BCE Inc. ($BCE)

Good morning. Welcome to the BCE conference call following the announcement of the Bell AI Fabric Data Center in Saskatchewan. I would now like to turn the meeting over to Kris Somers. Please go ahead, Mr. Somers.

Thank you, Matthew. Good morning, everyone, and thank you for joining on short notice. As you're aware by now, this morning, we announced that we are building a new 300-megawatt data center in Saskatchewan. The purpose of today's call is to provide you with additional details and perspective on this project. This morning, we will have prepared remarks from Mirko Bibic, President and CEO of BCE and Bell Canada; John Watson, Group President of Business Markets, AI Fabric and Ateko; and Curtis Millen, our Chief Financial Officer. We will also have with us today Dan Rink, President of AI Fabric, who will be on the line to support Q&A. Before we begin, I'd like to draw your attention to our safe harbor statement on Slide 2, reminding you that today's slide presentation and remarks made during the call will include forward-looking information, and therefore, are subject to risks and uncertainties. Results could differ materially. We disclaim any obligation to update forward-looking statements, except as required by law. With that out of the way, I'll turn the call over to Mirko.

Good morning, everyone. This morning, we've dialed into this call from Regina, where we announced a transformational expansion of Bell AI Fabric with a 300-megawatt AI data center in Saskatchewan, our largest investment ever in the province. For 145 years, Bell's investments in communications technology have been integral to Canada's economic prosperity. Now, our vision to lead in AI infrastructure and solutions and to drive the next wave of economic growth has clear momentum. I want to frame this announcement around 3 clear points. First, it reinforces and accelerates Bell's long-term growth profile and is incremental to the financial framework we laid out at Investor Day. Second, it materially strengthens Canada's sovereign AI capacity and AI infrastructure, supporting secure Canadian-controlled environments for governments, public institutions and enterprises. Third, this is a disciplined infrastructure investment, fully aligned with our clear strategic priorities and backed by a strong partnership with the Saskatchewan government, SaskTel, SaskPower and importantly, long-term contracted customers and contracted revenue. With this announcement, we now have line of sight to monetizing approximately 800 megawatts of power, including the 300 megawatts we're announcing today and the 73 megawatts we referenced at Investor Day back in October. We're also very aware that a project of this scale must be diligently managed, including construction and counterparty considerations. That's exactly why we structured this investment the way we have and why we are confident in our ability to execute. Bell has a long track record of delivering large-scale complex infrastructure projects across the country, including our multiyear, multibillion-dollar fiber build-out, which we executed with discipline, rigor and a clear focus on risk management. We're approaching this project with that same mindset. Phased, contracted and grounded in capabilities we have proven repeatedly at scale. From a financial perspective, this project meaningfully improves our revenue, EBITDA and free cash flow growth while strengthening our credit profile and dividend coverage over the medium term. What also differentiates Bell here is our ability to deliver solutions by bringing the right partners into our ecosystem. Through Bell AI Fabric, we combine infrastructure, connectivity, security, AI integration and services, working with best-in-class partners rather than trying to do everything ourselves. That partner-led integrated approach allows us to meet customer needs at scale while remaining disciplined on risk and capital. Let me start with the why. AI is rapidly becoming foundational infrastructure for governments, enterprises and research institutions, particularly in regulated data-sensitive environments. What customers are telling us very clearly is that they don't just need access to compute, they need secure, sovereign Canadian-controlled infrastructure integrated with connectivity, security and services. That's exactly the need Bell AI Fabric was built to address. With this project, Bell is investing to build the physical infrastructure, the space, the power, cooling and connectivity required to support AI workloads at scale. We're making this investment with clear visibility to significant contracted revenue, EBITDA and free cash flow. Importantly, we are not buying chips, and we're not speculating on future technology winners or cycles. We continue to partner with best-in-class compute providers, and we deploy capital only once we have customer commitments. This Saskatchewan facility is a clear example of the discipline we outlined last year, driving significant growth while managing risk. We've secured long-term tenant agreements with Cerebras and CoreWeave, 2 large global AI leaders with strong access to capital. They will supply and fund the compute hardware, and they will utilize the full 300 megawatts of capacity. Bell's role is the infrastructure layer, where we have a durable advantage and where returns are infra-like. We know that AI is here to stay, that it's transformational and that it will be a major input to economic growth. At the same time, it will continually evolve. And regardless of how that evolution unfolds, AI will always require secure power, space, connectivity and sovereign infrastructure, which is exactly where Bell is positioned. A key element, as I mentioned, is sovereignty. We have the flexibility to allocate a significant amount of capacity to sovereign AI workloads, ensuring that Canadian governments, public institutions and enterprises can run advanced AI workloads under Canadian jurisdiction with data residency and data ownership and control preserved. Until these sovereign workloads materialize, our customers will be using this capacity with other offtakers. This project also reinforces what we communicated at our recent Investor Day. Bell has renewed momentum backed by a unique set of differentiated assets working together and tightly aligned to our core strengths. Our broader AI-powered solution strategy is a great illustration of this momentum. AI Fabric does not stand alone. It complements Ateko and Bell Cyber, allowing us to participate across the AI value stack, infrastructure, security and integration, while remaining disciplined on risk. Before I hand it to John, I want to be very clear on one final point that matters, of course, to investors. This investment does not change our financial philosophy communicated at Investor Day. Today's announcement is contracted, it's phased and it's return driven. It strengthens our long-term EBITDA and free cash flow profile, and we remain laser-focused on achieving our 3.5x net debt leverage target by the end of 2027 and continuing to delever in 2028 and beyond. John will now walk through the operating model and execution.

Thanks, Mirko. Good morning, everyone. I'll focus on what we're building, how it comes online and how risk is managed. This is a 300-megawatt purpose-built AI center located in the rural municipality of Sherwood just outside of Regina. The site was selected for power availability, scalability and redundancy. This project is made possible because of the Saskatchewan government's strong partnership and engagement. Saskatchewan wants to be at the center of AI development in Canada, and we're proud to be part of their vision. We will construct the Saskatchewan data center in phases with each phase coming online in sequence. As we've outlined in the deck, the facility is expected to come online through 2027 and be fully operational by 2027 year-end. To support disciplined execution, we're in the midst of a competitive construction RFP process and have received bids from leading national data center construction firms. We are not self-building. We're deliberately selecting experienced partners with proven delivery track records at this scale. In parallel, we're actively managing procurement risk by placing early orders for long lead time equipment, including generators, chillers and key building materials and arranging for warehousing where appropriate. This approach is designed to reduce exposure to supply chain constraints and avoid scheduled delays. On the commercial side, Cerebras and CoreWeave are anchor tenants. They will fully utilize this data center. They contract with Bell for long-term access to space, power and infrastructure, and they're responsible for contracting with their end customers for the majority of workloads. For the sovereign component, Bell will work with partners, including SaskTel to support Canadian public sector and enterprise demand. From a risk mitigation perspective, there are 3 things to highlight. First, construction and execution risk. Through phased deployment, modular design, competitive procurement, early ordering of long lead time equipment and strong municipal and provincial engagement, we have taken steps to limit the risk associated with construction. We're building on experience from our recent AI data center developments that were delivered using standardized, reproducible designs, repeatable processes and experienced delivery partners, which gives us confidence in our ability to execute this project as planned. In addition, we will supplement Bell's internal teams with specialized on-site data center engineering and operations partners, reducing execution and uptime risk. Bell has been building infrastructure in Canada for 145 years. So this is a familiar path for us and one in which we'll execute as we always do. Second, counterparty and demand risk. In line with the plans we outlined at Investor Day, we have secured tenants before committing capital. This project is fully contracted with 100% of capacity already sold. Our contracts are long-term in structure to provide infrastructure-like cash flows and Bell does not take uncontracted exposure to AI utilization or AI pricing. Our pipeline is significant, and we'll continue to see strong demand for capacity from customers and potential partners, providing optionality beyond this fully contracted project. Third, technology and obsolescence risk. Bell does not only compute hardware. Our tenants do. That materially reduces technology risks and keeps our capital focused where we earn our returns and create the most value. Finally, on community and sustainability. The project includes commitments around local employment, indigenous economic participation through our memorandum of understanding with George Gordon First Nation and a water smart, energy-efficient design, including a closed-loop cooling system that does not draw water from municipal water sources as well as plans we're exploring around waste heat reuse with post-secondary institutions and others. Today is an exciting day for the team behind our strategy to lead in AI-powered solutions. Our full stack AI approach from Bell AI Fabric to Bell Cyber and Ateko has clear momentum. We're looking forward to meeting our enterprise and public sector clients' needs and scaling even further. With that, I'll hand it to Curtis to walk through the financial benefits of this project and its impact on our guidance.

Thanks, John. Good morning, everyone. I'll focus on capital investment returns and how this project improves our 3-year growth profile, while we maintain our 3.5x net leverage target by the end of 2027 and moving below that level in 2028. Construction of this facility will require approximately $1.7 billion of incremental CapEx with approximately $1.3 billion expected to be incurred in 2026. This CapEx is fully reflected in the updated 2026 guidance we released this morning. The $1.7 billion total spend will be partially offset by approximately $400 million of onetime setup fees and customer prepayments in '26 and '27. From a revenue perspective, we expect revenue to begin in the second half of 2027 as the initial phases come online. At full run rate, the Saskatchewan data center is expected to contribute approximately $500 million of revenue, $400 million of EBITDA and over $250 million of free cash flow for BCE. We've also engaged closely with the credit rating agencies on this project. They reviewed the structure, construction plan, contracted economics and risk mitigants and our current ratings and outlooks were maintained. From a returns perspective, we expect the project to deliver an IRR of approximately 20% at the data center level with additional upside potential from sovereign workloads, GPU as a service to enterprises as well as related services from Ateko and Bell Cyber. These returns are driven by long-term contracted economics, not by assumption around AI pricing, AI utilization curves or the success of any specific AI application. Because revenue is contract-based and infrastructure like, this project does not introduce incremental earnings volatility relative to Bell's existing profile. As the Saskatchewan data center comes online in a series of 4 phases through 2027, we expect it to generate positive free cash flow in 2027 and to exit the year operating at an EBITDA run rate. Our investment will be funded through a combination of debt and cash on hand. On a run rate basis, the transaction is leverage neutral and significantly enhances our deleveraging profile over time. Critically, this announcement does not change our capital allocation priorities. We remain committed to achieving 3.5x net debt leverage by year-end 2027 and moving below that thereafter. We're continuing to allocate funds toward our 4 strategic priorities. We will continue to pay a sustainable and attractive dividend and do not anticipate issuing common equity to fund this transaction. We've also increased our 2028 growth targets, reflecting the contribution from this project while keeping baseline capital intensity unchanged over the medium term. To be clear, the only reason for revising our guidance today is to reflect the Saskatchewan AI fabric project we announced this morning. There are no other changes to our underlying operating assumptions or outlook. We're also increasing our objective for AI-powered solutions revenue from approximately $1.5 billion by 2028 to approximately $2 billion by 2028. In short, this is exactly the type of investment we said we would pursue at our Investor Day, infrastructure-based, contracted, disciplined, accretive to free cash flow and fully aligned with our clearly communicated strategic priority to lead an enterprise with AI-powered solutions. Kris, let's open the line for questions.

Thanks, Curtis. [Operator Instructions] With that, Matthew, we're ready to take our first question.

[Operator Instructions] The first question is from Vince Valentini from TD Securities.

I'm going to ask a question and hopefully a clarification. Question is just if you are reserving space for -- first of all, congratulations. This is a great announcement. But the sovereign space that you're reserving, if you like max that out or reach close to your targets on that, does that take the $400 million of EBITDA up to $450 million or $500 million? Can you give us any context on what kind of upside if you're successful selling some of the capacity on a sovereign basis?

Yes. So we haven't -- so the sovereign demand will -- it's Mirko, by the way. Vince, thank you. Let me start again. So the sovereign demand will evolve over time. We've reserved more than enough. We've contracted for the ability to allocate demand or compute to sovereign demand as that evolves, and we've reserved the ability for far more than the near-term demand that we expect from sovereign workloads. And as those sovereign workloads materialize, yes, that will be over and above -- in some respects, will be over and above given the ancillary services that would come with that to the $400 million in EBITDA, but we haven't profiled that yet because that's a function of when demand comes and the exact type of demand.

Okay. And the clarification is just on the press release and material assumptions on Page 14, assumptions concerning BCE. I'm not sure I understand the 2 columns. One on the left-hand side, for 2026 assumptions, it doesn't look like interest expense or depreciation have changed from what you gave us in February. I assume there'd be a little bit of impact because of the $1.7 billion of CapEx coming in during the year that would boost interest expense. Am I reading that right that the left-hand side is '26 has not changed and the right-hand side is the '28 targets?

Yes, correct, Vince. We still assume -- you're right, firstly, there's going to be a little bit of incremental interest, but it still falls within our initial range. So we did not update that range.

Okay. So it's still -- it wasn't a big enough impact to change the ranges, but it should...

Our next question is from Maher Yaghi from Scotiabank.

I want to spend some time on the contract that you signed with CoreWeave and Cerebras. How long are these contracts? And for how much capacity will they be using?

So they've contracted for -- between the 2 of them, the full 300 megawatts are contracted over a long period of time.

How long?

Many years. Think like a decade.

I understand. Are these renewable leases or it's one full long-term lease?

Yes, the contracts are renewable and the facilities will have a long life in terms of the components that we're investing in.

Okay. So when you say it's a long-term lease, are you assuming that these, let's say, call them medium-term leases get renewed or it's multiple long-term leases?

I might have to follow up after the fact. But if I understand the question, we have 2 tenants. We've leased the entirety of the data center capacity to them over a long-term period with an ability to continue to extend those lease agreements as they come due again in the long term. But it's a lease with one tenant, a lease with the other tenant.

Okay. So if I take, for example, one of them, I don't care. It doesn't matter, but let's say, the first tranche of the first lease, would that be more or less than 5 years?

They're all long term. So it's -- maybe we're confusing things, but they're each taking 2 data halls each. And each data hall that they sign up for is a long-term lease. It's not 5 years, it's longer than that. And every data hall has the same long-term lease. I mean they're not staggered in any way. They're all long-term leases.

Okay. And is there any attached signed residual value to these contracts? Are there guarantees if the tenants stop their contracts for any reason that they pay you a residual value guarantee.

So Maher, our tenants are Cerebras and CoreWeave, as you know, they're best-of-breed. They're well capitalized with global scale. The contract structure that we have with them has clear revenue visibility, inflation protection. And so that protects the infrastructure like long-term cash flows given the long-term leases. In the -- when you're thinking from a risk mitigation perspective, the underlying infrastructure that we're building here will remain attractive to a broad universe of customers over time. So very, very fungible as AI, demand for AI continues to outstrip supply. And keep in mind, we own and develop the land here. We have the building, the power, the cooling, the connectivity. We're not taking the technology obsolescence risk on the compute hardware. And the last thing I'll say is the agreements that we have in place with these 2 customers have all the protections you would expect that we would build in, in terms of termination rights, renewal options at regular intervals and other credit protection mechanisms. I can't get into the details of those, but you can expect that we would have negotiated all the things you'd expect. And plus, Curtis mentioned in his opening remarks, some of the prepayments that they're making right now this year and next year that further derisks the project from our perspective.

Okay. And just on -- when do you expect the 300 megawatts to be running at a stabilized run rate? And when that happens, what's the expected cap rate?

Yes. As Curtis mentioned, as we exit 2027, we'll be at full run rate on revenue generation for the full 300 megawatts.

That's quite a fast build. Can you -- did you select already the construction project manager on this project?

So Dan Rink is going to answer that one.

Yes, this is Dan. So we've got the same team developing this that we've used to develop the prior data centers, and we're already in RFP and very close to final selection with a number of GC partners that will be building the various phases of this data center.

So you expect from date of first construction phase to be full running 300 megawatts in 1 year essentially?

Correct.

Yes. Maher, I answered that one. We're going to exit 2027 at full run rate on revenue generation, which means 4 data halls built and ready for service.

This isn't new for us. We've been actively doing this in these types of time frames, at scale building out other facilities. We've got some very novel ways to modularize and do a lot of work offsite and expedite the construction.

Finally, just one last -- 300-megawatt power, is it fully contracted already?

Yes, Maher. So fully contracted with -- through the Saskatchewan government and the Crown corporations that are part of this SaskTel for fiber and SaskPower and TransGas...

12 months, just to frame it, our customers are responsible for all of the work inside the halls. So we're delivering the facility. They will need to put the racks in and wire them and get that work done. So that should give you even more confidence in terms of the time line because they're responsible for that.

Yes, that's a great point, John. So we don't do any of the white space development. So that's a significant portion of the normal time of building a data center.

Yes, that's the delta that I was trying to figure out.

Our next question is from Stephanie Price from CIBC World Markets.

It's Sam Schmidt on for Stephanie Price. I wanted to talk about ongoing CapEx needs. Are there any associated with the data center? And does it change any longer-term capital intensity assumptions?

Yes. Thanks for the question. Look, this is a brand-new facility. We have modeled in some ongoing maintenance CapEx just to make sure that the facility is always up to code and always running smoothly. No expectation that it would change our $3.7 billion capital envelope once you get past the initial build phase in 2027.

Okay. And I'll just maybe squeeze in one more around margins. $500 million in revenue, $400 million in EBITDA implies quite high margins. Is that typical of how you are thinking about the data center opportunity? And then I'll pass the line.

Yes. Again, it's -- we do think these are attractive margins. It is a brand-new building. So the cost to operate a brand-new world-class facility are -- just drive efficient margins ultimately.

And also the facility of a complex of this size, brand new complex of the size that's brand new also gives us significant operating leverage given the scale.

Our next question is from Drew McReynolds from RBC Capital Markets.

Congrats on the announcement. Very interesting to get all the details. I appreciate that. Two for me. One, I guess, maybe for you, Mirko, you have line of sight on 800 megawatts. You've announced the 300 plus the 73 previously. Just how do you see that kind of playing out over the next few years? Obviously, this is a big one to tackle, but presumably, you may be working on others or maybe not. If you can comment on that? And then secondly, just with respect to this announcement, how do you begin to size up potential incremental opportunities for the likes of Ateko or Bell Fibre?

So I'll take the first one, Drew, and John will tackle your second question. On the first one, we're going to continue to take a very disciplined approach to the rollout of Bell AI Fabric. We -- certainly for the last 18 months, we've taken an approach that we're going to provide a road map to investors in terms of our strategy, our ambition and how the priorities are going to be funded, and we're going to be very transparent in that over multiple years, and we're going just to keep checking the boxes in terms of hitting key milestones. And that's what we're doing here. We're taking a very disciplined approach in terms of how we contract, who we contract with, demand-led, contract-based and returns driven. So that's what we decided we would do at Investor Day when we thought we had -- we knew we had kind of a line of sight to 73 megawatts monetization over the Investor Day horizon. We also have been saying more recently that the pipeline and the interest and the potential demand has been much stronger than we even anticipated back in October as we were building the plan. But again, we wouldn't change anything until we had those contracts in place. And now you see an example of a very significant data center is about to be built because of the approach that we've taken. It's going to be the largest AI data center in Canada once it's operational in a very short period of time. So to -- what do we do with the remaining 500 megawatts? We -- there continues to be significant demand. If -- we're not going to change the guidance beyond what we're saying today, we have clear line of sight to 373 megawatts being monetized. Should the clear interest allow us to pull some of that extra demand into the Investor Day time period between now and 2028, we'll certainly look at it. But the key factors will be returns driven, very strong IRR is going to be required. We're going to need well-capitalized partners, long-term agreements, and we don't plan to deviate from our deleveraging targets of 3.5x by 2027. And of course, if there's opportunities and we can bring in funding partners to help us in that, we certainly will look at that as well. And you've seen how we were successful in doing that in the U.S. with PSP and Ziply fiber. So we're not -- we're open to that kind of approach if it's an opportunity to pull in more demand sooner. That was a long answer, so I'll stop there, but it was an important question. Over to you, John.

Great. Thanks, Mirko. Drew, I think the way to look at it, if you had 100 companies in our position right now, 99 out of the 100 would say this is an amazing day. Let's go and celebrate. I think what's different about us and why we say we're one of one, things are just getting started here. We see the opportunity to bring AI to life for public sector organizations for private sector. The growth in sovereign, the growth in helping them with their mega platforms and their hyperscaler deployments to enable AI at scale is a massive need. And the assets we put together are really unique in terms of helping enable that. So I think you're going to see a lot of goodness that will come over the course of this year. We have very big targets. It started with the $1.5 billion, but the unique set of assets we have, the approach we have to stimulating demand, not just providing compute and consumption. Coveo, Cohere, SAP, 3 recent announcements to run on fabric in a sovereign domain, we're seeing tremendous interest. And it's now cascading into the largest private organizations with a desire to understand more about sovereign and how they could be beneficiaries of it. So we think the future is great across the full stack and today is a beautiful foundation to help accelerate that.

Our next question is from Jerome Dubreuil from Desjardins Securities.

First one is maybe back to Vince's question, trying to understand how the sovereign AI component works a bit better. I think you said the power, the 300 megawatts is fully allocated. But at the same time, the sovereign AI allocation is on top of the $400 million EBITDA that's expected. How exactly does that work? Do you have a bigger part of the pie if it's sovereign AI than other types of business?

Jerome, it's John. The sovereign side, as I mentioned a little bit earlier, we've essentially started to create demand across the board to consume it. And the challenge is not having the compute power. The challenge is stimulating demand, building an understanding, having the ecosystem brought together of a variety of companies. So you have a meaningful capability there that customers can take advantage of. And we've been going flat out to build that and have those capabilities that companies, public and private can take advantage of. So it's flat out on that domain. And the compute is not the challenge. It's creating demand across the landscape. And I can tell you, it's coming now. The conversations and activities, it's really gaining momentum across the board. So we're really hopeful. But today, it's nascent. But I think by the end of this year, you're going to see some really good things in this domain.

And then Jerome, it's Mirko. Just the point I was trying to convey in my conversation with Vince was on the sovereign side, there's, of course, the sovereign workloads, just like non-sovereign workloads. But with sovereign, there will come managed services, not just workloads. That's the part I was trying to convey would be incremental to the business case.

Makes sense. That's good to see we're able to carve out a bigger part if you bring more value. The other one I have, just a clarification. I'm looking at the run rate free cash that you're expecting to be [ $250 ] million. I'm looking at the updated 2028 guidance, and I don't see it quite being fully reflected there unless my calculation is wrong. So maybe you can reconcile what's not completed at the beginning of 2028 since you said the EBITDA run rate is achieved at the end of 2027, maybe there's more building cost in '28 or something.

Jerome, it's Curtis. No, your math is not wrong. Revenue and EBITDA exit 2027 at a run rate level. but free cash flow takes a little bit longer to reach run rate level. So still very strong free cash flow in '28, '29, but has not yet hit run rate simply because there's an amortization period of the prepayments we received in '26, '27.

Our next question is from Sebastiano Petti from JPMorgan.

I guess, Curtis, just following up on that last point. Is it $400 million in upfront payments in each year, '26 and 2027? Is that correct?

Total across. So I think $1.7 billion spend less $400 million kind of refunds and prepayments gets you to $1.3 billion net capital at risk.

Got it. Okay. That's the point -- over the build period through 2027. Got it. Okay. And then as we think about additional opportunities, and you guys have cited the 500 megawatts since the Analyst Day, but you now have this new facility coming online here. As we think about it, is this a one-off unicorn in terms of the AI data center opportunity in Canada? Or could you see additional opportunities evolving perhaps in different provinces, right, that you could perhaps see additional facility demand like this kind of popping up? Just as we kind of think about Bell's positioning and the multiyear opportunity, could something like this pop up incrementally between now and your 2028 guidance? Just trying to think about the demand or the conversations you're having from some of the end consumers.

Yes. So there's 500 additional megawatts of capacity that we'll be looking to monetize over time. As I mentioned to Drew, if we can pull that in, we will under the parameters I've already mentioned, I won't repeat them. So -- and the demand is very strong. The 500 megawatts could be monetized -- will be monetized, I should say, over time. And it will be a mix of large data centers and smaller ones like we already launched last June in Mission Flats BC at 7 megawatts. So it will be a combination, a mix of the mega scale like the one we announced today, the smaller ones and everything in between. In fact, we made a public announcement a few months ago that we have an MOU with Queen's University to monetize -- to build them a 40-megawatt data center. That's going to be dependent on Queen's University obtaining some funding. But if they do, they'll be paying commercial rates, but that gives you a sense of something between the 7 megawatts in Mission Flats and the 300 megawatts in Saskatchewan. So it's that type of mix that we're talking about. The demand is very strong. And again, we're sitting here in a nice position at the intersection of AI demand, network, security, AI integration and in many cases, the sovereignty upside. So it's a great position to be in. And we've got those ancillary services, as John mentioned, that's going to create that flywheel of AI industrialization in Canada. So we're really excited about it.

Our next question is from Aravinda Galappatthige from Canaccord Genuity.

On the announcement and the work so far. With respect to Mirko, the remaining 500 megawatts that you plan on monetizing, should we think of sort of the associated capital expenditure to be sort of proportionate to sort of the announcement today? And secondly, maybe a quick question for Curtis. Given the comments you made about when you hit $400 million in EBITDA run rate at the end of '27, but the lag in free cash flow, should we think of this sort of being either neutral for 2027 or perhaps slightly dilutive to free cash flow in 2027?

Well, I'll take the first one. I should clarify as we've been having the exchanges and we're talking about the 500, let me just -- just because I want to be clear on the fact. So we have a total of 800 megawatts that we have a line of sight to that we can -- that we seek to monetize. And in our guidance, we -- through to 2028, we've now kind of factored in monetization of 373 megawatts. So what's left to be monetized is 800 minus 373. I know we've taken shorthand here on the call saying 500 left to be monetized. In terms of the profiling of kind of the revenue generation and the cost to build, the remaining megawatts that we can monetize. It's going to -- it's hard to answer that now, Aravinda, because it's going to depend on the mix of small, mid and large data centers. When you have a large data center like the one we announced today, typically, your tenants are expecting a little bit of a better rate. On the other hand, you get way more operating leverage as you build greenfield, something this size. So you can see that in the flow-through as we discussed in response to an earlier question. And so the economics kind of vary a little bit depending on -- in terms of revenue and the flow-through and the capital cost depending on the size of the data centers, greenfield versus retrofit, et cetera. So it's hard to answer right now.

And then just on -- to follow up on that and then to get to your second question, Aravinda. So just to be clear, the 73 megawatts, the CapEx required to monetize that opportunity is baked into our capital envelope over the next 3 years. And importantly, we've said it a few times, but it bears repeating. We're only going to spend the CapEx once we have line of sight contracts into the actual revenue. So it's a different risk profile of CapEx spend. And ultimately, in the EBITDA side of the world, so yes, we'll be exiting 2027 at run rate revenue and EBITDA, but we're actually driving free cash flow accretion in 2027. So the last part of your question was free cash flow, but free cash flow in 2027 is going to be positive.

There are no further questions registered at this time. I would now like to turn the meeting over to Mr. Somers.

Matthew. And thank you, everyone, for your participation on the call this morning. Richard and I will be available throughout the day for follow-up questions or clarifications. Thanks, and have a great day.

Thanks, everyone.

Thank you. The conference has now ended. Please disconnect your lines at this time, and we thank you for your participation.