Applied Digital Corporation (APLD) Earnings Call Transcript & Summary

Good morning, and welcome to Applied Digital's Fiscal Fourth Quarter and Full Year 2023 Conference Call. My name is Donna, and I will be your operator today. Before this call, Applied Digital issued its financial results for the fiscal fourth quarter and full year ended May 31, 2023, in a press release, a copy of which will be furnished in a report on Form 8-K filed with the SEC and will be available in the Investor Relations section of the company's website. Joining us on today's call are Applied Digital's Chairman and CEO, Wes Cummins; and CFO, David Rench. Following their remarks, we will open the call for questions. Before we begin, Alex Kovtun from Gateway Group will make a brief introductory statement. Mr. Kovtun, please proceed.

Great. Thank you, operator. Good morning, everyone, and welcome to Applied Digital's Fiscal Fourth Quarter and Full Year 2023 Conference Call. Before management begins their formal remarks, we would like to remind everyone that some statements we're making today may be considered forward-looking statements under securities laws and involve a number of risks and uncertainties. As a result, we caution you that there are a number of factors, many of which are beyond our control, which could cause actual results and events to differ materially from those described in the forward-looking statements. For more detailed risks, uncertainties and assumptions relating to our forward-looking statements, please see the disclosures in our earnings release and public filings made with the Securities and Exchange Commission. We disclaim any obligation or any undertaking to update forward-looking statements to reflect circumstances or events that occur after the date the forward-looking statements are made, except as required by law. We will also discuss non-GAAP financial metrics and encourage you to read our disclosures and the reconciliation tables to applicable GAAP measures in our earnings release carefully as you consider these metrics. We refer you to our filings with the Securities and Exchange Commission for detailed disclosures and descriptions of our business as well as uncertainties and other variable circumstances, including, but not limited to, risks and uncertainties identified under the caption Risk Factors in our annual report on Form 10-K. You may get Applied Digital Securities and Exchange Commission filings for free by visiting the SEC website at www.sec.gov. I would also like to remind everyone that the call is being recorded and will be made available for replay via a link available in the investor election of Applied Digital's website. Now I will turn the call over to Applied Digital's Chairman and CEO for Wes Cummins. Wes?

Thanks, Alex, and good morning, everyone. Thank you for joining our fiscal fourth quarter and full year 2023 conference call. I want to start by thanking our employees for their ongoing hard work and service and advancing our mission. Before turning the call over to our CFO, David Rench, for a detailed review of our financial results, I'd like to touch on some recent developments across our business. I'd also share why we remain confident about the future and our ability to deliver long-term growth. Over the last year, we've been working toward providing digital infrastructure solutions that provide differentiated services from traditional data centers. Demand for our services from both traditional customers and emerging HPC applications remains robust, and we're excited about the year ahead. As we enter fiscal 2024, we are focused on three key strategic goals. First, we aim to have all 3 of our crypto hosting facilities fully online with high reliability and performance for our customers. Our 100-megawatt Jamestown facility continues to perform as expected and operated at full capacity with improved uptime throughout the quarter. We announced the initial energization of our 180-megawatt facility in Ellendale, North Dakota in March, and today, it's fully energized. This brings applied digital to 280-megawatts of hosting capacity across all our facilities in North Dakota all of which are contracted out to customers on multiyear terms. The high-voltage interconnection work began last week at our Garden City site and is expected to finish this week. Energization is expected after completion and approval of the facilities extension agreement, which is imminent. We expect these facilities to generate approximately $300 million in revenue and $100 million of EBITDA on an annualized basis. Having 3 facilities online with high uptime will provide us with steady cash flow, this will aid in the capital needed to fund the build-out of our HPC data centers as well as the purchase of GPUs to service our AI cloud customers. The second goal is to expand our AI cloud service business to support the next wave of AI-powered applications. With the launch of this service, we can expand our offerings and capitalize on the unprecedented demand we're seeing from customers. We initially provided this service from our 9-megawatt HPC Jamestown facility along with third-party colocation space as we continue to execute on the development of our dedicated next-gen HPC data centers. We continue to see extraordinary demand for our new cloud service offering. We recently announced two AI customers solidifying our position as a key player in the new cloud service provider landscape. During the quarter, we announced the signing and successful onboarding of our first customer, Character.AI with an agreement worth up to $180 million over 24 months. This includes the activation of the first compute cluster. We anticipate the service to be fully ramped by the end of 2023. This customer has already executed their option for the full $180 million agreement and made a significant prepayment. They have also signed an option agreement for an additional $180 million, which would bring the total value of the contract to $360 million if executed. We also secured our second AI cloud agreement in June, which is worth up to $460 million over 36 months. To help support these contracts and our go-forward capabilities, we have ordered over 26,000 GPUs and have secured the capacity to bring these online between now and April of next year. The GPUs will be financed through customer prepayments, vendor financing options and other financing options that have been structured specifically for this market. To ensure seamless service delivery, we have collaborated with industry-leading OEMs such as Supermicro and Hewlett Packard Enterprise to leverage their HPC expertise and support the execution of our cloud service for AI-powered applications. Our services are made available to customers through 2 distinct models: reserve capacity and on-demand capacity. Under the reserve capacity model, customers pay a predetermined amount for the entire contracted duration of the GPU usage. This option provides stability and allows customers to reserve capacity in advance at a discount to on-demand capacity. For on-demand capacity, customers have more flexibility in terms of usage, but pay higher rates. The typical customers for our AI cloud service are private VC-backed companies that have raised significant funding and will likely raise additional funding to help scale their AI applications. We tailor our agreements to these customers so that as they raise money, they can exercise options embedded in the contract to deploy GPUs and ramp up hosting capacity over time. Customers will typically make a prepayment on the contract, which helps fund a significant portion of the purchase price of the GPUs. Pipeline of business opportunities for AI cloud service remains robust, and we have significant opportunity in front of us. Our third priority is the development of our next-gen HPC data centers. We are well positioned for success in this space and believe our next-generation facilities are ideal hosting sites for HPC applications as they can accommodate the unique demands for this growing industry. Our data centers offer a more purpose-built solution offering lower costs combined with high computing power compared to traditional data centers that are typically focused on delivering low latency applications. We have 300-megawatts of capacity in development, which represents an additional 10-megawatts of capacity to what we previously disclosed. This capacity pipeline does not include the current 9 megawatts of capacity we have at our stand-alone HPC facility in Jamestown, which was initially commissioned in May to begin supporting our AI cloud service customers. This facility will be brought online in phases over the next few months. The 300-megawatts of capacity includes 200-megawatts of capacity available in North Dakota and a new facility we plan to build in Utah. We have a significant customer lined up for our new HPC facility in North Dakota and are currently planning to break ground in the coming months. We will continue to target states that have favorable laws and regulations for HPC application industries. We believe this further minimizes the associated risk with scaling our operations. To finance the build-out of these facilities, we're working with traditional data center lenders, along with alternative funding options. I will now turn the call over to CFO, David Rench, to walk you through our financials and provide guidance for the upcoming 2024 fiscal year before providing my closing remarks. David?

Thanks, Wes, and good morning, everyone. Revenues for the fiscal fourth quarter of 2023 were $22 million compared to $7.5 million for the fiscal fourth quarter of 2022. The results this quarter were attributable to our hosting operations in Jamestown, North Dakota, along with the increase in energized megawatt capacity at the Ellendale, North Dakota facility. The Jamestown site operated at full capacity throughout the quarter. Cost of revenue for the fiscal fourth quarter of 2023 was $16 million compared to $7.4 million for the fiscal quarter of 2022. The increase in the cost was attributable to the higher energy costs used to generate hosting revenues, depreciation and amortization expense and personnel expenses for employees directly working on our Jamestown and Ellendale hosting facilities. Adjusted gross profit for the non-GAAP measure that excludes depreciation embedded in the cost of revenue and onetime electricity charges was $7.8 million or approximately 36% of revenue for the fiscal fourth quarter of 2023, which was compared to adjusted gross profit of $1.1 million or 15% of revenue for the fiscal fourth quarter of 2022. Operating expenses for the fiscal fourth quarter of 2023 were $12.3 million, which included $5.2 million of stock-based compensation, $6.2 million in general and administrative costs and $0.9 million of depreciation and amortization expenses. For the year ago comparable period, operating expenses were $4.4 million, almost all of which were attributable to general and administrative costs. Net loss attributable to Applied Digital for the fiscal fourth quarter of 2023 was a loss of $6.5 million or a loss of $0.07 per basic and diluted share based on a weighted average share count during the quarter of approximately 95.1 million. This compares to a loss of $2.8 million or a loss of $0.04 per basic and diluted share in the fiscal fourth quarter of 2022 based on a weighted average share count during the quarter of approximately $76.6 million. Adjusted net loss attributable to Applied Digital, a non-GAAP measure for the fiscal fourth quarter of 2023 was a loss of approximately $300,000 or a loss of less than $0.01 per basic and diluted share based on a weighted average share count during the quarter of approximately 95.1 million. This compares to adjusted net loss attributable to Applied Digital of $1.4 million or a loss of $0.02 per basic and diluted share for the fiscal fourth quarter of 2022 based on our weighted average share count during the quarter approximately $76.6 million. Adjusted EBITDA, a non-GAAP measure for the fiscal fourth quarter of 2023 was $2.9 million compared to an adjusted EBITDA loss of $3.1 million for the fiscal fourth quarter of 2022. Lastly, on the balance sheet, we ended the fiscal year with $29 million in unrestricted cash and cash equivalents and $79.4 million in debt. During the fiscal fourth quarter of 2023, we received approximately $1.1 million in net deferred revenue due to the structure of our commercial agreements with our customers that incorporate prepayments. In certain contracts, the prepayments are amortized back to the customers over the first year of the contract with no impact to revenue recognition, but the timing of the cash flow with the upfront cash to us is a major benefit to the company and that it helps with our CapEx funding as we build our data centers. Now turning to guidance for the full fiscal year 2024, we expect revenue in the range of $385 million to $405 million and adjusted EBITDA in the range of $195 million to $205 million. That completes my financial summary. Now I will turn the call over to Wes for closing remarks.

Thank you, David. To add some detail around the guidance, we expect revenue to ramp significantly in every quarter with our fiscal fourth quarter approaching $150 million of revenue. We ended the fiscal year with significant momentum. We successfully energized our Ellendale facility and launched our AI cloud service to provide high-performance computing power for AI applications. We've been advancing our existing crypto hosting operations while simultaneously expanding our offerings to further capitalize on surging demand we're seeing from customers for our services. As we look to the year ahead, even amidst the dynamic and complex macro environment, it's increasingly clear that the secular tailwinds of digital transformation remains strong, and we are well positioned to capitalize on strong demand for both crypto and non-crypto customers for our services. We're now happy to take questions. Operator?

[Operator Instructions] Today's first question is coming from George Sutton of Craig-Hallum.

Congratulations on the great results and great outlook. So I'm curious if you could walk through what you mean by the pipeline being robust and just provide a little more detail relative to the 26,000 GPUs that you've discussed relative to NVIDIA. Can you just walk through sort of how those will lay out over the next couple of years?

Yes. George, so the 26,000 GPUs, so we have those on order -- actually, we've increased that order recently, but we have those on order. And as we said, the expectation is those deploy through April. We've already secured both our owned facilities and third-party colo facilities to deploy that, and that's kind of what we think the delivery schedule is from NVIDIA as well. So we have that locked in. And then from a customer standpoint, as we mentioned, Character has doubled their option on their contract. So we think just with the metrics they're seeing, I'm really optimistic about the demand from that customer. And then as our other customer, our second customer starts to ramp up later this year. That will -- those two combined, I think, will take the majority of the 26, if not potentially more than that. But then we have -- I think we shared back in June kind of the customer pipeline, which didn't include either one of those that we still are working on some of those contracts that I can close over the next couple of weeks.

As a follow-up on the competitive landscape, just so we better understand. When we're talking about reserve capacity deals and large language model training, what other options do these new players have relative to or kind of low-cost power options?

So you mean the other type of cloud options?

So for 3, 4, 5, 6, 7, et cetera, who are looking to run these large language models, but don't require ultra-low latency. What other options are there?

So there's other companies that are doing this, that are doing what we're doing. It's kind of the -- it's kind of a new class of CSPs that NVIDIA has invested in some of these. And so we -- it's a competitive process for us every time. That's the primary competition for us.

The next question is coming from Lucas Pipes of B. Riley Securities.

Wes, I wanted to follow up on the other financing solutions that you mentioned in your prepared remarks, specifically for AI. And I wondered if you could share some of the typical terms such as loan-to-value, amortization, schedules, borrowing costs, et cetera.

Sure. So it's a range. So you get typical vendor financing, right? So from the OEMs, we have terms from HPE. We actually have terms from Dell and from Supermicro. So there's typical vendor financing. And then there's another company that does the same style of vendor financing. And those terms with the one that we have used and we'll use already is kind of a 24-month term and fairly reasonable rates, high single digits on those. Then there's other options -- and the LPDs on that, Lucas, can be anywhere from 50% to call it 75%. And the GPUs, I think we've mentioned this previously, on the purchase price of the GPU, we recouped in the first 24 months of operating, so kind of the financing lining up with that number works fairly well. We're seeing other options that are more low double-digit kind of low teens financing. It's specialized specifically for this market, seeing those anywhere from 50% to 90% LTVs and the payback schedule is much less aggressive, I would say, on those, maybe 4- to 5-year type term. We haven't used that yet, but it's an option out there for us. But we're seeing quite a few different financing options and some like I said, fairly attractive rates.

I really appreciate that detail. That's very helpful. For my second question, I wanted to touch on the guidance. Very helpful. I appreciate the detail. I wondered if you could maybe share the breakdown of BTC hosting and AI and HPC respectively that would be embedded in that calendar -- sorry, fiscal year guidance.

Yes. I actually don't have that. I don't think we plan to share that, Lucas, but we will in the future as that ramps up.

The next question is coming from John Todaro of Needham & Co.

Congrats on the guide here. How many megawatts do you envision you're going to need to fulfill the $180 million of that initial contract with Character.AI. And then the second one with that other AI provider? And then on that, the third-party colo capacity you're using, is it already secured? And for how much of those contracts is it already secured for?

Yes. So the -- what I mentioned in the script, I believe, was that for the 26,000 GPUs between our owned facilities and third-party colo that we have secured, so we have space for all of those. And just for reference, John, so the first -- so I would just say that the 5,000 GPUs takes about 7.5-megawatts of IT capacity. So depending on which type of colo and what the PUE is in that colo, you can multiply into what the full amount of power you need. But for IT capacity, it will take 7.5-megawatts for the 5,000 or for 5,000 GPUs.

Got it. And then as we think about the margin profile over time, how would it change when you're using a third-party colo provider versus your own site, such as the megawatts at Jamestown?

Yes. Sure. So we have a slide that we actually used it at your conference that breaks down the three businesses and the margin we expect an op margin or an EBIT margin for the cloud piece of approximately 40%. And that is in third-party colo. And so when you think about it in our own facility, what you should think about is stacking our HPC margins that we outlined, plus the AI cloud margin.

The next question is coming from Mike Grondahl of Northland Securities.

First question, just you guys mentioned that the first customer, the original $180 million is you're working towards, and they signed an option for a second $180 million. Could you explain that second $180 million a little more? Does that sort of point to year 3 and 4? Or does that point to a 2-year contract starting near term and not really extending the terms?

Yes. No, it's not extending the term. It's pointing to additional GPUs. So the doubling of the number of GPUs.

Got it. And then earlier, you guys were asked a little bit to bifurcate the guidance, the $385 million to $405 million. I think before we've kind of updated models, we sort of -- we're in a rough range of about $300 million for the BTC hosting business. Are you still ramping towards that? Is -- if we were just to think about it roughly, is that about the level you think you can do in BTC hosting in '24? Or has that sort of been scaled back a little bit?

Well, it's scaled back just from the simple fact that Texas isn't on yet. And so we're going to miss 2 months of Texas running and then there will be some ramp time for Texas. We've talked about it coming online faster generally than is the construction is complete there. But there will be some ramp time. So annualized when they're fully operational at $300 million. And so obviously, we've already missed the June and July month for that facility. And then if you think about the AIP, I didn't give this to George, but I'm happy to give some color to this. But if you think about the AI compute side, you'll get a few million of revenue because we have the first piece that came online in July for character and so you get the month of August or first quarter, you should be thinking of a couple of million revenue there. And then as we deploy more with them, that will ramp. And then as we make all of the deployments up to the 26,000, you see that ramp through the April timeframe. And so if you think of a few million in the first quarter and then you think about the data center portion being that $300 million a year, $75 million a quarter, it gives you a pretty good idea if you're exiting at $150 million kind of where that -- where the AI cloud portion would be.

Fair. Yes. That's helpful. And then how do we -- your partnership with SMCI and Hewlett Packard, what are sort of the one or two things you gain the most from those partnerships? Is it basically sort of the OEM pipeline that they have? What -- just help us understand that.

Yes. So we get a couple of things. Obviously, getting delivery of these servers is fairly difficult right now. So getting kind of prioritized and getting delivery. But then the expertise that they all bring, I mean, Supermicro has been one of the largest manufacturers and shippers of this style of compute for a long time. And so their business at Southern renounce recently is ramping really well. But they have a lot of expertise as far as racking shipping, they have full solutions. They have tax come help set up, tune these, get everything operational. HP does the same thing. So there's a lot of expertise from that, but there is a pipeline that they -- the customers come to them and that they can bring to us, that's the way really we found a lot of our pipeline. So the biggest places we're finding it is OEMs, bringing customers to us. We have one other or another place that we find these customers. And then the venture capitalists that are -- have introduced the ones that they're invested in several of these companies. They have introduced us into other portfolio companies after us kind of onboarding our first customer fairly quickly. I mean, it wasn't all perfectly smooth, but I think we did that in a really short time frame and fast time frame, and everyone's happy with that. And so we've been introduced into other portfolio companies. So that's kind of how we get that pipeline. But the OEMs bring mostly those two things to the table, a lot of expertise.

[Operator Instructions] The next question is coming from Kevin Dede of H.C. Wainwright.

Yes, I was wondering -- I apologize if I sort of missed the boat on this one. But could you peel the onion back a little bit on the contracts themselves. I'm wondering, like with Character.AI, for instance, is it exclusive, binding? Is there any penalty for termination -- early termination on either side?

It's a standard contract. So we walked through kind of how these are structured and the first customer hits these option periods and the first customers executed the options, so we have the full contract for the $180 million. It's a noncancelable contract except for kind of standard contract terms if we fail to perform on the contract. But other than that, it's standard, it's not exclusive for either one of us. It doesn't -- obviously, it doesn't preclude us from doing business with anyone else and it doesn't stop any -- our customers from going and getting compute power elsewhere. But that's the basic structure.