Metro Connect 2026: CEOs argue fibre discipline will define the AI era

Against a backdrop of unprecedented AI-driven investment, with trillions of dollars flowing into digital infrastructure, the tone was notably measured.

Chair Jiva Jagtap, senior managing director at FTI Consulting, framed the debate with a warning from history.

During the dotcom boom between 1995 and 2000, an estimated $1–1.5 trillion was invested in fibre. “Half of that was washed out in bankruptcies,” he reminded the audience. With far greater sums being deployed today, the question was clear: how does the industry avoid repeating its past?

On stage were Kenny Gunderman, CEO of Uniti, Steve Smith, CEO of Zayo, and Chris Morley, CEO of Lightpath.

Anchors, not speculation

Gunderman was quick to draw a contrast with the late 1990s. “Back then, everybody and her cousin had a CLEC licence,” he said. Fibre was built “almost indiscriminately, certainly without customers”.

Today’s build model is fundamentally different. Instead of speculative long-haul routes with no committed revenue, projects are anchored. “We’re building for an anchor customer with good economics,” he said, “but then there’s the promise of that second or third or fourth customer to lease up the network and really get to economics.”

Demand, too, is structurally stronger. In the dotcom era, the applications simply were not there. “You didn’t have hundreds of millions or billions of cell phones driving demand. You didn’t have fibre to the home, you didn’t have satellite, you didn’t have all these mega use cases,” Gunderman noted. The proliferation of devices, cloud and now AI has created a demand base that did not previously exist.

Smith agreed, adding that the capital stack has matured. “There’s a lot more private capital involved now than there was 26, 27 years ago,” he said. More importantly, much of today’s investment is being funded by hyperscalers from operating cash flow. “The biggest, most sophisticated, smartest companies in the world are funding it from operating cash flow. That is not the case back in ’96, ’97, ’98.”

Bigger orders, tighter timelines

If the past was characterised by overbuild and loose assumptions, today’s cycle is defined by scale and urgency.

Smith described a dramatic shift in buying patterns. Where once fibre requirements might have been 12 or 24 strands, “today it’s in the hundreds… in the thousands.” Activation timelines are also tighter, with stringent penalties for delay. “Your feasibility studies when you’re building these routes today are super, super important,” he said. Investment committees are demanding rigorous proof that builds can be delivered “on time, on budget”.

Gunderman echoed the point. Hyperscalers are planning 10 years out, he said, and placing orders for 432 or 864 strands rather than modest counts. Crucially, he sees AI demand broadening beyond the training phase. “People equate AI with hyperscalers… but the reality is AI is going to proliferate through all of our customer segments.” As inference scales, enterprises, healthcare, universities and government will all require significantly more capacity.

Morley, for his part, rejected any narrative of underperformance. “I almost want to get up and apologise,” he joked, referencing industry charts showing historic value destruction. “But I don’t think we need to.” He pointed to double-digit revenue growth at Lightpath and strong investor returns in previous roles. “The opportunity right now in fibre… is probably better than it’s been in the 25 years that I’ve been in the business.”

Pricing power, but with discipline

Price compression has long been the industry’s structural headache. Jagtap pressed the panel on whether anything had truly changed.

Smith was pragmatic. “We live with price compression today. It’s never going to go away.” However, the nature of the product is evolving. The industry is shifting from pricing individual strands to pricing complex routes. In growth corridors linked to AI, there is clear pricing power, particularly where delivery speed and financing capability differentiate providers.

Morley argued that something more fundamental is happening. Historically, operators absorbed 3–5% annual revenue declines due to falling unit prices. At Lightpath, he said, pricing compression last year was more than offset by increases achieved in new, unique routes and locations, resulting in net business growth of around 2.5%. “That fundamentally changes and swings our growth rates,” he said.

Gunderman added that hyperscalers are disciplined but pragmatic customers. They prioritise speed, reliability and partnership alongside price. Fibre, he noted, represents a small fraction of total data centre investment compared to power and land. “Plus or minus a little bit on the fibre side… is not going to change the economics” of a hyperscale build.

There is also a mutual dependence. As one panellist observed, hyperscalers understand that connectivity providers must earn “owner economics” if the ecosystem is to function. “They don’t want to run these companies into the ground,” he said.

Selective expansion

When asked about remote training locations, from rural Texas to the Dakotas, Morley stressed selectivity. Past mistakes, he acknowledged, included building for single anchor tenants without sufficiently monetising follow-on demand. Today, new markets must offer a clear line of sight to commercialisation beyond the initial hyperscale customer.

Encouragingly, he said, where power is available, secondary demand often follows quickly. What might once have been a speculative bet is now underpinned by clustering effects.

Smith highlighted Zayo’s focus on specific long-haul growth corridors and metro densification, positioning for inference demand as it shifts closer to end users.

Enterprise fibre: quietly resilient

Amid the AI excitement, Jagtap challenged the panel not to forget enterprise fibre.

Gunderman was emphatic that the segment’s decline has been overstated. “Enterprise fibre has gotten a bad name,” he said, largely because it is lumped together with legacy copper and TDM services. Strip those out, and true on-net enterprise fibre is growing at 10–15% annually, with churn below 1%. “Those are very attractive characteristics.”

Enterprise also underpins lease-up economics. “Anytime we build something, there’s always a question of: are we going to have enterprise lease-up on this deal?”

Smith sees enterprise demand as the ultimate monetisation engine for AI. Training may be happening in remote facilities, but inference – and revenue – will be concentrated in tier-one and tier-two metros. As enterprises embed AI into operations, connectivity needs will intensify. He also flagged a potential re-architecture of the internet as personalised AI content challenges traditional caching models.

Learning from history

Throughout the discussion, the ghost of the dotcom bust lingered. The panellists did not dismiss the risk of winners and losers in the current cycle. But they argued that capital discipline, anchored builds, smarter customers and broader demand foundations set this era apart.

As a former banker, Gunderman offered a wry reflection: “I didn’t know how bad of a banker I was until I wasn’t one anymore.” The industry, he suggested, has had its own learning curve.

Whether AI proves to be electricity for the digital age, as Smith implied, or simply the latest overhyped wave, will become clear over time. For now, the CEOs on stage at Metro Connect 2026 were united on one point: fibre remains the essential substrate, and this time, they intend to build it responsibly.

Marc Ganzi at Metro Connect 2026: AI monetisation hinges on industrialisation