액스 컴퓨트(AGPU) 2026년 2분기 실적 발표회: 체결 계약액 32억 달러 돌파

핵심 요약

• 2026년 2분기 매출은 1분기 3만 5,000달러에서 증가한 320만 달러를 기록하며 엑스 컴퓨트(Axe Compute)의 컴퓨팅 매출이 본격 반영된 첫 전체 분기를 기록했습니다. 2분기 컴퓨팅 매출은 전액 액세스(Access) 모델에서 발생했으며, 빌드(Build) 매출은 아직 시작되지 않았습니다.
• 회사는 주로 1,310만 달러의 디지털 자산 손실로 인해 1,720만 달러의 순손실을 기록했다고 발표했습니다. 조정 EBITDA는 약 마이너스 490만 달러였습니다.
• 엑스 컴퓨트는 3분기 첫 6주 동안 추가된 29억 달러를 포함해 2026년 8월까지 누적 계약 가치 기준 32억 달러가 넘는 계약을 체결했습니다.
• 경영진은 연간 환산 매출(annualized run rate)이 2분기 말 3,700만 달러에서 4월 클러스터가 가동됨에 따라 약 1억 3,900만 달러로 증가하고, 2026년 4분기와 2027년 1분기에 걸쳐 체결된 전체 계약 물량이 배포되면 6억 9,600만 달러를 넘어설 것으로 예상하고 있습니다.
• 회사는 클러스터 확장을 위해 3억 1,700만 달러를 초과하는 고객 선수금을 수령했다고 발표했으며, 여러 지역에 걸쳐 55메가와트(MW)를 추가하는 계약을 듀오스 테크놀로지스(Duos Technologies)와 체결했습니다.
• 엑스 컴퓨트는 98개의 기회에 걸친 59억 달러 규모의 적격 파이프라인을 바탕으로 2026년 말까지 20억 달러의 신규 계약 체결을 목표로 하고 있습니다.

주요 재무 데이터

사업 및 영업 성과

엑스 컴퓨트는 두 개의 AI 인프라 사업을 운영합니다. 빌드(Build) 부문은 장기 고객 계약에 따라 전용 GPU 클러스터를 설계, 배포, 소유 및 운영합니다. 액세스(Access) 부문은 여러 지역에 걸쳐 GPU 용량에 대한 신속한 접근을 제공하며 2분기 동안 전체 컴퓨팅 매출을 창출했습니다.

2분기 말까지 누적 계약 금액은 3억 1,700만 달러였습니다. 회사는 7월에 발표된 3건의 빌드 계약을 통해 29억 달러를 추가하여 8월까지의 연초 대비 누적 계약 가치를 32억 달러 이상으로 끌어올렸습니다.

조지아주 콜럼버스 클러스터는 288개의 엔비디아 블랙웰(NVIDIA Blackwell) B300 노드와 고속 스토리지, 오케스트레이션 용량 및 전용 네트워킹을 기반으로 구축되었습니다. 경영진은 이 클러스터가 조만간 가동에 들어가고 이후 초기 규모의 3배로 확장될 것으로 예상된다고 밝혔습니다.

스웨덴 보덴에 건설 중인 별도의 클러스터에는 256개의 B300 노드와 예비품, RoCEv2 네트워킹 및 WEKA가 공급하는 스토리지가 포함됩니다. 경영진은 2026년 말경 가동될 것으로 예상하고 있습니다.

엑스 컴퓨트는 또한 여러 지역에 걸쳐 추가로 55메가와트(MW)를 확보하기 위해 듀오스 테크놀로지스와 계약을 체결했습니다. 양사는 모듈식 데이터센터 건설 방식을 사용할 계획이며, 엑스는 특수목적법인(SPV)을 통해 특정 시설의 소유 및 자금 조달에 참여할 예정입니다.

적격 매출 파이프라인은 98개 기회에 걸쳐 59억 달러 규모였습니다. 파이프라인 가치의 약 3분의 2는 B200, B300 및 GB300 하드웨어를 포함한 블랙웰급 GPU 또는 베라 루빈(Vera Rubin) 시스템과 관련되어 있습니다.

경영진 가이던스

경영진은 액세스 부문의 추가 성장을 포함하여 4월 클러스터가 가동에 들어갈 때 연간 환산 매출이 약 1억 3,900만 달러로 증가할 것으로 예상합니다. 2026년 4분기와 2027년 1분기에 걸쳐 체결된 전체 계약 물량이 배포된 후에는 연간 환산 매출이 6억 9,600만 달러를 넘어설 것으로 회사는 예상하고 있습니다.

경영진은 콜럼버스 사업장이 가동된 후 분기당 약 2,000만 달러에서 2,100만 달러의 매출을 창출할 수 있다고 밝혔습니다.

체결된 빌드 계약의 경우, 회사는 모델링된 매출총이익률 28%~44%, EBITDA 마진 62%~76%를 보여주는 예시적인 정상 상태(steady-state) 수익성을 제시했습니다. 이 수치들은 미래지향적이고 혼합된 추정치이며 정식 가이던스를 구성하지 않습니다.

엑스 컴퓨트의 연말 목표는 2026년 말 이전에 20억 달러의 신규 계약을 체결하는 것입니다. 회사는 또한 배포, 인프라 엔지니어링, 운영 및 고객 지원 부문에 걸쳐 20명 이상의 직원을 채용 중이며, 경영진은 이러한 채용이 이미 체결된 계약에 의해 뒷받침된다고 설명했습니다.

리스크 및 주시해야 할 사항

체결된 계약 가치와 연간 환산 매출은 일반회계기준(GAAP) 매출이라기보다는 영업 지표입니다. 매출 인식은 배포, 고객 승인, 서비스 개시 및 기타 계약 조건에 따라 좌우됩니다.

회사의 계획된 자금 조달 구조는 프로젝트 비용의 약 20%~45%를 충당하는 고객 선수금에 의존하며, 프로젝트 수준 파이낸싱으로 보완됩니다. 경영진은 부외(off-balance-sheet) 프로젝트 파이낸싱을 목표로 설명했지만, 파이낸싱 구조 및 조달 가능성은 여전히 불확실합니다.

디지털 자산 보유는 상당한 실적 변동성을 초래할 수 있습니다. 2분기에 기록된 1,310만 달러의 디지털 자산 손실 중 1,180만 달러는 미실현 비현금 공정가치 평가 손실이었습니다. 조정 EBITDA는 이러한 미실현 요소를 제외하지만 약 130만 달러의 실현 손실은 포함합니다.

엑스 컴퓨트가 계약 체결 단계에서 미국과 유럽 전역의 여러 대형 클러스터를 배포하고 운영하는 단계로 전환함에 따라 실행력 또한 매우 중요합니다.

실적 발표 컨퍼런스 콜 전문

전체 실적 발표 컨퍼런스 콜 녹취록

경영진 발표

Erin McMahon

Good morning, and welcome to Axe Compute's Q2 2026 Earnings Call and Business Update. I'm Erin McMahon, CMO and Head of Investor Relations at Axe Compute. Joining me today are Christopher Miglino, Chief Executive Officer; Jeremy Yaukey-Witter, Chief Financial Officer; and Kyle Okamoto, President. Today, we're hosting the call live from Columbus, Georgia, home to a data center that is the site of our 2K B300 cluster build we announced back in April that is set to go live in the coming weeks.

Before we begin, today's remarks include forward-looking statements as referenced on Slide 2. This presentation contains forward-looking statements within the meaning of safe harbor provisions of the Private Securities Litigation Reform Act of 1995, including statements regarding signed contracted value, anticipated customer prepayments, deployment timing, annualized run rate, expected margins and profitability, financing structures, potential future contract signing and future performance.

These statements are subject to uncertainties that could cause actual results to differ materially, including those described in the Risk Factors section of Axe Compute's most recent annual report on Form 10-K and subsequent quarterly reports on Form 10-Q filed with the U.S. Securities and Exchange Commission. Readers are cautioned not to place undue reliance on forward-looking statements, which speak only as of the date of this presentation. Axe Compute undertakes no obligations to update them, except as required by law.

Financial figures. Quarterly financial figures presented are as reported in the Form 10-Q filed on August 14, 2026. Annualized run rate means annualized monthly recurring revenue upon full deployment of signed contracts. Total contract value, or TCV, is an operating metric representing the aggregate estimated contractual spend under signed customer contracts. ARR and TCV are operating metrics and may not represent revenue recognized in a particular period as separately determined in accordance with U.S. GAAP. Signed contracts are subject to deployment, customer acceptance and other risks described in our SEC filings. Illustrative steady-state economics are modeled derived, blended across signed Build contracts and do not constitute guidance.

Non-GAAP measures. This presentation includes adjusted EBITDA, which is a non-GAAP financial measure. The company defines adjusted EBITDA as net income loss adjusted to exclude interest expense income, net income, tax expense benefit, depreciation and amortization, stock-based compensation expense, fair value adjustments on digital assets. Adjusted EBITDA is not a substitute for net income loss or any other measure of financial performance prepared in accordance with U.S. GAAP and may not be comparable to similarly titled measures used by other companies.

Management believes adjusted EBITDA is useful to investors because it provides supplemental measure of the company's core operating performance by excluding the effects of capital structure decisions such as interest expense and fair value changes related to digital asset holdings, noncash charges such as depreciation, amortization and stock-based compensation and tax impacts that can vary significantly between periods across companies. Management uses adjusted EBITDA to evaluate the company's performance, compare performance across periods and assist in the allocation of resources.

Investors are cautioned that adjusted EBITDA has limitations and is an analytical tool and should not be considered in isolation or as a substitute for analysis of the company's results as reported under U.S. GAAP. A reconciliation of adjusted EBITDA to the most direct comparables U.S. GAAP financial measures is included in this presentation. To the extent that the company provides forward-looking adjusted EBITDA guidance in connection with this release or related earnings call, a reconciliation of such forward-looking non-GAAP measures to the most direct comparable U.S. GAAP measure may not be available without unreasonable effort due to an inherent difficulty in forecasting and quantifying certain amounts, including, but not limited to, fair value adjustments on digital asset holdings, stock-based compensation expense and other noncash or nonrecurring items, the timing and magnitude of which may be significant.

No offer or solicitation. This presentation is for informational purposes only and does not constitute an offer to sell or a solicitation of an offer to buy any securities of Axe Compute or any of its affiliates. No securities are being offered or sold in any jurisdiction where such offer or sale would be unlawful.

Third-party information. Certain information contained in this presentation has been obtained from third-party sources. While Axe Compute believes such information to be reliable, it is not independent -- it has not independently verified the accuracy or completedness of such information and makes no representation or warranty expressed or implied as to its accuracy or completedness. Please review these statements alongside with our SEC filings available via investors.axecompute.com.

With that, I'll hand it to our CEO, Christopher Miglino.

Christopher Miglino

Good morning, everybody. I'm Chris Miglino, and I'm the CEO of Axe Compute. We're here in Georgia. I'm excited to be here. We're getting ready to launch a very large cluster. And for those that are new to the Axe Compute story, Axe Compute is an AI infrastructure platform with 2 growth engines. Build is our primary engine, which is part of what you're looking at here. We design, deploy, own and operate dedicated GPU clusters for enterprises. Then we have Access. Access is the complementary recurring stream, fast access to GPU capacity across many different locations across the world. Think of it like a top of the funnel, where we meet companies that are interested in GPUs immediately and then as they grow, we can help them, we can grow with them.

What the market sees for Axe is a new AI infrastructure company. But what the market doesn't realize is that we've been in this business for the past couple of years, and our sales team has been selling to the offtakers for that entire time. And our supply team has been talking to data centers and engaging with power owners that entire time as well. So while it seems like we're a new kid on the block that has all of a sudden signed all of these transactions, this has been an overnight success that's been in the making for the last 2 years.

Before we get into the quarter, I want to spend a couple of minutes on the market we're operating in because it's the single most important piece of context for everything that Jeremy and Kyle will be walking through later. It's no surprise that we're in the early innings of a generation build-out of compute. You can't turn on the news without seeing some kind of news about data centers, the growth of data centers, the amount of money that's being attributed to data centers.

In fact, McKinsey says that through 2030, they're just going to be spending around $6.7 trillion, which I think is a low number even at this point. Of that, about $5.2 trillion is attributed specifically to AI-capable capacity. So I want to be clear what that second number means. It's not a forecast of AI software revenue for like OpenAI or Anthropic, but AI spending on infrastructure. So it's the capital that's required for the physical, the power, the shells, the racks, the GPUs, and that's the market that we're in. That's the market that Axe Compute is in.

So the second thing that we want to point out is the shape of the demand, not just the size. So customers are not asking for generic cloud. They're asking for dedicated capacity. They don't want to just be in a shared location with a lot of people. They want to own a cluster like the one that you see behind me, and they want to own that and be able to engage with it on their own without anybody else interacting with them. They'll do that on a long-term basis as well. So that will enable us to sign 5- to 10-year transactions.

And they want a partner that can help them grow into that space. They want a partner that can help them acquire all the equipment, plan the network, design the network, deploy the network and work with the appropriate data centers to get the right power and to get the right building ready for their compute needs. So that's where our Build program comes in.

And we help these offtakers do exactly that. We help them do everything that's necessary to get live inside the data center and then we partner with the right data centers to execute. So not all data centers are made alike, and we do what we can to make sure that we have the right partners that are helping our clients. So when you hear the numbers we're about to walk through, more than $3 billion in signed contracts, $696 million in expected annual run rate when we're at full deployment and more than $0.5 billion in customer prepayments, I encourage you to read them against the backdrop of massive demand.

So we're not creating demand. The -- we're converting market where demand structurally exceeds the supply that's out there. So we're signing all these transactions on a take-or-pay basis with multiyear contracts. So with that context, let's turn to what happened this quarter.

We had an amazing quarter. If we could accomplish what we did this quarter, every quarter, everybody on this phone call would be beyond ecstatic. The -- as you can see, we're getting ready to launch the first cluster that we talked about a long time ago. This was the first deal that we told everybody about. This is getting ready to go live, and it's a beautiful piece of art that's there. Many, many hours of work have gone into bringing this to fruition.

But we've really done an amazing job at signing around $3 billion, close to $3 billion worth of additional agreements when we had anticipated that for the quarter, we would be happy to sign an additional $1 billion worth of transactions. So it far exceeded what we had projected. It really did. We really thought that it would be a little bit of a slower ramp. But I'll let -- I'm going to let Kyle talk a little bit more about the pipeline in a little bit, but you'll see that this momentum has not stopped at all.

So now we went from signing all these transactions, doing all these deals, getting all of the ARR set up for next year. So like next year, when these are all deployed, again, $696 million in ARR once they get deployed. Now we're in dual mode. We're in the execution mode. The team is working on putting these together and getting them up and running and managing them, making sure that we have all of the appropriate staff in place to manage these for our clients. So we're really excited about where we sit today from the momentum that we had this quarter.

One question we get a lot all the time basically is how are our margin on these Build projects. So we wanted to give you a slide that would give investors some insight into what that looks like. These are forward-looking, but this is our modeling, it gives the best shot as to where we believe them to be. And we think investors can take a look at these and can help utilize them to help them model going forward.

As you can see, the gross margins are between 28% and 44% and the EBITDA margins are 62% to 76%. So if you apply those against the $3 billion in announced transactions, you can see that the economics that we're going to be experiencing in '27 are significant. The only other question we get more than the question about the margins on these deals that we're doing is how are we going to fund these projects.

So I wanted to walk you through a little bit about how these projects are being funded and how we're working to fund these projects. So you can see that each project has a down payment from the customer. Customer puts anywhere from 20% to 45% of the project cost down upfront. What we then do is we have the ability to go out and seek project financing for that revenue stream.

Now when you have a credit -- A+ S&P credit client and that client is -- we're looking for financing for that revenue stream, there's a lot of lenders that are willing to lend against that, especially those lenders that understand the GPU market, which you can see from the news, there's a lot of them. There's a lot of them that have come up, a lot of them that are executing, a lot of them that are investing into the space.

So our goal is to get that prepayment, find the project financing for each project similar to what happened here, off balance sheet, we don't need to raise capital against it and then execute and deploy the cluster itself. So we think that when people understand that we can get these projects financed in this fashion, that's very favorable for the public company. And in the instances where we are looking to invest equity into these transactions, we could do so at a point when the stock is at a price where we deem it to be a good location to actually do a transaction if we wanted to. Otherwise, we can sit back and do project financings against each of these projects and leveraging the good credit of the clients that are building these things with us.

We have 2 really big announcements today. The first is that we received the first prepayment of $317 million plus for our cluster that we're expanding. And then we've signed an agreement for an additional 55 megawatts over course of a number of different locations with Duos Technologies. So we're excited about both of these things. I think these will give you an idea of where the business is going.

And in order to talk a little bit more about our partnership with Duos, we've asked the CEO of Duos, Doug Recker, to be with us here today so that we can give you a little bit of an insight of what our expansion will look like into that 55 megawatts.

One of the most important things in building data centers is having a good partner. And I'm here with Doug Recker from Duos Technologies, who has been a fantastic partner for us. Doug?

Frank Recker

Thank you for having me.

Christopher Miglino

No, thanks for being here. And I have to say that very glad that we are doing business with Duos. You guys have literally been an amazing partner in this process. So I wanted to thank you for that. I think today, we're excited to announce that we've signed a deal to do an additional 55 megawatts over multiple locations with Duos. So we appreciate that support.

Frank Recker

Great partnership. We're looking forward to it.

Christopher Miglino

I mean if this location where we are today is any indication of how everything else is going to go, I feel very confident in your hands that you will get everything done. So I appreciate your support. I mean what you've accomplished here since we got going is just incredible.

Frank Recker

It's incredible what we've done under, what, 45 days? If you could literally take the camera and walk around the facility, it's amazing how fast we've implemented this project. And it's kind of a testament of how we work as a company, right? We'll tell you we can do it and we'll deliver. Now we're not going to tell you we're going to do 100 megs in 3 weeks, right? We're not going to say that. But as you can see from our product, this is all under 45 days.

Christopher Miglino

Yes. So it's important to, like, I think a lot of people that watch this that are very interested in the data center space, but they don't understand all the components that go in and are involved in this process, right? So there's the data center owner, which is you guys. And you're also the operator of the building.

And then there's all this stuff that we have to get in here, and that's our role, right, is to get this stuff in here, get it up and running, manage it, manage the install, manage the process. But tell the people that are watching what -- how your business works, what you guys do and how that results for you guys because you guys are also a public company. DUOT, Duos Technologies. So explain how that works so that everybody understands.

Frank Recker

Sure. So what's great about our business is we've actually been in the business over 30 years, myself built many data centers, large data centers and even the modular approach, right? So what we're doing now in our partnership with Axe and what our business model is, is basically deploying modularly.

So what we'll do is we'll find a facility like this where there's the building and there's fiber to it, and we'll bring our modular approach to that, which basically is we build our e-houses, our chiller plants, everything modular. We'll build it off-site and deliver it. And then we can plug it in, in a matter of 60 days, 90 days, even quicker sometimes if you look at this environment. So we've really mastered that approach. We've been doing it for about 9 years.

Now we're really just focused on basically the 5- to 25-, 30-megawatt range. We're not going to go out and build 100-megawatt in 1 day. We'll do it modularly, and we'll build with the client, kind of like what we're doing with you.

Christopher Miglino

So -- and you guys, when you built this core here, you got everything ready for the next expansion.

Frank Recker

Exactly right. And that's going back to the modular approach, right? So we built a 10-megawatt, but we do the main infrastructure outside to be able to expand. So then we bring another 10-megawatt in of clusters, and we bring that in modular with our e-houses and our chiller plant and our electrical plant. All that plant and everything is done, and we build it and drop them as we go.

Christopher Miglino

So this is going to be interesting when we go to the expansion that's starting right now is how loud it's going to be in here. It's going to be a whole different world for everybody that's working in here because it is going to be screaming loud in here. And so I think they're going to charge us a little extra for putting that one together.

Frank Recker

But it is just amazing, isn't it just to see the -- a great partnership works in several ways, right? We're the infrastructure side of this house, right? Without the infrastructure, you don't have the operating side. You are the operating partner, and to work together and how we worked in this program together has been amazing. So the communication is key. You guys are masters at what you do, and we're good at what we do. You put those 2 minds together, and it's unbelievable. And hence, see what's going up now.

Christopher Miglino

And I think another exciting thing that we're doing together, and this is -- we put this information out today is our participation with Duos in owning the data centers. So if there's anybody that you want to own a data center with, it's Doug and his team. And so what we're bringing capital through SPV that is being invested into new buildings, new structures where we know that there's power, we know that we can get natural gas if it's going to be off the grid.

And so in these megawatts that we've signed up, we're also participating in the ownership along with Doug. That enables Doug to go a lot further in developing a lot more data centers with less capital, allows us to create an SPV that owns the -- that participates in financing the buildings, so Doug can build more data centers without having to raise a lot of additional capital.

Frank Recker

Equity. Yes. It's a perfect partnership. And what that allows us to do is to grow quickly as well. So what we're also good at, which we need to talk about, is the infrastructure side on the purchasing. So the key to building at this speed is to make sure that you're ordering the right equipment and getting it in on time. And we have a whole other division, the infrastructure division of our business, and that's what we do. So we'll buy our UPSs, our e-houses well in advance. We'll buy our generators, and we basically work them in parallel.

So with the partnership, that allows us to buy the assets now. So when we are ready for the actual build, we do the LEGO blocks immediately, and we're in, in 120 days, 90 days, 60 days depending on what the build looks like. When you come in at the 20-megawatt, 30-megawatt range, which is our sweet spot, the 10- to 30-megawatt, you're actually in a good place. You can deploy quick. You can do it right, you're environmentally friendly and also you're good for the community. And we can move quicker at that scale.

Christopher Miglino

Yes. And it doesn't require this 2-year build.

Frank Recker

That's right.

Christopher Miglino

Like a lot of stuff that we're looking at now will be deployed by the end of the year and into the beginning of the first quarter.

Frank Recker

That's correct. So speed -- what's key for your business and key for our business is revenue, so when we procure this product, we want it installed. We want that thing producing revenue. And this is key to our business and your business. We have to get it up. We need that revenue to come in.

Christopher Miglino

And you've got to -- everybody wants to be in the data center business right now, right? I get calls all the time. People want to be in the data center business. The first thing I say to them is, I know somebody that's in the data center business, and it is not as easy as you think. It is -- this is a very complex building process.

What you -- what I've seen happen here is unbelievable. Again, if you don't know what you're doing, you're losing -- you're losing. It is not going to happen. This is -- if there's any business where you need to know every detail, it's this business.

Frank Recker

Sure. And also in this business, you see this, it looks great, and anybody can put a cabinet in and plug power to it. It's actually after it's up is where the expertise needs to come in. They need to be able to manage the facility, know what you're doing. Anybody can build. It's after the build is what is the most important part of the business.

Christopher Miglino

Well, again, Doug, I really appreciate the partnership. We're excited -- we're excited to build with you, and we're looking forward to the future partnership with you. Thank you very much. Appreciate it.

Frank Recker

Thank you so much.

Christopher Miglino

Now I'd like to turn the call over to Jeremy. Jeremy is going to give us a breakdown of the financials for the quarter. Jeremy, take it away.

Jeremy Yaukey-Witter

Thank you, Chris. I'll start by covering the second quarter results, then we'll walk through the timeline for when the signed book is expected to translate to recognized revenue. Revenue for the second quarter was $3.2 million, our first full quarter of compute revenue, up from $35,000 in Q1. Bear in mind that all of our compute revenue this quarter came from the Access model. Build revenue has not yet started. That begins at go-live.

Net loss was $17.2 million, driven by $13.1 million of losses on digital assets, primarily unrealized fair value changes on our Aethir holdings that flow through the income statement each period as the token's market price moves. We generated $17.4 million positive operating cash flow for the first half of 2026, primarily driven by customer prepayments, which totaled $60.8 million at June 30. Both figures reflect the strength of our take-or-pay prepay first contract structure.

On the balance sheet, we ended the quarter with $21.9 million in cash, up from $6.9 million at the end of Q1. Our digital assets, primarily comprising the Aethir token, together with the current portion of our digital asset receivable, totaled approximately $18.8 million at quarter end.

During the quarter, we also made a number of prepayments to third-party compute suppliers to reserve compute for signed customer contracts spanning terms of up to 3 years, which are reported on the balance sheet as compute prepayments.

Contract liabilities dominate the other side of the balance sheet as of June 30, including $33.6 million expected to be recognized as revenue within 12 months, and another $27.1 million of long-term contract liabilities. As a reminder, these contract liabilities represent prepayments made by customers to us in connection with executed contracts for compute.

That brings me to adjusted EBITDA, which we're introducing this quarter. We believe adjusted EBITDA serves as a meaningful performance measure for investors to evaluate our operating performance and to compare the financial results between periods. I want to be specific about what that metric does and doesn't exclude. Adjusted EBITDA adds back interest, income taxes, depreciation, amortization as well as stock-based compensation, standard EBITDA add-backs.

It also excludes the unrealized noncash fair value adjustment on our digital assets, $11.8 million of the $13.1 million loss on digital assets this quarter. What it does not exclude is the realized portion, about $1.3 million, including ATH we actually used to pay for compute that we sold to customers. On that basis, adjusted EBITDA was approximately negative $4.9 million for the quarter, about $0.9 million of that amount related to our legacy Drug Discovery Service segment.

The takeaway, net loss this quarter is dominated by a noncash change in the fair value of our digital asset holdings, which are subject to notable volatility. Adjusted EBITDA strips out that valuation adjustment and the other add-backs noted, which management believes do not directly reflect our ongoing operating performance.

Now with the second quarter reviewed, let's turn to the quarters ahead and look at when our signed book is expected to begin recognizing revenue. Our business model involves both short- and long-term contracts with customers. Certain contracts, particularly under our Access model, commenced service delivery as quickly as 48 hours after contract execution, while others, particularly under our Build model, don't reach ready for service milestones until about 4 months from contract signing.

To model these contracts together, we utilize operating metrics, including monthly run rate and annualized run rate. A contract only enters our run rate once it's deployed and billing, not when it's signed, and that's the story that this slide tells. Through the end of Q2, we had signed $317 million in total contract value. Our first full quarter of compute services, including the $260 million landmark contract we announced in April.

Since then, in the first half of Q3, that's about 6 weeks, we've added an incremental $2.9 billion comprising the 3 Build contracts announced in July. Year-to-date through August, that's more than $3.2 billion in total contract value in under 8 months. Here's what that means for run rate.

Our exit run rate leaving Q2 was $37 million annualized. As the April cluster goes live in Q3, that climbs to roughly $139 million, including further growth to date under our Access model. And once the full signed book is deployed, spanning Q4 into Q1 of next year, we expect an annualized run rate north of $696 million. Q2 is just the start of the ramp.

Now that I've given you a glimpse of what's to come, Kyle will show you what's actually going to be deployed behind these numbers.

Kyle Okamoto

Signing contracts is only half the story. Delivering them at this scale is the other half. And that's where Axe earns its margin and returns value to shareholders. Every cluster starts with NVIDIA's best silicon, 8 B300 GPUs per node, 288 gigabytes of HBM3e memory on every single one. For our largest builds, we go further, the GB300 NVL72, 72 GPUs and 36 Grace CPUs fused into one rack scale system, sharing that same 288 gigabytes per GPU, times 72.

We rack it in high-density air-cooled, liquid-cooled or direct liquid cooling configurations. 64 to 155 kVAs per rack. That's not office IT, right? This is an industrial-scale AI factor. We connect it with InfiniBand, Quantum-X800 InfiniBand, 6.4 terabits per second per node or 57.6 terabits per second per rack on the GB300s. We feed it with attached storage at real scale. We're talking 20 or 40 petabytes and beyond through great partners like WEKA. And we house all of this in Tier 3 N+1 facilities, 5, 20, 25, 30 megawatts each, built with room to grow into the next contract before the ink is even dry.

Here's the point. It's the whole thesis of our business. At this scale, AI capacity cannot be bought off the shelf. It has to be engineered, or even better, co-engineered with our clients, rack by rack, watt by watt, node by node. That expertise is exactly what our customers are contracting us for, and it's exactly why they come to Axe instead of trying to build it themselves.

And while we're here in Columbus, Georgia, and we're giving you guys a tour of this data center, it's important to note that we're also building other clusters around the world. So we have a few pictures here on our Sweden cluster in Boden. We're very excited about this one. It's with a great partner, and that cluster is already being built right now. We'll have to do our next earning release from Sweden, although it might be a little cold that time of year. But we'll definitely keep the teams updated there.

What this cluster is, is 256 nodes plus spares of -- I'm sorry, B300s, full RoCEv2 on this one as well as a very large storage cluster also provided by our partner, WEKA, with dedicated circuits to data lakes across Europe, dedicated Internet for high-speed inferencing. And it's really a multipurpose cluster that supports both training across that RoCEv2 network fabric as well as high-speed inferencing. So it's a pretty exciting cluster that we expect to go live near the end of this year, and we'll definitely keep the teams updated on how that goes and provide some more videos and pictures as we go along.

Here to give you a live video walk-through of the latest build that Axe Compute is doing for our enterprise clients. We're here in wonderful Columbus, Georgia in a Tier 3 facility with an active ongoing build. So forgive me if you hear some construction noise in the background. There's a lot of folks working on this project to get it live for our client as soon as possible. This is an NVIDIA Blackwell B300 cluster of 288 nodes plus additional shared high-speed storage, CPU nodes for orchestration, Kubernetes, Slurm, et cetera, dedicated networking. So really everything you need for a true AI factory. And we'll get into each of those components throughout.

The real heart of this data center is right here, and these are the NVIDIA Blackwell B300 machines. Inside each of these machines are 8 individual GPUs, which all can speak to each other as if they were one unique computer. And then the really cool part is that each of these servers can all talk to each other across what we call an InfiniBand network, which basically turns each and every single GPU in this facility into one solid computer, right? So we can run very high-speed, very high-scale artificial intelligence use cases and workflows for our clients. So we'll get into a little bit about each one of those components throughout.

But basically, this is how an AI factory is built. These are not normal racks, these are not normal boxes. This is not an IT environment. This takes specialized cooling, specialized air handling, specialized plumbing and electricity work. So it really does take a village to build one of these with both the data centers, with the OEMs, the original equipment manufacturers, the data centers, the electricians, plumbers, it's definitely a very large effort. So we're going to walk through that today.

As I mentioned, all of these computers and chips are all connected together. So the way they do that is each pod or grouping of computers are all connected to an InfiniBand leaf architecture. So it's a spine-leaf architecture. These are the leaf switches that connect each of the pods together. And then these leaf switches ultimately all interconnect into a spine.

And so they are replicated for each pod of GPUs depending on capacity and redundancy purposes. We can fit so many in each of these leafs and spines. And then ultimately, those are all interconnected together. That allows each and every single GPU on this network to talk to each other as if they were physically connected to each other. It's about 6.4 terabits per second of communication between those GPUs, which is quite a lot of bandwidth.

In addition to that, over here, we have high-speed storage. Forgive the background noise, as I mentioned. This high-speed storage is provided by our partner, WEKA, who's been a great partner. They have their signature purple facade here on these plates. This high-speed storage is connected to each and every single one of these GPUs. So we can move very, very large amounts of data. This cluster is supporting over 20 petabytes of data, which is a lot of different iPhones. I can't do that math right now. But basically, it allows all of the GPUs in here to talk to this high-speed storage across that network that I mentioned before on both the north-south basis.

So now I'm going to walk you guys over to another part of the data center. We're walking through cold aisle containments right now that are being built. And you'll see a very similar infrastructure. You'll see all of the different servers of the Blackwell B300s, all connected to the leaf architecture, all connected to the spine architecture.

And over here, you'll see a bunch of management nodes. These allow us to do different things from automated provisioning, advanced monitoring, security hardening, setting up VPNs, managing different applications for our clients like Kubernetes or Slurm, et cetera. You'll see more purple boxes over here for even more high-speed storage.

And ultimately, behind all of the cabling activities that's going on right now, you'll see some more traditional servers here, right? You'll see the Dell boxes here that we're using for some of our bridges and some of our management capabilities. And ultimately, what these components do, they each play their role, right? So there's UFM nodes, there's OOB connectivity for different management capabilities.

All of this ties together with what we call an NVIDIA reference architecture or NVIDIA reference design, right? So full NVIDIA switching throughout, full NVIDIA GPUs throughout, with accoutrements that allow us to add ancillary services to our clients. And that's basically it.

So thank you guys for coming with us to Columbus, Georgia. We've got a lot of work to do from containment aisle finishing up tomorrow to all of the cabling happening now to get this cluster live as soon as possible. We look forward to the next one, as you can see behind me, and we fully intend to expand and this is the space that we are expanding this cluster into. So it will be one contiguous single spine architecture, 3x the size. So while the cluster that I mentioned is already quite large and quite powerful, it will triple up in the next few months.

Now let's shift gears to what's coming ahead. As of this month, we are tracking $5.9 billion in active qualified pipeline, 98 open opportunities across our sales organization. That's nearly double the size of our current signed book. And every week, it keeps growing as momentum continues to accelerate. By dollar value, roughly 2/3 of that demand is for Blackwell class GPUs or Vera Rubins. We're talking about B200s, B300s, GB300s and of course, the latest Vera Rubin chipset. Our customers aren't asking for yesterday's hardware. They're asking for the newest, fastest, most powerful silicon NVIDIA makes to really power their businesses for the future, not just for the current needs.

Also worth noting, clients understand that securing AI infrastructure on a long-term basis is a vital and strategic imperative for the business, which aligns nicely with our equipment ownership model and long-term data center investments. And keep watching this space. We're already fielding significant early demand for NVIDIA's Vera Rubins, the next architecture, ahead of volume shipment.

So our customers are already planning generations ahead, and so are we. A pipeline this size really isn't a hope, right? It's a queue. And our job for the rest of this year and beyond is pretty simple: convert that queue, contract that queue, and turn up additional clusters for our clients. So that's really the way we've converted the $3 billion that you just heard about. And as Chris has said to me multiple times, there's many more ahead.

Shifting gears to the organization. None of this, the contracts, the technology, the pipeline, means anything without the people who build it, which are the most important part of Axe Compute. So let's talk about the team a little bit. Right now, we are actively adding more than 20 people to this organization. And every one of those hires is funded by a contract we've already signed.

On deployment and operations, we're increasing our data center engineers, cluster commissioning specialists, our 24/7 operations staff, all in advance of standing up these clusters going live this quarter and next. And that's all under dedicated VP-level deployment leadership.

On the infrastructure engineering side, we're adding GPU, CPU, network fabrics, storage specialists, the people who really make 5-, 15- and 30-megawatt AI factories actually work across both our U.S. and European footprint.

On customer support and commercial operations, we're really enhancing our technical account teams, our program management organization, and broadening our enterprise coverage behind those signed contracts and that multibillion-dollar pipeline. And behind our next objective, an additional $2 billion in new signings before the end of the year.

Our philosophy is pretty simple. We don't staff up on hope. We scale fast to deliver on commitments, locked and loaded, and we hire experienced talent that can really deliver and execute on enterprise-grade support while still providing customers the choices that they deserve. Headcount follows committed revenue on a very disciplined and success basis. That discipline is pretty much what turns a hot pipeline into a durable, profitable company. And it's exactly where and why we're built to deliver on everything that you've just heard.

I'll now pass it back to Chris to close us out. Thank you.

Christopher Miglino

So where do we go from here? Our objective is an additional $2 billion in signed contracts before the end of 2026. And as you can see from what Kyle just presented, I don't think we're going to have a problem hitting that number. We're well on our way. And if things keep going the way that they did in this last quarter, we should exceed that goal. So we're excited about that.

Then just to summarize the quarter. This is the first quarter that we had revenue from the compute business. And you're just going to see that acceleration continue to kick in as we bring more clusters live. So as this cluster goes live, you're going to start to see around $20 million, $21 million a quarter just from this location come in.

And then as we bring on the other clusters into the end of Q4 and the beginning of Q1, you'll start to see us grow up to the $696 million in ARR for the year. So our job for the balance of this year is to continue to sign more agreements, but also to make sure that we have the infrastructure in place to deploy really solid amazing clusters globally.

So we appreciate you being with us here today to hear our story. We're excited about where we are. We're working really hard, and we're -- we think that we find ourselves in a really positive place. So thank you for being here and spending the time with us. We look forward to continuing to have you as an investor. And if you're new to our story, we continue -- we look forward to having you join us on this journey. So thank you very much.