Support, and we thank these investors for hanging around here late in the afternoon to hear a little bit about Cogent.
Appreciate it. So, Dave, maybe to kick off, obviously the big topic in Cogent over the last year and a half has been the Sprint integration. I think you've talked about being roughly 75% or so way through the cost synergies. Maybe you could talk about what's left, the timeline for achieving them, and maybe potential upside to that synergy target that you laid out.
Yeah, sure. Thanks, Eric, and maybe just to remind investors, Cogent has been a public company since 2005. In between 2005 and 2022, we looked at over 825 potential acquisitions and did a grand total of one of those, which was announced in September of 2022. That was the acquisition of Sprint Global Markets Group or Sprint GMG from T-Mobile. That is the original long-lines business of Sprint. When we acquired that business, there were effectively two somewhat decoupled divisions. One was an operating business selling 28 different product categories to about 1,320 customers, predominantly large enterprise. That business was doing $565 million in revenue and burning negative $300 million of EBITDA. We, when acquiring the business, directed T-Mobile to do a number of cost reduction efforts prior to closing.
They got the burn rate down to - $190 at closing in May of 2023, and the revenue rate had declined to $490. We then embarked on a more aggressive cost reduction program that was aimed at taking $220 million of costs out of the business over a 36-month period. Halfway through that, 18 months into it, we had already taken out $165 million. The primary areas that we've been able to achieve those cost reductions are: one, moving off-net customers to on-net; two, the elimination of non-core products, which are gross margin negative, meaning they were costing more than the revenue and direct cost of goods sold; and then third, general cost reductions in terms of headcount and facilities consolidation. We anticipate that we will beat that 36-month timeline to achieve the $220 million in savings, and beyond that, we do anticipate being able to increase that target.
We don't today have enough visibility to exactly how much above that, so I'm not willing to put a number out other than greater than $220, and we will continue to achieve cost reductions in all of these areas. The second part of the transaction, which was somewhat decoupled from this cost savings initiative, is the acquisition of the physical network of Sprint. That network was built between 1982 and 1991 at a capital cost of $20.5 billion. It was predominantly fallow. 93% of the revenues at Sprint did not even touch the network, and that network was transferred to Cogent for $1, so the cost savings have all occurred on the operating side of the business, not the network side.
No, fair enough. And, you maybe could talk about, I think you said you were hopeful to get that EBITDA burn rate back to break even by kind of mid-next year. Is that primarily from the cost savings, or are there other revenue-generating opportunities within the Sprint base that you can, where you can get back to, you know, positive growth, call it exiting, you know, second half of next year?
Yeah, the reduction in burn is exclusively coming from the cost reductions and not from the incremental revenue opportunities. We anticipate that the Sprint revenue streams have declined and probably will decline for several more quarters as we complete grooming in three dimensions: one, very low-capacity services that are not viable. Two, services and locations that are not economical. And then finally, these products that are gross margin negative. While we are through the majority of that, there will be some continued grooming that goes on. We've actually been somewhat pleasantly surprised that we've been able to actually grow certain portions of the enterprise revenue base. I think that is a direct result of our ability to commit to customers that we will support their legacy VPN services based on MPLS technology for at least a decade. It is an older, less efficient technology.
We are encouraging customers to migrate to a newer platform of Virtual Private LAN Service or VPLS, but many customers have been reluctant to do that, and for that reason, we've actually been able to go into that base and sell incremental locations. That is an upside that we were not expecting, but we do not anticipate that the acquired enterprise business will ever be a meaningfully growing part of our revenue, but rather a stable portion, and our revenue growth will come from two primary areas: the core Cogent IP business that existed prior to the acquisition, and then from the augmentation of the Sprint network to support wavelength services, which we anticipate to be a very rapidly growing part of our business.
Yeah, and I, I definitely wanted to touch on the waves business 'cause that's, I think, one of the primary justifications from, from the acquisition, right? This ancillary revenue stream. Maybe you could just update us on the provisioning process for getting the data centers, ready for the wavelengths revenue. I think you've talked about the early 2025, you'll start to see more of a ramp. Obviously, you have a long way to go to get to $500 million, but maybe you could give us an overview, your pathway to get to that $500 million. Is it, primarily coming from share gains against incumbent operators? I-is that a growing market where you see more opportunity, where you can take share yourself without, not at the expense of, Lumen or another operator?
Yeah, so in order to take the Sprint network that was initially designed to carry long-distance voice and modify that network to carry wavelengths or optical transport services, we had to undertake four major initiatives. We had to physically extend the network from Sprint proprietary facilities, typically 15-20 miles from downtown into the metro market. Two, we had to reconfigure our metropolitan networks to accept wavelengths and be able to deliver those wavelengths into 800 carrier-neutral data centers across North America. Third, we needed to deploy transponder shelves in those data centers to accept the wavelengths. And then finally, we needed to put reconfigurable add/drop multiplexers or ROADMs at the intersection of the long-haul and the metro network in order to be able to guide those wavelengths to the correct location.
So the industry, which is a $2 billion North American inter-city wavelength market, $7 billion globally, $3.5 billion of that is in North America, but within that $3.5 billion, $1.5 billion is pure metro waves, and about $2 billion are inter-city. That is the market that Cogent is primarily focused on. The current providers typically build a wave on a custom basis, one at a time, with average provisioning times of between 90 and 120 days. One of the differences in our architecture will be the ability to provision a wave in two weeks. In order to do that, we were able to start with a clean sheet of paper and then build a network that was optimized for wavelength services, much as we had done 25 years ago in the IP market. And as a result, going to market, Cogent will have multiple advantages over their competitors.
We will have unique routes in 90% of the cases. We will have defined mapping and latency on each of those routes. We will have rapid provisioning. We will have more endpoints, and then finally, because we're doing this on a network where our cost basis was $1, we will have the ability to be very aggressive on price. The total inter-city wavelength market in North America today is about 140,000 waves. We've only provisioned about 1,100 waves on a one-off basis, just like our competitors. Starting in the first quarter, we can provision those waves in a much more automated manner with the reduction in the number of field dispatches necessary from going from six dispatches down to two. That, I think, will be transformational for the industry.
I believe our growth will come both from gaining share from existing players, predominantly Lumen and Zayo on a national basis, regional players like Crown Castle, Windstream, and Uniti, and then on our ability to capture new business as well. Particularly for new applications, there's been a significant amount of incremental demand in the market from AI training that will be additive to the market that exists historically for wavelengths. But I think Cogent's stated goal of getting to $500 million run rate in revenue from today, about $20 million, is very realistic. While I know that represents a relatively rapid rate of growth, I think the competitive advantages we have vis-à-vis the others should allow us to capture that market share.
What does the sales process look like for wavelengths relative to the other products you sell? I mean, do you feel like your salespeople who sell things like transit or other enterprise services have, you know, the right capabilities to sell into that market? Is it a market that has a lot of incumbency advantage, or do you think that can be disrupted, like you said, especially given you can offer a lower price?
The switching costs are relatively low. These are delivered between data centers, and to switch providers, you just switch cross-connects in those data centers. Cogent has approximately 650 quota-bearing sales reps. These reps are organized into 72 teams that sit in 50 offices around the world. The sales organization is actually divided into four types of reps. Those that focus on very small businesses, those that focus on mid-sized businesses, those that focus on large enterprise. Those three groups represent just under 400, about 382 of our sales reps, and would predominantly be focused on selling IP services and would rarely sell waves. Our NetCentric or wholesale sales organization is 287 quota-bearing reps, and they are going to the same customers who buy transit services today and now selling them an optical transport or wave service. We have done extensive training of that sales organization.
Two, we have modified our CRM tool to integrate wave quoting directly into the tool. So something that we can do that is not possible with our competitors is a sales prospect gives the salesperson the two endpoints. They pick those endpoints from a dropdown real-time, and the tool picks the optimal path for that wave. It calculates the latency within one hundredth of one millisecond, and it gives a map based on GPS coordinates, a KMZ map that is accurate within one meter. That map and that latency calculation is part of the customer order. It also prices that order instantaneously. If the customer accepts that, an automated customer contract is generated with those terms and conditions. It is sent to the customer for e-signing, and then once executed, it comes back to Cogent, goes to our sales validation group.
This would be the exact same process an IP port would follow, at which point that team validates the billing information, the technical contacts, the demarc handoffs, all of the parameters associated with the service. Once that is completed, the order is then passed to a service delivery organization. This, again, is the same organization that today provisions about 3,000 IP ports a month. The difference with a wave is they have to provision two ports for one order, one at each end. They generate the work order tickets for the two field dispatches, and then they send the technical configuration details to the operations center to remotely configure that wave.
That level of automation, pre-engineering, and documentation is very different than the way waves have been sold in the industry, and I think will help us win confidence from customers and should ensure that the sales force we have is the appropriate one. We have built a backlog of about 3,400 orders, and in addition to that, we've actually installed about 1,100 orders since acquiring this business. All of that business is on a custom basis. We believe that as we have completed this network configuration work, that process will become more streamlined and accelerate. I think as we actually install what is in our funnel, we should see the funnel actually begin to grow at an accelerated rate due to the confidence that we can give customers that we can do what we say we're gonna do.
And as you alluded to, you know, a lot of the incremental opportunity could come from AI training use cases. One of your peers, Lumen, has obviously been very outspoken about some of the dark fiber IRU deals they've done with some of the hyperscalers related to AI training. Maybe you could talk about, you know, your interest, in those kind of deals, whether it's dark fiber, whether it's waves. What is kind of different about the deals you might try to do on your waves network versus what Lumen's announced?
Yeah, so three different points. The first one is AI is possible because of the internet. The roughly 800 zettabytes of data that are stored around the world that have been collected over the internet is the raw material for AI training. And now that data has incremental value and is being collected at an accelerating rate. So Cogent benefits as the largest carrier of internet traffic in the world in the collection of that inbound data. The second point is the training of that data in large language models typically occurs in a different location than where the data sits, oftentimes because of the power constraints of those locations. As such, there needs to be transport connectivity from those locations to the training location and then back to the original origin of the data. And that is facilitated through either wavelengths or dark fiber.
So in moving data a mile, the absolute cheapest, easiest way to do that is on the public internet. However, there are three characteristics that AI training values that the internet cannot deliver: defined latency, very large packet sizes, and high security. For that reason, companies are either buying wavelengths, buying dark fiber, or building their own dark fiber. Cogent is in the business of first selling wavelengths. Two, we do have excess dark fiber in our national footprint and will sell that off, but have chosen not to do so until we completed the wave enablement of the network. So sometime in 2025, we'll be more aggressive around those dark fiber sales. The deals that have been in the market so far have three components to them. Cogent would have participated in only one of those three, selling dark fiber out of inventory.
The two other components that have been requested by hyperscalers are new fiber along existing routes that would be expensive and not practical for us due to the fact our fiber is directly buried in armor along rail right-of-way. And then finally, building brand new fiber routes to new locations that never had fiber. And when we evaluated a number of these opportunities that were presented to us, when we parsed through them, the IRRs were low single digits, and they just did not make economic sense to us. Maybe the capital structure and cost of capital is different for our competitors, but for Cogent, you know, borrowing money at 7% to get a 3% return didn't make much sense.
Yeah, I think that's pretty clear. So, kind of moving on from that and related question. So, your data center footprint, you've talked about, you know, reconditioning some of those switching facilities. I think you've talked about roughly 100 megawatts of excess power that you're looking, you know, to sell into the market, and you've talked about maybe $10 million per megawatt of potential value. Maybe you could update just the process of re-provisioning those data centers to be ready for new tenants, when you think, you know, you'll start to execute on some of those sales, and then in terms of value, could you talk a little bit about the locations, the, you know, the quality of the facilities, you know, for what type of use cases we could see in those data centers?
Yeah, so, when we acquired Sprint, we acquired 482 fee simple-owned buildings around the country in 49 of the 50 states. There were no physical assets in Alaska. And we identified roughly initially 45. We've subsequently increased that to 48 of these facilities that are suitable for data center conversion. The 482 buildings in total had 230 megawatts of inbound power and 1.9 million sq ft. Many of those are too small or are in too remote of locations. The 48 facilities that we are converting have two different footprints within them. We initially, when we announced the deal, planned to put a Cogent data center in 10,000 sq ft and use one megawatt, or roughly 45 megawatts of the existing power, and leave everything else fallow.
As it became clear to us early this year that there was an acute shortage of power and the value of that power had increased, we decided to go into that footprint and begin to convert the remainder of the space and power in some of those facilities. We identified 21 facilities that have 1 million sq ft of excess space and 100 megawatts of inbound excess power. Those we are trying to market as either selling the facility or on a long-term wholesale lease. We will sell the facilities at $10 million a megawatt or lease them at a triple net $1 million a megawatt a year. What these facilities look like are brick-and-block buildings built in the late 1980s, typically about 15 mi from a downtown along a railroad track and an industrial park that housed telephone switches.
One of the tasks we had was to remove all of that old switchgear. There were 22,500 cabinets full of old obsolete telephone equipment in these facilities. We have now taken 90% of that out. We're still taking that last 10% out. These facilities needed some general housekeeping, and they needed the power systems inverted from negative 48 DC to AC 120. All of that work is ongoing. We probably have three of the facilities that'll be done this month, and then probably all 21 will be available by mid-year 2025. In accelerating this conversion of this excess space, we are spending $100 million of capital over a 12-month period, and we think that is money well spent based on the demand we have gotten for these facilities. We are talking to multiple counterparties around multiple facilities, and some are leases, some are sales.
But if you looked at one of these facilities, I think you're based in Chicago, right? We have a facility that's about three miles south of Midway Airport, off of Cicero on 4200 4th Street, I think. And it looks like a block warehouse. It's white painted, has a fence around it, it's about a seven-acre yard. It's 100 feet away from the train track, and it's got two substations feeding it. And basically, no one had used it for a decade. We went in and cleaned out that facility. We did some cosmetic work, upgraded the fire suppression system, upgraded the alarm system, tested the generators, and now have space available. Now, there is still some telephone equipment in that particular facility, but probably sometime in first quarter, it will be completely ready to be marketed.
All right, sounds like I need to do a property tour once I get home.
You should, absolutely should. Call the local sales team. And by the way, two of your competitors have done bus tours to these facilities without management to come in, and that offer stands for you as well.
All right, well, I appreciate that. It's a good idea. So maybe just going forward a little bit, as we think about the financial model for Cogent, maybe we could just touch base on the old Cogent, as I like to call it, the traditional corporate business. You know, obviously heavily impacted by the pandemic, but you've kind of reemerged where that business is now growing at a low to mid-single digit rate, I believe, in the last call. You know, where do you see that business going over the next couple of years? I know historically this was a double-digit growth business. There've been a lot of impacts from the pandemic. What do you need to see in that business to get it back to a growth rate that, you know, you've targeted longer term?
Okay, so pre-pandemic, roughly 94% of Cogent's revenues came from selling IP-based services, either dedicated internet access or VPNs over the internet. Roughly 16% VPN, and the remaining 78% was internet access. And that business had grown organically for 20 years at a compounded rate of roughly 10.2% per year. 60% of the business was selling to corporate end users and skyscrapers. 40% was selling wholesale bandwidth and data centers. When the pandemic hit, we saw two different things happen at the same time. We saw our corporate business, which had grown 11%, decelerate to -9% . It is now back to positive 4%. We saw our wholesale business, which had averaged 9%, but pre-pandemic was only growing at 3%, skyrocket to 26% year-over-year. It has since reduced its growth rate back to a more traditional 9%.
On top of that business, we acquired a new customer base from Sprint. We acquired a customer base that was buying VPN services and dedicated internet access, almost all off-net. We are jettisoning certain locations, as I described, and certain products and certain low-speed services, but we are continuing to keep those enterprise access customers. They historically have not grown. We expect that the combined business will probably grow in the high single digits, and with that growth, historically, we had delivered 75% of new services on-net, 25% off-net. With the inclusion of the enterprise base, it will be more like a 50/50 split. That will lower the rate of margin expansion to about 100 basis points a year, but we do anticipate that core business growing in the high single digits on a blended basis with roughly 100 basis points a year of margin expansion.
And I think related to that, obviously you've seen your leverage tick up kind of mechanically from the step down in the T-Mobile transit payment. But if you could just talk through kind of what steady state free cash flow should look like, what CapEx should look like, maybe stripping out some of these one-time expenses for the data center process. In addition, capital lease payments, I think you've prepaid some uneconomic leases, so that should step down as well. But just to give us a sense for what kind of the free cash flow, you know, ramp could look like over the next couple of years.
Yeah, so, Cogent's leverage for about 13 years had hovered between two and a half and three and a half times EBITDA on a net basis. When the pandemic hit, our leverage actually ticked up because of the slowdown in our corporate business and our commitment to continue to grow the dividend. We have increased our dividend sequentially for 49 consecutive sequential quarters. We did lower the rate of growth from two and a half cents per quarter sequentially to one cent per quarter. We'll be paying our dividend this Friday. It'll be $0.995 per share for the quarter, and we expect that to continue to grow. We have returned more than $1.4 billion to shareholders through buybacks and dividends. We have bought back roughly 23% of our float, and we have this growing dividend policy.
When the pandemic hit, our leverage got up to 4.2 times net above our stated range. When we initially acquired Sprint, we had an asymmetry in the subsidy payments from T-Mobile, and as a result, our leverage decreased. Actually, last quarter, on an LTM basis, our leverage was 3.17 times net. But that will go up. On an LQA basis, it actually peaked last quarter. Our leverage will breach five times over the next year due to the step down in those subsidy payments. So, in the first 12 months of the acquisition, T-Mobile paid us $29 million a month, and then for the next 42 months, those subsidy payments dropped to $8.3 million a month or $25 million a quarter. We've made up part of that ground through the cost reductions that we had talked about. We will continue to improve both through operating leverage and through growth.
So, our expectation is within a couple of years, we'll be back down in that three and a half to two and a half times net leverage. We will do that and continue to return more than 100% of free cash to shareholders. Now, let me touch on the capital intensity. The Cogent business needed about $35 million of maintenance CapEx, and we were spending about $30 million of expansion CapEx a year pre-acquisition. Those numbers haven't changed. The Sprint business needed $30 million of maintenance. Those three together are $95 million. We then spent this extraordinary $150 million, $50 million for network integration, $100 million for data centers. That work will be over mid-year, and we should be back to a roughly $100 million a year capital need. In addition to that capital, we spent about $20 million a year on principal payments on capital leases.
These are for IRUs, primarily Cogent, that built our original IP network globally. So, as an investor thinks about free cash flow, they take our EBITDA, $352 million last year, probably similar numbers in 2024 and 2025, and then really accelerating in 2026 and beyond. Subtract from that our interest expense and these two forms of capital, and that leaves the remainder that's available for shareholders. We're currently returning about $200 million a year to shareholders, and that does entail adding some leverage, but our EBITDA is growing in a way that makes us prudent.
All right, well, we're out of time, and as is usual, I only got through about half of my questions with you, Dave. But I appreciate your time today, and thanks for joining us.
Thank you, Eric. Thank you all very much.