Data center construction and buildout phases field guide

What a data center build is, and why it runs different

A data center build is a fast, MEP-heavy construction project that has to deliver a working power and cooling plant, proven under fault, on a schedule the owner set before the design was finished. The building shell is the easy part. Most of the cost, most of the risk, and most of the time live in the electrical and mechanical systems and the controls that tie them together, plus the proof that all of it works before a single server is at risk.

The lifecycle runs in a recognizable order: site selection and the utility power agreement, design, long-lead procurement, site work, the shell and core, the MEP fit-out, the white-space fit-out, and commissioning, ending at turnover and the first live IT load. The phases overlap hard on a real job. Procurement starts during design, commissioning starts at the factory, and one data hall energizes while the next is still steel.

What sets the job apart is the speed-to-market pressure laid over a plant that cannot be allowed to fail. The owner is racing to fill the building with revenue-generating compute, so the schedule is aggressive from day one, and the schedule keeps slamming into the lead time of the power gear and the hours the integrated test actually needs. This guide walks the phases. The white space and gray space guide covers how the building zones, and the data center commissioning levels guide covers how the plant is proven.

How data center projects get delivered

Three delivery models cover most of the work, and they differ in who carries the design risk and how early construction can start. Design-bid-build keeps design and construction in separate contracts: the owner finishes the documents, then bids the build. It gives the most owner control over the design and the least schedule speed, because the build cannot start until the design is done. On a market where the equipment lead time already sets the clock, that gap is expensive.

Design-build puts design and construction under one contract, so the two overlap and the build can break ground while the later packages are still being drawn. It is the common choice on fast-track and hyperscale work because it compresses the schedule and gives one party accountability for both halves. EPC and EPCM go further toward a single engineer-procure-construct scope, with EPCM keeping the owner in direct contracts with the trades while a manager runs the program.

Layer the owner model on top. An enterprise owner builds for its own use. A colocation provider builds a multi-tenant building and leases cages and suites. A hyperscaler builds campuses of 20 to 100-plus MW for its own cloud or AI platform, often with a repeatable baseline design rolled out phase after phase. The model shapes the team, the standardization, and how much of the design is frozen before the first phase even starts. Confirm the delivery method and the contract structure before you assume who owns a coordination gap, because that is exactly where the disputes land.

The front end: program, basis of design, and the power agreement

Preconstruction is where the project is won or lost, long before a shovel moves. The owner sets the program: how many megawatts of IT load, what redundancy, what density per rack, what the building has to do. That program drives everything downstream, so a vague program produces a building that fits nobody. The owner's project requirements, the OPR, state it in measurable terms, and the basis of design, the BOD, is the design team's documented answer. The commissioning levels guide covers the OPR and BOD in depth, because the whole verification chain traces back to them.

Site selection happens here, and so does the entitlement work: zoning, environmental review, and the local approvals that on a contested site can run 6 to 18 months on their own. None of that is fast, and none of it is the part owners want to hear about when they are chasing a go-live date.

The one front-end item that quietly governs the whole schedule is the utility power agreement. A data center needs tens to hundreds of megawatts, and the utility has to commit to deliver it. The interconnection study, the substation, and the transmission upgrades can take years, and on the current grid that timeline often beats the construction timeline. Sign the build before the power is committed and you have built a dark box. Get the power agreement locked, then schedule the building around when the megawatts actually arrive, not the other way around.

What picks the site?

Power availability picks the site, and in the current market it is not close. A site without a credible path to the megawatts the program needs is not a site, no matter how cheap the land or how good the tax deal. Developers now lead with power: find where the grid can deliver, then build the rest of the decision around it. In one owner survey, only about 3 percent ranked tax credits as the top factor. Power, land, and latency dominate.

After power, fiber and connectivity decide whether the building can actually serve traffic, which means diverse routes from more than one carrier so a single cut does not isolate the campus. Water for cooling matters where the cooling design uses it, and water access and water politics have killed deals. Land and climate set the cost: a cooler, drier climate cuts cooling energy, and flat, stable, low-disaster ground cuts the structure and the risk. Latency to the users the building serves sets how close it has to be to population. Tax and incentives are real money but rarely override the constraints above them.

The trap is treating site selection as a real-estate decision. It is a power and infrastructure decision wearing a real-estate costume, and the order of the drivers is the part the spreadsheet gets wrong.

The design phases and the redundancy target

Design moves through staged sets, each one adding detail and each one a place to catch a problem while it is still cheap to change. Conceptual and schematic design fix the big moves: the site plan, the building massing, the topology, the redundancy target. Design development sharpens the MEP systems, the room layouts, the equipment selections, and the clearances. Construction documents are the buildable set, the drawings and specifications the trades price and install from.

The redundancy and Tier target is the decision that sizes the whole plant. An N design has no spare. N+1 carries one extra component. 2N runs two full independent systems. Each step up multiplies the electrical and mechanical gear, which multiplies the gray space and the budget, so the redundancy call is a cost call as much as a reliability call. The white space and gray space guide covers how that redundancy reshapes the support footprint.

Two disciplines own the design here and have to stay coordinated: electrical for the power chain and mechanical for the cooling, with the controls layer tying them together. The expensive design failure is a clash nobody caught on paper, where the busway, the piping, the tray, and the duct all want the same overhead, and the crew finds it in the field. Catch the clash in a coordinated model during design, not with a grinder during the fit-out. Standards like BICSI-002 and TIA-942 commonly get written into the basis of design so substitutions later have a reference to check against.

Long-lead equipment is the real schedule

The long-lead equipment, not the building, sets the critical path on a data center, and that has gotten worse, not better. The generators, the medium and low-voltage switchgear, the UPS, the transformers, and the chillers all carry order books that now stretch well past the old build cycle. Order them late and the building sits finished and dark, waiting on a transformer.

The numbers move, so treat any lead time as a quote-and-verify figure, not a constant. As of 2026, large generators, switchgear, and UPS commonly run on the order of 12 to 18 months, and large power transformers have stretched much further, in some cases past three years. The binding bottleneck across the industry has been the electrical gear, the transformers and switchgear, more than the building or even the chips. That is why owners now place orders or supply agreements years ahead of when the gear is actually needed.

The practical move is to start the long-lead tracker during schematic design and order the critical gear off an early-release package, before the construction documents are even complete. The estimator who waits for a clean, fully coordinated design to place the transformer order has already lost the schedule. Procurement is a design-phase activity on this kind of job, and the owner who understands that is the owner whose building goes live on time.

Shell and core, warm shell and cold shell

The shell and core is the first physical phase you can see: the structure, the envelope, and the slab that make a weathertight building. Compared to what follows, it is fast and well understood. Earthwork, foundations, steel or tilt-up concrete, the roof, and the slab go up on a schedule any commercial builder would recognize. The complexity, and the schedule risk, arrives with the MEP after the box is closed in.

The trade splits the box into states of readiness, and the words matter on a colo or developer deal. A cold shell, sometimes called a cold dark shell, is the bare enclosed structure: foundation, frame, walls, roof, weathertight, with no power or cooling distributed and often no finishes. A warm shell carries the core mechanical and electrical infrastructure already installed, the switchgear, UPS, generators, and cooling, so the data halls can be fit out and energized far faster.

A powered shell is a related developer product: the building and the incoming power are in place, and a tenant or buyer finishes the fit-out to their own design. Which state the building is delivered in changes the schedule and the contract scope completely, so pin down what cold, warm, and powered actually include on your job. The same word means different things across two owners, and the gap between them is months of MEP work.

Site work, the utility entrance, and the substation

The civil and site work runs alongside and ahead of the building, and on a campus it is a project in its own right. Grading and earthwork set the pads, the access roads, and the drainage. Underground utilities go in: the power duct banks, the water and sewer, the fiber entrance, and the storm system. Get the underground sequence wrong and you are trenching across finished paving later, which is pure rework.

The power entrance is the heavy item. A data center at this scale usually takes a dedicated substation, owned by the utility or by the project, stepping medium or high voltage down to what the building distributes. The substation, the transmission tie, and the interconnection are often the longest pole on the whole job, which is why the power agreement in the front end matters as much as it does. The generator yard is the other big civil and electrical area: the standby generators, their fuel storage, the day tanks, and the paralleling gear, sized to carry the building through a utility loss.

Coordinate the yard, the substation, and the duct banks early, because they feed the gray-space electrical rooms and they are hard to move once they are poured. The white space and gray space guide covers how those electrical rooms lay out inside the building. The site point is that the power has to physically arrive, and the path it takes through the yard and the duct banks is set in the dirt before the gear ever shows up.

The MEP fit-out and rough-in

The MEP fit-out is the heart of the build and where the hours go. On the electrical side, the crews set and terminate the switchgear, the UPS and batteries, the transformers, the floor distribution, and the busway that carries protected power out to the rows. On the mechanical side, they set the chillers, the pumps, the air handlers or CRAH units, and run the chilled-water piping and the ductwork. The controls and the building management system get wired in to tie the two plants together.

This is the rough-in that makes or breaks the schedule, and coordination is the whole game. The power, the piping, the tray, and the duct all compete for the same overhead and the same wall space, and a clash that was missed in the model becomes a field stoppage while two trades argue over a chase. The crews that run a coordinated model and a clear sequence pull ahead here. The ones that field-coordinate fall behind and never catch up.

The deep technical content of the power chain, the protection, the transfer scheme, and the cooling, sits in the power and cooling material by topic. The phase point is that the MEP fit-out is the longest, most labor-dense stretch of the job, it runs in parallel with procurement deliveries and early commissioning, and it is where most of the budget physically becomes the building.

The white-space fit-out and the cabling

The white-space fit-out is the data hall coming together: the raised access floor or the slab finish, the containment, the cabinets and racks, the in-row or perimeter cooling, and the power and cooling landing at each rack position. It usually follows the gray-space MEP, because the hall is fed by the plant behind it, and on a phased job it is done hall by hall as each one is needed. The white space and gray space guide covers the hall layout, the hot-aisle and cold-aisle arrangement, and the raised-floor-versus-slab decision in full.

The structured cabling and low-voltage work goes in here and threads through everything: the fiber and copper, the pathways and tray, the patch fields, and the labeling that ties a cable to a rack to a port. It is governed by the cabling standards, commonly TIA-942 for the spaces and TIA-606 for the administration and labeling, so the grid coordinate, the rack ID, and the cable labels all agree on one scheme. Set that convention before the first label prints.

The discipline that separates a clean hall from a scavenger hunt is holding everything to the floor grid and the coordinate, not a tape off the wall. Walls are rarely square, and the drift down a long row pushes the last cabinet off its taps and its tile cuts. A hall where the coordinate is consistent from the floor tile to the patch panel is a hall the operator can manage. One where every trade invented its own is a permanent problem.

Building in phases, halls, and modules

Few data centers get built all at once. The capital and the demand both argue for phasing, so the building goes up in increments: hall by hall, pod by pod, or building by building across a campus. The owner energizes and fills one phase, starts earning revenue on it, and builds the next while the first is live. That phased-capacity model is how a developer matches spend to lease-up and how a hyperscaler rolls out a frozen baseline design over and over.

Modular and prefabricated construction pushes the idea further. Skidded power modules, prefabricated electrical rooms, and factory-built cooling units arrive at the site largely assembled and tested, which moves work off the critical path and into a controlled shop. It speeds deployment, it standardizes quality, and it suits the repeatable hyperscale rollout. The cost is that the design has to be frozen and standardized early, because a module is hard to change once the line is running.

Energizing one hall while building the next is the part that takes discipline. You now have a live, load-carrying facility sharing a site, and sometimes shared infrastructure, with an active construction zone. The work on the new phase cannot put the live phase at risk, which means the boundaries, the shared systems, and the tie-ins all get planned and isolated deliberately. A tie-in to a live bus done casually is how a new phase takes down a hall that was already serving customers.

Where does commissioning fit in the build?

Commissioning is not the last week of the job. It is a chain of witnessed checkpoints that runs through the entire build, from the factory before the gear ships to the integrated test that gates go-live. Level 1 is the factory acceptance test, witnessed at the manufacturer. Level 2 is receiving when the gear lands. Level 3 is the de-energized pre-functional checks. Level 4 is functional testing of each system on its own. Level 5 is the integrated systems test, the whole plant run together and put through a real failure at design load. The data center commissioning levels guide walks all of it in depth, so this is the short version.

The schedule mistake that defines a weak program is treating commissioning as a final inspection and squeezing it into whatever window is left at the end. It absorbs every slip ahead of it, so a program planned for twelve weeks gets eight, and the eight get eaten by rework nobody scheduled. The integrated test needs the entire plant, full load banks, and a long uninterrupted block of time, and that is exactly the thing schedule pressure attacks first.

Protect it by building commissioning into the master schedule as its own track, tied to construction milestones, front-loading the levels that can run early at the factory and on the bench, and fencing off the integrated test window so the only thing left at the end is the test that genuinely needs the whole plant. Skip a level to recover schedule and the defect does not disappear. It moves to a more expensive place to find, and the most expensive place is a live floor.

QA/QC and the daily record during construction

Quality control runs every day through the build, separate from commissioning and ahead of it. The trades check their own work against the contract, the inspections happen as the work is covered, and the documentation gets captured while the evidence is still visible. A torque that was witnessed and striped, a megger reading taken cold before energization, a weld inspected before it is buried: these are QC artifacts, and they are cheap to capture in the moment and impossible to reconstruct later.

The daily report is the spine of the field record. It logs the crews on site, the work put in place, the deliveries, the weather, the inspections, and the issues. On a fast job with hundreds of workers across many trades, the daily report is how anyone reconstructs what happened on a given day, which matters when a question or a claim comes up months later. The teams that keep it honest and current have a record they can stand behind. The teams that backfill it at the end of the week have fiction.

QC and commissioning are not the same thing, and conflating them is a common error. QC confirms the work was installed correctly. Commissioning confirms the systems actually perform together. Both belong on the job, QC continuously through the build and commissioning as the staged verification on top of it, and the inspection and daily-report discipline by topic feeds both.

The schedule, the critical path, and speed to market

The schedule on a data center is aggressive by design, because the owner is racing to fill the building with revenue-generating compute. Full construction commonly runs on the order of 18 to 30 months, and the whole development, including site selection, permitting, and the utility power, runs longer than that. AI campuses at gigawatt scale can stretch well past two years on the interconnection and transmission alone.

The critical path is rarely the building. It runs through the things that take the longest and that everything else waits on: the utility power agreement and the substation, the long-lead transformers and switchgear, and the integrated commissioning window at the end. Manage those three and you manage the schedule. Manage the drywall and you manage nothing that matters.

Speed to market is the pressure that distorts everything. It pushes owners to start the build before the power is committed, to skip the early commissioning levels to claw back weeks, and to compress the integrated test that needs the most time. The honest version of fast is to attack the real critical path: lock the power early, order the long-lead gear during design, phase the building so capacity comes online in increments, and protect the test window. The dishonest version is to compress the proof at the end, and that is the version that produces a building that goes live and then drops its load.

Safety on a fast MEP site

A data center build is a large, fast-moving, MEP-dense site with thousands of worker-hours stacked into a tight schedule, and that combination is exactly where safety gets pressured. Many trades work on top of each other in the same overhead and the same rooms, the schedule is relentless, and the energized work ramps up as the build matures. The hazards concentrate where the work concentrates.

Electrical is the headline risk. As the plant energizes phase by phase, parts of the building go live while construction continues around them, so arc-flash boundaries, lockout-tryout, and energized-work permits become daily realities, not paperwork. The electrical safety practices, NFPA 70E among them by topic, govern the energized work, and the discipline that keeps people alive is verifying a circuit is dead before touching it, every time, on a site where some of it is not.

The rest is heavy-construction safety run at speed: the cranes and rigging setting big gear, the work at height, the confined spaces in the underfloor and the tanks, and the fall and struck-by exposures that come with a crowded site. A real safety program plans the energization sequence and the live boundaries as deliberately as it plans the work itself, because the day a phase goes live next to active construction is the day the site got more dangerous, not less.

Turnover and go-live

Turnover is the handoff where a proven building becomes an operating facility, and it is a package, not a date. The turnover package is the record the operations team inherits and runs the building from: the commissioning reports at every level, the signed test scripts, the closed deficiency log, the as-built drawings, the operation and maintenance information, the warranties, and the training records. A turnover missing the as-builts, the closed log, or the training is missing its spine.

The as-builts have to reflect what was actually installed, including every field change, because the operations team will trust those drawings the day something fails at 2 a.m. A feeder that was re-routed and never redlined, a setting changed in the field and never captured, a device relocated off the plan: each becomes a trap for the operator reading a drawing that no longer matches the building. The commissioning and turnover discipline by topic carries this in detail.

Go-live is the first real IT load arriving on a floor that has been proven on load banks but never with live equipment. The careful version ramps the load in, watches the plant respond, and keeps commissioning support on site through the ramp, because the gap between a load bank and a hall full of real, variable server load is where the last surprises hide. The first customer server is the moment the months of work either hold or do not.

The project team and who does what

The owner sits at the top and inherits the building and the record at the end, so the owner's program and the OPR drive every decision below. The owner often runs the job through an owner's representative or a program manager who carries the owner's interest day to day. The engineer of record owns the design intent and answers when a test result does not match the sequence of operations.

The general contractor or construction manager holds the schedule, runs the site, and coordinates the trades, the access, and the energization sequence. The MEP subcontractors, the electrical and mechanical trades, do the heaviest lifting on this kind of job, because the building is mostly their scope. The equipment vendors run startup on their own gear and stand behind the warranty. The commissioning authority, the CxA, is the owner's independent agent who witnesses and accepts the tests rather than performing them.

The line that matters on any contract is who the CxA actually works for. The whole value of the check is that it is independent, so a CxA reporting to the general contractor instead of the owner is the party being checked controlling the checker. The commissioning levels guide covers the roles in depth. The construction point is that this is a large team with many parties, the coordination between them is the job, and the contract structure decides who owns the gaps between scopes.

Why the budget is mostly MEP

Most of a data center's cost is not the building. It is the power and the cooling. Industry benchmarks for 2026 commonly land on the order of 10 to 12 million dollars per MW for mainstream builds, with AI-specific scope and high-cost markets pushing well above that, sometimes past 20 million per MW. Treat any per-MW figure as a screening number, not a quote, because density, redundancy, region, and cooling approach move it hard.

Inside that number, MEP dominates. The mechanical, electrical, and plumbing systems commonly run around half of a standard build's total and a much larger share, often cited near three-quarters, on an AI-focused project. Within the MEP, electrical and power are the single largest line, the switchgear, UPS, generators, transformers, and distribution, with cooling the next, and the structure and finishes a relatively small slice. The building is the cheap part of a data center, which surprises people coming from commercial construction.

The cost lever owners actually control is the redundancy and density target set in the program. Every step from N toward 2N multiplies the gear and the gray space. Every step up in rack density pushes the power and cooling capacity, and the cost, with it. The expensive surprises come from sizing the gray space for today's density and running out as the load per rack climbs, and from a long-lead order placed late enough to force expedited freight or a schedule slip that costs more than the gear. Hedge every cost figure to the project and the market, because both move fast.

The AI and gigawatt buildout trend

The current wave is bigger, faster, and denser, and it has changed what a data center project is. Hyperscale campuses are now planned and financed at the gigawatt scale, not the tens or hundreds of megawatts that defined the last cycle. AI training and inference concentrate enormous power and heat into a few rows, which resizes the power chain, the cooling, and the structure all at once.

Cooling is the first thing to change. Air alone runs out somewhere in the tens of kilowatts per rack, and AI racks blow past that, so the dense rows move to liquid cooling, which brings coolant distribution units onto the floor, piping into the white space, and far more heat rejection into the plant. Forecasts have liquid cooling going from a small fraction of AI servers to the majority within a couple of years, and retrofitting an air-cooled hall to liquid is months of civil work, not a swap.

The buildout approach is shifting with the scale. Some AI campuses move away from the slow, even, phased rollout toward large increments delivered in tighter synchronization, because the compute wants to come online in big blocks. The constraints have not changed, though. The power agreement and the long-lead electrical gear still gate the schedule, and at gigawatt scale the grid interconnection is the hardest pole of all. Bigger and faster did not remove the bottlenecks. It moved more weight onto them.

Retrofit and expansion on a live site

Not every data center project is a new building. A large share of the work is adding capacity to a facility that is already running: a new hall in an existing shell, a power or cooling upgrade, or a conversion of an air-cooled hall to liquid for AI load. The difference from new construction is the one that changes everything. The building is live, carrying other people's revenue and data, and the work cannot put that load at risk.

Live-site work is slower and more careful by necessity. Tie-ins to an energized bus or an active chilled-water loop get planned as method-of-procedure documents, scheduled into maintenance windows, and isolated so a slip does not cascade into the running plant. The redundancy that protects the live load is also what lets the work happen: a concurrently maintainable design lets a path be taken down for the tie-in while the other path carries the floor. Where that redundancy is thin, the work window is thinner.

The expensive retrofit mistake is treating it like new construction and forgetting the building is awake. A casual tie-in, a breaker operated without confirming what it actually feeds, a cooling loop drained without checking what depends on it, and the expansion project takes down the production floor it was supposed to extend. On a live site, the construction and the operations both have to agree on every move that touches a shared system.

What to document across the phases

The paper runs the job. On a project this fast, with this many parties, the record is what lets anyone reconstruct what happened, defend a decision, or hand a working building to operations. The submittals prove the gear matches the design. The RFIs and their answers track the questions that changed the work. The daily reports log what got built. The commissioning scripts and reports prove the plant works. The as-builts capture what was actually installed, and the turnover package rolls it all up. A finding, its photo, its assignment, and its sign-off that scatter across separate inboxes are a record nobody can trust.

Capture it in the field as the work happens, not in the last two weeks before turnover. A submittal logged against the gear it belongs to, an inspection signed at the point it was covered, a commissioning step signed at the test, an as-built redlined as the change is made. Built that way, the turnover package assembles itself from the record instead of being stitched together from memory. This is the field tracking the tradeos workflow is built to hold, so the document, its evidence, and the sign-off stay attached to the same record across a long, fast job.

The table below maps the phases to what happens and the risk that bites each one, so the project plan and the record can be built around where the trouble actually lives.

Common mistakes

• Starting the build, or even committing the schedule, before the utility power agreement and interconnection are locked.
• Underestimating long-lead times and ordering transformers, switchgear, or generators after the design is fully complete.
• Squeezing commissioning into the end of the schedule, then skipping levels to recover weeks.
• Field-coordinating the MEP instead of running a coordinated model, so the power, piping, tray, and duct clash in the overhead.
• Weak documentation and a turnover package missing the as-builts, the closed deficiency log, or the training.
• Building with no phased plan, so capacity cannot come online in increments and the whole job rides one go-live date.
• Sizing the gray space for today's density and running out of electrical and mechanical room as the load per rack climbs.
• Treating a live-site retrofit like new construction and taking down the production floor with a casual tie-in.

Field checklist

Standards and references

A data center build answers to several layers of standards at once. TIA-942 is the data center infrastructure standard for the facility spaces, the layout, and the cabling, and TIA-606 governs the labeling and administration. BICSI-002 is the data center design and implementation best practice and is commonly written into the basis of design. The Uptime Institute's Tier classification is the common language for redundancy and concurrent maintainability, and on a Tier job the constructed-facility certification is a witnessed demonstration that the built plant meets its claimed Tier.

The discipline codes govern the installation underneath all of that. The NEC, NFPA 70, controls the electrical installation, the grounding and bonding, and the working clearances. NFPA 110 covers the emergency and standby power plant. The building, mechanical, and fire codes, the NFPA suite among them, govern the structure, the cooling, the suppression, and the life safety. NETA acceptance testing and the IEEE references sit under the electrical commissioning. ASHRAE TC 9.9 sets the thermal guidelines the cooling is designed to.

The commissioning process itself traces to ASHRAE Guideline 0 and Standard 202, covered in the commissioning levels guide. Across all of it, the owner's program, the basis of design, the project specifications, and the adopted code edition with local amendments control the specific numbers and the schedule. Confirm the current edition of any standard against the project documents and the jurisdiction, because titles, numbers, and lead times all shift between cycles. Cite the document that controls the point, not a figure carried from the last job.

Terms and abbreviations

A data center build runs on a stack of terms that cross construction, electrical, mechanical, and IT, and the same word can read differently across a contract, a spec, and a vendor sheet. Pin the term to the phase and the document before you act on it.

OPR / BOD

Owner's project requirements and basis of design, the requirement and the design team's documented answer to it

SD / DD / CD

Schematic design, design development, and construction documents, the staged design sets from concept to buildable

Long-lead equipment

Gear with a long order time, the generators, switchgear, UPS, transformers, and chillers that gate the schedule

Cold / warm / powered shell

Bare enclosed structure, shell with core MEP installed, and shell with incoming power ready for tenant fit-out

White space / gray space

The conditioned data hall for IT gear, and the support area for the mechanical and electrical plant

MEP

Mechanical, electrical, and plumbing, the systems that carry most of a data center's cost and schedule

Cx / CxA

Commissioning, and the commissioning authority who plans, witnesses, and signs off the staged verification

IST

Integrated systems test, the full-plant Level 5 test that proves the building rides a failure at design load

N / N+1 / 2N

Redundancy levels, from no spare, to one extra component, to two full independent systems

$/MW / $/kW

Cost per megawatt or kilowatt of IT capacity, the common way data center capital cost is benchmarked

Related tools

Calculators and readiness checks for this work

Compare your options

FAQ

People also ask

• What is data center commissioning?
• What is white space in a data center?
• What is an RFI in construction?
• Cast-in vs post-installed anchors: which should I use?
• What is the difference between cold and hot commissioning of a BESS?