Lumen chief executive Kate Johnson has penned an open letter to enterprise CEOs (and the whole telecoms industry): that there is a golden opportunity to connect the AI revolution, but legacy networks are not up to scratch, and the industry must move fast.
In sum – what to know:
More than plumbing – Johnson argues the network is now the “nervous system” of the enterprise AI stack, directly shaping performance, cost, and ROI – which opens the door for telcos to reposition themselves.
New network design – AI workloads are in constant flux between the cloud and edge, with 50% of internet traffic now M2M; static legacy point-to-point networks don’t work; cloud-like programmability is required.
Wasted GPU capacity – Constrained bandwidth means inefficiencies in short-life AI compute systems, where ROI is stretched to start with; conversely, flexible connectivity is an ROI enabler – and a smart sales pitch.
Everyone in telecoms knows this – or everyone in telecoms says the same thing, at least. There is nothing new here; and yet an open letter from Lumen chief executive Kate Johnson yesterday (April 13) to “fellow CEOs” captures very well the urgent mood in the telecoms industry to reimagine itself for the AI era – and to shout about it. The same message was trumpeted at PTC, MWC, GTC, and everywhere in between: that connectivity is king – type of thing. Or that connectivity is critical, anyway, and maybe even cool again. For one thing, it ain’t boring anymore.
In ways, the letter is in an echo of Johnson’s more-personal address at the company’s investor day in New York in February, that the “dark days are over” for Lumen.
And in truth, it has been the same message forever, since way before the levee broke in late 2025 / early 2026 – when infrastructure investments skewed towards hyperscale compute and interconnect, enterprise strategies gambled on agentic automation, and global geopolitics went kersplat, raising sovereignty and localisation up the new techno-nationalism agenda. This industry has been trying to reinvent itself for a decade, at least; the difference now is it is presented with a golden opportunity, in the fractured form of this thing called AI – as consumed by enterprises.
The “competitiveness of our country” depends on it, writes Johnson in the letter. If only the industry can just take the chance – because “yesterday’s networks”, she suggests, “can’t handle the job”; and so there is work to do. She is talking to enterprises, ostensibly – as an enterprise leader. “Many of us are trying to build the future on infrastructure that cannot support it.” She talks about “commitments to our boards”, placing herself in the fray, and about the value of “your AI investments” and the tendency for “procurement departments” to “constrain bandwidth for cost control”.
But it is also a call to arms for the whole telecoms sector to build better networks – and to tell the world about them, so enterprises seek programmable high-performance bandwidth for their AI workloads. If they know where to look, they might just find it already – on bits of Lumen’s fiber in the US (which is the pitch, of course). But really, Johnson is presenting a roadmap for what telcos should deliver – which critiques legacy architectures (as static, point-to-point, constrained), and proposes new capabilities (more speed and flex) and models (cloud-like, consumption-based).
It is a classic broad-brushed address, of the sort plenty of tech chiefs have produced in 2026 keynotes at PTC, MWC, GTC, and elsewhere. But it is a lean and punchy stew with some heady top notes, and it works as a kind of tasting menu for this feast of telecoms in the new AI kitchen. So in brief, here are five takeaways from Johnson’s letter just to whet the appetite – in case we aren’t full of it already. The full letter is copied at the bottom of this article. Note, Johnson is appearing for fireside chat (on AI’s Next Chapter) at Semafor World Economy today.
Lumen’s message – five points about networks-for-AI
1 | The network is no longer just “plumbing” – it’s the product
Johnson reframes the network as the “nervous system” of AI-driven enterprises, not a background utility. Operators have a rare opportunity to reposition connectivity, from commodity bandwidth to mission-critical infrastructure in the AI era. They can also turn it from a cost centre to a strategic enabler of enterprise ROI. Which supports its more premium positioning: plentiful and programmable bandwidth with latency guarantees and other AI-geared SLAs.
2 | AI workloads will explode demand for dynamic connectivity
AI systems are moving data in volume between the cloud and edge; they engage autonomous agents to generate continuous traffic at every turn. Static legacy point-to-point networking is fundamentally mismatched to AI traffic patterns. As well as upgrading bare-bones fiber (and mobile) systems, operators must accelerate their as-a-service (NaaS) models with support for on-demand bandwidth and API-driven programmability – to match the familiar flex of IT cloud-compute systems.
3 | Bandwidth constraints are a business risk, and not just a cost issue
Johnson makes a sharp point, made elsewhere in these pages of late: limiting bandwidth to save cost can idle expensive GPUs and degrade AI performance (such as with slower TTFT). This is a powerful commercial lever: telcos can tie connectivity directly to ROI in AI workloads, and position high-capacity on-demand services as a means to protect enterprise AI investments. An under-provisioned network means wasted AI spend, she says. Which is a much stronger sales story, of course, than traditional capacity arguments.
4 | Enterprises need fully adaptive, workload-aware networks
The letter emphasizes shifting workloads across locations (clouds, regions, edge), and in line with external factors (weather conditions, energy pricing, latency needs). There is rising demand for intelligent routing and orchestration, real-time network reconfiguration, and deeper integration with cloud platforms. Operators should move beyond connectivity into orchestration layers and ecosystem partnerships (with hyperscalers, neoclouds, and others in the developing AI infrastructure ecosystem), she says.
5 | AI traffic (including bots/agents) will dominate network usage
Johnson quotes a single stat, from an Imperva/Thales report about ‘bad bots’ (Bad Bot Report, 2025), and it is a powerful one: more than half of internet traffic is already machine-generated. This signals a structural shift. Networks must evolve for new-style machine-to-machine (not just IoT) comms at scale, allowing for unpredictable ‘bursty’ traffic patterns and massive east-west interconnect traffic. Which, again, has implications for network design (capacity distribution), pricing models (consumption-based), and security (where autonomous traffic represents a new and expanding attack surface).
…
Lumen’s address – ‘An open letter to my fellow CEOs’
“Ok, let’s be honest. How often do you think about your network architecture? Do you just delegate it to your CTO?
Who can blame you? Most of us have spent the last few years all-in on AI. We’ve debated models, talent, chips, and data. We’ve made commitments to our boards. We see AI as the defining technology of our time. Exciting stuff.
“Networks, on the other hand, are boring. That is, until they become critical. Until they can’t handle the job. Until they fail. And that leads to a hard truth: many of us are trying to build the future on infrastructure that cannot support it.
For decades, networking has been in the background – like plumbing. Today, that framing is not only dated, but also detrimental. In an AI-driven enterprise, the network is more like the nervous system. It controls and coordinates. It determines how fast you can move, how much you spend, and whether your AI investments produce value.
“This isn’t a tweak. It’s a hard reset. One that requires you to treat the network as an AI ROI lever: fund it, govern it, and measure it.
“AI systems don’t operate in a single place. They require constant data movement between clouds, data centers, and edge points. The new corporate workforce is comprised of AI agents and bots. They’re proliferating rapidly, operating continuously, insatiably consuming and generating data, and dynamically interacting with other agents, bots, and humans. And despite the early days of AI adoption in most businesses, today, more than 50% of internet traffic is created by these autonomous workers.
“A mind-blowing statistic like that reveals a dramatic disruption in business operating models and the demands they place on network infrastructure. Success will depend on the ability to move huge amounts of data from anywhere to anywhere, in real time, while delivering great customer experience at a reasonable cost. For your network to keep up, it must be big enough, fast enough, intelligent enough, and secure enough.
“Consider that chip investment – the asset with the shortest economic half-life on earth. If it sits idle, money is wasted. Still, it’s common practice in procurement departments to constrain bandwidth for cost control. Doing that also constrains data flow, starving precious GPUs. That’s like filling your F1 race car with low octane fuel.
“Bandwidth matters for customer experience, too. Employees, customers, partners – they don’t like to wait. How quickly an AI system responds with insights, measured by time to first token (TTFT), is a key metric. Less bandwidth means longer time to insight…and gratification.
“In networking, bigger is better. But who can afford enough bandwidth to support the near-infinite combinations of connections that AI agents demand? No one at scale. At least not in legacy network architectures where connections are predetermined, point-to-point, and static, and bandwidth allocation is fixed.
“To support the brave new world of AI, networks need to be completely adaptable, programmable, and consumption-based, just like cloud. Fire it up, shut it down. Make sure it’s there, wherever and whenever you need it. Pay only for what you use. Your network should be able to adapt to your business while being resilient to the environment around it. Hurricane heading towards that data center? Move the workload. Energy spot prices spiking? Move the workload. AI model needs data from a different cloud? Move it!
“You can’t do any of that on yesterday’s networks. So, make sure your network supports the future you’re building. Your business depends on it. And so does the competitiveness of our country.”
Kate Johnson, CEO, Lumen Technologies
