Retail vs Wholesale Colocation Cost in 2026

The old colocation rule was easy to remember: retail costs more, wholesale costs less. It is still directionally useful. It is also much less reliable than it used to be.

McKinsey's June 2026 work still shows the familiar split: retail colocation at roughly $200–$380 per kW per month and wholesale leases around $150–$200. But Colliers' 2026 U.S. market report adds an important wrinkle: once power costs are normalized, the historical pricing gap between retail and wholesale has largely closed in many U.S. markets.

That is the interesting part. The market has not stopped having retail and wholesale products. What has changed is that scarce deliverable power, low vacancy and expensive new construction are now influencing price so heavily that the product label alone does not tell you what the better deal is.

McKinsey's 2026 business-model ranges. Useful as orientation, not as a universal rate card.

Retail and wholesale are really two different ways of buying infrastructure

The simplest distinction is scale, but scale is only the beginning. Retail colocation packages a smaller slice of a facility with more flexibility around space, connectivity and services. Wholesale gives a customer a much larger capacity commitment and generally strips the economics closer to the underlying infrastructure.

The labels themselves are not regulated specifications. One provider may call a 300 kW deployment wholesale; another may treat it as a large retail cage. Colliers' 2025 market work used 500 kW as a practical dividing line in its pricing charts, but I would not turn that into a universal definition. In real procurement, the contract structure and facility configuration matter more than the heading on a sales deck.

The part most comparison articles miss: power accounting

A retail rate can look far more expensive simply because electricity is already inside it. A wholesale rate may be quoted net of electrical charges and then receive a separate power bill.

This is exactly why Colliers' 2026 observation matters. Its current report says existing multi-tenant wholesale and colocation deployments in primary U.S. markets generally transact around $160–$200/kW/month net of electrical charges, while the old retail-versus-wholesale gap has largely closed after adjusting for power costs. The report is unusually explicit about that point.

The $70 in that example is deliberately illustrative, not a market benchmark. It shows the arithmetic problem. Comparing $175 net-of-power with $250 power-inclusive is not a $75 price gap until the actual electrical economics are added back.

What the current market is doing to the traditional volume discount

Bigger used to mean cheaper almost automatically. Commit more capacity, give the operator a longer contract and the unit price falls. The logic still exists, but the market is interfering with it.

Lightyear's current colocation pricing research notes that requirements above 1 MW are no longer consistently receiving the volume discounts buyers would historically expect, because large contiguous blocks of power are difficult to find in primary markets. Its H2 2025 retail benchmark reached a median $380 per usable kW, up 17.3% in six months, with the 90th percentile at $663. Lightyear's methodology includes space and power but excludes cross-connects and connectivity.

This is one of those market changes that sounds counterintuitive until you think about what the provider is actually selling. A 50 kW customer can fit into many more pockets of residual capacity than a 2 MW customer. The larger buyer may be operationally attractive, but it also consumes an increasingly scarce asset: a large block of deliverable power in one place.

A 100 kW buyer and a 2 MW buyer are not shopping the same way

I would be wary of forcing the 100 kW buyer into a wholesale-style decision purely to get a lower nominal $/kW. If that saves $20 per kW but adds network complexity, weaker support or a second facility later because expansion is unavailable, the cheaper infrastructure rate may lose at the system level.

The reverse is equally true. A 2 MW tenant paying for a retail service bundle it barely uses can be carrying unnecessary cost for years.

Put the contract beside the price

Retail and wholesale economics diverge in the contract as much as in the headline rate. A lower starting price can be overtaken by escalators, operating-cost pass-throughs or expensive expansion terms.

Colliers reports that triple-net structures have become more common in large single-tenant environments as operators and customers try to allocate increasingly volatile taxes, insurance and maintenance costs more clearly. That is another reason a wholesale lease can look inexpensive on a base-rent comparison while transferring more operating-cost risk to the tenant.

There is also an investor-side reason retail has historically priced higher

McKinsey's 2026 analysis puts retail colocation equity internal rates of return around 20%–25%, compared with roughly 13%–18% for wholesale. Its explanation is intuitive: retail customers pay more for flexibility, redundancy and managed services, while wholesale facilities accept lower unit pricing in exchange for large, long-duration commitments and lower lease-up risk.

I would not use those return ranges to reverse-engineer a provider's margin on a specific quote. They are investment-model outputs, not a disclosure of what an operator earns on one customer. They are useful because they explain why two apparently similar facilities can pursue very different pricing strategies.

The “cheaper” option changes when connectivity enters the picture

Retail colocation can be expensive for a reason that is easy to underestimate: some customers are buying access to an ecosystem, not just floor space and electricity.

A carrier-dense facility can put multiple networks, cloud connections, partners and counterparties within one building. If those interconnections reduce wide-area network cost or improve architecture, the premium belongs in the total-cost calculation rather than being dismissed as expensive rack space.

This is also why the term “retail premium” can be misleading. A premium exists only relative to a comparable alternative. If the wholesale option requires separate network services, a different operations model and a second site to get equivalent connectivity, the true premium may be smaller than the rate card suggests.

A worked comparison: 250 kW retail versus wholesale

The following example is intentionally simplified. It is not a quote and does not attempt to model every contract term. The goal is to show how quickly the comparison changes once power is normalized.

On the original rate card, wholesale looked 40% cheaper: $180 versus $300. After adding an illustrative $70/kW effective power component, the normalized gap becomes about 17%. Add different cross-connect, support or escalation assumptions and it can move again.

This is the sort of comparison where I would stop looking at the first-year $/kW rate and build a three- to five-year cash flow. The arithmetic is not difficult. The hard part is making sure the inputs represent the same service.

So which model is economically better?

The market itself is making that distinction less tidy. Colliers' 2026 vacancy data shows exceptionally tight conditions in several U.S. hubs, and its pricing analysis says scarcity, tenant credit, lease term, facility condition and execution certainty now explain more of the dispersion than the retail/wholesale label by itself. McKinsey still finds a meaningful business-model difference, but it reaches the same broader conclusion: pricing depends on customer type, contract structure and supply-demand balance, not geography alone.

If I were reviewing a colo shortlist today, I would still keep a column for “retail” or “wholesale.” I just would not let that column decide anything. The columns beside it — usable kW, electricity treatment, contract term, escalator, interconnection cost and expansion rights — are where the 2026 economics actually show up.