管理層發言
Greetings. Welcome to Core Scientific Fourth Quarter Fiscal Year 2025 Earnings Conference Call. Please note, this conference is being recorded. I will now turn the conference over to Jon Charbonneau, Vice President of Investor Relations. Thank you. You may begin.
Great. Good afternoon, and welcome to Core Scientific's Fourth Quarter and Full Year 2025 Earnings Call. Before we begin, I need to remind you that statements made on this call other than historical facts are forward-looking statements within the meaning of the Private Securities Litigation Reform Act of 1995 and are based on our current expectations. Words such as anticipates, estimates, expects, intends, and believes and similar words and expressions are intended to identify forward-looking statements. These statements are subject to risks and uncertainties that could cause actual results to differ substantially. For further information on these risks and uncertainties, we encourage you to review the risk factors discussed in the company's reports on Form 10-K, 10-Q and 8-K filed today with the Securities and Exchange Commission and the press release and slide presentation contained therein. The forward-looking statements we make today speak as of today only, and we do not undertake any obligation to update any such statement to reflect events or circumstances occurring after today. Today's presentation is available on our website, investors.corescientific.com. The content of this conference call contains information that is accurate as of today, March 2, 2026. Joining me today from Core Scientific are our CEO, Adam Sullivan; our Chief Operating Officer, Matt Brown; and our Chief Financial Officer, Jim Nygaard. We will conduct a question-and-answer session after management's remarks. We will now begin with remarks from Adam.
Good afternoon, everyone, and thank you for joining us. On our October 30 update call, we laid out four specific deliverables for this earnings call. First, we expect to sign at least one new customer, an important step towards diversifying our customer base. Second, we plan to sign one new power expansion contract at an existing site. Third, we expected to sign a new large land and power agreement. And fourth, we plan on making a financing announcement. These are the building blocks for our future growth, expanding contracted revenue, increasing our power optionality, widening our footprint, and funding growth in a way that is responsible and repeatable. Before we talk about those four priorities, let's talk about execution. The complexity of these build-outs is enormous, and we've made incredible progress on our CoreWeave build-out. Despite challenges in the market and the evolving criteria for operating the newest generation of GPUs, this requires the deep bench that Core Scientific has to adapt to changes in real time. Matt will cover the details, but here are the facts. As of this week, we'll have energized approximately 350 megawatts of capacity, of which close to 200 megawatts are currently billing. This puts us well at the halfway mark of the CORE contract. When we say energized, we mean power has been delivered and is generally within 90 days of billing. The natural lag between energization and billing commencement varies by site and customer requirements. Now let's put that in perspective. Last year, we energized as many megawatts as our closest publicly traded peers combined. We were building and delivering while they were still signing their first AI contracts. And going forward, to keep this simple and consistent, we'll report megawatts when they start billing. In this business, the market will always talk about demand. Investors should stay focused on what actually matters here: execution. Schedules will always move. It's easy at mile 2 of a marathon to say that you're on pace, but these are large and complex projects that expose who can actually execute. We've shown we can build and turn on capacity at scale and deliver for our customers. This leads directly into our pipeline. As this industry matures and evaluates opportunities against a more stringent criteria, we're confident we check those boxes through proven execution and true site readiness. And we're disciplined; we only sign contracts we know we can deliver on time and done right. That discipline is how we're building a durable, long-standing company, one defined by execution and known for being a great partner to our customers. That's what we set out to do, and it's exactly what we're doing today. And over time, that's what will separate us from the rest of the industry and position us to be a market leader for years to come. Now let me start by providing an update on the most visible item on our priority list: a new customer contract. We did not sign one by this call, and we are not satisfied with that. But the demand is there, and we have two sites under short exclusivity arrangements. We expect that this exercise will result in colocation leasing agreements in the near future. Our funnel is larger and broader than it was a few months ago, and we are in active discussions with hyperscalers, neoclouds, and large enterprises. This is a timing issue, not a demand issue. While we are operating under the merger agreement, hyperscalers simply would not engage with us. Those conversations restarted following the termination, and we have made significant headway. Deals of this level are not a one meeting exercise. It's a rigorous multistep process. While hyperscalers have a longer contracting process, the path to project financing and delivery can often be more straightforward. The bottom line is we are engaged, moving through the process, and competing for the right long-term opportunities. Now on neoclouds, there is meaningful demand across the industry. However, for those deals to work for us, there needs to be a strong balance sheet standing behind the contract. In most cases, that means a hyperscaler, chip manufacturer, or another investment-grade guarantee. Putting that structure in place requires the due diligence of both the neocloud and investment-grade guarantee, which has more coordination and steps. Because these guarantees are new for many parties, they take time to negotiate and finalize. This is one reason we believe you've seen fewer neocloud deals announced across the industry recently. We are not going to compromise on counterparty strength because that protection matters over the life of the contract. Second, we said we plan to add new power at an existing site, and we delivered in Dalton, Georgia. Dalton will expand to 450 megawatts of total gross power capacity, including 120 megawatts of uncommitted leasable customer capacity. Our Dalton site is strategically located about 90 miles from Atlanta, sitting in the middle of an incredibly attractive demand corridor. We've been working towards this expansion with local stakeholders for over a year. And to support it, we've secured an additional 175 acres of land. This is what execution looks like: long-term planning, deep coordination, and strong partnerships with the utilities and the local community. Late last year, we also increased leasable customer capacity in Pecos, Texas to 200 megawatts, an area that has seen significant traction for high-density compute. Given that demand, we're moving forward with the conversion of Pecos from Bitcoin mining to colocation. Pecos is in the Goldilocks zone for customer signing, meaning we've secured a general contractor, locked in long lead equipment, and conversion work is underway with a time line to RFS within 12 months. This means Pecos is within a time frame that customers are actively trying to solve for right now. Stepping back, our strategy remains the same. We expect every megawatt in our portfolio to be dedicated to colocation within the next 3 years. Third, we delivered on signing a new large land and power agreement through our contract to acquire a major new site in Hunt County, Texas. This site represents approximately 265 acres that we expect can support roughly 430 megawatts of gross power capacity or 285 megawatts of customer leasable capacity. This location is about 45 miles outside of Dallas in one of the fastest-growing colocation markets. We expect this to close by the end of Q1. And importantly, this site has a clear interconnection path. The ERCOT energization schedule was approved in 2024, and with power expected to begin coming online in 2027 and ramp through 2029. You've also seen the headlines around ERCOT. In our view, more discipline and transparency in that process is constructive. It helps reduce speculation and rewards companies with real sites, real plans, and the ability to execute. We believe this makes our two leasable sites in Texas, both Hunt and Pecos, even more attractive in the market given the clarity around their ability to deliver. As we sit here today, our pipeline is approximately 1.5 gigawatts of customer leasable capacity. This number is not a speculative position. It is not inflated with load studies. It only includes real opportunities with a clear line of sight to development, existing power under contract, new sites like Hunt and available incremental power at both new and existing sites. Power matters, and we stay disciplined on it. But in the broader market, power is often treated as the bottleneck, and we think that can be overstated. In practice, the bigger constraints are often securing long lead equipment and lining up experienced general contractors and subcontractors. The reality is simple. We already have more power in our pipeline than we can build over the next several years. And fourth, on financing. Our balance sheet remains strong, and we have a variety of financing options that Jim will cover here shortly. Looking at 2026, our priorities are straightforward: diversify our customer base and execute on the CoreWeave contract. We are focused on delivery, disciplined growth, and doing what we said we would do. I'll now turn it over to Matt to give an updated construction overview.
Thanks, Adam. Through 2025, our teams executed with intensity and precision. We stayed focused on what matters: our customers and building the infrastructure powering the Fourth Industrial Revolution. Our mission is simple: design and deliver AI factories at scale, purpose-built for accelerated computing. Every quarter felt like new architecture cycles, new GPUs, higher power densities, and new cooling paradigms that created extraordinary opportunity and real complexity. We maintained operations through unprecedented weather events across multiple regions, iterated designs in real-time to support the newest GPU platforms, and applied lessons from prior deployments to better align infrastructure delivery with evolving customer needs. Each challenge refined the system. Each build made the thinking machine better. Now let me take a step back and frame the magnitude of what the team accomplished over the last 14 months. Alongside our design-build partners, we broke ground on five AI factories supporting our 590-megawatt commitment to CoreWeave: two brownfield expansions in Denton, Texas and Marble, North Carolina; three greenfield campuses in Muskogee, Oklahoma, Dalton Phase 1, and Phase 2 in Georgia. In 2025, these five sites represented 1 million square feet of data center shell, nearly $2 billion of installed infrastructure assets, more than 5 million labor hours supported by an average of 3,300 workers on site, and over $5 billion in total project investment. This is one of the most significant AI expansions underway anywhere in the world. Let's start with Texas. Our 262-megawatt 400,000 square foot Denton campus made remarkable progress. By the end of Q4, Denton delivered 67 billable megawatts across three buildings with roughly half the campus energized. Today, Denton North is fully operational, running production GPU workloads, and represents 90 billable megawatts. At Denton South, the first 40-megawatt data hall has commenced building, and as of today, our next 41-megawatt data hall will begin the energization process. The remaining buildings on the South campus remain on track for Q2 energization with full campus completion by midyear. Denton alone currently represents approximately 130 billable megawatts, actively supporting more than 50,000 Grace Blackwell GPUs. In North Carolina, our 65-megawatt 250,000 square foot Marble data center achieved full site energization in 2025. Two of the three data halls were delivered by the end of the fourth quarter, representing 36 billable megawatts supporting approximately 15,000 Grace Blackwell GPUs. The third and final data hall is currently in commissioning and is expected to be delivered in the second quarter. Our customer is actively accelerating GPU deployments at the site this week. Next, at our Muskogee, Oklahoma campus, Phase 1, a 70-megawatt, 138,000 square foot data center has completed vertical construction and is now fully energized and has advanced into commissioning, remaining on track for full delivery in the second quarter. Finally, our Dalton, Georgia campus Phase 1, a 30-megawatt 52,000 square foot data center has also completed vertical construction and is now fully energized. Commissioning is progressing and preparing the facility for high-density liquid-cooled AI systems with full delivery expected in the second quarter. Then at Dalton Phase 2, a 145-megawatt, 250,000 square foot data center, vertical construction is currently underway with full delivery targeted for early 2027. This facility will serve as the final AI factory supporting CoreWeave's 590-megawatt commitment. Looking ahead, I want to outline our development and go-to-market strategy, Operation Forward Observer. This strategy is straightforward: advance development across multiple sites through the first commission data hall while simultaneously securing long lead equipment to enable rapid expansion. By progressing sites to this advanced stage before contract signing, we position ourselves ahead of our peers and winning colocation agreements. This approach provides customers with a high degree of certainty around RFS timelines, not only for the initial delivery, but also for seamless expansion into subsequent data halls. Executing this strategy strengthens our competitive positioning, enhances our leverage in negotiating favorable terms with a broad base of creditworthy customers. Let me walk through our initial forward observer sites. First is the Hunt campus, a planned 285 leasable megawatt AI campus strategically located near the Dallas-Fort Worth market. Our development teams are actively engaged in predevelopment work to deliver the full 285 megawatts across multiple buildings with initial delivery currently planned in the second half of 2027. Next, our Pecos campus, a planned 200 leasable megawatt campus in West Texas. Our development teams are mobilized and advancing early civil work and engineering on Phase 1, which is designed to deliver 185 megawatts of leasable capacity across multiple data halls with initial delivery expected to begin in early 2027. At Dalton, Phase 3 will consist of approximately 250,000 square foot greenfield data center planned to deliver 120 megawatts of leasable capacity across multiple data halls with initial delivery targeted for the second half of 2027. The development teams are mobilized and progressing through early civil work and engineering. Finally, construction is underway on the first phase of our 30-megawatt leasable data center in Auburn, Alabama. The site remains on track for its first 10 megawatts in the second half of 2026. Auburn is designed as a Tier 3 facility with dense connectivity positioned to serve multi-tenant enterprise AI customers. Engineering, preconstruction, and permitting are complete, and all long lead equipment is on site. As we close, the takeaway is simple. We've built a repeatable execution engine for AI infrastructure at scale. We've delivered more than 185 meaningful billable capacity, progressed multiple campuses through energization and commissioning, reached 350 megawatts energized, and expanded our development pipeline by 600 leasable megawatts to support the next wave of accelerated computing, all while continuing to enhance how we design, build, and onboard customers. Entering 2026, our priorities are clear: maintain alignment with customer GPU deliveries, stay ahead of the technology curve, and keep transforming megawatts into production-ready AI factories. We're proud of what the team accomplished in 2025 and even more focused on the path ahead in 2026. With that, I'll turn it over to Jim.
Thanks, Matt. 2025 was a transitional year for the company. While the vast majority of our revenue continued to come from our Bitcoin mining operations, our primary focus was on scaling the colocation business, including the ongoing build-out of capacity for CoreWeave. At the same time, mining activities continue to support the funding of the company as we progress through the transition. Although colocation revenue in 2025 was limited, we expect to reach an important inflection point in the coming months as we begin billing for additional megawatts, bringing colocation revenue to a level that will not only cover our operating costs but also drive significant margin expansion going forward. In terms of Bitcoin mining, we remain focused on operational optimization, and we'll continue to mine to cover contractual power costs. We finished the year with a very strong balance sheet with total liquidity of approximately $530 million. We also opportunistically sold just over 1,900 Bitcoin for approximately $175 million in January at materially higher prices above current market levels. At this time, we hold under 1,000 Bitcoin and expect to remain opportunistic going forward. In terms of a broader capital formation strategy, we have a full range of financing options available that we will continue to evaluate in the coming months and quarters as our needs evolve, including both sizable alternatives at the corporate level and the up to $4 billion that we can raise against our contracted capacity with CoreWeave at stabilization. These capital sources will fund investments in our pipeline sites going forward. At these sites, we will also utilize project-based financing structures with 60% to 85% advance rate on build costs, depending on customer credit quality and site characteristics. Finally, I want to address the historical restatement outlined in our 10-K filing today. In early 2025, we changed auditors to KPMG from Markham. As part of KPMG's normal audit procedures and our ongoing review of the conversion of legacy mining sites to colocation, we identified an error in our historical accounting going back to 2024 for certain property, plant, and equipment that was demolished as part of those conversions. Under the historical accounting treatment, demolition costs were capitalized, and existing carrying values were maintained. It was determined that these values and expenses should have been written off in certain historical periods. We have filed amended statements to correct the error. Please refer to the SEC or the Core Scientific Investor Relations website for today's filings. To clarify, there was no impact to revenue, adjusted EBITDA, or our net cash flow. While you will see a material weakness noted in our filings for the next four quarters, rest assured, we have taken the appropriate steps to strengthen our controls over non-routine accounting items going forward. As we look ahead, we are incredibly excited about the trajectory of our business. The demand backdrop for high-performance infrastructure remains strong, and we've positioned Core Scientific to capitalize on that opportunity with scale, operational discipline, and a clear strategic vision. We are building a differentiated data center platform with the capabilities and balance sheet strength to compete at the highest level. With an experienced and focused team, a growing pipeline, and a commitment to disciplined capital allocation, we believe we are not only well positioned for the coming year, but structurally set up to create meaningful long-term value for our shareholders. With that, I'll turn the call over to the operator for questions.
分析師問答
Our first question is from Jon Petersen with Jefferies.
Adam, you mentioned two deals that are currently in discussion. Can you provide more details about the potential sizes of those deals and the locations they might be in? Additionally, I'm curious about your selectivity regarding the types of tenants you are willing to engage with at this time, specifically how important credit quality is in your considerations.
Yes, happy to, and thanks for the question, John. It's helpful to look back at October 30 when we first emerged from the termination of the merger agreement. One of the things that we talked about is that we were engaged with a number of different counterparties, including neoclouds. But at the time, hyperscaler customers and certain large investment-grade counterparties were not willing to speak with us during that time, which was understandable. Where we have migrated our sales pipeline over the course of the past four months is we've continued or we've engaged with those large counterparties once again, which has been great to see. We're in discussions across a number of them with a number of our sites. The one thing I'll say is today, we sit with 500 megawatts under exclusivity arrangements with a large investment-grade counterparty that we're excited to continue to advance forward, and we're really looking forward to hopefully signing one of those over the near future.
Okay. All right. And as a follow-up, in your presentation, you list 700 megawatts of unannounced leasable customer power opportunities. Is that additional power you might get at existing land sites? Is that new land sites? How do we think about what that bucket is exactly?
Yes. That's really just a combination of both of those items. There are places where we may be waiting to sign certain extensions on power at an existing site due to certain collateral requirements until we're closer on customer signing, or it might be sites that we have under exclusivity that are completing due diligence but have the confidence to be able to bring those to customer conversations as opportunities that we can present to them.
Our next question is from Brett Knoblauch with Cantor Fitzgerald.
Adam, on the new site in Hunt County, I think for the most part, the deals we've seen get signed kind of are on maybe sites with energized power today. Obviously, this isn't going to be energized until next year. Could you talk about maybe the level of confidence you have in being able to get a lease signed for that site, even though power is going to be delivered at a later date?
Yes. I mean, for us, really, it doesn't matter if the power is available today because it's going to take time for us to construct and build that site. So as long as the power and the ramp schedule that we've been provided by the utility matches with our construction schedule, that is acceptable to potential customers. So we feel great about the Hunt site. We see it today that the site is not impacted by Senate Bill 6 or any of the recent ERCOT changes. We've also been told that this project will not be restudied by ERCOT. So it gives us a lot of confidence in that site and that project. We're excited about building out another large-scale campus just outside of Dallas.
Awesome. And then maybe just as one follow-up from a demand perspective and maybe pricing perspective, it feels like the deals kind of have gotten better month after month, quarter after quarter. Are you guys seeing that on your end when you are having conversations with prospective tenants? Just curious kind of your thoughts for the overall pricing environment and where we're heading.
Yes. We've definitely seen pricing continue to shift. Part of that's driven by equipment prices and labor prices continuing to rise in the market. And so you're seeing a similar move in terms of leasing economics. That's one of the reasons why we launched our project with going forward with securing long-lead equipment, securing trades at sites, and beginning civil work across a number of different locations was so that we could lock in economics at those sites, essentially locking in what our costs are going to look like while we're still in an environment where we're seeing lease rates continue to move higher. So that's something that's more protective for our business, but it's also an offensive approach for us to continue to attack the market and put ourselves in a position to really compete on deals with hyperscalers because they're expecting capacity delivered sub-18 months and in some cases, sub-12 months. For us to be able to put ourselves in those positions, we have to be making the moves that we're making today related to really securing site readiness around these new locations.
Our next question is from Darren Aftahi from ROTH Capital Partners.
Two, if I may. Just on the Hunt County site, could you talk about what the rough payment was? And then I know you said it'd be energized by 2027. How does that kind of energization scale up? And then a follow-up, Jim, you've made a comment about financing against the CoreWeave deal when there's 'stabilization,' can you just kind of enlighten us what that actually means and perhaps that six months after all the campuses are built out?
Yes. Thanks, Darren. So we'll be announcing further details related to the Hunt County site as that site gets to close later in this quarter. As it relates to energization, as we look at the megawatts, there are tail megawatts here, but really the energization schedule ramps alongside what our construction schedule looks like. We feel how that site looks today in terms of our site readiness and our ability to deliver against the ramp schedule provided by the utility. We think that puts that site in a very strong position in terms of checking a number of different criteria that both hyperscalers as well as other large offtake companies may have. So we're excited about that site and how that continues to move forward. I'll let Jim take the last question you had.
Thanks, Adam. When you look at the size of our contract with CoreWeave at 590 megawatts, it represents somewhere between $5 billion and $5.5 billion of total infrastructure. So when I say stabilization, and I indicate the availability of capital up to $4 billion, I'm referencing the full stabilization of the contract relative to the asset base that we're constructing. The reality is that this is different than what is more commonly structured in project finance terms where you're borrowing the money upfront and building later. We have substantial availability under that asset base to borrow a good portion of that $4 billion today. But the scaling is quite fast because we are already at such significant progress on the billing; we will get through the vast majority of that before the end of this year.
Our next question is from Henry Hearle, and he is from B. Riley Securities.
This is Henry Hearle on for Nick Giles. I wanted to follow up on the new Hunt County site. Specifically, what does the site kind of look like today? And are there any preliminary permits that are needed before construction can begin?
Yes. I appreciate the question. So today, the site is essentially what we have to do to energize that site is there's still a substation that needs to be built. When we look at the utility energization schedule and the construction schedule, we feel like we can start energizing that site in late '27, but that's going to require us to start to get the process rolling both in terms of our preconstruction activities and getting the substation going here in pretty short order.
And then just on preliminary permits for construction, any color on that?
So we've gone through pretty much all of the EFA sort of Phase I studies and geotechnicals for the site. We have schematic designs in place, and so we have a pretty good idea on the development strategy for that site. It's really just a matter of finalizing design docs, getting to IFPs, and then releasing those for permitting.
Our next question is from George Sutton with Craig-Hallum Capital Group.
Matt, you mentioned new architectures, cooling, and new GPUs, which seems a bit frustrating. I'm curious about how much change you're experiencing compared to previous sites as you develop new ones.
Yes. So the evolution, I think, on the technology stack, obviously, when you're building projects that are taking 12 to 18 months to sort of get out of the ground and you're going through sort of multiple technology changes, and trying to adapt to those in real-time. In our case, we started our very first data center with CoreWeave with H100. Then we quickly evolved into NVIDIA's first iteration of Grace Blackwells, NVL36s, and then the NVL72s, and then from the GB200 platforms into the GB300 platforms and sort of having to adopt the data halls kind of in flight to those technology iterations. Recently, NVIDIA released its reference architecture for Vera Rubin, and we have a pretty good idea of what that paradigm shift is going to start to look like both from a cooling and power distribution standpoint. I would say our teams are starting to factor those changes into our new projects, and so we're pretty future-proofed in terms of what we're expecting next to happen. Additionally, Google's TPUs and TPUs becoming more prevalent in the market, we're also evaluating how we take our standardized base of design so that we have a very predictable, repeatable approach to getting products to market that is both adaptable to what we're doing today and what we're expecting from NVIDIA tomorrow, and these new chipsets coming to market like TPUs as those become more prevalent even outside the Google ecosystem. A lot of moving parts, a lot of things happening in the technology ecosystem, but I think our engineering team and our development teams are well ahead of the curve in thinking about how to adapt to those.
Very helpful. And then just a follow-up for Adam. You mentioned the broader and larger funnel of groups that you're talking to, but you also have suggested that we need investment-grade guarantees at some point. How broad can that funnel really be relative to those guarantees? Are you seeing that availability?
Yes. I would say we're seeing a pretty wide range, and that range has continued to expand. It's helpful to think back on 2025. Demand from both neoclouds and from AI labs in the first half of 2025 did not really see many new deals announced. That changed dramatically in Q3 as guarantees were introduced in the market. We saw some deals backed by Google. I believe over time, we will continue to see the evolution of these guarantees, and the ones we've seen so far come from a wide variety of sources. We've listed hyperscalers, chip manufacturers, and other large investment-grade guarantors. That's really what we're seeing in the market; it's a wide range. They look different, covering the full spectrum from debt guarantees to full lease guarantees. I think what we're going to see in 2026 is really a centralization on terms related to those guarantees and wrappers. Some of the delay and pause you've seen in some of the neocloud and lab signing over the past three months has been related to what those guarantees are going to look like because I don't think they're going to look like they have in the past. We're in the process of negotiating with certain guarantee counterparties here, and we're hopeful that these counterparties begin to centralize on the right guarantee for data center developers to fund these developments.
Our next question is from Mike Donovan with Compass Point.
In prepared remarks, you stated you have more power in your pipeline than you can build over the next few years. Can you share a range of megawatts you are confident you can bring online per year? And are there any areas of concern today around supply chain or labor availability?
Yes. I appreciate the question. In terms of what we think kind of an order of magnitude of what we can develop, a lot of that is customer-driven. We announced in our strategy that we're progressing multiple sites through the design-build process into 2027. As customers step into those, that's really going to drive what the scalability and pace of acceleration in terms of how many megawatts we build out. In previous quarters, I think we've guided around the idea that we could, in theory, build out as much as 500 megawatts in a single calendar year. I think that's certainly possible, but it's going to require customers stepping into these projects early so that we can line up the financing and the supply chain in order to scale those out to that sort of order of magnitude. Another way of saying that is our internal capacity to take on 0.5 gigawatts in a year and an 18-month time horizon is something we feel comfortable with. It's really a matter of lining up economics and financing to support that strategy. Hopefully, that answers your question.
Our next question is from Jon Hickman with Ladenburg Thalmann.
Could you elaborate on your comments like right upfront? You talked about the billing. There's a 90-day delay in when you energize and when you get the bill. Is that what you were telling us?
I can handle the first part of that. This is Matt. Kind of what we're referring to, as we turn on power to a building, we achieve our energization milestone, essentially hydrating all the equipment with electrons. From that point, each data hall within the building has to go through its subsequent commissioning phases, the fully commissioned test commission, and go through all the integrated testing to operationalize each of those data halls within that structure. So from the time we start energization to the time that we fully commission those data halls could roughly range into the 90-day time horizon for which we would expect to turn on revenue.
Okay. And then I think I missed this number, but as of the end of the year, how many megawatts had you delivered to CoreWeave?
Yes. So we had energized 213 megawatts by the end of the calendar year. We had fallen just slightly short of our one data hall goal. But as we said in this earnings call, we have more than made our way back ahead of schedule with 350 megawatts energized and nearly 200 megawatts billing. The step function of progress over the last couple of months has been pretty remarkable.
And you said that by midyear you'd have it all energized?
By the end of 2027 or the early part of 2027, all contracts should be fulfilled and fully delivered to CoreWeave.
Our next question is from Kevin Dede with H.C. Wainwright.
Adam, can you drill in a little bit about Alabama? It just seemed to be a little bit of an outlier at 30 megawatts. I'm just wondering how you sort of process that in this grand scheme of landing hyperscalers.
Yes, absolutely, Kevin, and thanks for the question. The 30-megawatt site in Alabama is a site that we saw early on as a location that could move quickly. We also recognize the power constraints in Georgia and the low latency that Auburn has available to it. That's a site that we believe, sitting here today, based on our customer conversations, has interest across a number of different potential counterparties. We're using it as leverage for larger contracts. It's important to note that hyperscalers are not only focused on the larger sites and larger campuses; they're also looking for backfill across certain locations to serve specific markets. We think Alabama and Auburn specifically serves that very well, so we're excited about that project and we're looking forward to landing a customer there as well.
So do you think it fits the bill for an inference type solution? And if that's the case, is Matt sort of reorganizing the way sites are constructed and fitting potential use case changes?
Yes. It's definitely a site that's going to be utilized for inference use cases. But Matt, I'll let you take it related to infrastructure design.
Yes. Auburn is unique from a couple of different standpoints. One, it sort of has the makings of a much more traditional multi-tenant data center, Tier 3 type facility, with multiple 10,000 square foot data halls, high degrees of security, and a massive amount of connectivity. More akin to what you might find in Digital Realty or Equinix or a modern Equinix type facility. From that standpoint, we think we look at Auburn as really an entry point for inference AI loads but also in the enterprise segment since the enterprise segment tends to be on smaller deal sizes and constructs, with the needs for dense connectivity, multicarrier-neutral environments, making it ideal for a number of different customer segments, both on AI and non-AI segments. In terms of adjusting to inferencing versus large language model sites, the technology standpoint, we don't see a ton of difference. The power density needs between those may be slight variances, and I think the cluster sizing might be slightly different between an inference cluster and a large language model cluster, driving a little bit more segmentation potentially within the campus at some point.
Okay. Adam, before I get the hook, and nobody has really asked about Bitcoin mining. And I know it's not a priority, but it's still the lion's share of revenues for a little while anyway. Can you give us some insight on how you expect this year to fall out from a hash rate perspective and whether or not you're chasing down those block miners that you thought you might look at the end of last year?
Yes, absolutely. It's been interesting to watch what's happening in the mining environment, right? We're seeing hash prices go to levels that have never been seen before, dipping below $0.03 was definitely something that I think folks thought might happen probably in 2027 or 2028, just given where machine efficiency is today. For us, that business is still essentially in runoff today. We're trying to manage our machine fleet based on what our minimum power draw requirements are across a number of different sites. That's something we'll continue to operate in that mode over the course of this year. We're really just optimizing now to ensure that we are hitting our minimum power draw requirements across our portfolio. In terms of the block units, those units are getting installed today. That's something that's going to help us maintain productivity across our mining portfolio and really help us hit our minimum power draw requirements profitably, given that a majority of our machine fleet is anywhere from 4 to 5 years old today, which will allow us to do.
Our next question is from John Todaro with Needham & Company.
Progress so far. There's been some conversations of NVIDIA backstopping a number of neoclouds. It would just potentially open up the type of customers you guys could sign with by quite a bit. Just wondering if that's starting to happen in some lease discussions. Any commentary there?
Yes. I think we're going to see all chip manufacturers start to play the game of guarantees to help them secure their customers moving forward and really lock in architecture in the data center around their GPU chipset. It's definitely something we've seen, and I think it's going to continue to evolve, as I noted earlier. But I would expect both hyperscalers and chip manufacturers to continue to march down the path of looking to provide guarantees for both neoclouds as well as labs.
Our next question is from Mike Donovan with Compass Point.
In prepared remarks, you stated you have more power in your pipeline than you can build over the next few years. Can you share a range of megawatts you are confident you can bring online per year? And are there any areas of concern today around supply chain or labor availability?
Yes. I appreciate the question. In terms of what we think kind of an order of magnitude of what we can develop, a lot of that is customer-driven. We announced in our strategy that we're progressing multiple sites through the design-build process into 2027. As customers step into those, that's really going to drive what the scalability and pace of acceleration in terms of how many megawatts we build out. In previous quarters, I think we've guided around the idea that we could, in theory, build out as much as 500 megawatts in a single calendar year. I think that's certainly possible, but it's going to require customers stepping into these projects early so that we can line up the financing and the supply chain to scale those out to that sort of order of magnitude. Another way of saying that is our internal capacity to take on 0.5 gigawatts in a year and an 18-month time horizon is something we feel comfortable with. It's really a matter of lining up economics and financing to support that strategy. Hopefully, that answers your question.
Our next question is from Jon Hickman with Ladenburg Thalmann.
Could you elaborate on your comments like right upfront? You talked about the billing. There's a 90-day delay in when you energize and when you get the bill. Is that what you were telling us?
I can handle the first part of that. This is Matt. Kind of what we're referring to, as we turn on power to a building, we achieve our energization milestone, essentially hydrating all the equipment with electrons. From that point, each data hall within the building has to go through its subsequent commissioning phases, the fully commissioned test commission, and go through all the integrated testing to operationalize each of those data halls within that structure. So from the time we start energization to the time that we fully commission those data halls could roughly range into the 90-day time horizon for which we would expect to turn on revenue.
Okay. And then I think I missed this number, but as of the end of the year, how many megawatts had you delivered to CoreWeave?
Yes. So we had energized 213 megawatts by the end of the calendar year. We had fallen just slightly short of our one data hall goal. But as we said in this earnings call, we have more than made our way back ahead of schedule with 350 megawatts energized and nearly 200 megawatts billing. The step function of progress over the last couple of months has been pretty remarkable.
And you said that by midyear you'd have it all energized?
By the end of 2027 or early 2027, all contracts should be fulfilled and fully delivered to CoreWeave.
Our next question is from Kevin Dede with H.C. Wainwright.
Adam, can you drill in a little bit about Alabama? It just seemed to be a little bit of an outlier at 30 megawatts. I'm just wondering how you sort of process that in this grand scheme of landing hyperscalers.
Yes, absolutely, Kevin, and thanks for the question. The 30-megawatt site in Alabama is a site that we saw early on as a location that could move quickly. We also recognize the power constraints in Georgia and the low latency that Auburn has available to it. That's a site that we believe, sitting here today, based on our customer conversations, has interest across a number of different potential counterparties. We're using it as leverage for larger contracts. It's important to note that hyperscalers are not only focused on the larger sites and larger campuses; they're also looking for backfill across certain locations to serve specific markets. We think Alabama and Auburn specifically serve that very well, so we're excited about that project and we're looking forward to landing a customer there as well.
So do you think it fits the bill for an inference type solution? And if that's the case, is Matt sort of reorganizing the way sites are constructed and fitting potential use case changes?
Yes. It's definitely a site that's going to be utilized for inference use cases. But Matt, I'll let you take it related to infrastructure design.
Yes. Auburn is unique from a couple of different standpoints. One, it sort of has the makings of a much more traditional multi-tenant data center, Tier 3 type facility, with multiple 10,000 square foot data halls, high degrees of security, and a massive amount of connectivity. More akin to what you might find in Digital Realty or Equinix or a modern Equinix type facility. From that standpoint, we think we look at Auburn as really an entry point for inference AI loads but also in the enterprise segment since the enterprise segment tends to be on smaller deal sizes and constructs, with the needs for dense connectivity, multicarrier-neutral environments, making it ideal for a number of different customer segments, both on AI and non-AI segments. In terms of adjusting to inferencing versus large language model sites, the technology standpoint, we don't see a ton of difference. The power density needs between those may be slight variances, and I think the cluster sizing might be slightly different between an inference cluster and a large language model cluster, driving a little bit more segmentation potentially within the campus at some point.
Okay. Adam, before I get the hook, and nobody has really asked about Bitcoin mining. And I know it's not a priority, but it's still the lion's share of revenues for a little while anyway. Can you give us some insight on how you expect this year to fall out from a hash rate perspective and whether or not you're chasing down those block miners that you thought you might look at the end of last year?
Yes, absolutely. It's been interesting to watch what's happening in the mining environment, right? We're seeing hash prices go to levels that have never been seen before, dipping below $0.03 was definitely something that I think folks thought might happen probably in 2027 or 2028, just given where machine efficiency is today. For us, that business is still essentially in runoff today. We're trying to manage our machine fleet based on what our minimum power draw requirements are across a number of different sites. That's something we'll continue to operate in that mode over the course of this year. We're really just optimizing now to ensure that we are hitting our minimum power draw requirements across our portfolio. In terms of the block units, those units are getting installed today. That's something that's going to help us maintain productivity across our mining portfolio and really help us hit our minimum power draw requirements profitably, given that a majority of our machine fleet is anywhere from 4 to 5 years old today, which will allow us to do.
Our next question is from John Todaro with Needham & Company.
Progress so far. There's been some conversations of NVIDIA backstopping a number of neoclouds. It would just potentially open up the type of customers you guys could sign with by quite a bit. Just wondering if that's starting to happen in some lease discussions. Any commentary there?
Yes. I think we're going to see all chip manufacturers start to play the game of guarantees to help them secure their customers moving forward and really lock in architecture in the data center around their GPU chipset. It's definitely something we've seen, and I think it's going to continue to evolve, as I noted earlier. But I would expect both hyperscalers and chip manufacturers to continue to march down the path of looking to provide guarantees for both neoclouds as well as labs.
There are no further questions at this time. That will conclude today's conference. You may disconnect your lines at this time and have a wonderful day.