Telcos shoulder big risk building fiber for AI data centers

• AI data center demand is pulling regional fiber providers into bigger, riskier builds as hyperscale campuses move into harder-to-serve markets
• Fallout for costs overruns and failed projects is increasingly landing on network operators
• Fiber providers say discipline is key to surviving the data center gold rush

Regional fiber providers are being pulled deeper into the AI data center boom as data center campuses move off the beaten path. But DQE Communications CEO Mike Sicoli told Fierce the economics of building high-capacity fiber routes to hyperscale sites are getting more complicated — and the risk is landing increasingly on the telecom operator.

Sicoli said hyperscalers initially understood they would need to shoulder much of the cost to get new fiber routes built. But as the market has matured, contract terms have tightened, with customers pushing providers to absorb more exposure if projects run late or cost more than expected.

“As time has gone on and the proposition for telecom operators has gotten a little bit more challenging in terms of the risk profile. You have to be really confident that you have scoped it properly and that you have enough cushion for things that could go wrong, both in terms of the cost and the time frame” Sicoli said. “Because most of these contracts put the burden on the telecom provider if you’re late or if it costs more than you thought.”

“All of us are being asked to take an incredible amount of risk and trust that if something does go wrong you’ll still be able to work it out in a way that doesn’t kill you,” Sicoli added.

Careful calculations

Part of the reason these projects are so risky is their size. For small or regional operators, the scale could end up being make or break if they’re not very careful.

Jason Adkins is CEO of Bluebird Fiber, a regional provider of business fiber in the Midwest. As Fierce has reported, that area is seeing a huge increase in data center activity. 

For the most part, Adkins said his experience working with hyperscalers on data center projects has been good. And in many ways, he said data center contracts are not that different from any other contract. But he acknowledged that “the scale of this is so big. You’re talking hundreds of millions of dollars on a single project.”

Why does that matter? Because a 5% cost overrun on a $10 million project looks very different than 5% on $100 million. 

“We have been appropriately conservative and disciplined in that we only build where it’s a win-win for both parties,” Adkins said. That means building in areas it is familiar with, where it knows the permitting authorities and where it feels comfortable with the cost estimates that have come in. 

“I think where people could get into trouble on this is if you’re going to a completely new state that you’ve never built in before and have no knowledge of how to build there,” he said. 

Scott Bergs, CEO of communications infrastructure developer Kirkwood Infrastructure Group, told Fierce he too has had good experiences with hyperscalers, but noted something similar. Given the scale of what has to be built, legacy networks – and the processes for connecting those to data centers – are no longer relevant, he said. 

“The biggest caution is your processes you’ve used historically are not the processes you’re going to have to use here. You’re going to have to look at them differently, and resource them differently,” he said.

Location, location, location

But there’s more to the risk equation than just cost. 

Bergs said the reason companies like his exist is that the data center infrastructure currently being developed is in areas “that just don’t have anything close to adequate communications infrastructure to be able to support them.” It turns out that is both an opportunity and a potential pitfall. 

Both Bergs and Sicoli said that data center operators are looking for as many as four physically diverse network paths to connect their facilities. But given many of these new facilities are popping up in the middle of nowhere and increasing resistance from local communities, it’s a gamble for network providers as to whether they’ll be serving the next big data center hub or building a bridge to nowhere.

“The impact that we’re seeing today is sites that would otherwise be approved, or in some cases have been approved and are ready to go…go away because of community opposition,” Bergs said. “If one of those sites goes away…we’ve got an asset that can’t be utilized, and so that is a big risk.”

That’s true even if no fiber has been laid. Berg said even in cases where it hasn’t yet started construction, it has spent the time and money on engineering studies for projects that have then “just gone away.” 

Asked about the sunk cost in that kind of instance, Berg said an initial feasibility study run in the tens of thousands of dollars while advanced engineering and work acquiring easements and completing permitting drawings can run into the hundreds of thousands or millions of dollars. 

Making it work

Both Sicoli and Adkins likened the data center boom to the tower rush, but they said operators learned hard lessons then that are helping operators take the current moment in stride. 

Back then, operators were “falling all over themselves” to build to as many towers as possible, and there ended up being nowhere near as many second tenants or enterprise leaseups as had been expected. Now, Sicoli said, “the industry is being more disciplined this go around. 

Based on their experiences, the execs said there are a couple of things operators can do to help ensure success when tackling data center projects. The first is scope carefully. All three executives stressed due diligence and careful project scoping around all deals before signing contracts, pointing to that restraint as necessary despite what feels like a gold rush going on all around. Operators who build where and what they know have a better chance of sidestepping avoidable pitfalls.

Second, Bergs said, is understand the capacity and the size of the conduit that should be put in. “The worst thing would be to build a package that’s too small,” he said. It’s not enough to meet today’s needs, he added. Operators should also be looking at future needs so they don’t end up having to build the same route twice.