers, to source the power contracts and the managed services, to put that deal together and to come with an end price of a lease. On top of that, some of these clients need us to do the firmware and software work.

Some of them already have firmware and software stacks that they prefer, we layer that on top of our hardware and provide it to them. As I mentioned, these are usually two- to five-year contracts. These days, we're mostly negotiating four- and five-year contracts as it locks in the use of our hardware at 100% capacity for the majority of the lifetime of the machines. As I mentioned briefly in my former question-answer, each of these contracts is usually a minimum of $16 million-$20 million per year. That usually means at least 500 B200 or B300 NVIDIA Blackwell GPUs that are deployed as an enterprise cluster. This is a smaller enterprise cluster. Of course, some of the inquiries we have are for 2,000, 5,000, 10,000 chips at a time. But the very minimum that we have requests for is 500.

As you have seen, ALPHA-01 is 504 chips, ALPHA-02 is 576 chips, and our demand is getting larger and larger. As an example, ALPHA-01 is a two-year-long contract. That's $16.1 million in annual revenue. It's a two-year contract, so it's a $32.2 million in revenue overall over the two years together. Many of our other contracts, as I said, we're really negotiating four- and five-years at this time. The deals are usually 2x to 2.5x that size or more, because sometimes we're talking about NVIDIA B300 chips instead of B200s, which are newer and more expensive, or GB300s, which are huge stacks, which contain many more GPUs and end up being even more expensive than the B200s or the B300s. These deals are growing in size. They're growing in the amount of demand in our pipeline.

As we said, we believe that this $200 million pipeline, in just a few more deals, ALPHA Clusters 01, 02, 03, and 04 will likely be over $200 million in annual revenue without even thinking about other clusters beyond that. For us, we're very excited about that opportunity, and we believe that there will be a very high conversion rate because we're only considering a qualified pipeline when we're already in end deal negotiations where we have our MSAs and our SLAs and final contract terms that have already been verbally or agreed in writing and are undertaking either final redlines or final pricing negotiations.

Okay. This concludes our Q&A session for today. Thank you all so much for joining us, and we look forward to keeping you updated soon.