Hyperscale vs Wholesale vs Retail Colocation

Introduction

Ask most people to describe the data center market and they will describe one product. In reality the sector serves three distinct demand segments, and the differences between them, in deal size, lease term, pricing, and margin, are large enough that they behave like separate businesses sharing a roof. A banker who lumps them together will misread both the revenue quality and the risk of any given portfolio. The three are retail colocation, wholesale colocation, and hyperscale, and the most surprising thing about them is that the smallest deals, not the largest, carry the highest percentage margins.

The cleanest way to separate the segments is by the size of the commitment, which then drives almost everything else about the deal. Retail customers take small footprints, often a single rack or a few cabinets, and pay largely for connectivity and proximity to an ecosystem. Wholesale customers take dedicated rooms or suites and operate them as their own. Hyperscalers take floors, whole buildings, or entire campuses, frequently on a build-to-suit basis.

Colocation

A model in which a data center operator owns the building, power, and cooling and leases space and electrical capacity to customers who install and run their own equipment. It spans everything from a single rack (retail) to a multi-megawatt suite (wholesale), and is distinct from the cloud, where the customer rents computing rather than space.

The numbers attached to each segment make the distinction concrete:

These map directly onto the two public REITs: Equinix anchors the retail-and-interconnection end while Digital Realty anchors wholesale, a split laid out in the data center REIT landscape. The detailed lease forms each segment uses, from monthly retail service contracts to long hyperscale build-to-suit leases, are covered in the article on data center lease structures.

Here is the counterintuitive part. Retail colocation deals carry the highest gross margins in percentage terms, because a single rack sold with premium interconnection commands a high effective rate per kilowatt, and the customer is paying for the ecosystem, not just the power. But retail volume is small and fragmented across many customers. Wholesale and hyperscale deals carry lower percentage margins because large tenants have the scale and the market power to negotiate hard, yet they generate far more total revenue per signed contract and lock it in for a decade or more.

That trade-off explains why most operators have pivoted toward wholesale and hyperscale even though the margins are thinner. A single hyperscale lease can deliver more contracted revenue than hundreds of retail deals, with predictable, recurring cash flow over a fifteen-year term. The revenue is lower-margin but far more bankable, and bankable cash flow is what supports the debt and the development capital the sector runs on.

The cleanest way to size that retention gap is the churn rate, the share of capacity lost to non-renewal over a period. Retail tenants on 12-to-36-month service contracts can leave on short notice and frequently do, so retail churn runs structurally higher than the near-zero churn of a hyperscaler locked into a fifteen-year build-to-suit lease. That single ratio is why the same dollar of revenue is worth more at the hyperscale end: bankable cash flow is partly a statement about how rarely the tenant walks.

Interconnection: The Retail Moat

Interconnection is the hidden economics underneath retail, and it is what lets the segment defend its pricing while raw capacity gets commoditized. A facility full of networks, clouds, and enterprises becomes a marketplace, and the value of joining it rises with every participant already inside.

Cross-Connect

A direct physical or virtual link between two customers inside the same data center, bypassing the public internet. Cross-connects typically rent for $100 to $300 per month each, and a dense campus can carry thousands of them, generating high-margin, recurring revenue that has little to do with how much power a tenant consumes.

Because that revenue is sticky and largely independent of power draw, the retail-and-interconnection model earns durable margins even as wholesale per-kilowatt rates rise and fall with supply. It is the clearest example of why the three segments cannot be valued on a single FFO or AFFO multiple: a dollar of interconnection revenue is worth more than a dollar of commodity wholesale rent.

The AI build-out has pushed demand decisively toward the wholesale and hyperscale end, because training and large-scale inference need contiguous blocks of high-density power that only those segments can deliver. The pricing data shows the result plainly. Average wholesale asking rates for a 250-to-500 kW requirement rose 6.6% year over year to a record of roughly $196 per kW per month in the second half of 2025, and in Ashburn, Virginia, the largest market, wholesale rates breached $215 per kW per month, the highest on record. Pricing for the largest 3-to-10 MW requirements jumped by 12.5% year over year as tenants competed for scarce contiguous space.

That surge is a direct function of the power constraint: when deliverable capacity is scarce and pre-leasing timelines run three to four years out, landlords gain pricing power even against the largest tenants. Lease rates across the sector have climbed 20% to 35% year over year, a pace unheard of in traditional real estate and a symptom of demand chronically outrunning supply. The broader AI demand backdrop driving all of this is laid out in the data center demand surge overview.

Segment is only half of what sets price; the other half is where the power sits. Constrained, high-demand hubs command multiples of cheap-power markets for the identical product. US wholesale rates run from roughly $120 per kW per month in lower-cost markets like Atlanta to about $250 in Silicon Valley, and international hubs go further still: Singapore, long supply-capped by a government moratorium on new builds, has reached as high as $450 per kW per month. That spread is not a quality difference in the buildings. It is the price of deliverable power on a timeline a tenant can underwrite. In the most constrained metros, colocation has stopped behaving like a real estate product priced on space and started behaving like a power-access product priced on scarcity, which is why a banker now reads a wholesale rate as much as a signal of grid tightness as of building quality.

Segmenting a portfolio is the first analytical step, because each segment implies a different risk and cash-flow profile. Retail revenue is high-margin and diversified but operationally intensive and exposed to short-term churn. Wholesale revenue is contracted and predictable but lower-margin and dependent on a smaller tenant set. Hyperscale revenue is the most bankable in dollar terms but the most concentrated, which makes the hyperscaler tenant credit the assumption the entire model turns on. Lease duration follows the same gradient, running from months at the retail end to two decades at the hyperscale end, which is why these assets price more like long-duration net lease than like traditional commercial real estate, a comparison developed in the discussion of gross, net, and NNN lease structures.

The discipline that ties the three segments together is refusing to read them as one. An operator's earnings quality is a function of its mix, high-margin but churny retail, bankable but concentrated hyperscale, and the wholesale middle, and the AI cycle is actively rewriting that mix beneath every portfolio. Read which of the three businesses an operator actually runs, and how fast the hyperscale end is crowding out the rest, before trusting any single growth or margin number it reports.