Lumen is rebuilding around AI, pairing a vast US fiber footprint with programmable networking through Alkira. The bet is that enterprise connectivity is moving from connecting sites to connecting clouds, data centers and AI factories.
In sum – what to know:
Survival story – Lumen has reset its balance sheet, shed non-core assets, and shifted to high-capacity services with ‘strategic revenue’ rising 14%, and its NaaS base doubling to 2,000-plus enterprises.
Physical layer – Lumen has 17m strand miles of US fiber, with plans for another 40m; enterprise sales of 100G/400G wavelengths are up 35%; 90% of hyperscaler on-ramps will be 400G by the end of 2026.
Control plane – $475m Alkira deal brings its programmable network strategy forward “by two years”, and adds a carrier-agnostic control layer to stitch Lumen into third-party carriers via cloud providers.
“Let’s start at the top of the house,” says Jim Fowler, chief technology officer at Lumen Technologies. What he means actually, in telco terms, is the bottom of the house: the network foundation, which, in Lumen’s case, stretches to 340,000 route miles and 17 million strand miles of fiber optic cable in the US. Indeed, the firm claims “more than half” of metro, inter-city, and long-haul enterprise fiber in the US; it is set to treble its strand miles, to 58 million, by 2031, it suggests. And its physical build project is not even its top-of-the-house strategy item, it seems. Lumen’s house is being stacked ever-higher, and so there’s a lot to get through – and we should let the tape roll a little.
“We’re building the nervous system for AI,” says Fowler. “The models are important, but if they don’t get the data – if the brain is disconnected from the eyes and the ears – then they don’t work. People ignore this piece – that once the AI makes a decision, it has to do something about it. Which is not new. Fifty percent of internet traffic is between machines, transacting back and forth – and has been since 2024. I don’t know if people get that.” The machine-share has (only) inched up a little between times; it now stands at 53 percent, suggests Fowler. Nevertheless, almost all the demand is in the cloud, going between data centers and AI factories. Lumen has cleared up, it seems.
“We have more miles of fiber in the ground across the US than any other carrier. We’ve done $12 billion of deals with all the hyperscalers to put fiber in the ground – to support the people running the AI. We have led that build, and all of that investment. We have far more than anybody else. And the deals you’re seeing coming up now are deals that we, frankly, don’t have the economics for us in the long run – because we’ve already done this across the board for almost all the hyperscalers. Nobody matches us for miles in the ground or connectivity with hyperscalers. Because we’ve been the one building out their internal networks as well, and overlaying our network on top of that as well.”
Turnaround and build-out
Fowler joined Lumen in January after 30 years in enterprise-based CIO roles, including with General Electric and Nationwide Insurance. Louisiana-based Lumen is deep into its own business transformation, and coming up for new air. Only a few years ago, it was in “survival mode” and “playing not to lose” – reflected its chief executive, Kate Johnson, in February – with declining revenue, heavy leverage, and a tough debt profile. In its own assessment, it has restructured and “rebooted”, refinancing more than $15 billion of debt in 2024, and concluding the sale of its consumer fiber-to-the-home business to AT&T for $5.75 billion in February. “The dark days are over,” said Johnson.
Last quarter, “strategic” revenue reached 53 percent of its total sales, up from 45 percent a year earlier, growing 14 percent year-on-year. For Lumen, “strategic” services basically means high-capacity services for cloud hyperscalers, model builders, data center operators, and large enterprises. Traditional “legacy” revenue (from older voice, data, transport products) fell by 15 percent, and is being harvested rather than revived. Its network-as-a-service (NaaS) base has doubled since late 2025, to more than 2,000 enterprises, apparently. It is growing at “faster than the market”, says Fowler – “one-to-two percent, whereas ours is at three times that”.
Revenue from enterprise 100G and 400G wavelength services grew nearly 11 percent, with sales (bookings, contributing revenue over a longer period) up almost 35 percent. The physical network, firstly, is at the heart of its reset – and has been attended to consistently by Lumen, it seems, even through the lean years, for longer than most others’, certainly versus telcos in the business of mobile and fiber access. Fowler says: “Let’s start with the physical network. We’ve upgraded our photonics across the US – 400G in the inner city, 400G from the metro out to the hyperscalers. We’ll have 90 percent of hyperscaler on-ramps with 400 G by the end of this year.”
He goes on: “So we’re there already, and building for the future. We have 17 million miles of fiber today, and another 40 million will be in the ground by 2031. So that physical layer – check.” But the bigger opportunity, as Lumen tells it, is not simply to build more fiber; indeed, as above, some of the new-build fiber contracts are just too rich, it suggests. Instead, it is to capture value on top. Enterprise network spending is shifting, notes Fowler, from conventional north-soutth premise-to-cloud connectivity towards high-capacity east-west interconnections between cloud cores, data centers and AI factories – from 36 percent of spending today to 84 percent in three years, by its reckoning.
Programmable and on-demand
It wants to supply both sides of that transition – the physical infrastructure and, through its NaaS platform and Alkira acquisition, the programmable control layer on top. Which is really the context for Fowler’s analogy about the AI “nervous system” at the start. Lumen has already built much of the physical network it says the AI economy needs. The challenge is to make it behave less like a collection of carrier assets and more like a programmable platform – which also goes ‘off-net’ across the planet on third-party cables for multi-national firms. “If cloud 1.0 changed how we consume compute, cloud 2.0 changes how we consume connectivity. AI is the force behind that shift,” says Fowler.
Lumen finalized a $475 million deal for California-based Alkira in May, calling it a “bullseye” purchase to unify ‘east-west’ and ‘north-south’ AI traffic as a programmable ‘fabric’ in a single ‘pane-of-glass’ platform. Which is what every telco out there is trying to do, of course – to build programmable networks for on-demand AI traffic services. But Alkira – already part of the Lumen Connect platform, being brought closer (single account management by the end of 2026; a reverse integration “into Alkira” by late 2027/early 2028) – is producing results. “It has advanced our programmable network strategy by two years,” says Fowler.
He offers up a neatly-sketched case study to explain the Alkira upside – already, “without any sort of integration in our physical network”. He says: “The first is multi-cloud optimization. Ninety percent of enterprises are multi-cloud. That is the future. With Alkira, it takes 84 percent less time to add a new cloud environment / region with 47 percent fewer FTEs (staff) to manage the cloud networks. Veritas unified 600 VPCs across AWS, Azure, and Google on one control plane, and avoided hiring about six FTEs. So huge ramifications.” He has a couple more, about spinning up shared extranet services and about consolidating cloud co-location sites.
He continues: “The second is they get an instant extranet – so Alkira customers can open and orchestrate network connectivity to third parties. So think about it: you need to share data in a secure fashion because you did a merger, or just with partners; [this brings a] 91 percent reduction in calendar time [for a full project setup] to add a new extranet partner, [and] 98% less time [in hands-on configuration hours] to add a new extranet partner. So you’re talking days to be able to get a new M&A [business] integrated to be able to share data back and forth. In the past, you would have put network lines in place, set up the firewall rules etc. You can now do that on day one.
“And third is co-location simplification. One of the hardest parts about a cloud hybrid infrastructure is working through the co-location connection sites. [With Alkira, customers are] 93 percent faster to connect to new data centers… [via] 44 percent fewer networking devices. You can consolidate your firewall infrastructure as virtual firewalls inside the cloud environment and reduce a large amount of the connectivity infrastructure. We have an enterprise customer that set up a data center co-lo in one month – versus what used to take them a year. That ability to better-manage across co-locations has been one of the biggest benefits for Fortune 50 customers.”
Acquisition and adventure
The latter redesigned and reduced a half-dozen co-location facilities, he says, to optimize how traffic “sprawls around the world”. The Alkira brand is being retained (as ‘Alkira, a Lumen Connect Technology’) for its brand awareness with Fortune 50 firms. Asked about the segments/verticals adopting AI fastest, and coming to Lumen to rearchitect their traffic flows, he suggests it is a mixed bag of enterprises, better defined by their CIO/CTO leadership. “They have strong leaders who understand the impact AI will have, get that the models are important, and know the networks are too – because of data sovereignty and location, and the fact they work with multiple SaaS providers.”
He adds: “Lots came to Alkira early on.” And are they coming to Lumen in busier droves than ever – away from Zayo and Verizon, and whoever else? “A large percentage of the NaaS growth is new logos, driven by our NaaS strategy, which existed before Alkira. So yes, we’re taking share in prem-to-cloud and cloud-to-cloud because companies want to move to a consumption model, and want to modernize their wide-area networks for AI. You’ll see more new logos added now, post-Alkira. We have overlap with about three quarters of its base today. So we picked up some new logos with it, but we have a global footprint now, as well, so we have the ability to pick up international logos.”
Which is important, and gives perspective about its aggressive fiber build out in the US, and its apparent readiness to stand back from certain deals. As well, it explains its new architectural flex to serve the trio of use cases discussed above. With Alkira, as the control plane on top, Lumen does not need to own every piece of the network anymore, and can sell an end-to-end global service using patches from third-party carriers. It changes its economics, and its addressable market – because Lumen can sell a consistent network experience without replicating its US fiber footprint around the world. (It also justifies the $1.8 billion sale of its EMEA assets to Colt in 2023.)
Fowler explains: “The Alkira platform doesn’t care who the carrier of record is. It should just be the right one to provide the best connectivity. What Alkira gives you is a way to bring it all together with one set of policies, one architecture to manage, one control plane to manage it in. Previously, we’d do off-net partnerships with [overseas] carriers through our NNIs (network-to-network interfaces). We have one of the most peered networks in the world. But Alkira gives us the ability to do it through cloud providers – via their control planes, in their clouds globally. So we can address any network connection they can create out of their physical locations around the world.”
The firm is still betting on physical infrastructure; 40 million strand new miles are not going anywhere. But it wants to sell the system, and not just the pipe. Alkira is a mechanism for that, to manage connectivity where the workloads sit, including in foreign climes where it has no network at all. But it raises a question about where they sit? Because the AI opportunity for telcos is not just about north-south prem-to-cloud access anymore. Well-wired enterprises want to connect globally to far-flung hyperscale facilities and multi-national branch offices, and run workloads to and from AI training clusters and who-knows-where inference engines; and they increasingly work from home anyway.
Fowler has more to say. To be continued.
