When most people hear "data center," they picture a single category — a big building full of servers. In reality, data center real estate spans four distinct investment types, each with different tenant profiles, lease structures, cap rates, and risk characteristics. Layer on top of that the Uptime Institute's Tier I through IV rating system, and you have a framework that determines how assets are priced, what tenants they attract, and how they perform across economic cycles.

This lesson breaks down each category so you can evaluate any data center opportunity with precision.

The four investment types

Data center investments fall into four broad categories. Understanding where a deal sits in this matrix shapes every other decision — underwriting assumptions, debt structure, exit strategy, and expected return.

| | Hyperscale Campus | Powered Shell | Colocation | Edge / Micro | |---|---|---|---|---| | Typical size | 100 MW+ | 20–200 MW | 1–50 MW | 1–5 MW | | Typical tenant | AWS, Azure, Google, Meta | Single cloud/hyperscale tenant | Multiple enterprise tenants | Telecom, edge cloud, local enterprise | | Lease structure | 20-year NNN, corporate guarantee | Long-term NNN (BTS or pre-lease) | License agreements, per-rack/cage/suite | Short-term, managed service | | Cap rate (stabilized) | 4.5–5.5% | 7.0–8.5% (yield on cost) | 6.0–7.5% | 7.5–10.0% | | Development complexity | Extreme | High | Moderate–High | Moderate | | Who builds / buys | REITs, sovereign wealth, large PE | Sponsors, REITs, developers | REITs, PE platforms, large operators | Telecom companies, emerging platforms |

Each category occupies a different risk/return position. Hyperscale delivers the most compressed cap rates and the most creditworthy tenants. Edge delivers the most upside but the least institutional trading history.

Hyperscale campuses

A hyperscale data center campus is a large-scale facility purpose-built for a single cloud or technology giant. The defining characteristic is scale: these campuses are typically 100 megawatts and above in total critical IT load, often built out in phases across hundreds of acres.

The four dominant hyperscale users are Amazon Web Services (AWS), Microsoft Azure, Google Cloud, and Meta. These four companies account for the majority of global hyperscale capacity demand and have each committed tens of billions of dollars in capital expenditure annually to expand their infrastructure footprints.

How hyperscale deals work

Hyperscale users typically approach development in two ways. In a build-to-suit transaction, the developer or REIT acquires the site, permits the project, and constructs the facility entirely to the tenant's specifications before handing over keys. In a long-term pre-lease, the tenant commits before groundbreaking and takes the completed shell at stabilization.

Either way, the economics are predictable and institutional:

Lease term: 20 years is the market standard; some deals are structured as 15-year terms with multiple 5-year renewal options
Rent escalations: 2–3% annual fixed bumps, compounding over the life of the lease
Guarantee: Investment-grade corporate guarantee from the parent entity — AWS guarantees are backed by Amazon.com, Inc.'s balance sheet
NNN structure: Tenant pays power, cooling, and operating costs above base rent
No TI exposure: The tenant is often finishing their own interior fit-out; the landlord delivers power and shell

Why do hyperscale assets command the lowest cap rates in data center CRE? The combination of long lease term, investment-grade credit, fixed escalations, and NNN structure mimics a long-duration corporate bond — but with real property as collateral. In a 4.5–5.5% cap rate environment, hyperscale leases often trade tighter than comparable industrial NNN assets because the credit and lease term are superior.

The trade-off: hyperscale development requires enormous capital, deep utility relationships, and site control in markets where hyperscale users are actively expanding. Virginia's Loudoun County corridor, Phoenix's West Valley, and the Dallas–Fort Worth Metroplex absorb billions of square feet annually.

Powered shell development

The powered shell model is the most accessible data center strategy for sponsors and institutional developers who want hyperscale exposure without the capital intensity of full-turnover delivery.

In a powered shell deal, the developer builds the physical shell of the data center — the concrete tilt-up structure, the roof, the electrical infrastructure from the utility interconnection to the switchgear, and the cooling infrastructure to the raised floor level. The hyperscale tenant then finishes the interior with their own proprietary equipment, servers, and fit-out. The tenant's capital contribution is typically 3–5x what the developer spends on the shell.

Why this model works

The powered shell approach optimizes the risk/return equation:

Lower all-in cost for the developer: A powered shell might cost $8–12M per MW to deliver, versus $18–25M+ for a fully fitted hyperscale turnover
Still captures hyperscale credit: The tenant is the same — AWS, Azure, Google — so the lease guarantee is identical to a full turnover
Yield on cost: Stabilized yields on cost typically run 7.0–8.5%, well above where these assets trade at disposition (5.0–5.5%), creating a strong spread on exit
Development advantage: By delivering less finished product, sponsors can start construction faster and reduce the cost exposure during the development window

Development risks

Powered shell development is not without complexity. The two primary risks are:

Utility interconnection: Large data center projects require substation capacity and transmission upgrades that utility companies control. Lead times for new interconnections in constrained markets like Northern Virginia or Silicon Valley can run 24–36 months. Sponsors must secure power commitments before breaking ground or risk delivering a building with no power to offer tenants.

Equipment lead times: Critical infrastructure — transformers, switchgear, generators, cooling towers — has experienced supply chain disruption. Lead times that were historically 16–26 weeks have stretched to 52–80 weeks in recent years. Sponsors who do not order equipment early in the development cycle face delivery delays that delay lease commencement and income.

Despite these risks, the powered shell strategy remains the dominant model for data center sponsors. The margin of safety — delivering less, capturing the same credit, and selling to REITs or pension funds at compressed exit caps — is hard to replicate in other asset classes.

Colocation facilities

A colocation data center is a multi-tenant facility where the operator owns and manages the building, power, cooling, and physical security infrastructure, while tenants own their own servers and equipment. Tenants rent space by the rack, cage, or private suite, paying for the "real estate" of their deployment rather than the equipment itself.

This is fundamentally different from hyperscale. Instead of one tenant on one lease, a stabilized colocation facility may have dozens or hundreds of tenants with individual license agreements. The operator handles facility operations; tenants handle their own technology.

Major colocation operators

The colocation sector is dominated by large platforms:

Equinix (EQIX) — the world's largest colocation provider, known for network-dense interconnection hubs in major markets
CyrusOne — acquired by KKR in 2022; large-scale enterprise colocation
QTS Realty Trust — acquired by Blackstone in 2021; hybrid hyperscale and colocation
DataBank — regional enterprise colocation, majority-owned by DigitalBridge
Iron Mountain (IRM) — enterprise colocation with strong data management customer base

The acquisitions of QTS by Blackstone and CyrusOne by KKR signal how seriously institutional capital has moved into data center real estate. These were not small bets — Blackstone paid $10B for QTS; KKR paid $15B for CyrusOne.

Colocation economics

Colocation facilities command cap rates in the 6.0–7.5% range, higher than hyperscale because:

Management intensity: Multi-tenant operations require staffing, security, 24/7 operations teams, and active leasing
License agreements, not leases: Colocation agreements are often shorter-term than traditional real estate leases and structured as service contracts rather than property leases, adding rollover risk
Tenant concentration risk: Losing a major customer matters more in a facility with fewer tenants

The premium to generic industrial is justified by the significant infrastructure investment per square foot, the specialized tenant base, and the barriers to entry for new colocation supply (power, permitting, network density).

For investors, colocation can be accessed through REITs (Equinix, Iron Mountain), or through direct investment in smaller regional colocation platforms that may not be institutional-quality today but have a path to scale or acquisition.

Edge and micro data centers

Edge computing describes a model where computation happens closer to the end user rather than in a centralized cloud region. A traditional hyperscale campus might be 800 miles from the user; an edge data center might be in the same metro, 20 miles from the user's device.

Why does proximity matter? Latency — the time it takes for a signal to travel from the device to the server and back. For many applications (video streaming, e-commerce transactions), latency of 50–100 milliseconds is acceptable. For emerging applications — autonomous vehicles, industrial robotics, real-time AI inference, augmented reality — latency requirements drop to 5–20 milliseconds. Physics limits how far a signal can travel in that time window.

The edge thesis

Three megatrends drive edge data center demand:

5G densification: 5G base stations generate and require processing of massive data volumes. Edge compute nodes co-located with or near cell towers reduce latency and backhaul costs
Autonomous vehicles: A self-driving car cannot tolerate a 100ms round-trip to a cloud server when making real-time driving decisions
Real-time AI inference: AI models running at the edge (analyzing factory video feeds, personalizing content, detecting fraud) need compute nodes near the data source

Investment characteristics

Edge and micro data centers are smaller (1–5 MW typical, sometimes less) and trade at 7.5–10.0% cap rates because:

Less institutional trading history: The asset class is newer, pricing discovery is still developing
Operational complexity: Many small sites rather than one large campus
Tenant base: Less dominated by investment-grade cloud giants
Shorter lease terms: More service-oriented agreements

The edge thesis is compelling but early-stage for institutional investors. Most edge infrastructure is owned and operated by telecom companies or cloud providers directly, not as third-party leased real estate. The evolution of edge toward a landlord/tenant model — similar to how hyperscale evolved from owned to leased — is a multi-year transition that is actively underway.

Tier I through IV ratings

The Uptime Institute is the global authority on data center performance standards. Their Tier certification system is the industry benchmark — referenced in lease agreements, used to qualify tenants, and incorporated into underwriting assumptions.

The Tier system rates infrastructure based on two concepts: redundancy (backup systems that can take over if primary systems fail) and concurrent maintainability (the ability to service infrastructure without shutting the facility down).

| Tier | Classification | Uptime SLA | Annual Downtime | Typical Users | |---|---|---|---|---| | Tier I | Basic | 99.671% | 28.8 hours | Small business, IT closets | | Tier II | Redundant components | 99.741% | 22.0 hours | Mid-market enterprise | | Tier III | Concurrently maintainable | 99.982% | 1.6 hours | Colocation, large enterprise | | Tier IV | Fault tolerant | 99.995% | 26 minutes | Financial services, gov, healthcare |

What each tier means in practice

Tier I is a basic data center: single path for power and cooling, no redundant components, no generator backup required. Downtime for planned maintenance takes the facility offline. Annual downtime tolerance is 28.8 hours — nearly a full day per year. Tier I is appropriate for small businesses with limited uptime requirements.

Tier II adds redundant components — backup power and cooling equipment — but still uses a single distribution path. Maintenance on that single path requires facility downtime. Annual downtime tolerance of 22 hours is meaningfully better than Tier I but still unacceptable for mission-critical operations.

Tier III is the most commercially relevant tier for colocation and enterprise data centers. The key differentiator is concurrent maintainability — all components can be serviced, repaired, or replaced without taking the facility offline. Tier III facilities have redundant power and cooling paths (typically N+1 or 2N), multiple utility feeds, and generator backup. Annual downtime is 1.6 hours. Most enterprise colocation facilities target Tier III certification.

Tier IV adds full fault tolerance: the facility can sustain any single unplanned failure — equipment failure, short circuit, fire suppression activation, human error — without impacting operations. Every component is fully redundant (2N or 2N+1), with redundant distribution paths that are simultaneously active. Annual downtime is 26 minutes. Tier IV is required for financial trading systems, government agencies, and healthcare networks where even minutes of downtime create regulatory exposure or financial loss.

Why tier rating affects cap rates

Tier rating directly affects asset value for two reasons:

Tenant quality: Mission-critical tenants — financial institutions, federal agencies, healthcare networks — require Tier III minimum and prefer Tier IV. These tenants are typically stronger credits and sign longer leases. Facilities that can demonstrate Tier III or IV certification attract a deeper, higher-quality tenant pool.

Construction cost: Tier IV infrastructure costs meaningfully more than Tier III. The redundant systems, dual distribution paths, and fault-tolerant design add to the all-in development cost. Higher cost basis pushes yield-on-cost economics and exit pricing.

A Tier IV stabilized colocation facility in a primary market will trade at a tighter cap rate than a Tier II facility because the tenant base is superior, the infrastructure is more defensible, and the replacement cost is higher.

What to take away

Data center real estate has four investment types, each with distinct cap rates, tenant profiles, and lease structures: hyperscale campus (4.5–5.5%), powered shell (7.0–8.5% yield on cost), colocation (6.0–7.5%), and edge/micro (7.5–10.0%)
Hyperscale is dominated by AWS, Azure, Google, and Meta; standard deal structure is 20-year NNN with investment-grade corporate guarantee and 2–3% annual escalators
Powered shell is the most attractive development strategy for sponsors: lower cost basis, same hyperscale credit, strong spread between yield on cost and exit cap
Colocation is multi-tenant, more management intensive, and trades at a premium to industrial because of infrastructure barriers to entry; major platforms include Equinix, QTS (Blackstone), CyrusOne (KKR), and DataBank
Edge data centers are driven by 5G, autonomous vehicles, and real-time AI inference; the asset class is earlier-stage with less institutional pricing history
The Uptime Institute Tier system rates facilities I through IV on uptime availability: Tier I (99.671%, 28.8 hrs downtime/year) through Tier IV (99.995%, 26 min downtime/year)
Tier III (concurrently maintainable) is the most common standard for commercial colocation; Tier IV (fault tolerant) is required for financial, government, and healthcare users
Tier rating affects cap rates and tenant quality — higher tier facilities attract mission-critical users who are better credits and sign longer leases
Utility interconnection and equipment lead times are the primary development risks in powered shell deals
When evaluating any data center asset, identify the type, verify the tier certification, analyze the tenant credit, and understand where the facility sits in the local power and network ecosystem

Next lesson: power infrastructure and why megawatts are the unit of value in data center real estate.

Data Center Types and Tier Ratings — What Investors Need to Know