Wholesale vs. Colocation Data Center Development: What Changes in the Business Case

The product type decision shapes every downstream assumption — lease structure, power density, capex, and exit.

Two Products, Two Business Models

From the outside, a wholesale data center and a colocation facility look similar: large building, lots of power, raised floors or cold aisles, generators outside. From a developer's perspective, they are fundamentally different businesses with different risk profiles, different tenant bases, and different financing structures.

Getting the product type decision wrong early is expensive. Correcting it mid-development is more expensive.

Wholesale Data Centers

Wholesale facilities lease power capacity in large blocks, typically starting at 1 megawatt and ranging to 20 or more megawatts per tenant. Leases are long-term, usually 10 to 15 years, triple-net. The tenant builds out and operates their own infrastructure within the shell. The developer provides power, cooling infrastructure, physical security, and a reliable building envelope.

What changes in the business case:

Tenant credit is everything. Wholesale leases are essentially credit instruments. A 15-year lease to a hyperscaler — Microsoft, Amazon, Meta, Oracle — is close to investment-grade paper. A lease to a smaller cloud provider or enterprise tenant introduces credit risk that changes how lenders and equity underwrite the deal. Most institutional lenders apply different coverage and LTV thresholds based on tenant quality.

Capex is higher upfront. Developers building wholesale facilities finance the shell, power infrastructure, and cooling to a much higher specification than colo. Transformer capacity, redundant power paths, and cooling head room are all pre-built to tenant specification before a lease is signed — or, in pre-leased deals, built to a hyperscaler's technical standard simultaneously with lease execution.

Power density drives design. Traditional wholesale leases have been priced and designed around 200 to 400 watts per square foot. GPU and AI inference workloads are pushing requirements toward 600 to 1,000+ watts per square foot. Developers building to 2024 specifications are already behind. Design assumptions made today need to accommodate the density trajectory over a 15-year lease term.

Exit is straightforward when stabilized. A stabilized wholesale facility on a long-term hyperscaler lease trades at compressed cap rates — 4% to 5.5% in primary markets as of early 2026, depending on market, tenant, and remaining lease term. The investment thesis is income stability and inflation passthrough in leases with annual escalators.

Colocation Data Centers

Colocation facilities lease smaller increments — individual cabinets, cages, or suites — to a larger number of tenants. The operator provides the shared physical infrastructure (power distribution, cooling, network connectivity) and tenants bring their own servers. Leases are shorter, typically one to five years, and priced per kilowatt of power capacity or per rack unit.

What changes in the business case:

Operational complexity is higher. Colo operators are managing dozens or hundreds of tenant relationships simultaneously, with varied power requirements, security requirements, and support needs. The developer/operator model requires a service layer that wholesale development does not — NOC staffing, customer success, and 24/7 support are operating costs that factor into NOI.

Revenue per square foot is higher, but so is volatility. Colo facilities command higher revenue per square foot than wholesale when fully leased. But lease tenures are shorter, tenant concentration risk is distributed differently (no single tenant can walk away and take 20 MW with them), and vacancy affects revenue immediately. Underwriting a colo facility requires more granular demand assumptions.

The network matters. Colo tenants choose facilities based on connectivity density — the number of carriers, cloud on-ramps, and IX access points available in the building. Facilities with rich connectivity ecosystems (Equinix-style interconnection hubs) command premium pricing. Developers entering colo without a clear interconnection strategy compete on price, which is a losing position.

Cap rates are different. A stabilized colo facility with diversified tenancy and high interconnection density trades at different multiples than wholesale. In primary markets, well-connected colo assets can achieve sub-5% cap rates based on revenue quality. Enterprise-focused colo without a connectivity moat trades closer to 6.5% to 7.5%.

The Decision Framework

Neither product type is inherently superior. The right answer depends on market conditions, capital structure, and the developer's operational capability.

Wholesale fits when:

• A hyperscaler pre-lease or letter of intent is available

• The developer does not want to operate a services business

• Market conditions favor long-term lease income over operational upside

• The capital structure prioritizes low-risk, long-dated cash flows

Colocation fits when:

• The target market has strong enterprise and mid-market demand

• The developer has or can acquire operational expertise

• Network connectivity density can be established at the facility

• The investment thesis depends on revenue upside from a growing interconnection ecosystem

The hybrid path — building wholesale shell with colo capacity reserved in a meet-me room — is common in campus developments. It captures both income streams but increases complexity in design, operations, and lease management.

How AI Models the Business Case

For developers evaluating both options at a given site, AI can run parallel pro forma models with different lease-up assumptions, power density scenarios, and cap rate ranges. The output is not a single answer — it's a sensitivity table showing where each product type wins under different market conditions.

The most common use case: stress-testing wholesale underwriting against a scenario where the hyperscaler lease comes in 12 months late or at 15% lower power pricing than modeled. That stress test, run against a colo alternative with a 12-month lease-up curve, often clarifies the product type decision faster than months of committee debate.

The business case for each product type is knowable. The risk is assuming the answer without modeling the question.