For decades, commercial real estate has been underwritten on a familiar premise: the building is the asset. Location, structure, tenant turnover, and lease rates define value, and over time, the property can be repositioned, re-leased, or sold to generate returns.
That model breaks down when applied to modern AI data centers.
In this new paradigm, the building is not the asset it’s a wrapper. The real value lies in the infrastructure that enables compute: power delivery, fiber connectivity, and the ability to continuously produce and scale digital workloads. Investors who continue to evaluate data centers through a traditional real estate lens risk fundamentally mispricing both opportunity and risk.
The Fallacy of the “Reusable Building”
In office, retail, or industrial real estate, the structure itself has enduring utility. Tenants come and go, but the building remains adaptable. Its value is tied to location and its ability to generate rent across cycles.
AI data centers don’t behave this way.
These facilities are:
- Highly specialized in design and density
- Tightly coupled to specific power and cooling configurations
- Built for purpose, often aligned to a single tenant or workload profile
Repurposing a data center for a new tenant is not as simple as signing a new lease. In many cases, it requires significant reconfiguration or isn’t feasible at all. That reality shifts the center of gravity away from the structure and toward the infrastructure supporting it.
Compute Is the Product
AI data centers are not passive real estate assets they are production environments.
The “output” is compute: training models, running inference, and supporting massive data flows. Revenue is driven not just by space, but by the ability to deliver reliable, scalable, high-density compute capacity.
This reframes value in a critical way:
- The building enables operations
- The infrastructure powers the operations
- The compute is the value
If the facility cannot deliver power at the required scale or connect to high-capacity networks, the building itself becomes irrelevant.
Power: The Primary Constraint
Power interconnection is the single most important determinant of value in an AI data center project.
It dictates:
- How much compute can be deployed
- How quickly the facility can go live
- Whether future expansion is possible
A site with a secured, scalable power interconnection is exponentially more valuable than one without—regardless of the quality of the building. Conversely, a state-of-the-art facility without adequate power is effectively a stranded asset.
Investors should be underwriting:
- Confirmed megawatt capacity (not theoretical availability)
- Delivery timelines from utilities
- Long-term scalability and grid reliability
Without this, projected returns are built on assumptions, not infrastructure.
Fiber: The Multiplier
If power enables compute, fiber monetizes it.
High-capacity, low-latency connectivity is essential for AI workloads, particularly as training and inference become increasingly distributed. Fiber determines how effectively a data center can integrate into broader ecosystems cloud regions, enterprise networks, and global backbones.
Key considerations include:
- Proximity to diverse fiber routes
- Access to both dark and lit capacity
- Scalability to support future bandwidth demands
A facility with robust fiber connectivity can fully leverage its compute capabilities. Without it, even abundant power cannot translate into meaningful output.
Infrastructure Over Improvements
Traditional real estate valuation emphasizes “improvements” the physical enhancements made to land. In AI data centers, the hierarchy is inverted.
The most valuable components are:
- Power interconnection rights and capacity
- Fiber routes and network access
- Land positioned to support both
- The building itself
This is a fundamental shift. The infrastructure is not supporting the building the building is supporting the infrastructure.
The Risk of Misaligned Underwriting
When investors apply legacy real estate frameworks to AI data centers, several risks emerge:
- Overvaluing the structure: Assuming the building retains long-term flexibility and resale value
- Undervaluing infrastructure timelines: Ignoring the complexity of securing power and fiber
- Misjudging exit strategies: Expecting traditional buyer pools for highly specialized assets
- Underestimating obsolescence risk: As compute density evolves, facilities without scalable infrastructure fall behind
These gaps can lead to mispriced acquisitions, delayed projects, and compressed returns.
A Required Mindset Shift
To accurately evaluate AI data center opportunities, investors must shift from a real estate mindset to an infrastructure mindset.
This means:
- Viewing power and fiber as core assets, not utilities
- Prioritizing capacity, connectivity, and scalability over square footage
- Aligning investment horizons with infrastructure deployment timelines
- Understanding that compute output not occupancy is the revenue driver
In practical terms, the question is no longer “What is this building worth?” but rather “What can this site produce?”
Conclusion: The Asset Is the Output
AI data centers represent a convergence of real estate, energy, and network infrastructure but they should not be mistaken for traditional property investments.
The building may house the operation, but it does not define the value.
Power interconnection and fiber connectivity are the true assets. They determine whether compute can be delivered, scaled, and monetized. Everything else is secondary.
Investors who recognize this shift and underwrite accordingly will be positioned to capture the real value of AI infrastructure. Those who don’t may find themselves holding buildings that look impressive on paper, but fail to deliver in practice.
