Home / Transcripts / Axe Compute Inc. (AGPU) · August 17, 2026

Axe Compute Inc. (AGPU) Earnings Call Transcript

August 17, 2026

NASDAQ US Health Care Health Care Equipment and Supplies earnings 54 min

Earnings Call Speaker Segments

Erin McMahon executive
#1

Hello everyone, and thanks for joining Axe Compute's second-quarter 2026 earnings call and business update. I'm Erin McMahon, serving as CMO and Head of Investor Relations here at Axe Compute. With me on the line today are Christopher Miglino, our Chief Executive Officer; Jeremy Yaukey-Witter, our Chief Financial Officer; and Kyle Okamoto, our President. We're coming to you live from Columbus, Georgia, which is where a data center sits that hosts the 2K B300 cluster build we first told you about in April, and that facility is scheduled to begin operating within the next few weeks. Prior to getting started, I want to note that remarks made today contain forward-looking statements, as noted on Slide 2; this presentation includes forward-looking statements as defined under the safe harbor provisions of the Private Securities Litigation Reform Act of 1995, covering things like signed contracted value, anticipated customer prepayments, deployment timing, annualized run rate, expected margins and profitability, financing structures, potential future contract signing and future performance. Uncertainties surround these statements, and actual results could differ materially, including the uncertainties outlined in the Risk Factors section of Axe Compute's latest annual report on Form 10-K along with later quarterly reports on Form 10-Q filed with the U.S. Securities and Exchange Commission. We want to warn readers not to lean too heavily on forward-looking statements, because those statements reflect circumstances only as of today's presentation date. Unless the law demands otherwise, Axe Compute isn't required to refresh them. Turning to the numbers: the quarterly results we're showing come straight from the Form 10-Q that was submitted on August 14, 2026. When we talk about annualized run rate, we mean monthly recurring revenue converted to a yearly figure, calculated once signed contracts have been completely deployed. As for total contract value, or TCV, that's an operating metric capturing the combined expected contractual spending tied to customer agreements we've already signed. Both ARR and TCV are operating metrics, so they may differ from revenue actually recognized in any given period, which is determined separately under U.S. GAAP. Before we go further, a few important notes. Contracts we've signed still carry risk — they depend on deployment happening, the customer signing off, and other factors we've laid out in our SEC filings. The steady-state economics shown here are illustrative only; they're modeled and blended from the signed Build contracts, and they aren't guidance. A word on non-GAAP measures: adjusted EBITDA appears in this presentation, and it is a non-GAAP financial measure. We calculate it by taking net income or loss and removing interest expense or income, net income, tax expense or benefit, depreciation and amortization, stock-based compensation expense, and fair value adjustments on digital assets. Adjusted EBITDA shouldn't be treated as a replacement for net income or loss, or for any other U.S. GAAP financial performance measure, and it may differ from similarly named measures at other companies. On this call, management wants to explain why we think adjusted EBITDA is helpful for investors. The reason is that it gives you an extra way to look at how the company's core operations are doing. It does this by leaving out certain things: the impact of how we've chosen to finance the business, like interest expense, plus changes in fair value tied to the digital assets we hold; noncash items such as depreciation, amortization and stock-based compensation; and tax effects, which can swing a lot from one period to the next and from one company to another. Management relies on adjusted EBITDA when assessing how the company is performing, when making comparisons between different periods, and when helping decide where resources should go. That said, we want to remind investors that adjusted EBITDA has its limits. It's just one analytical tool. You shouldn't look at it by itself, and you shouldn't treat it as something that takes the place of reviewing the company's results as reported under U.S. GAAP. A reconciliation showing adjusted EBITDA alongside the most directly comparable U.S. GAAP figures is included in this presentation. Should the company provide forward-looking adjusted EBITDA guidance during this release or on the accompanying earnings call, it may be impossible to reconcile those forward-looking non-GAAP figures with the closest U.S. GAAP equivalent without unreasonable effort, since some amounts are inherently hard to predict and measure — including fair value adjustments tied to digital asset holdings, expenses from stock-based compensation, and additional noncash or nonrecurring items, the timing and size of which could be substantial. This is not an offer or a solicitation. What follows is purely for information — it doesn't constitute a proposal to sell, or an invitation for anyone to make an offer to purchase, any securities belonging to Axe Compute or any of its affiliates. We aren't offering or selling securities in any jurisdiction where that would break the law. Now, a note on third-party information: portions of what's in this presentation were drawn from outside sources. While Axe Compute regards those sources as dependable, they aren't independent — nobody has separately confirmed that the details are accurate or complete, and we give no guarantee or representation, whether explicit or implied, about either their accuracy or their completeness. We'd ask you to consider these remarks in conjunction with the filings we make with the SEC, which you can access at investors.axecompute.com. On that note, I'll hand the floor to Christopher Miglino, our CEO.

Christopher Miglino executive
#2

Hello, everyone, and good morning. My name is Chris Miglino, and at Axe Compute I hold the position of CEO. We are located in Georgia, and it's great to be here. A very large cluster is nearly ready for us to launch. For anyone just learning about Axe Compute, our company is an AI infrastructure platform that has 2 growth engines. The primary one is Build, and that's what you see in front of you here. For enterprises, we handle the design, deployment, ownership, and operation of dedicated GPU clusters. After that comes Access. That's our complementary recurring revenue stream, delivering rapid GPU capacity access across numerous locations worldwide. Picture it as a top-of-funnel mechanism: we connect with businesses that want GPUs right away, and afterward, as those companies expand, we're able to assist them and expand alongside them. The market views Axe as a brand-new AI infrastructure firm. However, what people in the market fail to recognize is that this business has been ours for the last couple of years, and throughout that whole period our sales team has been selling to the offtakers. In that same span, our supply team has been conversing with data centers and working with power owners too. So even though we appear to be a newcomer that has unexpectedly closed all these deals, this overnight success has actually been 2 years in the making. Before discussing the quarter, I'd like to take a couple of minutes on the market where we operate, since it's the most crucial context for all that Jeremy and Kyle will cover afterward. It should come as no shock that we're in the early innings of a generational compute build-out. Any time you switch on the news, you'll encounter something about data centers, their expansion, and the sums of money being linked to them. As a matter of fact, McKinsey projects that through 2030, spending will reach roughly $6.7 trillion, a figure I consider low even now. Roughly $5.2 trillion of that is tied specifically to AI-capable capacity. Let me be clear about what that second figure represents. It isn't a projection of AI software revenue for companies such as OpenAI or Anthropic, but rather AI infrastructure spending. That means the capital needed for the physical side, the power, the shells, the racks, the GPUs — that's the market we compete in; that's the market Axe Compute competes in. The second point we want to highlight concerns the shape of demand, not merely its size; customers aren't requesting generic cloud. What they want is dedicated capacity. Being in a shared facility alongside lots of other people isn't what they're after. What they're after is owning a cluster similar to the one visible behind me — having it as theirs and being able to work with it independently, with no other party involved. That's something they'll do over the long haul, which opens the door for us to sign agreements spanning 5 to 10 years. They also need a partner capable of helping them expand into that area. Such a partner would assist them in obtaining every piece of equipment, planning and designing the network, deploying it, and coordinating with the right data centers so the proper power is secured and the correct building is prepared for their compute requirements. That's precisely where our Build program fits in. We assist these offtakers in doing exactly that. Whatever is needed to get them live inside the data center, we help them accomplish, and then we collaborate with suitable data centers to carry it out. Data centers aren't all the same, so we make every effort to ensure we've got the right partners supporting our clients. So when the figures we're about to go through come up — over $3 billion in signed contracts, $696 million in expected annual run rate at full deployment, and more than $0.5 billion in customer prepayments — I'd urge you to view them in light of enormous demand; we aren't manufacturing that demand. What we're doing is converting a market where demand structurally outstrips the available supply. Every one of these transactions gets signed on a take-or-pay basis, with contracts running multiple years. With that framing in mind, let's move to this quarter's results; it was a phenomenal quarter for us. If every quarter brought what this one did, everyone listening on this call would be thrilled beyond words. As you can see, we're preparing to launch the first cluster we discussed quite a while back. That was the initial deal we made public. It's nearly ready to go live, and what's sitting there is a gorgeous work of art. Enormous effort, countless hours, went into making it a reality. Still, we've truly excelled by signing roughly $3 billion — nearly $3 billion — in extra agreements, when our expectation for the quarter had been that adding $1 billion in transactions would have made us pleased. So it went well beyond our projection; that's the truth; we genuinely believed the ramp would be somewhat slower. I'll hand things over — I'm going to let Kyle speak a bit more about the pipeline shortly, but you'll notice this momentum hasn't slowed in the slightest. So now we've moved from signing all those transactions and completing all those deals to having all the ARR arranged for next year. Meaning next year, once deployment is done on all of them, again, $696 million in ARR after they're deployed. At this point, we've shifted into a dual approach — we're now in execution mode. Assembling these assets, bringing them online, and overseeing them is what the team is focused on, along with ensuring the right personnel are in position to run them on behalf of our clients. So the momentum we carried through this quarter has us genuinely enthusiastic about our current standing. A question that comes up constantly is what kind of margin these Build projects carry. That's why we chose to include a slide offering investors visibility into that picture; while these are forward-looking, they reflect our modeling and represent our best estimate of where things stand. Our view is that investors can review them and put them to use in shaping their own forward projections. As shown, gross margins fall in the 28% to 44% range, while EBITDA margins land between 62% and 76%. Apply those figures to the $3 billion in announced transactions, and it becomes clear that the economics awaiting us in '27 are substantial. The one thing we're asked about even more often than the margin question on these deals is how we intend to fund these projects.

Christopher Miglino executive
#3

Okay, I'd like to take you through some details on how we're getting these projects funded and what our approach is to securing that funding. As you'll see, every project includes a down payment made by the customer. The customer pays somewhere between 20% and 45% of the project's cost upfront. After that, we're able to go out and pursue project financing against that revenue stream. When you're dealing with a client that carries an A+ S&P credit, and you're seeking financing tied to that revenue stream, plenty of lenders are ready to lend against it — particularly lenders who know the GPU market well, and as the news shows, there are many of them. A lot have emerged, a lot are delivering, a lot are putting money into this space. So what we aim to do is collect that prepayment, secure project financing for every project along the lines of what took place here, off balance sheet, without needing to raise capital against it, and then carry out the buildout and deployment of the cluster itself. So we believe that once people grasp that we're able to finance these projects this way, it's quite advantageous for the public company; and where we want to put equity into these transactions, we could do that at a moment when the stock price strikes us as a good spot to actually complete a transaction if we chose to. If not, we can hold back and pursue project financings against each of these projects while drawing on the strong credit of the clients constructing these things alongside us. We have 2 truly major announcements today. The first: we've gotten the initial prepayment of $317 million plus for the cluster we're expanding. And second, we've signed an agreement for an extra 55 megawatts spread across a number of different locations with Duos Technologies. So both of these have us excited. I believe these will show you where the business is headed; and to speak a bit more about our partnership with Duos, we've invited the CEO of Duos, Doug Recker, to join us today so we can offer you some insight into what our expansion into that 55 megawatts will look like. A good partner is among the most crucial elements when building data centers. And I'm here alongside Doug Recker of Duos Technologies, who has been a fantastic partner for us. Doug?

Frank Recker executive
#4

Thank you for having me.

Christopher Miglino executive
#5

No, I appreciate you being here; and I must say I'm extremely happy that Duos is a company we work with. Throughout this process, you folks have truly been a fantastic partner. So for that, I wanted to express my gratitude. I believe that today, we're pleased to share that an agreement has been signed for an extra 55 megawatts across several sites together with Duos. So that backing means a lot to us.

Frank Recker executive
#6

Great partnership; we're looking forward to it.

Christopher Miglino executive
#7

If what we're seeing at this spot right now tells us anything about how the rest will unfold, then I've got a lot of faith that you folks will take care of every last thing. So thank you for backing us. Honestly, what you've pulled off here from the moment we started is simply amazing.

Frank Recker executive
#8

What we've accomplished in — what was it — 45 days is just unbelievable. If you were able to grab a camera and stroll through the plant, you'd be stunned by how quickly this project came together. In a way, it shows how our company operates, doesn't it? When we say we can get something done, we get it done; that said, we're never going to claim we'll build 100 megs in 3 weeks, right? That's not something we'd say. Yet looking at our product, all of this happened within 45 days.

Christopher Miglino executive
#9

Right. So here's the thing — plenty of folks following this call who are really into the data center world, but they don't get all the pieces that come into play and are part of this whole process, correct? So you've got the owner of the data center, and that's you folks. On top of that, you run the building as the operator. Then there's everything we need to bring inside, and that's what we do, right — bring that equipment in, get it operational, oversee it, oversee the installation, oversee the whole process. But tell the audience how — what your company does and how that translates into results for you all, since you're a publicly traded company too. DUOT, Duos Technologies; go ahead and break that down so everyone gets it.

Frank Recker executive
#10

Okay, so the thing that really stands out about our company is that we've been operating in this space for more than 30 years. I personally have constructed numerous data centers — big ones — and have even worked with the modular method, correct? So at present, through our collaboration with Axe and under our current business model, what we're essentially doing is deploying in a modular fashion. What happens is we locate a site such as this one, where a building exists and fiber is available, and then we apply our modular strategy there — which essentially means we construct our e-houses, our chiller plants, all of it in modular form. We assemble everything off-site and then ship it. After that, we're able to connect it within 60 days, 90 days, or occasionally even faster depending on the situation you're looking at. So we have truly become experts at that method. It's been roughly 9 years that we've been doing this. At this point, our focus is really just on the 5- to 25-, 30-megawatt span; we're not about to go build 100 megawatts in 1 day. We'll handle it modularly, and we'll construct alongside the client, similar to what we're doing with you.

Christopher Miglino executive
#11

So — and look, everyone, when you put this core together here, you had it all set up for whatever expansion comes next.

Frank Recker executive
#12

That's precisely it. And this ties back to the modular strategy, correct? So we constructed a 10-megawatt facility, though the core infrastructure is placed outside so expansion remains possible. Then we add a further 10-megawatt of clusters, and that gets introduced in modular fashion using our e-houses along with our chiller plant and our electrical plant. That entire plant and all of it is complete, and we assemble it and install them progressively.

Christopher Miglino executive
#13

So once that expansion kicks off right now, things are going to get pretty interesting around here — it's going to be seriously noisy in this place. For everyone working in here, it's going to feel like an entirely different world, because the sound level is just going to be screaming loud. So my guess is they'll tack on a little extra charge when they put that one together.

Frank Recker executive
#14

Yet it's truly remarkable, isn't it, simply to witness how a strong collaboration functions across multiple dimensions, correct? On this house, we represent the infrastructure portion, right? If that infrastructure isn't there, the operating side doesn't exist. You serve as the operating partner, and collaborating the way we did throughout this program has been incredible. So communication matters most. All of you are experts in your field, and we excel in ours. Bring those 2 minds together, and the result is extraordinary. And so, look at what's rising at this moment.

Christopher Miglino executive
#15

Also, one more development we're pursuing jointly that I find really thrilling — this news went out earlier today — involves us taking an ownership stake alongside Duos in the data centers. So, when it comes to picking a partner to share data center ownership with, Doug and his crew are the ones you'd want. What we're doing is supplying capital via an SPV, and that money goes into constructing new buildings and new structures in locations where power availability is confirmed and, for sites operating off the grid, natural gas access is confirmed. Consequently, across these megawatts we've committed to, we're taking an ownership position together with Doug as well. That lets Doug extend his reach much further and develop many more data centers while using less capital, and it lets us set up an SPV that holds a stake — that takes part in funding the buildings — meaning Doug can construct additional data centers without needing to bring in a large amount of extra capital.

Frank Recker executive
#16

Equity. Yeah. This is an ideal collaboration, and it lets us expand at a fast pace too. Something else we excel at, and it deserves mentioning, is what we handle on the infrastructure side when it comes to procurement. So the secret to constructing this rapidly is ensuring you're placing orders for the correct gear and receiving it on schedule; and we've got an entirely separate arm, the infrastructure arm of our company, and that's exactly what we handle. So we'll purchase our UPSs, our e-houses far ahead of time; we'll purchase our generators, and essentially we operate them simultaneously; so thanks to this partnership, we're able to acquire those assets right now; so once we're set for the real construction, we do those LEGO blocks right away, and we're in, within 120 days, 90 days, 60 days depending on how that build looks. When you arrive at the 20-megawatt, 30-megawatt range, which happens to be our sweet spot, the 10- to 30-megawatt, you're genuinely in a strong position. You can roll out fast. You can do it properly, you're eco-friendly and additionally you're beneficial to the community. And at that scale we can act more swiftly.

Christopher Miglino executive
#17

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

Frank Recker executive
#18

That's right.

Christopher Miglino executive
#19

A good number of the items we're currently reviewing should roll out by year-end and carry into the start of Q1.

Frank Recker executive
#20

Exactly right. So with speed — the crucial thing for both your operations and ours is revenue, meaning once we buy this product, our goal is to have it installed. We need that item generating income; this matters greatly to us and to you; it must be brought online; those dollars have to start flowing.

Christopher Miglino executive
#21

And you have to — at this moment, everyone's dying to get into data centers, correct? I receive phone calls constantly. Folks are eager to join the data center industry. My initial response to them is, I'm acquainted with someone who operates in that space, and it's far from simple like you'd imagine. This is — constructing these facilities is an extremely intricate undertaking. What you — what I've witnessed take place in this area is extraordinary. Once more, if you lack the expertise, you end up losing — you end up losing; success simply won't come. This is — if any sector demands that you understand each and every aspect, it's this one.

Frank Recker executive
#22

Yes, and on top of that, within this sector, it appears impressive, and just about anyone is able to install a cabinet and connect electricity to it. The real point where skill becomes necessary is once it has been set up. They have to be capable of running the site and understand the work involved. Construction itself is something anyone can do. What matters most in this line of business is what happens once construction is finished.

Christopher Miglino executive
#23

Once more, Doug, I truly value the collaboration. We're thrilled — thrilled to be building alongside you, and we look forward to what lies ahead in this partnership with you. Thanks so much. I appreciate it.

Frank Recker executive
#24

Thank you so much.

Christopher Miglino executive
#25

At this point, I'll hand things over to Jeremy. He'll walk us through the quarter's financial results. Jeremy, the floor is yours.

Jeremy Yaukey-Witter executive
#26

Thank you, Chris. To kick things off, I'll go over what we saw in the second quarter, and after that we'll take you through the expected schedule for converting the signed book into revenue that we can recognize. For Q2, revenue came in at $3.2 million — that's our first complete quarter of compute revenue — compared with $35,000 in Q1. Keep in mind that the Access model accounted for 100% of compute revenue during this quarter. We haven't begun generating build revenue yet; that kicks off once we reach go-live. We posted a net loss of $17.2 million, and the main driver was $13.1 million in losses on digital assets — mostly unrealized fair value movements on the Aethir holdings we carry, which hit the income statement every period as the token's market price fluctuates. Operating cash flow for the first half of 2026 came in positive at $17.4 million, largely because of customer prepayments, which reached $60.8 million as of June 30. Both of those numbers show how strong our take-or-pay, prepay-first contract structure is. Turning to the balance sheet: we closed the quarter with $21.9 million of cash, which is up from $6.9 million when Q1 ended. Our digital assets — mostly the Aethir token — plus the current portion of our digital asset receivable came to roughly $18.8 million at the close of the quarter. We also prepaid several third-party compute suppliers during the quarter in order to lock in compute for signed customer contracts with terms reaching as long as 3 years; those show up on the balance sheet as compute prepayments. On the other side of the balance sheet, contract liabilities are the dominant item as of June 30, and they include $33.6 million that we expect to recognize as revenue inside 12 months, along with an additional $27.1 million in long-term contract liabilities. As a reminder, these contract liabilities are customer prepayments tied to executed compute contracts. That leads me to adjusted EBITDA, which is a metric we're rolling out this quarter. We think adjusted EBITDA is a useful performance measure for investors when assessing how we operate and when comparing financial results across periods. Let me be precise about what this metric excludes and what it doesn't. Adjusted EBITDA adds back interest, income taxes, depreciation, amortization, and stock-based compensation — the usual EBITDA add-backs — and it also removes the unrealized noncash fair value adjustment on our digital assets, which is $11.8 million out of the $13.1 million digital asset loss this quarter. What it does not remove is the realized piece, roughly $1.3 million, which includes ATH we actually spent to pay for compute that was then sold to customers. On that basis, adjusted EBITDA was approximately negative $4.9 million for the quarter, with about $0.9 million of that tied to our legacy Drug Discovery Service segment. The bottom line: the net loss this quarter is mostly a noncash swing in the fair value of our digital asset holdings, which can be quite volatile. Adjusted EBITDA takes out that valuation adjustment along with the other add-backs I mentioned, which management views as not directly reflecting our ongoing operating performance. So now that we've gone over the second quarter, let's shift our attention to the upcoming quarters and examine when revenue recognition is expected to start from our signed book. Our business model includes contracts with customers that are both short-term and long-term. Some contracts—especially those under our Access model—began delivering service as soon as 48 hours after being executed, whereas others—especially under our Build model—don't hit ready-for-service milestones until roughly 4 months after signing. To model these contracts together, we rely on operating metrics such as monthly run rate and annualized run rate. A contract is added to our run rate only after it's deployed and billing, not at signing, and that's what this slide conveys. By the close of Q2, we had signed a total contract value of $317 million. That was our first full quarter of compute services, which included the $260 million landmark contract announced in April. Since then, during the first half of Q3—roughly 6 weeks—we've added an incremental $2.9 billion made up of the 3 Build contracts announced in July. Year-to-date through August, that amounts to more than $3.2 billion in total contract value in fewer than 8 months. Here's what that means for run rate. As we left Q2, our exit run rate stood at $37 million annualized. When the April cluster goes live in Q3, that rises to approximately $139 million, including additional growth to date under our Access model. And after the entire signed book is deployed, stretching from Q4 into Q1 of next year, we anticipate an annualized run rate exceeding $696 million. Q2 is only the beginning of the ramp. Now that I've shown you a preview of what's coming, Kyle will demonstrate what will actually be deployed behind these numbers.

Kyle Okamoto executive
#27

Putting signatures on agreements is just one piece of it. The remaining piece is being able to deliver at this magnitude. That is the reason Axe captures its margin and gives value back to shareholders. Each cluster begins with NVIDIA's finest silicon, 8 B300 GPUs in every node, with 288 gigabytes of HBM3e memory on each and every one; for the biggest builds we do, we take it further with the GB300 NVL72, 72 GPUs plus 36 Grace CPUs merged into a single rack scale system, all sharing that identical 288 gigabytes per GPU, multiplied by 72. We install it in high-density air-cooled, liquid-cooled or direct liquid cooling setups. 64 to 155 kVAs per rack. That isn't office IT, correct? This is an industrial-scale AI factor. We link it through InfiniBand, Quantum-X800 InfiniBand, 6.4 terabits per second per node or 57.6 terabits per second per rack on the GB300s. We supply it with attached storage at genuine scale. That means 20 or 40 petabytes and beyond via excellent partners such as WEKA. And we place all of this inside Tier 3 N+1 facilities, 5, 20, 25, 30 megawatts apiece, constructed with space to expand into the following contract before the ink has even dried. Here's the key. It's the entire thesis behind our business. At this magnitude, AI capacity can't be purchased off the shelf. It must be engineered, or ideally, co-engineered alongside our clients, rack by rack, watt by watt, node by node. That expertise is precisely what our customers sign with us for, and it's precisely why they choose Axe rather than attempting to build it on their own. And since we're here in Columbus, Georgia, and we're showing you guys around this data center, it matters to mention that we're also constructing other clusters across the globe. So we have a few pictures here on our Sweden cluster in Boden. We're quite excited about this one. It's alongside a great partner, and that cluster is currently under construction right now. We'll need to do our next earning release from Sweden, even though it could be a bit cold that time of year. But we'll certainly keep the teams updated there. What this cluster is, is 256 nodes plus spares of — I apologize, B300s, full RoCEv2 on this one along with a very large storage cluster also supplied by our partner, WEKA, with dedicated circuits to data lakes throughout Europe, dedicated Internet for high-speed inferencing; and it's truly a multipurpose cluster that handles both training across that RoCEv2 network fabric and high-speed inferencing. So it's a pretty exciting cluster that we anticipate going live near the end of this year, and we'll certainly keep the teams updated on how that turns out and offer some more videos and pictures as we proceed. Here to give you a live video walk-through of the newest 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. I apologize in advance if construction sounds reach you while I'm speaking. A large number of people are putting effort into this project so it can go live for our client at the earliest possible moment. What we have here is an NVIDIA Blackwell B300 cluster made up of 288 nodes, along with extra shared high-speed storage, CPU nodes handling orchestration, Kubernetes, Slurm, and similar tasks, plus dedicated networking. Essentially, that covers all the pieces required for a genuine AI factory, and we'll go through each of those components as we proceed. The true core of this data center sits right here, and what you're looking at are the NVIDIA Blackwell B300 machines. Every one of these machines contains 8 separate GPUs, and all of them can communicate with one another as though they made up a single distinct computer. Then comes the genuinely impressive bit: every one of these servers is able to talk to the others over something we refer to as an InfiniBand network, which in effect transforms every individual GPU in this facility into one unified computer, correct? That means we can handle extremely fast, extremely large-scale artificial intelligence use cases and workflows for our clients, so we'll touch on a bit about each of those components as we go. But fundamentally, this is the way an AI factory gets constructed. These aren't standard racks, they aren't standard boxes. This isn't an IT environment. What's needed here is specialized cooling, specialized air handling, specialized plumbing, and specialized electrical work. So building one of these genuinely requires a village, involving the data centers, the OEMs, meaning the original equipment manufacturers, the data centers again, the electricians, the plumbers — it's certainly a massive undertaking, so we're going to walk through that today. As I said earlier, all of these computers and chips are linked together, and the method for doing that is that every pod, or grouping of computers, connects into an InfiniBand leaf architecture. In other words, it's a spine-leaf architecture. These are the leaf switches whose job is connecting each pod to the others. Then those leaf switches all ultimately link into a spine. And so they get replicated for every pod of GPUs, based on capacity and redundancy needs; we're able to fit a certain quantity into each of these leafs and spines. And then in the end, all of those are linked together. That lets every individual GPU on this network communicate with the others as though they were physically wired to one another. It comes to roughly 6.4 terabits per second of communication among those GPUs, which is a substantial amount of bandwidth. On top of that, over here, we've got high-speed storage. Pardon the background noise, as I mentioned. This high-speed storage comes from our partner, WEKA, who has been an excellent partner. Their distinctive purple facade appears here on these plates. This high-speed storage links to every individual one of these GPUs. That way we're able to shift extraordinarily large volumes of data; this cluster is supporting more than 20 petabytes of data, which works out to a great many different iPhones. I'm not able to do that math at this moment. So essentially, what this does is enable every GPU inside this facility to communicate with that fast storage system over the network I brought up earlier, working in both the north-south direction. Alright, at this point I'll take you all to a different section of the data center. Right now we're passing through cold aisle containment structures that are currently under construction. You're going to notice an infrastructure that looks quite comparable. What you'll observe is every one of the various servers — the Blackwell B300s — linked into the leaf architecture, and all linked into the spine architecture as well. Then, over on this side, there's a group of management nodes. Through these, we're able to carry out a range of tasks, including automated provisioning, sophisticated monitoring, security hardening, VPN setup, and handling various applications on behalf of our clients, such as Kubernetes or Slurm, and so on. You'll spot additional purple boxes right here, providing yet more high-speed storage. And in the end, past all the cabling work taking place at this moment, there are some more conventional servers, correct? You'll find the Dell boxes right here, which we rely on for certain bridges along with some of our management functions. And in the end, each of these pieces has a specific job to do, correct? So you've got UFM nodes, and you've got OOB connectivity serving various management needs.

Kyle Okamoto executive
#28

Everything here comes together through what we refer to as an NVIDIA reference architecture, or an NVIDIA reference design, correct? That means NVIDIA switches everywhere, NVIDIA GPUs everywhere, plus the extras that let us bolt on additional services for our clients. That's really the whole picture. So thanks, everyone, for traveling with us to Columbus, Georgia. There's plenty left to do — from the containment aisle, which wraps up tomorrow, to all the cabling going in right now, so this cluster can go live as quickly as we can manage. We're excited about the next one, which you can see behind me, and we have every intention of growing — this is the space into which we're expanding this cluster. It'll be a single contiguous spine architecture, 3x the size. So even though the cluster I just described is already fairly large and fairly powerful, within the next few months it will triple. Now let's switch topics to what lies ahead. As of this month, we're tracking $5.9 billion in active qualified pipeline — 98 open opportunities throughout our sales organization. That's almost twice the size of the book we've currently signed. And it keeps climbing each week as momentum builds faster and faster. Measured by dollar value, about 2/3 of that demand is for Blackwell class GPUs or Vera Rubins. We're talking B200s, B300s, GB300s, and naturally the newest Vera Rubin chipset. Our customers aren't requesting yesterday's hardware. What they want is the newest, fastest, most powerful silicon NVIDIA produces, to genuinely drive their businesses into the future, not merely satisfy today's requirements. Another thing worth pointing out: clients recognize that locking in AI infrastructure over the long term is a strategic necessity that's vital for the business, which dovetails well with our equipment ownership model and long-term data center investments. And keep your eyes on this space. We're already seeing meaningful early demand for NVIDIA's Vera Rubins, the next architecture, before volume shipment; so our customers are planning generations out, and so are we. A pipeline of this size isn't wishful thinking, right? It's a queue. And our task for the remainder of this year and beyond is quite straightforward: convert that queue, contract that queue, and bring up more clusters for our clients. That's essentially how we converted the $3 billion you just heard about. And as Chris has told me many times, there are many more ahead. Switching topics to the organization. None of this — the contracts, the technology, the pipeline — amounts to anything without the people who build it, and those people are the most important part of Axe Compute. So let's discuss the team a bit. At this moment, we're actively bringing more than 20 people into this organization; and each of those hires is funded by a contract we've already signed. On deployment and operations, we're growing our data center engineers, cluster commissioning specialists, and our 24/7 operations staff, all ahead of standing up these clusters going live this quarter and next. All of that operates with committed VP-level leadership for deployments. When it comes to infrastructure engineering, we're bringing in specialists for GPU, CPU, network fabrics, and storage — the individuals who truly ensure 5-, 15-, and 30-megawatt AI factories function properly throughout our footprint in both the U.S. and Europe. Regarding customer support and commercial operations, we're strengthening our technical account teams and program management organization, while expanding enterprise coverage to support those signed contracts and that multibillion-dollar pipeline; and to support our next goal, an extra $2 billion in new signings prior to the year's end. Our approach is quite straightforward; we don't hire based on optimism. We grow quickly to fulfill commitments, locked and loaded, and we bring on seasoned talent capable of delivering and executing enterprise-grade support while continuing to give customers the options they deserve. Staffing levels track committed revenue in a highly disciplined and success-driven manner. That discipline is essentially what transforms a strong pipeline into a lasting, profitable business. And that is precisely where and why we are positioned to deliver on all that you've just heard. I'll hand it over to Chris now to wrap us up. Thank you.

Christopher Miglino executive
#29

So what comes next for us? Before 2026 wraps up, we're aiming to get another $2 billion signed up in contracts. And based on everything Kyle just walked through, I really don't see us struggling to reach that figure. We've got good momentum going, and assuming the current quarter follows the same pattern as the one we just closed, we ought to go past that target. That's something we're thrilled about. Now, to wrap up on the quarter — this marks the first time we've booked revenue from the compute business, and you'll watch that speed-up really take hold as additional clusters come online. Once this cluster is live, you can expect roughly $20 million to $21 million each quarter coming from just this site. Then, as we activate the remaining clusters toward the close of Q4 and into the start of Q1, you'll see us climb toward $696 million in ARR across the year. So for the rest of this year, our task is twofold: keep signing additional agreements, and ensure the infrastructure is ready so we can roll out truly strong, impressive clusters around the world. We're grateful you're here with us today listening to our story. We feel great about where we stand. We're putting in serious effort, and we believe we've landed in a genuinely strong position. So thanks for attending and giving us your time; we hope you'll stay with us as an investor. And for those of you just learning about our story, we hope — we'd love to have you along for this ride; so thank you so much.

Read the full transcript via the API

You're viewing the first half of this call. Get the complete Axe Compute Inc. transcript - plus 132,000+ transcripts from 7,000+ companies, speaker segments and full-text search - through the EarningsAPI REST API or hosted MCP server.

Get an API key View API docs →

For developers and AI pipelines

Programmatic access to Axe Compute Inc. earnings transcripts and 132,000+ others is available through the EarningsAPI REST API and the hosted MCP server. Quarterly plans from $162 - full transcripts, speaker segments, full-text search, and the /api/v1/transcripts/recent polling endpoint for ETL pipelines.