Colocation vs building your own data center.

Where colocation usually wins

Colocation is the cleaner answer when the business needs resilient power, cooling, security, and connectivity but does not want to develop and operate the physical plant. It converts a facility program into a contracted service and lets internal teams focus on the IT environment.

• The deployment date matters more than owning the real estate.
• The requirement benefits from established carrier and cloud ecosystems.
• The organization wants a defined recurring cost rather than a major construction program.
• The internal team is not staffed to operate critical electrical and mechanical infrastructure around the clock.
• The footprint may change over the next contract cycle.

Where a private build can make sense

Ownership can be rational at very large, stable scale, especially when the company has specialized infrastructure requirements, long-horizon demand, in-house operating expertise, or strategic reasons to control the entire site. The analysis needs to include more than construction cost.

• Land, entitlement, utility, design, construction, commissioning, and financing.
• Generators, UPS, switchgear, cooling, fire protection, controls, physical security, and network entrances.
• Operations staff, maintenance agreements, testing, spare parts, compliance, and lifecycle replacement.
• Schedule risk, stranded capacity risk, and the cost of designing for growth before that growth is certain.

Advantages and disadvantages, side by side

The sections above describe when each model fits. This one lists what each model actually gives up, because a business case that only lists the benefits of the option it prefers is not a business case. Where a point rests on a published figure, the figure is linked to the page that sources it.

Advantages of colocation

• The power is already energized. You are buying capacity that exists, in a building that has cleared its utility queue. In Northern Virginia that queue runs four to six years for a new site.
• No capital outlay. Space and power become an operating expense that can be sized to the requirement and ramped as it grows, instead of a facility paid for on day one.
• The operator carries the plant. Generators, UPS, cooling, fire protection, maintenance contracts, spares, 24 by 7 staffing and the audit evidence behind SOC and ISO reports are theirs to keep current.
• The ecosystem is in the building. Carriers, cloud on-ramps and other tenants are reachable by cross-connect rather than by a fiber build across a county.
• The exit is priced. A term has an end date and a known termination cost. A building you own is sold or written down.

Disadvantages of colocation

• The rent never stops, and it rises. Disclosed wholesale contracts carry annual increases of about 3% to 3.5%, and Digital Realty reported renewal rents up 25.4% on a cash basis in a single quarter. What a renewal increase is made of, and what is negotiable.
• The operator's rules apply. Draw caps, overage charges, access procedures, change windows and cabinet standards are set by the landlord. In one published set of terms, exceeding the draw cap bills at twice the contract rate.
• Growth depends on someone else's capacity. Primary market vacancy ended 2025 at 1.4%, a record low. Expansion rights that are not written into the contract at signature may not exist when you need them.
• Density is bounded by the hall you are in. A hall built for air cooled racks has a sustained ceiling per rack, and moving to liquid cooling can mean moving buildings.
• Every move costs money. Migration, overlap and restoration are real costs that a lease pushes to the end of the term rather than removing.

Advantages of building your own

• Total control. Site, design, density, cooling method, security posture, change windows and who is allowed on the floor are all yours to decide.
• No landlord, no escalator, no renewal. The largest single budget risk on a multiyear lease, the renewal clause, does not exist.
• You keep the premium. A stabilized, leased Northern Virginia asset traded at about $27.1M per MW against a replacement cost of $11M to $14M. An owner who reaches stabilization holds that value; a tenant pays for it.
• Lower cost per kilowatt at high, sustained utilization. Once the capital is sunk, the marginal cost of running a fully loaded building is small. One operator's disclosure puts data center operating expense at about $15 per kW per month.
• A specification nobody sells. Some regulatory, security or engineering requirements are not available in a multi-tenant building at any price.

Disadvantages of building your own

• The capital. A standard build averaged $11.3M per MW in 2026 before land, and an AI-optimized, liquid cooled build runs $15M to $20M or more.
• The wait. Utility interconnection is the long pole, and a build cannot start earning until it is energized. Phoenix measures its queue in months; Northern Virginia measures it in years.
• The politics. Denver stopped accepting data center applications in May 2026 for a year, and neighboring Aurora kept its door open by a single council vote. Water rules, noise buffers and large-load tariffs are being written in real time in most metros.
• The people. A facility needs round the clock operations, maintenance contracts, testing and specialist electrical and mechanical skills that a tenant never has to hire.
• Stranded and obsolete capacity. You pay for the megawatt you built whether you fill it or not, and a hall designed for 8 kW racks cannot host 135 kW racks without substantial rework.
• Escalation during construction. Build costs rose 6% in a single year. A multiyear project is exposed to the whole of that.

The middle ground: dedicated infrastructure inside colocation

The choice is not always cage colocation or a ground-up building. Private suites, dedicated halls, build-to-suit capacity, and wholesale colocation can provide greater control and scale while leaving the base facility with an operator. For many large enterprises, that middle ground captures the useful control without creating a new facility operations business.

Five questions that usually settle the direction

1. How much power is required now, and how certain is the five-to-ten-year growth curve?
2. What is the latest acceptable in-service date?
3. Which facility functions are strategically differentiating, and which are simply necessary?
4. Does the organization already operate critical electrical and mechanical infrastructure at this scale?
5. How should capital flexibility, residual value, and technology change be priced into the decision?

What building actually costs per megawatt

The build side of this comparison is usually the weakest part of a business case, because the capital number is guessed rather than benchmarked. Published figures are available and they have moved sharply.

Standard build average from JLL's 2026 Global Data Center Market Outlook; ranges from industry cost benchmarking. Costs rose from roughly $7M per MW in 2020, driven by electrical density, liquid cooling, redundant power trains and commissioning requirements. Figures exclude land, which in Northern Virginia has run $3.5M to $4.0M per acre.

Ten years, one megawatt: the arithmetic in full

The one-line comparison above is the version that fits on a slide. This is the version that survives a finance review. Every input is a published figure from elsewhere on this site, the assumptions are stated, and the arithmetic is ours, so check it.

Inputs: $11.3M per MW from JLL's 2026 Global Data Center Market Outlook; $156 per kW per month is Digital Realty's Americas base rent on new leases above 1 MW in its second-quarter 2026 results; $15 per kW per month of data center operating expense is from Galaxy Digital's disclosure on its CoreWeave lease; the 3% increase is the escalator disclosed in most of the wholesale contracts on the colocation pricing page. The 8% cost of capital is an assumption, not a source. A level ten year payment at that rate on $11.3M comes to about $1.68M a year.

Read the two ten year rows together and the point of the page falls out. Before financing, building wins on paper by about $5M to $8M. Add the cost of the money and the totals land within a few percent of each other. Land and residual value then pull in opposite directions, and neither is small. What actually decides it is the two lines at the bottom of the table, which no spreadsheet prices well: who holds an asset at the end, and who holds the risk at the start.

The costs that make a build look cheaper than it is

Build business cases tend to compare a capital number against a lease rate and stop there. These are the items that most often move the answer once included.

• Land: excluded from the per-MW figures above. Northern Virginia has traded at $3.5M to $4.0M per acre, and the best powered sites above $8M.
• The interconnect queue: four to six years in Northern Virginia, and some Chicago projects face dates out to 2031 or later. A build you cannot energize is not capacity.
• Cost escalation during construction: build costs rose 6% in a single year and about 47% since 2020. A multiyear project is exposed to that.
• Commissioning and functional testing: scripted testing is one of the four factors cited for cost growth, and it is frequently underbudgeted.
• Staffing: a facility needs 24 by 7 operations, maintenance contracts and specialist skills you do not need as a tenant.
• Refresh and obsolescence: a facility designed for 8 kW racks in 2020 cannot host 135 kW racks today without substantial rework.
• Utilization risk: you pay for 1 MW of built capacity whether you use 300 kW or 900 kW of it. A lease can be sized and ramped.

That last point is usually the largest single factor and the least discussed. Build economics only beat lease economics at high, sustained utilization. Most enterprise builds run well below the utilization assumed in the model that justified them.

What the market itself is telling you

There is a useful external check on this decision. Sophisticated investors buy and sell data centers constantly, and the prices they pay reveal what capacity is worth against what it costs to build.

Transaction detail from Digital Realty's purchase of Blackstone's interest in three Northern Virginia data centers. Full sourcing on the pricing benchmarks page.

A leased, stabilized asset traded at roughly twice its replacement cost. That premium is not paid for concrete. It is paid for energized power and a signed tenant, which are precisely the two things a private build does not have on day one and must acquire at risk. If professional buyers pay double to avoid construction and interconnection risk, an enterprise should think carefully before taking both on to save on paper.

Five conditions that actually favor building

1. You already control energized power. An existing site with capacity and a substation relationship removes the largest single risk.
2. Utilization will be high and sustained. Above roughly 80% of built capacity for most of the asset life, not in year seven of a projection.
3. The requirement is genuinely unusual. A specification no operator will build, for regulatory, security or engineering reasons.
4. Scale justifies the overhead. Several megawatts and enough of them to make dedicated operations staffing rational.
5. Your cost of capital is low and your horizon is long. Build economics are a financing decision as much as a facilities one.

If fewer than three of those are true, the analysis usually favors leasing, and the middle ground of a dedicated suite or cage inside a colocation facility captures most of the control benefit without the capital or the queue.

What each path looks like on a calendar

The cost comparison gets the attention. The date comparison usually decides. These are the stages each path has to pass through before the first server is powered, in order, with the published figure wherever there is one.

The honest summary is that a colocation deployment is limited by how fast you can decide, and a private build is limited by how fast a utility and a planning department can. Only one of those is inside your control.

When the answer is both

The build or lease question is usually asked as if it had one answer for the whole estate. It rarely does. Four patterns come up repeatedly, and each has a published price point behind it.

• Lease now, build later. Take colocation for the requirement that exists, on a term that ends around the date the utility queue clears, and keep the build option open. The lease is the bridge, and expansion rights at signature are what keep the bridge from becoming permanent.
• Own the core, lease the edge. A private facility for the stable, high-utilization base load, with colocation for disaster recovery, for markets you enter, and for growth you cannot yet size. This is what the utilization arithmetic above argues for.
• Buy the shell, run it yourself. A powered shell on a net lease is the priced middle ground. Ionic Digital leases land and finished buildings to Nscale at $65 per kW per month, about a third of a full colocation rate, and the tenant carries the plant, the staff and the risk. You get most of the control of ownership without the interconnect queue.
• Build to suit inside an operator's campus. A dedicated hall or building designed to your specification, delivered and operated by a colocation provider. Control over the design, none over the land, and a term that behaves like wholesale colocation.

For the parallel comparison against public cloud, which is the third option most of these business cases actually contain, see colocation versus cloud. For where the lease side of the case sits against the market, the pricing benchmarks page holds every published figure used above.