QumulusAI, Inc. (QMLS) Q3 2026 Earnings Call Transcript

Hello, everyone. Thank you for joining us, and welcome to the Cumulus AI Third Quarter 2026 Earnings Call. At this time, all participants are in listen-only mode. After today's prepared remarks, we will host a question and answer session. If you would like to ask a question, please press star 1 to raise your hand. To withdraw your question, press star 1 again. Today's call is being recorded. I will now hand the conference over to Zachary Grave, Head of Investor Relations. Zachary, please go ahead.

Thank you. Good afternoon and welcome to Cumulus AI's third quarter 2026 earnings conference call. Joining me today to discuss our results are CEO, Mike Maniscalco, and CFO, Scott Krasnowski. Before we get started, I would like to remind you that our remarks today will include forward-looking statements. These include statements about our strategy, capacity and deployment plans, customer agreement value and expectations, our market, and future financial and operational performance. Please be advised that these forward-looking statements are covered under the Safe Harbor provisions of the Private Securities Litigation Reform Act of 1995, and the Cumulus AI decides to avail itself of the protections of the Safe Harbor for these statements. These statements are based on our current expectations and assumptions, and they are subject to risks and Thank you. The press release and an accompanying investor presentation are available on our website at investors.cumulusai.com. A replay of this call will also be available there. With that, I would like to turn the call over to Mike.

Thanks, Zachary, and thank you all for joining us. I'm pleased to report that revenue more than doubled sequentially in the third quarter to $14.8 million. Adjusted EBITDA was positive for the first time and signed customer contracts passed half a billion dollars. Here's the simplest way to read this quarter. We did $3.4 million of revenue in the first quarter. We doubled that in the second to $6.7 million. And doubled it again in the third to $14.8 million. Additionally, the remaining performance obligations disclosed in our Q3 10Q already schedule another doubling into the fourth quarter. That's our business model working and designed. We move quickly into smaller pockets of power and then bring them online in months, not years, allowing us to close the gap between signing and billing faster than most. Our most recent large deployment went from signed contract to full production in under three months. At any given moment, we are securing the next block, deploying the last one, and billing the ones before it, all at once and continuously. I'll cover three things today. What we do, where our capacity and demand stand, and how we're built for the market we're in. Some of you may be new to the story, so let me take a moment to explain what we do. We deploy GPUs. We don't build models. We build the compute that powers them. And as models get cheaper, I believe the value shifts to the infrastructure that runs them. Open models now carry nearly a fifth of enterprise tokens at a fraction of the price. And customers want that running on dedicated capacity they control. So we find pockets of existing power, typically 2 to 50 megawatts, mostly in co-location. And we bring the AI compute in those facilities online in months, not years. We're using today's capacity to meet today's demand while building the pipeline to power what's coming. Our current customers are AI inference platforms and the teams building on them. And they open every conversation with the same question. When can I be running? In a market that is compute constrained from the largest providers down to the startups, speed is the decision. And the next wave is AI agents doing real work. An agent uses far more compute per task than a chat query. For example, Amazon said its customers deployed twice as many agents in the third quarter as in the prior three quarters combined. And physical AI is the wave behind that. So we are distributed for speed now and positioned for inference next, building what our customers need today on the way to the distributed platform the next generation of workloads will require. That is why we built this company around hyperspeed rather than hyperscale. Let me give you the capacity picture the way we'll report it from here. Active, contracted, and powered land. Active capacity, which is energized in billing, is 8 megawatts across five co-location sites, with the final deployments we described in August now complete in billing. Contracted capacity, which is secured, not yet billing, is 5.45 megawatts, 3.75 megawatts in metro Atlanta, expected to be ready for service in the second quarter of 2027. Half a megawatt in Columbia, South Carolina. Ready for service in December, and 1.2 megawatts in Denver, Colorado, which we just signed, ready for service in Q1. Our final takedowns for the year are in late stages of negotiation. And beyond that, we're already working towards the next generation expansions at a site we're in today. We have a right of first offer on 7 megawatts in Atlanta, which could take that site to 10.75 megawatts. That's 7 megawatts of expansion rights on top of the 13.45 that is active or contracted today. That's expansion we've already lined up at a site we already know. And we are selected about where we go. And when a site proves out, we want first call on the next block of it. These are rights, not obligations, and we'll exercise them when it's right for the business, but the runway is there, and it's ours to take. And for powered land, our long-term lease optionality is our Oklahoma site, where we have 10 megawatts of power active today, running our legacy mining operation, and up to 7 megawatts of additional grid power available. That's the supply side. On demand. We signed $352 million of new direct contracts in the third quarter, more than double the second quarter, bringing our 2026 signings to $529 million, including a number of smaller agreements we don't announce individually. Contracts keep getting larger. Deposits keep arriving ahead of delivery, and pricing on new generation capacities rising, not falling. Our most recent three-year contracts are journeying around $19 to $21 million of annualized revenue per megawatt, up from the $18 to $20 we reported in August. The market is moving the same direction. Rental rates on three-generation-old H100s are up more than 40% since December. We continue to sell capacity faster than we can energize it. Demand is not our constraint. Capacity is. So, let me turn to our sector where two things stood out this earnings season. The first is the capacity we just spoke about. Two weeks ago, Microsoft, Amazon, and Google reported earnings. Together, they disclosed about $1.8 trillion in contracted backlog. Microsoft said customer demand continues to exceed available supply. Google said it has started running third-party compute because its own is constrained. The largest buyers of compute in the world cannot build fast enough, and it's still early. The Census Bureau counts about one in five U.S. businesses using AI today. That is the market we sell into, and it's why two of our customers came back this quarter for another $241 million in total contract value. The second. It's the question the market has been asking for the last several weeks. The question investors have asked us since the day we listed. How is the growth financed is now the question being asked of the entire sector and its largest names. The market spent weeks repricing AI infrastructure companies. Their credit first, and then their equity. And the worry underneath it is simple. Who is paying for the GPUs, and what happens to the debt if the customer doesn't show up? So the question has shifted from how large the customer contract is to whether we can deliver, finance, and run it at a profit. This quarter, we answered all three. Eight megawatts completed in billing. Every deployment financed against its own contract. An adjusted EBITDA turned positive. Here's how we do it. A customer signs a term, take or pay contract, and puts down a deposit. That deposit covers the portion of the GPU cost that isn't financed, and equipment financing tied to that contract covers the rest. The contract revenue services the financing behind it. We use a mix of loans and leases depending on the terms and overall cost of capital. At the end of a loan, the equipment is paid off free and clear and ours to keep. At the end of a lease, We retain flexibility to acquire the equipment at fair market value, with most capped at a relatively small percentage of its original purchase price, preserving significant residual asset value and upside, or we simply return the equipment with no further obligation. Each financing is tied to a deployment, and each deployment to a contract. We have no unsecured bonds, the customer's commitment funds to the hardware, and the hardware earns the customer's commitment back. That's the model. On the customers themselves, they're mostly AI native inference platforms growing fast, and they pay ahead of delivery. We structure every contract for the credit we take. Take or pay terms, deposits up front, and remedies if a customer doesn't perform. We're deliberately building towards a mix where a meaningful share of our capacity is anchored by credit-worthy enterprise customers alongside of the AI natives. That work is underway, and you'll hear more about it as it lands. Let me pull it all together. Three months ago, we told you we had 8 megawatts sold with final deployments in process. Those deployments are done, and they're billing. Revenue doubled. Adjusted EBITDA turned positive for the first time. And the remaining performance obligations disclosed in our Q3 10Q already schedule another doubling into the fourth quarter. Today, eight megawatts are active in billing. Nearly five and a half more are contracted. Seven megawatts of expansion rights sit a site we are already in with 17 megawatts of power land in Oklahoma behind them. And we sold every megawatt we've energized before was turned on. Half a billion dollars of signed contracts is a starting point, not a ceiling. In a market that's repricing scale. We are built on sold, funded capacity delivered in months rather than years. Secure the next block, deploy the last one, build the ones for it, then do it again. All three moved this quarter. They are moving again now, and our conviction in this model has never been stronger. With that, I'll turn it over to Scott to walk through our financial results in greater detail. Scott?

Thanks, Mike. Good afternoon, everyone. I'll take revenue first, then walk down the income statement, then cash in the balance sheet, and finish with our outlook. First, revenue. Third quarter revenue was $14.8 million, up 376% compared to the prior year period, and more than doubled the second quarter of 2026. The growth came from capacity we energized during the quarter, much of which billed for only part of the period. Compute accounted for 96% of revenue, up from 47% in the prior year period. Next, remaining performance obligations. Our PO, direct customer compute contract signed as of quarter end, ended the third quarter at $515 million, nearly three times the $173.1 million at June 30th, with a weighted average term of about two and a half years. That is the contracted revenue in front of us. Under the terms of those contracts, $29.1 million is scheduled to be recognized in the fourth quarter and the remaining $486 million from 2027 through 2030. Down to the operating line. Operating loss was $41.6 million. That includes a $26.4 million non-cash charge for shares issued as a professional services fee in connection with our public listing, a one-time item that is added back in adjusted EBITDA. Beyond that charge, the loss widened as compared to the second quarter for a favorable reason. Depreciation and amortization was $14.9 million, up $8 million quarter over quarter as the capacity Mike described was placed in service. That is the cost of the capacity that doubled revenue. Adjusted EBITDA was $3.7 million, representing a 25% margin in our first positive quarter, compared to a loss of $0.8 million in the second quarter and a loss of $0.6 million a year ago. Depreciation and interest are expected to step up again in the fourth quarter, with a full period on the equipment placed in service during the quarter. And stock-based comp will normalize moving forward as the initial issuance of employee and director RSUs, including one-time grants that vested on issuance, was recorded in September. And below the operating line, third quarter net loss was $14.4 million, or $0.43 per share. That includes $32.3 million of non-cash gains for marketing and convertible notes, the related option and our warrants to fair value, gains that move inversely with our share price, and the $26.4 million non-cash charge for shares issued as a professional services fee in connection with our public listing. Net interest expense was $5.4 million, up $3.3 million quarter-over-quarter, driven by equipment financing for deployments placed in service. Turning to cash. Customer prepayments are funding to build. Contracts signed in the third quarter carried $71 million of customer deposits, more than double the second quarter, and deferred revenue rose to $79 million from $30.5 million at June 30th. On our major multi-year contracts, we typically require prepayments of 15% to 20% of the total contract value. That money supports the deployment it's attached to and reduces what we need from debt. As a result, operating cash flow was positive $32 million in the quarter. Capital expenditures were approximately $100 million in the quarter, primarily for GPU servers, and we added approximately $94 million of equipment and facilities under new leases, with about $194 million placed in service, which drove the step up in depreciation. Financing activities provided $44.6 million in Q3, and we ended the quarter with $31.1 million of cash and restricted cash, of which $9.2 million

is restricted under our loan agreements.

On the balance sheet. Total assets with $394.1 million at September 30th, up from $215 million at June 30th, as equipment was placed in service. Debt and finance lease obligations totaled $248 million. The large majority is equipment financing tied to a specific deployment and amortizing against the underlying contract. In the quarter, we closed a $51.5 million loan and completed a third draw under a USD.AI facility. After quarter end, we expanded our lease financing with TFC to $123.6 million, all without new equity. Shareholders' equity was $25.3 million at September 30th, up from a deficit of $0.8 million at June 30th. Finally, turning to our outlook, we are reaffirming the guidance we issued in July for 18 megawatts of year-end contracted capacity and year-end forward ARR of $300 million. We have 8 megawatts active and another 5.45 megawatts under executed agreements, bringing our total footprint to 13.45 megawatts with the balance and late-stage negotiation. About $209 million of forward ARR is under signed contracts today, and we are now marketing our newest capacity. We are also evaluating how we provide guidance, and we expect to update the way we provide forward guidance when we report the fourth quarter as we look ahead to 2027. With that, I'll hand it back to Mike.

Thanks, Scott. Let me finish off where I started. The question in our sector. has become whether capacity can be delivered, financed, and run at a profit. We now have eight megawatts fully active in billing and five and a half more under contract. We financed every deployment against its own contract. Our Q3 adjusted EBITDA was positive for the first time, and remaining performance obligations in our 10Q already schedule another sequential doubling of quarterly revenue in the fourth quarter. Step back for a moment. This company listed in July. It went through the hardest stretch the sector has seen in years, and it came out the other side with revenue doubling, contracted revenue tripling, and adjusted EBITDA turning positive. For a company of our size, those are real accomplishments, and the team here earned every one of them. And we're still early. Early in AI adoption and early in this company's growth, customer demand and AI use cases driving it remain extremely strong, and we believe we're incredibly positioned for significant growth ahead. We're just getting started. Moderator, we're ready to take questions.

We will now begin the question and answer session. Please limit yourself to one question and one follow-up. If you would like to ask a question, please press star 1 to raise your hand. To withdraw your question, press star 1 again. We ask that you pick up your handset when asking a question to allow for optimum sound quality. If you are muted locally, please remember to unmute your device. Please stand by while we compile the Q&A roster. Your first question comes from the line of Michael Donovan with Compass Point. Your line is open. Please go ahead.

Thank you, Operator. Hi, guys. Congrats on the execution, and thanks for taking my questions. For the two B300 agreements signed in September, how much capital is still needed, and then how much is covered by customer prepayments and committed financing, and how do you expect to fund the balance?

Thanks, Michael. Appreciate it. I'm going to let Scott chime in on that one.

Hi, Michael. One of the orders is fully paid for, and then the other, we did a 20% prepay.

But the RFS stayed slipped. This is not even completely financed.

Yeah, we have a 20% prepay and the balance is in process.

And on that one, as we noted, the RFS slipped slightly on that, so we're aligning the timing on that final financing to where the RFS actually lands in that data center capacity coming online.

Appreciate that. That's helpful. With the additional 1.2 megawatts in Denver expect to be ready in the first quarter, is that capacity tied to existing customer contracts or new demand?

The new capacity is that we just signed. That's currently being marketed. It is not yet sold.

And I believe that all of the capacity with the exception of Columbia is being marketed right now.

Great. Columbia was already sold and announced. Appreciate that. You mentioned interest from potential customers beyond AI native platforms. Can you provide more color on the types of customers you're seeing and the workloads they're looking to run?

Yeah, and I think it's important to point the current pockets of power that we're growing into and how those fit the various customers that we're selling and marketing towards. As you know, some of the larger customers want some of the larger footprints. So as we look forward into our pipeline of larger sites, that starts to change the shape of those customers and what they look and feel like. But right now, for the smaller pockets that we have been successfully finding and acquiring that capacity and then marketing it to our customer base, a lot of that is happening within the AI natives. Still a lot of healthy demand there. But at the same time, we are working the pipeline of the enterprises. They tend to make decisions on different timelines. So we're trying to work and align those timelines appropriately. But to give you an example, the financial services sector is still a pretty hot one for us as we announced the DRW announcement. In the last few months here, I think that's a good example of those types of customers. But general enterprise, I mean, it runs the gamut as far as who we're targeting within those segments.

Thank you. Best of luck moving forward. Thank you.

Your next question comes from the line of Fyodor Chabalin with B. Reilly. Your line is open. Please go ahead.

Thank you very much, operator, and good afternoon, everyone. Congrats, Alyssa Quater, solid and complete progress. And my first question is on public kind of research nature. Near cloud and just related infrastructure that followed Corvive several weeks ago. Have you seen any change in the cadence or maybe even quality of inbound demand for new capacity? Maybe just number of RFPs, deal size, or maybe like any different in contract terms. And on pricing, our new B300 increase. Still clearing around the $4.30, something like per GPU hour level of your current resin contracts, or is the market softening? Thank you.

Yeah, I'll answer the first part of your question first, and thank you for the questions. So we are continuing to see a very healthy amount of demand and haven't really seen a big shift since last month, one way or the other. What we are seeing is, I think, Trying to think of the right way to explain it, but a little higher sense of urgency of our existing customers and their desire to lock down capacity because they, I believe they know that the scarcity of that capacity in the market. And I would also say that that's pointing towards the. Stability or even increasing of the GPU per hour pricing and GPU prepayment terms. So generally speaking, that's what we're seeing in the market right now, that the pricing on the B300s continues to pinch up as pricing of everything has continued to increase slightly. The demand is remaining strong and the terms are improving in our favor currently.

Thank you for that. And just a little follow-up on the Atlanta contract. What slipped there from early 27 to second quarter, 2027? What is it? Power, collocation provider, or maybe hardware delivery? And does it customer's contract start, date, move with it?

I might have misunderstood the question, so feel free to ask it if I misspeak here. But I think that what we're seeing right now as far as scarcity is the data center capacity itself. So that is really the biggest driver of the kind of early 2027 growth and potential from a supply chain perspective on the GPUs and associated equipment themselves. We've done a good job of logistics on pre-planning and getting that supply allocated and arranged ahead of time in those orders and ahead of time. So that side of things is a little faster right now. And that's how we're managing the... The placements. Did I answer your question?

Yeah, like partially. Yeah, but just maybe just like to frame it shortly, what was the reason of this timing shift?

Yeah, I can speak to that. I mean, when the timing's shifting, it's usually a number of factors, and just speaking more generally about all of the capacity out in the market, sometimes it's the supply chain on the data center owner-operator side of things where they estimated a role of cooling or electrical equipment, and maybe we're just a little too aggressive, so we're trying to do a better job of giving buffers there. As you have heard on these, we're also narrowing in a little more on quarters within months so that we have that flexibility given the supply chain. It's just so dynamic right now. And then we're also, as I think I've mentioned on some of these calls and on conversations with the market, the way we've been finding this data center capacity has been in Generally, going back to existing data centers, oftentimes those are legacy data centers. And oftentimes those legacy data centers have existing tenants. So a few things need to happen. The tenant, the existing tenant needs to have a proper vacation date and move out on time. And then the data center owner operator needs enough time in between that window to retrofit and renovate to accommodate the GPU servers. Rather than some of the legacy equipment that was in there with the tenant that's vacating and then they're also they have to arrange the financing and move that through. So there are several different levers that have to move to these dates. And what you're seeing with some of these delays is just slight shifts in some of those levers.

Thank you very much for all this caller. I'll turn it over for now and continue. Best of luck.

Your next question comes from the line of Bill Papanestazou with Chardon Capital Markets. Your line is open. Please go ahead.

Good evening, gentlemen. Congrats on a strong quarter, and thanks for taking my questions. For the first one, I was hoping you'd be able to share a boots-on-the-ground perspective of the supply-demand environment for data center colo capacity and how you expect it to evolve. I'm curious, in particular, if you're seeing the same level of availability for the sub-5 megawatt colo opportunities, or are you seeing that dynamic shift at all, given the continued regulatory headwinds that are putting pressures on larger sites? I guess long-winded question of, do you expect to see new entrants to go smaller-scale deployments, given what we're seeing on the regulatory front? Thank you.

No, it's a great question, and thank you for the question, Bill. I would say that... It's not necessarily getting more competitive for that space. And a lot of these smaller pockets that we have been moving into, because they're existing, haven't necessarily been impacted by some of the headwinds facing the data center market. I will say some of the bigger sites that we're eyeing into the future may be impacted a little more greatly by that. But that's factored into. Later into 28 and late half of 27 right now, and that's something we're thinking through as we're thinking through our options for growth. But I think there's a few different points I want to make here. We have found success looking for even smaller pockets of power, and I think that the Columbia, South Carolina site is a great example of that. So we're continuing to be very flexible and opportunistic of where we're finding pockets of power because a 500kW site accommodates roughly 32 GPU nodes, which is a healthy and marketable footprint for a lot of our inference customers. So it's still a very marketable pocket of compute and power that we can sell to our customer base. And it also is still somewhat significant to our revenue in that if we can monetize that at the $20 million round number, that's still $10 million of potential revenue, and we can replicate that at several sites and bring that online quickly. So it fits our model really well. So that's one thing we're doing. The other side of things is the credit crunch has certainly put a crunch on the data center capacity in the smaller scale data centers. I think that's been more. Impactful than, say, a lack of supply. But right now, it feels like that was a broad stroke overreaction by some of those data center owner operators and investors, and we are watching closely to see if that loosens up. And to give you a little more specificity there, some of those smaller co-location pockets. We've made a pretty fast adjustment to say that they do not want to sell to non-IG in those instances. And as you know, most of the IG customers are looking for much bigger pockets than a megawatt or two here or there. So I think they're going to see that problem pretty quickly. So that's been one thing that's impacted our ability to find some of that data center supply. But we've also been looking at creative ways to work through that through increased customer prepayments. On the upfront deposits and on our contracts, and also looking at various instruments in the market for a credit wrapper, to give you a specific example. So that's how we're working through the current conditions.

Appreciate that. That's really good color. And then just as a follow-up, with respect to the seven megawatts, Under the right of first offer at Metro Atlanta. Can you just refresh the scenario there? When could that come online and be ready for service? Is there an expiry date attached to that ROFO? And is there enough demand today to sign that contract if it's possible to enter?

Yeah, I mean, the demand still is extremely strong. The pipeline the sales team has and the customers that are calling them consistently asking for updates in our capacity in 27 is certainly not a limitation to our growth right now. It really is aligning the right supply. As to the specifics of the Atlanta property, I'll answer it to. As much detail as I can give you top of mind, and then if we have follow up information, we can certainly address that. But that one is a expansion that the owner of that data center is planning to do earmarked for 2027 delivery that is at their discretion. So, as we get more information from that owner and operator on their ability and timelines to expand that, I'll have more specifics on the actual timeline of RFS.

Great. I appreciate it. Thanks for answering my questions, and that's a lot. Thank you for it.

Thank you. Appreciate it.

We have reached the end of the Q&A session. I will pass the call back to Michael Maniscalco for closing remarks.

I think through my commentary on the Q&A and my parting comments coming out of the earnings report really highlighted the points. I think the demand in the market is still really strong. Supply is continuing to be constrained, a lot of that by external factors, and I think those factors are changing pretty consistently. And we're, as I mentioned before, looking at creative ways to work through those in the near term. So really the goal right now is to continue executing in the way we've been doing and bringing that capacity online in smaller packets of power, lining up with customer demand and continue to grow because that's what the customers are asking for. So I appreciate everybody's time today. I appreciate all the questions and have a great evening.

This concludes today's call. Thank you for attending. You may now disconnect.