Colocation Lease
Definition
A colocation lease, also called a colocation agreement or colo contract, is a contract under which a customer places its own servers, storage, networking or GPU equipment inside a data center owned and operated by someone else, and pays that operator for the space, the electrical power delivered to the equipment, the cooling that removes its heat, physical security and network connectivity. The tenant owns and manages the hardware; the operator owns and runs the building. Pricing is quoted per kilowatt of committed power per month, commonly $100 to $200 per kW for standard retail colocation and $130 to $250 per kW for high-density, liquid-cooled AI capacity, plus metered electricity, cross-connect fees and installation charges. Terms run one to three years for retail cabinets and cages and five to fifteen years for wholesale suites and halls, with ramp schedules, expansion rights, uptime service levels backed by credits, and, for tenants financing their equipment, the landlord waivers and access rights their lenders require.
Why it matters
The colocation lease is the contract on which most AI infrastructure investing rests. The data center REIT or developer earns its revenue from it; the GPU cloud tenant's ability to deliver compute depends on it; the lender to that tenant can only reach its collateral through it; and the ABS or CMBS that finances the building is underwritten on its term, its tenant and its termination rights. When a GPU cloud announces a multi-billion-dollar capacity deal, the number is usually the sum of colocation lease commitments over their term. Reading the lease tells an investor whether that number is contracted, ramped, cancellable or aspirational.
Common misconceptions
- •Colocation is not cloud. The tenant owns the hardware and bears its obsolescence; the operator sells power, cooling and space, and its revenue does not fall when the tenant's GPUs age.
- •Square footage is not capacity. A facility can be full on power or cooling with half its floor empty; AI tenants buy kilowatts and cooling, and a lease that quotes space without density is not a modern one.
- •A long lease is not automatically good collateral. Ramp schedules, delivery contingencies, broad termination rights, weak guarantors and uncapped SLA credits can make a ten-year lease worth far less than its face.
- •Electricity pass-through is not margin. Operators report power reimbursement as revenue; an investor normalizing a data center's income separates the per-kW infrastructure charge from the metered electricity, which carries little or no margin.
Technical details
Worked example: a 2 MW AI colocation lease
Tenant: a GPU cloud provider committing 2 MW of liquid-cooled capacity across 20 racks at 100 kW each, for five years, at $160 per kW per month, with metered electricity passed through at cost times a PUE of 1.25.
Base charge: 2,000 kW times $160 = $320,000 a month, $3.84 million a year, $19.2 million over the term. Ramp: 25 percent of capacity billable in month one, 100 percent by month seven as GPUs are delivered, so year-one billings are about $2.9 million.
Electricity: at 85 percent average utilization the racks draw about 1.7 MW; with PUE 1.25 the facility bills about 2.1 MW of consumption, roughly 1.55 million kWh a month, at $0.07 per kWh about $109,000 a month passed through.
Other charges: $250,000 of installation and cabling, six cross-connects at $300 each per month, remote hands at $150 an hour.
Lender terms: the tenant's GPU lender obtains a landlord waiver giving it 60 days of access after a tenant default to remove the equipment, with the tenant's rent for that period paid by the lender.
The clauses that carry the economics
Committed power: the kW the tenant pays for whether or not it draws them; the facility's revenue base and the tenant's take-or-pay obligation.
Ramp and delivery: when capacity is delivered and when billing begins; delay remedies for the tenant if the operator cannot energize on time, which is the most common failure in the current build cycle.
Density and cooling specification: kW per rack and the cooling method (air, rear-door, direct-to-chip liquid); who pays to upgrade if the tenant's next GPU generation needs more.
Electricity: fixed, metered at cost plus PUE, or metered at a fixed rate; who carries utility price volatility and demand charges.
Term, renewal and expansion: the initial term, renewal options with rent resets to market or a fixed escalator (commonly 2 to 3 percent a year), and rights of first offer or refusal on adjacent capacity.
SLAs and credits: availability targets, credit schedules, caps, and the chronic-failure termination right.
Assignment and change of control: whether the tenant can assign to an affiliate, an acquirer or a lender without consent; whether the operator can sell the facility subject to the lease.
Retail, wholesale and powered shell compared
Retail: cabinets to small suites; 1 to 500 kW; one to three years; operator runs power, cooling and often hands-on services; $100 to $250 per kW.
Wholesale: data halls or buildings; 1 to 100+ MW; five to fifteen years; tenant fits out and operates the interior; $90 to $160 per kW for air-cooled, more for liquid-cooled; the market hyperscalers and large AI clouds transact in.
Powered shell: the building, the utility feed and often the generators and chillers; the tenant installs electrical distribution, cooling loops and everything inside; longest terms, lowest per-kW rent, highest tenant capital.
Build-to-suit: the operator constructs a facility to the tenant's specification under a pre-lease, financed against the lease; the structure behind most 2024 to 2026 hyperscale announcements.
What the tenant's lender needs from the lease
A landlord waiver or collateral access agreement: the operator subordinates any lien on the tenant's equipment, agrees to give the lender notice of tenant default, and grants a period (30 to 90 days) to enter and remove the collateral, usually against payment of rent for that period.
Assignment on foreclosure: the right to have a buyer of the repossessed equipment step into the tenant's space, or at least to leave the equipment in place while it is sold.
Confirmation of the lease's term and the absence of defaults (an estoppel), so the lender knows the equipment has a home for the loan's life.
Without these, a lender's security interest in GPUs inside a third-party facility may be unenforceable in practice, which is why GPU-backed loans price the landlord waiver into their conditions precedent.
Reading a colocation lease as an investor
Find the committed kW, the price per kW and the electricity mechanism; rebuild the monthly bill from them and compare with reported revenue.
Find the ramp and delivery dates and the remedies for delay; the current shortage is in delivered power, not signed contracts.
Find the density and cooling specification and who funds upgrades; a five-year lease at 40 kW per rack does not serve the GPU generation after next.
Find termination rights, SLA caps and the chronic-failure clause; these define the downside of the operator's revenue.
Find the tenant's credit and guarantor, and whether the tenant's own customers are contracted; a GPU cloud's colocation commitment is only as good as its compute contracts.
Frequently asked questions
- What is a colocation agreement?
- A contract to house your own computing equipment in someone else's data center. The operator provides a secured space (a cabinet, a cage or a whole suite), a committed amount of electrical power, the cooling to match, and connections to carriers and other tenants; you provide and operate the servers. You pay for the power you commit to, the electricity you use, and the connections you order, for a fixed term.
- How is a colocation lease priced?
- By power, not floor space. The core charge is a monthly rate per kilowatt of committed capacity, with electricity either included at a fixed rate, metered and passed through at cost plus the facility's overhead (PUE), or metered at a fixed price per kWh. On top come one-time installation fees, monthly cross-connect fees per connection, remote hands billed hourly, and charges for additional space or security. A 10 kW cabinet at $150 per kW is $1,500 a month before electricity; a 5 MW wholesale hall at $130 per kW is $650,000 a month.
- What is the difference between retail and wholesale colocation?
- Scale and who runs what. Retail colocation sells cabinets, cages and small suites (a few kW to a few hundred kW) on one-to-three-year terms with the operator providing power, cooling and often managed services inside the space. Wholesale colocation leases whole data halls or buildings (1 MW and up, often 10 to 100 MW for AI tenants) on five-to-fifteen-year terms, with the tenant fitting out and running the interior. A powered shell goes further: the tenant leases a building with power and cooling infrastructure and installs everything else.
- What SLAs does a colocation lease include?
- Power availability (commonly 99.99 or 99.999 percent), temperature and humidity ranges at the cabinet inlet, cooling capacity per rack, network availability for the operator's own connections, and response times for remote hands and security. Breaches are remedied by service credits, typically a percentage of the monthly fee scaled to the length of the outage and capped at the monthly charge; chronic failure (several breaches in a period) usually gives the tenant a termination right. Credits rarely cover a tenant's actual losses.
- What do GPU tenants need in a colocation lease that others do not?
- Density and cooling. A standard cabinet draws 5 to 10 kW; a rack of current AI accelerators draws 40 to 130 kW and needs direct liquid cooling, which most facilities were not built for. AI tenants negotiate committed density per rack, liquid cooling specifications and who pays for the retrofit, low-latency cross-connects to their cloud or storage, fast delivery schedules (capacity in months, not years), ramp schedules that match GPU deliveries, and expansion rights on adjacent capacity. Their lenders add landlord waivers and access rights so the GPUs can be repossessed.
- What is colocation financing?
- Two different things. For the operator, financing the facility: construction loans, ABS and CMBS secured by the building and its leases, where the lease's term, tenant credit and termination rights determine what can be borrowed. For the tenant, financing the equipment inside: GPU-backed loans and leases whose collateral sits in a building the lender does not control, which is why those lenders require a landlord waiver from the colocation operator before funding.
- Can a colocation tenant break the lease?
- Usually only by paying. Retail agreements typically allow termination for convenience only at renewal, with early termination fees equal to most or all remaining fees; wholesale leases are effectively non-cancellable for their term, which is what makes them financeable. Tenants get out through the chronic-SLA-failure right, through assignment to another customer with the operator's consent, or by negotiating a buyout. Operators sometimes allow termination when they cannot deliver committed power on time.