Wholesale Colocation: How to Lease Data Center Capacity at Scale, and Get the Right Terms
Wholesale colocation is the lease of dedicated data center capacity, typically 500 kW and above, in a private hall or suite with committed power, a ramp schedule, and a multi-year term. It is priced per kilowatt rather than per cabinet, negotiated rather than quoted, and the contract carries terms that retail colocation never touches: power floors, expansion rights, delivery remedies, and escalation. Consider this your independent wholesale colocation review, written by an advisory practice that takes requirements of this size to operators across North America and internationally, is paid by the operator you choose, and has no interest in which one that is.
Summary: At wholesale scale, the operator negotiates these leases every week and you negotiate one every five years. That gap is where the money goes. A 2 MW lease at market rate versus list rate is a difference of several hundred thousand dollars a year, and the terms around the rate, the power floor, the ramp, the expansion right, the delivery date and what happens if it slips, are worth more than the rate itself over the life of the contract. Having someone on your side of the table who has seen what operators actually sign, and who can put three of them in competition for your requirement, is the difference between a good rate and a market rate.
If you have a wholesale requirement now, tell us what it is. Capacity, density, market, timeline. It comes to the principal directly and you will have a first read within 24 hours on where it can be served and what it should cost.
How we get you a better deal
Five things, and they are the same five every time.
1. We put operators in competition. A requirement that goes to one operator gets one price. The same requirement taken to three or four operators who know they are competing gets a different price and different terms, because the operator’s flexibility appears when the alternative is losing the tenant. Managing that process, so each operator is answering the same specification on the same timeline, is most of what a wholesale placement is.
2. We know what they actually contract at. In the last ninety days the practice has been in front of more than fifteen operators with a live multi-megawatt requirement. What came back is what they will sign, not what they publish. That is the benchmark you negotiate against.
3. We know which halls can serve what you need. A rack-scale AI requirement at 140 kW or above eliminates most of the market before pricing starts. The operators who say they can serve it and the operators who can are different lists, and the gap only shows up when you ask the specific questions: facility water temperature at the rack inlet, flow rate per cabinet, floor loading, when the liquid loop is actually commissioned. Data Center Knowledge used the practice as its source on exactly this in August 2026, from a live search across more than ten operators.
4. We negotiate the terms that matter, not just the rate. Power floor, ramp schedule, expansion rights and how long they hold, the delivery date and the remedy if it slips, escalation, renewal mechanics. Each of those is worth more than a few dollars per kW over a ten-year term. Operators lead with the rate because it is the term buyers focus on and the one that costs them least to concede.
5. It costs you nothing. The operator you choose pays the placement fee as a standard part of its channel program. It is built into their pricing whether you use an advisor or not. Going direct does not save it; it leaves it with the operator.
If that is the help you need, send us the requirement. We work across North America and internationally, and the first conversation is where it can be served and what it should cost.
What is wholesale colocation, and how is it different from retail?
Wholesale colocation is dedicated capacity leased by the kilowatt. The tenant takes a private hall, suite or cage, commits to a power level, and pays a monthly rate per committed kilowatt plus metered consumption. Retail colocation is shared space leased by the cabinet, with power bundled into a per-rack price and the operator managing the environment around you.
The dividing line is usually around 500 kW, though it moves. Below that, retail. Above about 1 MW, almost always wholesale. In between, either structure can work and the choice depends on how much control you want and how long you are committing. When a buyer searches for data center space for lease or a data center for lease, this is the market they are in, and the data center wholesale market is where the capacity, the pricing and the contract structure all differ from what the retail buyer sees.
| Retail colocation | Wholesale colocation | |
|---|---|---|
| Typical size | 1 to 50 cabinets, up to a few hundred kW | 500 kW to 20 MW and beyond |
| How it is priced | Per cabinet, power bundled or metered | Per committed kW, plus metered consumption at cost |
| What you get | Cabinets in a shared hall | A private hall, suite or cage with dedicated infrastructure |
| Term | One to three years | Five to fifteen years |
| How it is negotiated | Quoted from a rate card | Negotiated, every term, every time |
| Who is on the other side | A sales rep | A deal team and legal |
| What matters most | Rate, connectivity, location | Power floor, ramp, expansion rights, delivery, escalation, then rate |
| Fit | Standard enterprise workloads | Enterprise colocation at scale, AI clusters, service providers, cloud repatriation |
The practical difference is that a retail buyer compares prices and a wholesale buyer negotiates a contract. The rate is one line in that contract. The rest of it is where operators build margin, and it is where a tenant who has not done this before gives it up without noticing.
What does a wholesale data center lease have to say?
Five terms carry most of the money over a multi-year lease. The rate is not one of them.
1. Power floor. The minimum kW you pay for regardless of use. Operators want it at contracted capacity from day one; tenants want it to track the ramp. On a 5 MW lease, a floor at full capacity from month one when the ramp takes a year is millions in unused power.
2. Ramp schedule. How capacity steps up, on what dates, and what happens if the tenant is early or late. Tied to the floor, and the term where phased AI deployments most often go wrong.
3. Expansion rights. Right of first refusal or first offer on adjacent capacity, for how long, at what price. Operators now bound these tightly, two to three years and often conditioned on utility confirmation, because power is scarce. A tenant planning to grow needs the right to be real, dated and priced.
4. Delivery date and remedy. When the capacity is commissioned, and what the tenant gets if it is late: rent abatement, liquidated damages, a termination right at an outside date. Operators write “commercially reasonable efforts.” Tenants need a date.
5. Escalation. Annual increases, typically 3 to 4 percent fixed or CPI-linked. Over ten years, the difference between 2 and 4 percent is roughly 20 percent of total cost. Negotiate it with the rate, never separately.
The data center lease guide covers each of these in detail, including what operators currently accept and what they hold.
Why does the wholesale data center market favor the operator?
Three things tilt the table.
They do this constantly and you do not.
An operator’s deal team negotiates wholesale leases every week. They know current market rates in every hall they run, which terms buyers push on and which they accept, and exactly where their own flexibility sits. A tenant negotiates a wholesale lease once every five to ten years, usually with a team that has never done it before.
Power scarcity has flipped the leverage.
In most primary markets, available capacity is spoken for years ahead. Operators negotiate from strength, and terms that were negotiable in 2022, power floors, ramp schedules, take-or-pay, are now standard in the operator’s first draft. A buyer who does not know which of those can still move accepts all of them.
Published data does not help you.
CBRE’s H2 2025 survey put average primary-market wholesale pricing at $196 per kW per month for 250 to 500 kW deployments, up 6.6 percent year on year, with 3 to 10 MW requirements up 12.5 percent. That is a useful benchmark and it is also the number you will never be quoted, because published averages describe closed deals and your quote reflects what the operator thinks you will accept. The only way to know what a specific hall is contracting at right now is to have been in front of the operator with a live requirement recently.
That last one is the whole argument for an independent advisor at this scale. Not because a broker is cheaper than going direct, since the operator pays either way, but because the information you need to negotiate well is not available to you and is available to someone who does this for a living.
What does wholesale data center capacity cost per kW?
Wholesale pricing is quoted per committed kilowatt per month, and data center cost per kW depends on market, density, term, and how much capacity the operator has left. Rough ranges for 2026:
| Tier | Range, $/kW/month, net of metered power | Basis |
|---|---|---|
| Wholesale, 1 to 10 MW, existing capacity | $150 to $220 | Colliers 2026: primary US markets transact $160 to $200 net of electrical. McKinsey June 2026: $150 to $200. |
| Wholesale, 250 kW to 1 MW | $190 to $240 | CBRE H2 2025: $196 average for 250 to 500 kW in primary North American markets, up 6.6 percent, expected to exceed $200 in 2026. |
| Hyperscale, 10 MW and up, long term | Typically below the wholesale band | Negotiated individually; CBRE reports 3 to 10 MW pricing rose 12.5 percent year on year on scarcity. |
| Liquid-cooled, rack-scale AI | Premium over the wholesale band | CBRE H2 2025: operators with AI-optimized facilities capture rent premiums over conventional space. Limited supply, often pre-leased with delivery past twelve months. |
Three things move the number.
- Market: Silicon Valley is the most expensive primary market and volume discounts there have been reduced or eliminated; Atlanta is the most affordable; Northern Virginia sits below the primary-market average despite being the largest, because of its scale. Canada often prices below the US on power cost.
- Density: liquid-cooled capacity carries a premium because there is little of it.
- Term: a ten-year commitment gets a better rate than five, and a take-or-pay structure gets better again, but both cost you flexibility that has a value of its own.
Vacancy in primary North American markets reached a record low of 1.4 percent at the end of 2025. That is why published averages are rising and why the quote you receive reflects scarcity as much as cost.
The rate is also not the whole cost. Metered power is passed through at the operator’s cost, and that is typically 25 to 40 percent of the total bill at a fully utilized deployment. Cross-connects, remote hands and any facility-supplied equipment sit on top. A quote that leads with a low per-kW rate and a high metered rate is a common structure, and it is one reason the comparison between operators has to be done on total cost rather than headline.
Who are the wholesale data center providers?
Wholesale colocation providers fall into four kinds, and they behave differently.
The large operators with wholesale and hyperscale colocation halls.
Equinix, Digital Realty, and to a lesser extent CoreSite and DataBank. They run retail and wholesale in the same campuses, have the most mature channel programs, and are the most disciplined negotiators. Their advantage is interconnection density; their limit is that wholesale capacity in primary markets is largely committed.
The pure wholesale developers.
Vantage, Aligned, QTS, Stack, CyrusOne, Compass, Sabey. Built for multi-megawatt tenants, priced per kW from the start, and more likely to have capacity in secondary markets. They negotiate on volume and term and are the natural home for a 2 to 20 MW requirement that does not need to be inside a carrier hotel.
The regional and specialist operators.
Flexential, TierPoint, Cologix, Evocative, Colovore, plus market-specific names in every city. Smaller campuses, often more flexible on structure, and sometimes the only operators with liquid-cooled capacity available inside a year. This is where the AI requirement usually lands.
Developers with powered land.
Not operators yet, but a growing source of wholesale supply: a developer with a site, a utility commitment and a design, looking for an anchor tenant before construction starts. That is a build to suit or powered shell conversation rather than a lease of existing capacity, and it is the route to a purpose-built facility when nothing existing fits.
The practice holds channel relationships across the first three categories and works directly with the fourth. Which category fits a requirement depends on density, timeline and market, and that is the first thing the placement establishes.
How do you run a data center RFP for wholesale capacity?
A wholesale RFP is a specification plus a process. Both have to be right.
The specification.
Committed capacity at go-live and at each ramp step. Density per rack and total. Cooling: air, liquid, or both, and if liquid, facility water temperature, flow rate per cabinet, and whether CDUs ship with the racks. Redundancy: N+1, 2N, and on which systems. Connectivity: carriers required, cloud on-ramps, cross-connect volume. Term, with expansion rights and how long they need to hold. Timeline, stated as a commissioning date and an outside date. Compliance: SOC 2, PCI, HIPAA, FedRAMP, whatever applies.
The process.
Three to five operators, chosen because they can plausibly serve the requirement rather than because they are well known. Same specification to all of them on the same day. A written question-and-answer window so everyone gets the same clarifications. A response format that forces comparability: rate per kW, metered rate, one-time charges, and each contract term answered yes or no. Two weeks to respond for existing capacity, longer if construction is involved.
What goes wrong.
Sending the RFP to operators who cannot serve the density, which wastes a month. Accepting each operator's own proposal format, which makes them impossible to compare. Not stating the outside date, which lets an operator win on a commissioning date they will not hit. And negotiating the rate before the terms, which is the operator's preferred sequence and the wrong one. Running this well is most of what the placement engagement is. The specification comes from the requirement, the operator list comes from knowing who can actually serve it, and the comparison comes from forcing everyone into the same format.
Wholesale colocation for AI and high density
Rack-scale AI is the fastest-growing wholesale requirement and the hardest to serve. A GB300 NVL72 rack draws 140 to 160 kW and needs direct-to-chip liquid cooling; the Rubin generation behind it draws more. Most halls marketed as liquid-cooling-ready cannot take it on the timeline a buyer needs, because the constraint is not the rack, it is the facility loop, the CDUs and the commissioning date.
For a wholesale AI requirement the questions are specific: is the facility water loop live, what temperature does it deliver at the rack inlet, what flow rate per cabinet, does the operator supply CDUs or does the tenant, what does the floor carry, and when is the liquid-cooled cage actually deliverable as opposed to when the building opens. The AI and GPU colocation page covers the facility requirements in detail, and the practice has taken this exact requirement to the market at 5 to 20 MW.
Where is data center space for lease at wholesale scale?
Data center space for lease at wholesale scale concentrates in a handful of markets, and the primary ones have the most supply and the least available. Northern Virginia remains the largest wholesale market in the world and the tightest. Dallas, Chicago, Atlanta and Phoenix have capacity coming online through 2027. The New York metro is constrained by power and a permitting pause on facilities above 50 MW.
Outside the US, Canada has become a serious wholesale market on hydro power and sovereign-AI demand, with Quebec, British Columbia and the Greater Toronto Area all building at scale. The practice works across North America and internationally; where a requirement can be served depends on density, timeline and where the tenant can operate.
What this page cannot tell you
It cannot tell you which operators have capacity for your requirement on your timeline, because that changes monthly and depends on density, market and how much power the operator has left. It cannot tell you what a specific hall will contract at, because that is negotiated, not published. It cannot tell you which of the five lease terms the operator in front of you will concede, because that depends on how full they are and how badly they want the tenant. And it cannot run the RFP.
Those are the placement. This page is the framework it uses.
How the practice works on wholesale
Metro Colo Advisory is an independent data center advisory practice. There are two sides to it and they are paid differently.
- Placement. You give us the requirement. We take it to the operators who can plausibly serve it, across North America and internationally, run the RFP so the responses are comparable, and negotiate the terms before the rate. The operator you choose pays us, as a standard part of its channel program. You pay nothing. If nothing fits, we tell you that.
- Consulting, when the requirement is not yet defined. Sometimes the question is earlier: whether to move off cloud at all, what the right footprint is, or how the P&L changes under different structures. That is the consulting side, paid by you at a fixed fee, and it produces the specification the placement then runs. The cloud repatriation guide covers the decision that usually precedes a wholesale requirement.
The two are separate. The consulting fee does not depend on the placement, and the placement does not depend on having done the consulting.
Frequently Asked Questions
What is the minimum size for wholesale colocation?
Around 500 kW, though it depends on the operator. Below that, most operators sell retail by the cabinet. Between 500 kW and 1 MW, either structure is possible and the choice comes down to whether you want a private hall and per-kW pricing. Above 1 MW, wholesale is standard. Some operators will structure a 250 kW private suite as wholesale if the term is long enough.
How is wholesale colocation priced?
Per committed kilowatt per month, plus metered power at the operator’s cost, plus one-time charges for installation and cross-connects. The per-kW rate covers space, cooling, redundancy and the facility. Typical 2026 ranges run from $100 per kW for hyperscale commitments to $300 or more for liquid-cooled AI capacity. Total cost includes metered power, which is often a quarter to 40 percent of the bill.
What is the difference between wholesale colocation and a data center lease?
Very little in practice. Wholesale colocation is the industry’s term for leasing dedicated capacity by the kilowatt; the document you sign is a data center lease or a master services agreement with a capacity schedule. The terms that matter, power floor, ramp, expansion rights, delivery and escalation, are the same either way.
How long does it take to lease wholesale capacity?
For existing capacity, eight to sixteen weeks from RFP to signature, with commissioning following in weeks to a few months. For capacity under construction, the lease can be signed a year or more before delivery. For liquid-cooled AI capacity, availability is the constraint and timelines are set by when the facility loop is commissioned, not by the negotiation.
Can wholesale capacity be split across two sites?
For most workloads, yes. For a single AI training cluster, no, because the GPUs have to be within a fabric that does not cross buildings. Steady inference, storage and general compute can split. Whether a requirement can be phased across sites is one of the first questions to settle, because it changes which operators are in play.
Do I need a data center broker for a wholesale lease?
You do not need one to sign a lease. You need one if you want to know what the operator would actually accept. A broker who has taken requirements of this size to operators recently knows current contracted rates, which terms move, and which halls can serve the density. The operator pays the broker, so the cost to you is nothing and the alternative is negotiating blind.
What is take-or-pay in a data center lease?
A structure where the tenant pays for committed capacity whether it is used or not, from the commencement date. It has become standard for AI anchor leases because operators are financing construction against the commitment. The negotiable parts are the ramp, so take-or-pay tracks deployment rather than starting at full capacity, and the outside date, so a late delivery does not trigger payment for capacity that does not exist.
Does wholesale colocation work for cloud repatriation?
At scale, it is the destination. A company moving seven figures of steady workload off cloud usually lands in a wholesale lease at several hundred kilowatts to several megawatts, often after a first wave on bare metal. The cloud repatriation guide covers the decision; this page covers the lease that follows it.
Related reading
- The lease and the contract: the data center lease guide for the five terms in detail, and the colocation pricing guide for what facilities charge at every scale.
- The decision before the lease: the cloud repatriation guide for whether to move off cloud, and data center consulting for when the requirement itself needs defining.
- High density and AI: the AI and GPU colocation page for what a facility has to have, and the site selection guide for the criteria that apply at any scale.
- Building instead of leasing: the powered shell and build to suit pages for developers and for tenants who need a purpose-built facility.
- Markets: Northern Virginia, Dallas, Chicago, Atlanta, and the New York metro. Or compare all providers. For how the placement side works, see how it works and why companies use an independent advisor.
Send us the requirement
Capacity at go-live and at full ramp. Density. Cooling. Market or markets. Commissioning date and outside date. That is enough for a first read on where it can be served and what it should cost, and you will have it within 24 hours.
North American and international coverage. Paid by the operator you choose. If nothing fits, we say so.
