Build-to-Suit Data Centers for AI Tenants: What Site Owners Need Before Design

A build-to-suit data center is developed around an identified customer's requirements. For a site owner planning to host an AI operator, the starting point is an agreement about the computing equipment, usable capacity, operating responsibilities and handover conditions. Those decisions turn land and power into a facility brief that engineers can price and develop.

The facility may use prefabricated modular construction. Its commercial structure may involve ownership, a lease or a hosting agreement. Establish each separately. H5 describes a development-and-lease arrangement, while Stream discusses tailored facilities with several ownership and operating options. The term alone leaves important contract details open. H5 Data Centers · Stream Data Centers

For a prospective owner, three questions deserve an answer before detailed design begins.

• What can the site actually supply, and from what date?
• What equipment and service conditions will the tenant require?
• Who will pay for, operate and accept each part of the facility?

An incomplete answer can still support a feasibility study. Record the uncertainty and the decision that will resolve it before ordering affected equipment.

What the owner and tenant are each buying

A compute tenant may own its servers, network and software while the facility owner funds the building and its supporting infrastructure. The tenant then buys an agreed service that keeps that equipment powered, cooled, connected and accessible. A different arrangement may place some IT equipment or operating duties with the owner. The technical brief should identify which arrangement applies.

A facility supplier needs to know the equipment interfaces. An investor needs to know which party has committed to pay for the capacity. An operator needs to know who responds when a pump, power feed or server fails. Assigning those duties early avoids a facility that satisfies a drawing but leaves an operational task unowned.

Use a responsibility table at the first joint meeting.

These are working fields for a project discussion. Contracts, electrical design and local permissions require the appropriate professional review.

For the distinction between the condition of the facility at delivery and its commercial structure, see our powered shell versus turnkey comparison.

Qualify the site beyond its headline megawatts

A site described as having “5 MW available” still needs a defined measurement point. It might refer to a generation rating, a proposed connection, an import limit or supply available only under particular operating conditions. The first-phase IT requirement must be reconciled with usable electrical capacity and the other loads on that supply.

Document the supply location, voltage, capacity, restrictions, protection requirements and date of availability. Ask what evidence supports each field. An application for capacity and an executable connection offer represent different stages of readiness. The project schedule should show the remaining steps and their owners.

At the same time, establish space for external heat rejection, transformers and other plant, delivery and crane access, maintenance clearance, fibre entry, drainage and future works. A module footprint is only part of the site footprint. Read our data center site-selection guide for the broader evaluation.

A useful early site pack contains the following.

An owner with missing documents can begin by marking the fields as provisional. The design team can then target surveys and studies at the unknowns that could change the project.

Obtain a tenant equipment brief before fixing the facility

A GPU count says little about rack geometry, weight, service access or the division of cooling duty. Request the selected system configuration, equipment documentation, planned rack population and support equipment. Record both the initial deployment and the credible expansion case.

For example, networking and storage can add rack positions and air-cooling duty even when the main compute equipment uses direct liquid cooling. Power paths, piping, service clearances and heat rejection must fit the complete installation. NVIDIA's deployment guidance links equipment density to facility constraints, including situations in which cooling limitations require more racks for the same server population. NVIDIA deployment planning

Ask the tenant to identify the inputs it can commit to now.

• Equipment model, configuration, quantity and procurement stage.
• Peak design electrical load, expected operating profile and acceptable power architecture.
• Rack dimensions, operating mass, delivery configuration and service access.
• Liquid and air cooling requirements, including water-side conditions where applicable.
• Network connections, security, access and operational responsibilities.
• Acceptance criteria for the facility interfaces used by its equipment.

When the tenant has several hardware options, keep them as separate cases. Determine which facility choices remain valid across all cases and which depend on the final selection. A provisional equipment list can support an option study while the order waits for a controlled design baseline.

Worked example — separating a 5 MW site ambition from a 1 MW first phase

Consider an illustrative development with land and a stated 5 MW future power opportunity. A prospective tenant wants an initial 1 MW of IT capacity. The owner expects to expand after that tenant commits to a second block.

The figures below are hypothetical planning inputs. The exercise tests what must be established before the owner orders a first phase.

The owner should commission three connected pieces of work.

First, establish the connection basis. Confirm whether 5 MW is import capacity available at a defined point, whether it arrives in stages and what constraints apply. Determine the initial facility demand from an electrical load schedule that includes cooling, power-system losses and auxiliaries. A 1 MW IT requirement alone cannot establish the required connection rating.

Second, prepare the tenant interface schedule. An agreed 1 MW IT allocation still leaves equipment density unresolved. Compare the tenant's candidate configurations before setting rack count, room dimensions, cooling plant and maintenance access. If one option needs a liquid connection and another uses air-cooled servers, preserve that decision explicitly.

Third, draw the first phase and its expansion works together. Establish which shared infrastructure must be installed now, which can be added later and how construction will proceed while the first tenant operates. Include the route for future power cables and pipes, isolation points, access and any temporary arrangements.

This exercise may reveal a sensible first phase even while the ultimate site development remains conditional. It can also reveal that shared upstream works make the proposed first phase impractical. Both results are useful before procurement.

The first phase should have its own release condition. For example, the owner may continue site surveys while the tenant evaluates two hardware configurations, then release cooling equipment only after the selected configuration and water-side requirements are accepted. Record which work can proceed under the provisional brief and which purchase depends on that decision. This keeps the 5 MW ambition from becoming an assumption in a 1 MW order.

Freeze the decisions that affect manufacture

Prefabrication gives the team an opportunity to integrate systems away from the site. Its value depends on knowing the interfaces before fabrication and having the receiving site ready when delivery occurs.

Identify the design releases that allow long-lead procurement, electrical assembly, piping, enclosure manufacture and controls work to begin. Tie each release to approved inputs. The whole project need not wait for every minor choice, but a change to rack geometry or cooling conditions can affect several packages.

A change log records the old assumption, the revised input, affected drawings and equipment, cost and programme consequences, and the person authorized to accept them. Establish who can request changes on the tenant's behalf.

Sequence site works, factory work, tenant equipment delivery and testing on one programme. A supplier's manufacturing duration covers a particular scope. The overall date also depends on site permissions, connection, civil works, logistics, commissioning and equipment readiness. Discuss those dependencies alongside the module proposal.

Make acceptance reflect the tenant's operating conditions

A facility needs an agreed state at which the tenant can install, energize and operate its equipment. These may be different milestones.

Define what will be tested at the factory, after installation and during integrated operation. Specify the relevant load conditions, measurement points, response to agreed failure scenarios and the documents provided. Establish how the tenant's equipment interfaces will be verified if its servers arrive after facility testing.

The handover pack should include approved drawings, test results, operating and maintenance instructions, setpoints, alarm information, asset records and open items. Agree the treatment of those open items before they become a dispute at the delivery date.

Include ownership of routine maintenance and emergency response. A tenant can be responsible for server faults while depending on another party to maintain the facility cooling. The contract and escalation plan should describe how those teams coordinate.

What to send for an initial project discussion

Send the site location, available-power evidence, equipment list or candidate configurations, required initial IT capacity, expansion intention and target dates. Add a simple responsibility table showing the facility owner and the intended compute operator. Those documents let a supplier identify the interfaces that need clarification and prepare an appropriately scoped proposal.

ModulEdge develops prefabricated modular data centers around the selected equipment and site. An owner can discuss the facility configuration and, for suitable opportunities, potential tenant introductions. Tenant commitment, project financing and commercial terms require their own agreements.

Frequently asked questions

Can a modular data center be build-to-suit?

Yes. Build-to-suit describes a facility developed for the customer's requirements, while prefabrication and modular construction describe how parts of that facility are produced and assembled. Scope, ownership and operations still need agreement.

What if the tenant has not selected its GPU systems?

Use a bounded set of candidate configurations for feasibility work. Record the inputs that remain provisional and identify which design releases depend on the hardware choice. Confirm the actual interfaces before ordering affected systems.

Does owning land with power secure a compute tenant?

Tenant commitment depends on the location, service, equipment fit, pricing and commercial terms. An infrastructure proposal and an introduction to a prospective tenant each require further technical and commercial qualification.

Should later phases use exactly the same equipment?

Reserve space and interfaces around a defined expansion case, then revalidate later equipment against it. Future rack, cooling and power requirements may change, so agree who funds and authorizes any redesign.