Prepared remarks
Hello, and welcome to the White Fiber Second Quarter 2026 Earnings Conference Call. Good morning, and thank you for joining us. We will begin with prepared remarks from management. The operator provided instructions to participants. As a reminder, today's conference is being recorded. I would now like to turn the call over to your host, William Schnier, Senior Vice President of Capital Markets and Corporate Strategy at White Fiber. William, please go ahead.
Thank you, and welcome to the White Fiber Second Quarter 2026 Earnings Call. Joining me today are Samir Tabar, our Chief Executive Officer; and Justin Zhu, our Chief Financial Officer. Before we begin, I'd like to remind everyone that some of the statements we make on this call are forward-looking in nature and subject to risks and uncertainties that could cause actual results to differ materially. Such risks and uncertainties include, but are not limited to, those factors described in today's earnings press release, our Form 10-Q for the quarter ended June 30, 2026, filed today, as well as other filings we may make with the SEC from time to time. Our remarks today may also include non-GAAP financial measures. Reconciliations to the most directly comparable GAAP measures can be found in our Form 10-Q and in the earnings press release posted on our website. Following our prepared remarks, we will open the line for questions. With that, I'll turn the call over to Sam to discuss our performance. Sam?
Thank you, William, and thank you, everyone, for joining us. Last week marked the first anniversary of White Fiber's initial public offering. Over the past year, we've made substantial progress toward the company we set out to build. Most notably, we signed a transformational 10-year agreement, representing approximately $865 million of contracted revenue for 40 megawatts of IT workload at NC1. We've since advanced the project through construction and now into active customer deployment. We've also started operations and turned on revenue at our Montreal 3 location under our Cerebras agreement, expanded our development pipeline, strengthened our capital base and repositioned our cloud services business around larger, longer-duration opportunities. We are proud of what we've accomplished in our first year, but we aren't satisfied. We remain in the early stages of what White Fiber can become. Our most significant accomplishments remain ahead of us. As we enter this next phase, I'd also like to welcome Justin Zhu as White Fiber's Chief Financial Officer. Justin previously served as Senior Vice President of Finance and Chief Accounting Officer. He has been with White Fiber since its formation. He has a deep understanding of our business, financial operations and growth strategy. Eric is stepping away from his executive role at White Fiber to focus fully on Bit Digital. We thank Eric for his important contributions to White Fiber's development. Eric will continue to support White Fiber as a senior adviser and nonvoting observer to our Board. He will provide additional continuity through the transition. We believe this structure provides each company with increasingly dedicated financial leadership as both businesses continue to grow. Turning to our operating update. I'll begin with NC1, which remains our most important near-term operating and financial priority. NC1 has moved into active customer deployment. As of today, approximately 20 megawatts of IT capacity is available to support the installation and testing activities of Nscale and its investment-grade offtaker. Initial billing to our customer has now commenced for the initial tranches of capacity. Remaining equipment start-ups and testing are progressing very well. We expect the remaining capacity to be turned over progressively through August. By the end of this month, the full 40 megawatts of contracted IT load will reach a full run-rate billing. As we discussed last quarter, the pace of the ramp was affected by delivery and commissioning issues involving certain switchgear equipment. Those issues have since been resolved. Final deployment also requires tight coordination between the commissioning of our infrastructure and the installation and testing of customer equipment. We worked closely with Nscale on a phased turnover schedule that sequences the work being completed by both parties. While the ramp has taken a touch longer than we originally anticipated, the contracted economics of the agreement remain unchanged. The results speak for themselves. It took disciplined coordination across our team, our customer, the utility, our equipment vendors and our construction partners, all amid persistent supply chain constraints. We believe NC1 shows what White Fiber can do. It demonstrates our ability to execute complex, large-scale AI projects. Just as importantly, we have expanded an experienced operating team on the ground. The team spans facility operations, engineering and customer support. This is not simply a development project or a piece of powered real estate to us. It is a mission-critical facility built to operate continuously and support customers over long-term contracts. The people, systems and operating capabilities now in place reduce execution risk as NC1 moves toward full contracted operations. We established a foundation for continued expansion of the campus. We also established a long-term and positive presence with the local community. Ultimately, we're building a durable operating business in North Carolina. The initial 40-megawatt deployment is only the first stage at NC1. We expect Duke Energy to provide a delivery schedule for the next 45 megawatts of gross capacity in the near term. At that point, Nscale will receive priority notification of the available capacity in accordance with our existing agreement. We also received extremely strong inbound interest for this new upcoming tranche. We'll evaluate the path forward based on what we believe will deliver the best outcome for White Fiber. Beyond this, we are working with Duke Energy in further evaluating the potential delivery of an additional 200 megawatts of incremental power to the site. Together with the initial phases, that would bring NC1 to approximately 300 gross megawatts. This is a longer-term opportunity and remains subject to the utility process. We believe it shows how NC1 could scale over time. It also shows why securing the site early was strategically important. NC1 is our flagship facility. It validates White Fiber's ability to acquire, develop and operate large-scale AI infrastructure. We intend to repeat that capability across our pipeline. Turning to our Canadian portfolio. The most significant update is at NTL2. We had paused development while we evaluated the best use of that site. We've now decided to move forward. We plan to develop approximately 5 megawatts of gross capacity targeting completion around year-end. This decision is supported by active discussions with certain prospective customers. We're evaluating two deployment paths. The first is traditional colocation. The second is a vertically integrated deployment combining our data center infrastructure and our cloud services capabilities. We'll provide more details soon as customer discussions and the commercial structure progress. Moving on to our other sites. MTL1 continues to perform steadily. Recent customer renewals support a stable outlook, and we're evaluating a modest expansion of that facility. At MTL3, the Cerebras deployment continues to perform well. We're also pursuing additional utility capacity for that site that could support a meaningful expansion over time. The approval process remains ongoing. Beyond our existing portfolio, demand for power-ready, high-density AI infrastructure remains very strong. Demand is particularly acute for 2027 deployments. This reinforces our view that capacity able to reach the market within the next 12 to 18 months will remain extremely scarce. This is where our retrofit-first approach has a clear advantage. We prioritize sites with existing infrastructure and a credible path to power. That allows us to bring capacity to market faster than in traditional greenfield development. Speed to market is a key competitive advantage for White Fiber. We've built a substantial development pipeline. We're concentrating on the opportunities we can advance toward definitive commitments. We remain disciplined with capital. We prioritize sites with clear current and future power visibility, strong customer alignment, attractive return potential and a path toward project-level financing. We're also deliberate about sequencing our investments. As permanent financing for NC1 progresses, we expect greater flexibility to advance the next opportunities in our pipeline. We remain focused on moving forward on the right terms and in a way that supports disciplined, repeatable growth. Turning to cloud services. We made substantial progress in transforming the business around larger, longer-duration customer engagements and a more capital-efficient operating model. We streamlined the organization and concentrated our resources on the areas where White Fiber provides the greatest value, that being sourcing next-generation hardware, deploying complex clusters and operating infrastructure over the life of a customer engagement. We're encouraged by early results. Our commercial pipeline has expanded considerably. We are increasing our conversion of that pipeline into larger scale multiyear contracts. These agreements are supported by firm customer commitments. Customer prepayments and third-party equipment financing significantly reduced the equity capital required from White Fiber's balance sheet. We're also seeing an important shift in how customers select infrastructure partners. Larger buyers are consolidating their deployments among a smaller group of providers capable of supporting them at scale. While pricing remains important, customers are increasingly prioritizing engineering credibility, deployment execution and reliable ongoing operations. We believe these are the areas White Fiber is particularly well positioned. Since our last earnings call, we've entered into new multiyear cloud services agreements representing more than $540 million in aggregate contract value over their initial terms. Based on contracts signed to date, our cloud services portfolio is expected to generate more than $200 million of annualized revenue once fully deployed. One of the new agreements is with Base 10, an AI infrastructure platform focused on production inference workloads. Under the three-year agreement, we will deploy 1,392 NVIDIA B300 GPUs at a third-party data center in Ontario. The agreement represents approximately $165 million of contract value over its initial term with service targeted to commence in November of this year. Phase 1 also has the option to extend the deployment for up to two additional years, creating potential upside beyond this committed initial term. Separately, we entered into a three-year agreement with Prime Intellect, an AI-focused platform on large-scale model training and distributed compute. Under the agreement, we'll deploy 576 NVIDIA Vera Rubin 200 GPUs in Canada, marking White Fiber's first Vera Rubin deployment. The agreement represents approximately $108 million of contract value with service targeted to commence in the second quarter of 2027. This Vera Rubin deployment demonstrates the technical depth and expertise of our engineering team. It also aligns to our strategy of focusing on current and next-generation GPUs. Both of these deals expand existing customer relationships, and that illustrates our customers' confidence in White Fiber's engineering and operational capabilities. We also continue to advance our previously announced five-year deployment in the Paris region, which represents over $160 million of contract value. Following the completion of procurement and site-level arrangements, we're targeting an end of September ready-for-service date. Additionally, we entered into a five-year agreement with an existing customer supporting the deployment of 576 NVIDIA V300 GPUs in Iceland. The agreement represents approximately $87.5 million of contract value over its initial term with additional potential upside through revenue sharing. We expect deployment to commence later this year. Beyond these dedicated infrastructure deployments, we're seeing meaningful demand for our managed services offering. Under this model, customers fund the underlying hardware and data center capacity, while White Fiber applies its technical and operating capabilities to deploy and operate the infrastructure on their behalf. Managed services allow us to generate revenue without funding the underlying equipment, creating a hyper capital-efficient path to growth. This model will also leverage systems and personnel and expertise that are already in place. This creates the potential for attractive incremental margins with limited additional direct operating expense. We're in active discussions regarding several potential managed services engagements, including larger scale opportunities. We believe managed services can become an increasingly important capital-light extension of our business. To support cloud growth in 2027 and beyond, we've entered into an agreement with data center developer and operator Krambu. The agreement provides White Fiber with exclusive access to 100 megawatts of liquid-cooled colocation capacity beginning in 2027 with the potential to expand over time. Access to deployable power remains a key constraint across the industry. This agreement provides an important pathway to additional capacity for our cloud services business. Taken together, these developments demonstrate the progress we're making toward a scalable cloud services model. We can secure access to deployable capacity. We can provide dedicated infrastructure through long-term customer commitments. We can access third-party equipment financing, and we can apply our technical expertise to customer-funded infrastructure through managed services engagements. These models allow us to pursue longer-duration revenue while maintaining discipline around White Fiber's capital investment. Finally, we continue to advance our cross data center networking initiatives. During the quarter, we successfully demonstrated 111.2 terabits per second of bandwidth with guaranteed sub-millisecond latency across 83 kilometers. We believe our patent-pending technology has the potential to create significant platform value for White Fiber. By enabling certain AI workloads to operate across geographically separated facilities, it could allow us to aggregate smaller blocks of power and compute into a single integrated environment, thereby creating a virtual super cluster under one logical system. This could expand the commercial utility of capacity that might otherwise be difficult to monetize independently. This would also increase the value of White Fiber's broader site portfolio. We're now validating specific customer cases for this technology. We're targeting an initial commercial launch of this new technology by this September. Given the proprietary nature of the architecture and the early stage of commercialization, we're not disclosing all aspects of the technology and commercial model for now. Over time, we believe this opportunity could extend beyond White Fiber's own infrastructure through licensing and other commercial structures involving third-party facilities. Across both colocation and cloud services, the demand backdrop remains extraordinary. We're being deliberate about how we grow. Our priority is to pursue the right sites, customers and deployments. We will scale at a pace that allows us to execute consistently, maintain a high standard of service and continue building White Fiber's reputation as a trusted infrastructure partner. I'll now turn the call over to our Chief Financial Officer, Justin, to discuss our financial results. Go ahead, Justin.
Thanks, Sam. Second quarter revenue was $28.8 million, an increase of 54% from $18.7 million in the second quarter of 2025. Cloud services revenue was $23.8 million compared with $16.6 million in the prior year period. Revenue for the quarter included approximately $12.3 million associated with the previously disclosed customer termination. The termination also resulted in approximately $4 million of related expenses payable to the GPU lease provider, which was recorded in cost of revenue. Underlying cloud services results also reflected a temporary downturn between the termination of the prior contract and the commencement of the newly signed replacement contract. Colocation revenue was $4.7 million compared with $1.7 million in the prior year period. The increase primarily reflects the contribution from MTL3, which commenced operation under our agreement with Cerebras in the fourth quarter of 2025. Gross profit, excluding depreciation and amortization, was $17.1 million, representing a gross margin of approximately 59%, and this compared with gross profit of $11.5 million and gross margin of approximately 61% in the prior year period. G&A expense was about $14.8 million, down from $17.8 million in the first quarter. The sequential decline primarily reflected lower professional and consulting expenses and lower share-based compensation expense. G&A for the quarter also included approximately $2.2 million of bad debt expense associated with the previously disclosed customer termination. Adjusted EBITDA was about $5.5 million compared with $3.3 million in the prior year period. A reconciliation of adjusted EBITDA to net loss is included in our earnings release and Form 10-Q. Net loss was $15 million or $0.39 loss per diluted share. The net loss reflects a higher depreciation and interest expense associated with the expansion of our infrastructure and related financing activities. We ended the quarter with $56.1 million of cash and cash equivalents. Deferred revenue was approximately $143 million and primarily reflects customer prepayment associated with our NC1 site and cloud services deployments. During the quarter, we added approximately $83.2 million of project-level equipment and bridge financing to support the continued development of our colocation and cloud services infrastructure. As Sam discussed earlier, completing the permanent financing for NC1 will further strengthen our financial capacity and allow us to recycle capital into future development. Overall, the quarter reflected continued positive adjusted EBITDA and substantial investment in infrastructure supporting our contracted growth. We remain focused on converting the investment into recurring revenue and cash flow while maintaining discipline around capital deployment. I will now turn the call back to Sam.
Thank you, Justin. Before we open the call for questions, I want to leave you with a few thoughts. Last quarter, we said the pieces of our development model were beginning to come together. They are. Since then, NC1 has moved into active customer deployment and toward full contracted operations. We've also focused our pipeline on the opportunities best positioned to move forward. Importantly, we have recently entered into exclusivity with a consortium of well-known lenders for the proposed secured financing for NC1. The parties have commenced diligence, are negotiating definitive documentation and are working toward closing subject to customary approvals and conditions. This financing process has taken longer than we initially anticipated. But finally, reaching exclusivity and negotiating definitive documentation represent meaningful progress. If completed, the financing would return a significant portion of the capital invested in NC1. It would also allow us to advance the next site in our pipeline. But as my lawyers have advised me to say, there could be no assurance that the financing will be completed on favorable terms or at all. This financing would also complete the first turn of the development flywheel we've described. We acquire power-advantaged infrastructure, we secure long-term customer commitments, we develop and stabilize the asset, we then access institutional capital and recycle our equity into the next project. Completing that first turn would represent an important inflection point for our colocation business. We believe our next opportunity is also becoming increasingly tangible. Several sites have advanced significantly through our diligence process. Among the most actionable is a site that could support approximately 60 megawatts in 2027 and scale to more than 250 megawatts over time. The site has passed substantial diligence. We are now actively negotiating a purchase agreement as we complete the final stages of our evaluation. Power available at this scale in 2027 is scarce. Our retrofit-first approach can bring capacity to market faster than traditional greenfield development, creating a meaningful speed-to-market advantage in a supply-constrained environment. We believe this combination of scarcity and speed to market should support premium economics. Across our pipeline, we're increasingly prioritizing opportunities with investment-grade credit support. We believe this will enhance project finance ability and execution certainty. This reflects the same disciplined sourcing approach that produced attractive economics at NC1—advantaged power, speed to market and strong customer demand. We're not pursuing growth for its own sake. We're focused on opportunities that combine advantaged power, credible customer demand and a financeable contract structure. Completing the NC1 financing would strengthen our ability to act on opportunities that meet those standards. In cloud services, we're also converting strategy into signed contracts. The multiyear agreements we've signed since our last earnings call meaningfully expand our contracted revenue base and improve revenue visibility. These deployments are structured around firm customer commitments and are designed to be funded through customer prepayments and third-party equipment financing. This limits the capital required from White Fiber while allowing us to retain attractive economics. For the most part, for most of the past year, we've been building the individual pieces of this strategy. We're now beginning to demonstrate how they all work together. In colocation, we're moving toward a repeatable model for developing and financing long-term contracted infrastructure. In cloud services, we're pursuing longer-duration customer engagements designed to generate attractive returns with limited White Fiber capital. There is still important execution to be done ahead on NC1 and on the financing, but completing this first turn of the flywheel would position us to enter 2027 with greater financial capacity, a larger contracted revenue base and a more actionable development pipeline. We remain focused on execution, capital discipline and building durable value for our shareholders. With that, we're ready to take your questions. Joining us today for Q&A are White Fiber President, Billy Krassakopoulos; Chief Financial Officer, Justin Zhu; Eric Lang, an adviser to White Fiber and our former Chief Financial Officer; and Michael Francisco, Vice President of Cloud Services. Operator, please open the line.
Questions and answers
The operator provided instructions to participants. We'll take our first question from Nick Giles with B. Riley Securities.
It sounds like demand is really strong for the remaining available capacity at NC1. Just hoping you could speak to that commercial process and when you would ultimately cut it off, or if you would be willing to entertain other potential counterparties at this point?
Sure. Thanks, Sam. Nick, we're still in the early phases of that. It's a little early to comment on the timing of when we would be able to make that available for clients right now.
Fair enough, Bill.
I was going to add to that, but go ahead, Billy.
It is imminent. We're fully focused on, as Sam said, completing that first turn of the flywheel and Phase 1 of North Carolina. The next step is marketing and putting together a full project plan for Phase 2.
Great. I appreciate that. And then maybe just on the new site side — when you eventually acquire the site, where it stands today and what type of development work you would be willing to complete before any commercial signing to ensure 2027 delivery?
Bill, do you want to take that?
Sure. We're looking at similar situations to North Carolina One, buildings that we can go into quickly and retrofit. Our key advantage here is speed for ourselves and for our clients as well. The more quickly we develop these properties and get clients in them, it serves both purposes. The overall strategy that we're looking at is very similar to what we've accomplished at NC1.
We'll go to our next question from Greg Lewis with BTIG.
I was hoping we could talk a little bit about the cloud services business. Congratulations on bringing on a couple more customers. One of the things we've been hearing is that there's ample opportunities to bring on prepayments. How do you balance those upfront prepayments as you're thinking about structures versus the overall return on a multiyear cloud services business? Just trying to understand how you're thinking about that as you continue to build out the cloud business.
Glad you asked that question. We have Michael Francisco here, who is in the weeds on the cloud side. Michael, go ahead.
Thanks, Sam. The way we think about this is we evaluate all of our deals at the project level and look for healthy terms across the life cycle of each deal. When we restructured the cloud business earlier this year, we set a framework that forced us to think about how we run this business differently from much of the market, focusing on high-quality customers and deals that are cash flow positive throughout and that limit the amount of capital we must deploy from our own funds. Prepayments are a mechanism we can leverage to reduce the capital we deploy in support of these deals. We then look at the structure of the deal across its life cycle to ensure it is cash flow positive and get creative around the later years of those deals. For example, Base 10 has a two-year option for the customer to extend the deal. We can structure that to both benefit the customer over the term and allow us to adhere to the parameters we set out up front.
Okay. Super helpful. And then my other question was around the pipeline. How do we think about opportunities where at a given location you might have a colocation customer, but also, perhaps in another building at the same campus, run GPU-as-a-service? Is that something you envision to scale the business?
Yes — in fact, Michael, you should take this. We'll be doing that, I think, sooner rather than later.
Getting to the vertically integrated model has always been the goal. To make that happen, we need our development pipeline on the data center side to align with our customer pipeline on the cloud services side and customers of the right quality that allow us to get the right cost of capital to make the arrangement attractive. Billy and I have been talking about how to get this done, and the timeline on this has gotten shorter rather than longer. Ultimately, I think we will see a move to that. I don't think all cloud services will roll into the data center business, nor will all data center customers come from the cloud business, but we'll be opportunistic about how we do that and look for other opportunities such as the arrangement with Krambu to identify and build capacity for customers we don't place in our own data centers.
Super helpful. And congrats on getting NC1 off and running.
We'll next go to Raimo Lenschow with Barclays.
Congrats. I have two quick questions. One, how should we think about the long-run mix between cloud services, colocation, and managed services? The second is a quick follow-up after that.
Is your question about how we view the mix and the pros and cons of integrating colocation and cloud versus keeping them separate?
Yes — and whether to pursue one more than the other, or keep them slightly separate.
Aside from the fact that the market assigns different multiples to colocation versus cloud, and they require different skill sets, we have two separate teams working on those businesses by design. Integrating the two could affect multiples, but there's a strong reason to integrate because we can capture margin in both businesses. Michael, do you want to add?
There are benefits to both models. The way we have it structured today actually creates a healthy tension. For the cloud team to become a customer of the data center team and partner on vertically integrated projects, we have to earn that opportunity. The data center team is not beholden to the cloud organization; instead, the cloud team needs a compelling customer and economic case to displace demand. That tension is healthy because it forces clear goals and parameters around vertical integration. I could see benefits for both sides, and the desire for closer partnership is good for all of us.
Okay, perfect. And on the managed services approach, it helps you avoid deploying capital — what sort of margin profile should we think about there?
The managed services model is interesting. We built it anticipating enterprise adoption of AI, but demand has been greater than expected, including from other cloud providers, financial partners and frontier AI labs. From a margin perspective, we see a very healthy profile. I will pause on specifics for now, but one key benefit is that we start driving meaningful margin from day one because customers fund the CapEx and we are providing services on top. We have adjusted how we think about software development, focusing internal software work primarily on projects that help drive revenue and balance sheet growth. We also layer on third-party services to provide APIs that developers use with our bare-metal solutions. That structure creates layers of sales opportunities on top of managed services deals, producing incremental revenue and margin. In general, the margin profile will look more like a software offering than a hardware offering.
We'll take our next question from Brian Dobson with Clear Street.
You've been signing many contracts recently. As you discuss with clients, are they leaning more toward longer-duration contracts? And how is the execution at NC1 impacting those conversations?
I don't think execution at NC1 is directly related to the cloud contracts; they're separate businesses. Michael, do you want to speak about longer-duration contracts on the cloud side?
On the cloud side, we are seeing a shift. Last year and into this year, customers often wanted shorter-term arrangements, but the GPU pricing dynamics are leading customers to reevaluate procurement duration. For example, the cost per GPU hour for H100s is higher today than when they were released. Customers are considering total cost of ownership and are looking to preserve access to GPUs and bring down costs over time. Longer-duration contracts allow us to adjust the deal economics in ways that are favorable to the customer and help improve our margin across the life cycle. So market dynamics and customers' procurement thinking are driving longer-term deals.
Great. And about the next opportunity you mentioned being in late-stage diligence — would you favor a single campus opportunity or a multi-campus portfolio-type development?
Those are great questions. Billy, do you want to take that?
It's leaning toward a single-tenant opportunity. Think of the process we went through for North Carolina One—similar approach here. We're in the final stages of due diligence and will marry the opportunity to a client shortly if the technical due diligence goes well and we can execute the same game plan we used for NC1.
We'll take our next question from George Sutton with Craig-Hallum.
A lot of great updates. One challenge in the market recently has been 'not in my backyard' sentiment. It seems retrofits and your cross-DC initiatives could be answers to that. Can you walk through that in terms of what you're looking at?
Absolutely — and it's important to engage with the community. We had a Community Day at NC1 where we discussed how we're taking 85% less water than the previous tenant and that there will be less noise than the previous tenant. The retrofit format really does solve a lot of the pushback related to data center build-out, particularly versus greenfield projects. Engaging with the community and using retrofit rather than greenfield helps mitigate a lot of the objections being raised nationally. In fact, the approach has received media attention discussing how it helps mitigate pushback. Billy, do you want to add a few more points?
Sure. The GPU division and the data center division work hand-in-hand. Our cross data center platform will enable deploying that technology across our sites. The projects in our pipeline are intentionally modest size—30 megawatts, 60 megawatts, 99 megawatts—so we can execute quickly and bring capacity online faster than competitors, as we've proven with NC1. With the cross data center platform, we can bundle smaller sites into larger clusters. This work is a strategic, ongoing cooperation between the GPU division and the data center division.
I want to highlight that the technology is patent-pending and unique. The impetus for it is that if we have smaller modular sites and we can create a virtual super cluster across them, we can solve for disparate smaller sites by connecting them under one logical system. This technology could be transformational for the industry as well as for White Fiber.
Makes great sense. One other question on Krambu: you have access to 100 megawatts in 2027. Can you walk through how that's influencing your pipeline and the deal?
We have been working with Krambu for a while, and we have a couple of deals in late stages where we hope to leverage their facilities. Customers face the dual problems of GPU and power access with mismatched timelines. By partnering with Krambu, we create a single phase that aligns sales pipelines and supply chain timelines to provide clear availability to customers. The technical collaboration began because as we've shifted to the physical layer of GPU infrastructure, we've been working with them on cluster density and footprint. That collaboration, combined with aligned timelines, creates a compelling offering for customers with defined scaling plans. Because we are just announcing this partnership now, we'll see how it grows the pipeline, but it makes some of our pipeline more realistic in terms of fulfillment in the near and long term. We've already expanded a couple of customers this quarter and have others interested in partnering across the next 12 months. An arrangement like Krambu allows us to give a clear timeline and help fulfill those needs.
We'll take our next question from Paul Golding with Macie Capital.
Congrats on the progress. On the 200-megawatt incremental opportunity at NC1 that Duke is evaluating: how would you expect that capacity to come on? Would it be phased as with the first tranche, or could the load study lead to the full 200 coming on at once? And how might you market that?
Billy, do you want to take that?
It's a little difficult to say right now. It's too early in the process. There will more than likely be one or multiple steps before any schedule for that 200 megawatts is released.
Understood. And from a cloud perspective, on GPU availability: given your relationships, how confident are you about forward-looking availability of GPU supply, particularly for the Vera Rubin and V300 deployments?
I'll address this in two parts. First, our relationship with Krambu helps with availability. Leveraging our OEM and NVIDIA connections plus Krambu's work around cluster density and data center design improves our access to allocations. Second, when we restructured the business, we put a strategy in place to prevent chasing near-term GPU demand. Our customers appreciate that we are willing to say no to commitments we cannot absolutely meet. Many discussions are for deployments far enough out that we can be certain we can secure GPU allocations. For customers partnering with us to scale, we look out a timeline of about 12 months, which gives us time to source the right infrastructure.
Do you expect White Fiber to buy GPU compute speculatively in the marketplace based on visibility into demand and committed contracts?
Perhaps in the future, but I don't see that happening in the near term. It would have to be the right opportunity. Our restructuring parameters today don't favor speculative purchasing. There may be future circumstances where it makes sense, but today we are focused on fulfilling real customer demand based on real contracts with high-quality customers. Speculative purchasing is not a priority given our current capital allocation.
We'll take our next question from John Todaro with Needham & Company.
Can you give more color on the financing market for NC1 and future sites? Who are the most likely guarantors—chip manufacturers, banks, hyperscaler leases? How do you think about that?
Overall, our priority is solving for the lowest cost of capital. We've learned lessons in the NC1 financing process about contractual features that make assets more financeable. We want a structure with a counterparty that is financeable from day one and firmly underwritten. Whether that's directly with a hyperscaler, backed by a chip manufacturer, or another structure, the priority is the best financing terms and lowest cost of capital. There's not one type that we prefer above others—it's about the outcome for cost and certainty.
Got it. One more: on the cloud segment, pricing looks attractive. Could margin expand sooner than expected?
As we look across deal types, we expect margin improvement over time as mix shifts toward deployments we're targeting. Customers—especially mature AI labs and enterprises—understand the balance between cost and quality and the ROI of a reliable cluster. Many have been burned pursuing the cheapest options and have suffered downtime and missed SLAs. Because of our engineering capabilities, we deliver high-quality deployments, and customers are willing to pay more for that. We expect margin improvements for several reasons, but it will take time for that to develop.
We'll next go to Michael Donovan with Compass Point.
You now have operations across the U.S., Canada, Iceland and France. How do you think about geographic expansion from here? What other markets look attractive and what factors drive expansion decisions?
Those are different answers by business unit. Billy, do you want to talk about colocation?
On the colocation side, it's simple: match client opportunity with available power and timing. Right now, most opportunities we're looking at are in the United States and some in Canada.
On the cloud side, location is driven by customer demand—compliance, latency, performance all matter. Most deals today are in North America, but we are seeing interest in European deployments for GDPR and other reasons. Economics also influence where we place GPUs, and the U.S. is the most attractive market given current customer demand. Regarding Project Redwood and cross data networking: today we’re building it to support multiple use cases. We selected Modal as the customer for our R&D cluster because they were doing intense training workloads; training is the most demanding workload we can put on the system. So the technology can support both training and inference. There will also be use cases in telecommunications, edge computing and others. We'll see how customers want to use it once we bring it to market and finalize testing across the full fiber span.
And our last question comes from Nathan Francovitz with Cantor Fitzgerald.
On the cross data center product, how do you think about the long-term opportunity and the pathway to monetization? Is it primarily an internal capability, or can this be commercialized at scale?
It could be both internal and commercialized; we're even thinking about licensing it. Michael is leading the work stream and can elaborate.
Sam is correct. We see compelling internal use cases, such as connecting multiple large sites to create more economic clusters. There's value in aggregating fragmented power resources into a single cluster. If power becomes more constrained or if regulatory regimes create thresholds or taxes for large sites, this technology allows us to deploy strategically to preserve economic value by bringing sites together. In terms of how many sites it can aggregate, we will validate that in testing, starting with a hub-and-spoke model. We believe it can expand significantly, and we'll present real data as testing progresses.
Sorry, I was on mute. Thank you, everyone, for joining us today. We look forward to the next quarterly call. Until then, we'll be working very hard and delivering results. Thank you.
This concludes today's call. We thank you for your participation. You may now disconnect.