Joel Pacheco Gonçalves
The conversation about data centers has collapsed into one image. It’s the wrong one.
Somewhere in northern Virginia, a resident named John Steinbach opened his electricity bill in January and found $281 staring back at him. His usual bill: around $100. He’d lived in his house for nearly 40 years. The explanation he kept hearing was the same everywhere: data centers. AI. The build-out.
He’s not alone in his frustration. At least 75 data center projects worth around $130 billion were blocked or delayed in the first quarter of 2026 alone, according to research firm Data Center Watch. Moratoriums are passing in cities and states across the country. More than 800 groups are now working across 49 states to oppose some 1,500 planned data centers. The backlash is real. It’s bipartisan. And the concerns motivating it are legitimate.
But here’s what’s getting lost in that conversation: not all data centers are the same.
The facility making headlines is a specific thing
When people picture a data center today, they’re picturing a hyperscale AI campus. These are massive facilities purpose-built for the most compute-intensive workloads on earth: training large language models, running cloud platforms at global scale, storing and processing data for billions of users. The companies building them, the major cloud providers and AI developers, are pouring hundreds of billions of dollars into this infrastructure because demand is real and growing fast.
So is the footprint. Some of these campuses span thousands of acres. They draw significant power from regional grids and require substantial water for cooling. Communities are concerned about rising electricity rates, enormous water use, and public handouts in the form of tax breaks going to data center developers. That’s a legitimate policy debate. The industry knows it. Many of the largest players are actively working on more sustainable approaches, from liquid cooling to renewable energy commitments to community agreements that address local impact directly.
But the image of a warehouse-sized AI campus has become the mental model for all data centers. That’s where the conversation breaks down.
There’s an ecosystem, not a monolith
Most people don’t realize that “data center” is an umbrella term covering at least five meaningfully different types of facilities. Enterprise data centers are private facilities companies build and operate for their own use. Managed and hosted facilities are outsourced versions of the same. Hyperscale serves the cloud and AI workloads. Edge facilities push compute closer to end users to reduce latency.
And then there’s the one nobody is talking about in the backlash narrative: the interconnection hub.
These facilities, historically called carrier hotels, operate as the physical meeting point of the internet. They’re where networks exchange traffic directly with each other, without routing it through the public internet. Interconnection hubs serve as the physical home for internet exchange points, cloud on-ramps, and content delivery networks, creating a single location where networks, cloud providers, and content platforms can interconnect directly. That concentration reduces latency, eliminates unnecessary routing hops, and keeps traffic local when it doesn’t need to travel globally.
They’re not measured in acres. They’re measured in connections.
The fabric nobody sees
Think of the internet as a city. The hyperscale AI campuses are the power plants being built on its outskirts. Essential. Transformative. Changing the landscape in ways that deserve serious debate. But the city only functions because of something else: the roads, the intersections, the junctions where different systems hand off to each other.
Interconnection hubs are those junctions. Today’s internet exchange points host multiple carriers that interconnect across a shared switching fabric. The largest exchanges in the world have hundreds of participants and span multiple buildings and colocation facilities across a city. They don’t serve one company’s AI workload. They serve the whole ecosystem: ISPs, carriers, content delivery networks, cloud providers, streaming platforms, telecom operators, enterprises. Everyone who needs to reach everyone else.
The footprint is fundamentally different. So is the purpose. An interconnection hub isn’t defined by megawatts consumed or acres occupied. It’s defined by the density of networks present and the quality of the paths between them.
At MDC, this is the infrastructure we’ve built our work around. MEX-IX, the internet exchange we operate across the US-Mexico border, handles over 500 Gbps of peak traffic. That traffic represents real connectivity for carriers and ISPs serving real people in underserved markets on both sides of the border. The footprint is measured in fiber and cross-connects, not cooling towers.
Why the distinction matters
The data center backlash is going to produce regulation. Some of it will be good. Some of it will be blunt. And blunt regulation that treats a large AI training campus the same as a carrier-neutral interconnection hub will break the wrong things.
The infrastructure that connects the internet should not be collateral damage in a fight about the infrastructure that trains AI models. They are not the same category. They don’t draw the same resources. They don’t carry the same community obligations.
Understanding the difference isn’t just a technical matter. It’s a policy one. And the industry has been too quiet about making that case.
The conversation about data centers needs a more precise vocabulary.
