Your AI Campus Has a Power Date. It Needs a Network Date.

Permitting, geological conditions, and fiber and equipment lead times can stretch a network build to 2 years.

By Jesse Jones, Director, Neocloud and Superscale Sales, Lightpath

Most AI campus announcements I read lead with megawatts and acreage. The mockup is always beautiful. Fiber rarely makes it into the press release, and for most of my career that was fine because fiber was available at the site before the data center broke ground. That isn’t the world anymore, and the announcements still don’t say anything about fiber readiness.

Data center developers, neoclouds, and joint-venture operators I work with are standing up AI infrastructure in geographies where the existing fiber map thins out. Yes, long-haul runs through those places. What is often missing is the lateral into the actual site, the diverse entrance a working campus needs, and the fiber count AI workloads require. Everyone in this industry knows where fiber isn’t. The harder question, and the one ten years working in wholesale, carrier, and enterprise fiber taught me to answer, is how long it actually takes to build fiber where none exists yet. When you add permitting, easements, county and railroad crossings, and equipment sitting on order, does the number still leave your Ready-for-Service (RFS) date intact, or is your go-live already impossible?

Data centers used to be built where fiber already was

For most of the last two decades, data centers didn’t build the fiber map. They followed it. Ashburn, Silicon Valley, Dallas, Denver, Chicago, and Atlanta were already telecom junction points, dense with interconnect and carrier presence. A developer who signed a lease in one of those markets could call a local carrier 90 days before their RFS date and turn service up in the last quarter of the build.

That sequence has flipped. The largest AI campus builds moving through the current pipeline are being sited on the availability of power and land, not on the presence of network services. Long-haul often passes within tens of miles of these sites, but long-haul is not what serves a campus. The lateral into the site is, the diverse-entrance path is, and neither is built yet. Without them, the tenant’s applications can’t run at commit levels on the day the doors open, and no amount of long-haul availability upstream fixes it. A developer planning around the 90-day carrier turn-up is planning a building that will not commission on the date the tenant expects.

State reviews are shrinking the number of buildable AI projects

The other pressure squeezing that timeline is state policy. State-level oversight is tightening across the biggest AI markets. In Texas, Governor Abbott’s August directive orders ERCOT to audit data-center projects advancing through the interconnection queue and requires denial of grid connection to any project that fails. In Pennsylvania, Governor Shapiro’s executive order ended fast-tracked approvals, required data centers to bring their own power, and blocked state permits without local municipal approval. In Michigan, Governor Whitmer’s ten-point pledge asks data-center companies to protect ratepayers or face legislation that codifies the same protections.

Every one of those reviews takes projects off the buildable list, but the tenant demand doesn’t disappear. That demand consolidates onto the projects still moving forward. Surviving build projects may have to absorb capacity they weren’t originally scoped for, toward RFS dates that are less forgiving now that tenants have fewer alternatives. Every one of those build project teams is bidding for the same fiber crews and the same permit slots as every other surviving site in the market. That is why the network planning effort sitting in front of them just got harder.

Fiber and equipment lead times are longer than the schedule has room for

That squeeze lands on top of a supply chain that was already stretched. Industry reporting from Global Data Center Hub this spring puts fiber cable lead times at up to 60 weeks, the longest since the early-2000s build-out cycle. Network equipment lead times sit alongside cable at 60+ weeks for preferred optical equipment manufacturers. Fiber and equipment together can consume more than a year of a construction schedule, even while design, permitting, and conduit installs are underway.

Then there’s the design load. Global Data Center Hub has AI campuses drawing 10x to 36x the fiber of a conventional deployment. Per-rack fiber counts now run into the hundreds, compared to 15 to 30 on a legacy design. The same stretched supply chain is being asked to serve a much thirstier build.

If a project has a go-live date 15 months out and no one has started the network conversation, the math already doesn’t work.

How to keep fiber on the critical path from day one

The teams working on AI campus builds and avoiding this pattern are the ones running the network conversation in parallel with power and site. They ask what long-haul routes exist today and what has to be built new. They ask where the diverse entrance comes from, and whether the fiber count they need matches what the site is designed for (and what their tenant needs). They know their live date and they count backward from it.

The starting point

If you’re working on an AI campus with a power date and a compute date but no network date yet, the fastest first move is a conversation. Our team walks through your site plan in relation to fiber and equipment lead times that stand between now and your RFS date: long-haul availability, the lateral into the campus itself, and where the diverse-entrance path comes from. That is the starting point.

When you’re ready to run the numbers together, reach out to the Lightpath Major Infrastructure Solutions team, or find me on LinkedIn.