When we talk about the future of artificial intelligence, the conversation often centers on models, algorithms, and GPUs. But beneath every breakthrough in AI lies something far more tangible—the physical infrastructure that makes it all possible. In New Albany, Ohio, a first-of-its-kind engineering project is quietly redefining what’s possible in data center connectivity: the NAO AI Mega Cluster.
This isn’t just a network upgrade. It’s a ground-up reimagination of how distributed data centers communicate—delivering 100X the fiber capacity of traditional builds and connecting a constellation of facilities into a single, unified AI training machine within a five-kilometer radius.
The challenge: connecting an AI supercluster
Meta’s AI ambitions demand infrastructure at a scale never previously attempted. The NAO Mega Cluster is a multi-region data center campus that brings together self-built facilities, leased space, cloud capacity, and purpose-built tents into one cohesive training environment.
The numbers tell the story:
- 2.1 million kilometers of fiber—enough to stretch to the moon and back 3 times—needed to make the cluster function as a single machine within a 10-km radius
The fundamental challenge: How do you build fiber connectivity at this scale, in a region already teeming with construction from other large companies all within a 10-km radius?
Civil engineering and fiber innovation: a new playbook
Rethinking the trench
Traditional telecom builds use modest conduit configurations—typically 2×1.25″ conduits with a 7-way microduct arrangement. The NAO Mega Cluster demanded something radically different.
The solution: thirty-six 6-inch conduits carrying 14×27/20mm microducts each, installed on almost every road in the New Albany Business Park. The trench itself reaches dimensions that defy telecom norms:
- 6 to 9 feet wide and 14 to 20 feet deep—large enough to fit big SUVs stacked on top of each other
- concrete casing (flowable fill/CDF encasement) protecting the conduit bank
- construction methods borrowed from utility civil works rather than traditional telecom
This isn’t a fiber build—it’s civil engineering at industrial scale.
Fiber without compromise
One of the most innovative aspects of the project is our approach to fiber placement:
- hundreds of 864-fiber and 1,728-fiber cables installed across the conduit bank, delivering over 200,000 fibers per path at scale
- no splice enclosures—cables are pulled in continuous runs using 16,000-foot and 26,000-foot reels, eliminating splice points that would introduce signal loss and complexity
- ultra-low-loss fiber selected for specific locations requiring maximum performance
- AFL fiber chosen through a competitive Kaizen process for its hardened design, reducing breakage risk during the massive installation effort
- more than 400 cables already installed, with more planned
The decision to pull fiber without splices was driven both by performance requirements (AI training demands ultra-low latency between GPUs) and practical constraints—there simply wasn’t physical space for traditional splice enclosures at this cable density.
Innovation born from necessity
The project team introduced fiber-splicing huts—purpose-built structures positioned between the cable vaults (CAVs) and meet-me rooms (MMRs)—to handle the unprecedented volume of fiber terminations. Each GPU tent holds 19,008 GPUs consuming 34 megawatts of power, and the fiber infrastructure must deliver connectivity at matching density.
Key lessons learned
Measure twice, don’t fear the cut
The project team adopted a clear decision-making philosophy: solve permitting risk for known needs, and build optionality for everything else. This meant:
- securing permits and right-of-way for the full 36-conduit configuration upfront
- installing cables in a portion of conduits on Day 1
- keeping empty conduits reserved for future growth or technology changes
This approach acknowledges a fundamental truth in infrastructure: The hardest part is getting permission to dig. Once the trench is open, the marginal cost of additional capacity is minimal. Don’t fear committing to the physical footprint—fear having to go back and do it again.
Human dynamics are critical
A project of this scale doesn’t succeed on engineering alone. The team emphasized:
- strong internal collaboration across network design, DEC (data center engineering & construction), civil engineering, and fiber supply teams
- external relationship management with the City of New Albany, utility companies, and construction partners like Congruex
- robust communication channels—when you’re open-cutting active traffic lanes in a booming tech corridor, every stakeholder needs to be aligned
The city granted permission to cut through a lane of traffic for the 14- to 20-foot-deep trench, a testament to the trust built between Meta and local authorities.
Revisit old ways of working
Perhaps the most culturally significant lesson: Don’t be afraid to challenge traditional telecom approaches. The NAO project broke numerous conventions:
- using civil-construction methods (concrete encasement, flowable fill) instead of direct-buried telecom conduit
- eliminating splice cases entirely—heresy in traditional fiber networks
- running a competitive Kaizen (continuous improvement event) at a test track to evaluate fiber vendors in real-world conditions
- treating the build as a utility-scale civil project rather than a telecommunications project
The team proved that when the problem changes (from connecting telephone offices to connecting AI supercomputers), the solution must change too.
Future implications
Partnerships are everything
The NAO Mega Cluster required deep partnerships—with construction firms , fiber manufacturers, municipalities (City of New Albany), and internal teams across Meta. As AI infrastructure scales further, no single organization can deliver alone. The future belongs to those who can orchestrate complex, multi-party construction programs at unprecedented speed.
Move fast with stable infrastructure
The project timeline was aggressive: from MSA execution in late 2024 to fiber completion of the first site in early 2026. In an industry where fiber builds typically span years, this team delivered in months—while innovating on nearly every aspect of the design. Speed came not from cutting corners, but from eliminating unnecessary complexity and making decisive architectural choices early.
Is this the new normal?
Perhaps the most provocative question the team raised: Is this scale of builds the new baseline for AI infrastructure?
With Meta’s Prometheus reaching 1GW and Hyperion (Louisiana) targeting 2+GW by 2030, the demand for hyper-dense fiber connectivity between distributed AI facilities will only accelerate. The NAO project may not be an outlier—it may be the template for every major AI campus that follows.
Beyond construction: engineering the AI backbone
The NAO AI Mega Cluster represents more than a fiber construction project. It’s a proof point that the AI revolution requires a corresponding revolution in physical infrastructure. The team behind this work—spanning civil engineering, fiber optics, network design, and program management—demonstrated that when you bring fresh thinking to old problems, extraordinary outcomes follow.
100X fiber capacity. Zero splice points. One unified AI training machine.
The ground beneath New Albany, Ohio now holds the nervous system of one of the world’s most powerful AI clusters. And the lessons learned here—about scale, speed, partnership, and the courage to challenge convention—will echo through every mega-cluster that follows.
By Fabrice Ouandji & Sebastian Gault | Network Investments
I approve subject to someone approving the messaging of GPU, MW of power, etc externally
