Over the past year, as AI infrastructure and AI data centers became the dominant topic at every industry conference and in every boardroom conversation, the question I kept hearing was the same: “How do we connect the data center?”
It’s a good question, and the industry responded well. Vendors, consultants, and operators all leaned into the connectivity opportunity: fiber connectivity to AI data centers, transport upgrades, and backhaul capacity. That work is important, and it represents real revenue.
But the more time I spent with operators exploring this space, the more I found myself asking a different question: Are we leaving something bigger on the table by thinking too small? If all we see in this moment is a connectivity sale, are we missing the chance to participate in the AI infrastructure economy at a level that fundamentally changes what a rural broadband company can be?
So when Verizon announced its dark fiber deal with Google last week, worth more than $1 billion, using fiber it already had in the ground, I knew it was a significant signal. What stood out to me, though, was that many people in our industry seem to be drawing the wrong conclusion about what this moment actually means for rural broadband providers and the future of AI infrastructure.
The Headline Isn’t the Real Story for Rural Broadband Providers
Verizon’s deal is straightforward: Google needs high-capacity connectivity between data centers, and Verizon has fiber to sell. CEO Dan Schulman said on the earnings call that he expects to announce additional deals by year-end worth “multiple billions” more.
The same week, AT&T posted its strongest broadband quarter on record with 646,000 net additions, while Verizon added 348,000 broadband subscribers and reached record EBITDA margins.
The carriers are collectively discovering that the most valuable thing they own right now isn’t spectrum or subscribers. It’s fiber infrastructure already in the ground. If you’re a rural broadband provider, that should get your attention. You own fiber in the ground too. But if you read this headline and thought, “Maybe I can sell dark fiber to a data center someday,” you’re looking at the AI infrastructure opportunity from only one angle.
The AI infrastructure buildout, which Schulman described as “one of the largest capital cycles of our lifetime,” is creating demand for an entirely new layer of physical infrastructure. And the market’s biggest constraint isn’t capital or hardware. It’s power.
Goldman Sachs projects U.S. data center power demand will roughly double from 31 to 66 gigawatts between 2025 and 2027. JLL’s 2026 outlook identifies speed-to-power as the number one site selection criterion. Grid interconnection queues in major markets still stretch two to five years, and Bloomberg/Sightline Climate reported in April that roughly 30% to 50% of large data center projects planned for 2026 have been delayed or canceled.
The hyperscale model is hitting structural limits. And that’s opening the door to a different approach to building and deploying AI infrastructure.
Why AI Infrastructure Is Moving Beyond Hyperscale
Rethinking What “Data Center” Means
Here’s where most rural operators eliminate themselves from the conversation. They hear “data center” and picture a 100-megawatt hyperscale campus spread across hundreds of acres. They look at that image, conclude they have no role to play, and move on. Many also anticipate concerns from their board members about the disruption these large facilities could bring to their communities.
But that picture is incomplete, and it’s a relatively recent way of thinking.
A decade ago, a 5-megawatt data center was a perfectly normal commercial facility. Enterprise data centers under 5 MW still account for roughly half of all servers in operation today, according to the Electric Power Research Institute.
What changed isn’t that smaller facilities stopped being useful. It’s that hyperscalers built at such massive scale that they redefined what the industry considers a “real” data center. They moved the goalposts, and much of the market accepted that new definition without questioning it.
Meanwhile, the Neoclouds, the specialized providers offering GPU-based AI compute as a service that I wrote about in a previous issue, are scaling rapidly. ABI Research forecasts more than 2,200 Neocloud-operated facilities globally by 2035, up from roughly 560 today.
The largest Neoclouds are building at significant scale. But the broader AI compute market isn’t becoming more concentrated. It’s becoming more distributed.
As inference workloads grow, already representing roughly two-thirds of AI compute according to Deloitte, demand is shifting toward locations that can deliver power quickly, not simply locations with the most available power.
Here’s what matters for rural operators. The power envelope that supported a large data center ten years ago, roughly 1 to 5 megawatts, is now well within the range that can support meaningful AI compute and edge AI workloads.
That’s not a projection. We’re working with operators right now on facilities in that range. The infrastructure that seemed too small during the hyperscale era may actually be the right size for distributed AI workloads that need to be closer to where data is generated and consumed.
The AI Infrastructure Opportunity Rural Broadband Providers Are Missing
I’ve been spending a lot of time at the intersection of rural telecom and digital infrastructure, and here’s what I keep seeing. Operators who engage with this opportunity often frame it as a connectivity play: “We could provide fiber to a data center.” And yes, they could. But connectivity alone captures only a fraction of the value available in this AI infrastructure market.
The market isn’t buying bandwidth. It’s buying deployable capacity: the combination of available power, utility alignment, fiber, and a site that can become operational quickly.
The operators that stand to capture the most value from this wave won’t simply sell a circuit to a facility someone else builds. They’ll recognize that the assets they already own, including fiber networks, land, power access, utility relationships, existing facilities, and community trust, are exactly the inputs the AI infrastructure market is looking for.
The better question is: What combination of fiber, power, land, facilities, and local relationships could position a rural broadband provider to support AI compute in its market?
Who Is a Good Fit for AI Infrastructure?
I want to be direct about something: not every rural market is a fit for this.
To participate meaningfully in AI infrastructure, you need some level of compute demand within reach — whether that’s a Neocloud looking for fast deployment, an enterprise customer with latency-sensitive workloads, or even local institutions like hospitals, agricultural operations, or school districts that could benefit from distributed computing and edge AI.
If you’re hyper-rural with limited commercial activity and no realistic path to power at scale, this may not be your opportunity. That’s not a failure. It’s an honest assessment that can save you from an expensive mistake while allowing you to focus on opportunities that better fit your market.
But here’s what I’d encourage you to consider. Your best first customer may not be a Neocloud at all. It may be your own community. Local healthcare systems running AI-assisted diagnostics, precision agriculture platforms that need edge processing, and county services that can’t afford the latency of a data center three states away are all examples of demand that already exists closer to home than most operators realize.
Those conversations about what compute your community actually needs are often the right place for many operators to begin evaluating their AI infrastructure opportunities..
Why Rural Broadband Providers Should Look Beyond Connectivity
Here’s the dynamic I’m watching most closely.
The AI infrastructure market is taking shape right now. Developers, investors, and Neocloud operators are actively scouting locations, signing power purchase agreements, and securing sites.
Rural operators who engage now — even if that simply means evaluating their fiber infrastructure, available land, power access, existing facilities, and utility relationships — will be well positioned as the market matures.
The operators who wait will almost certainly still have a role, but it will likely be a familiar one: selling connectivity to someone else’s facility, on someone else’s terms.
Maybe that’s not a bad outcome. Fiber connectivity to a data center is real revenue. But it also means giving up the larger AI infrastructure opportunity, and the economic impact that comes with it, to someone else. That value could have remained in your community.
Twenty years ago, large carriers told rural operators that broadband was their game. You proved otherwise by leveraging local assets, local relationships, and a business model that worked at your scale.
Today, we’re seeing a remarkably similar strategic pattern. The application has changed. Now it’s AI compute and digital infrastructure instead of connectivity. But the underlying opportunity looks very familiar.
The first move isn’t construction. It’s evaluation. And the window to lead that evaluation in your territory is open right now for rural broadband providers willing to look beyond dark fiber and connectivity and consider their role in the emerging AI infrastructure market.
If you’re trying to determine whether this opportunity applies to your market, that evaluation is exactly the kind of work we do at Sunstone Associates.
Stay connected to us through our monthly email newsletter. If you want to keep up with what we’re doing at Sunstone Associates, please sign up here(opens in new tab).