Data Center Infrastructure: A Practical Guide for Buyers Who Have to Get the Power Right

Data Center Infrastructure: A Practical Guide for Buyers Who Have to Get the Power Right

Why do so many data center projects stall before a single rack goes in? Because the hardest part isn't the building or the servers. It's securing enough reliable power, and the cooling that power demands, on a timeline your board or council can live with.

This guide is for the people who have to make that call: corporate infrastructure leads, municipal and public-agency planners, and government teams weighing a data center project in their jurisdiction. You're probably being pitched by developers, utilities and equipment vendors, each describing a different slice of the problem. You need the whole picture.

Here's what you'll get:

• A plain breakdown of the six physical layers that make up a facility, and which ones carry the most risk
• A 12-week planning sequence you can start this quarter
• A site checklist and a set of questions to put to any developer, word for word
• An honest look at where projects go wrong, including the cases where the usual advice doesn't apply

I'm Dwayne Corbitt, founder and CEO of Volarix Global, a company that develops advanced energy technologies, infrastructure and sustainable power systems, including data-center infrastructure through its Bahrain platform. I've spent more than 30 years building companies across manufacturing, energy and technology, and the lens I bring here is the energy side, because that's where data center projects are won or lost.

The short version: power decides everything

Your power strategy is your data center strategy. Floor space, fiber and real estate are mostly solvable problems. Grid capacity often isn't, at least not quickly.

Demand is rising fast enough that utilities in many regions are rationing new connections. The International Energy Agency's Energy and AI report lays out how data center electricity demand is climbing alongside AI workloads, and the 2024 United States Data Center Energy Usage Report from Lawrence Berkeley National Laboratory projects substantial growth in U.S. data center consumption through the end of the decade.

What that means for you is simple. If you start with the building and work backwards to power, you'll lose months. Start with power and work forwards.

The six layers you're actually buying

A data center is a stack of systems, each depending on the one beneath it. When a vendor quotes you a "turnkey facility," it helps to know exactly which of these layers they're covering and which they're quietly leaving to you.

The top three layers absorb most of the capital and most of the schedule risk. The bottom three matter, but they're well understood and widely available. If your project team spends the first month debating the security system, they're working on the wrong layer.

Power: where most projects stall

Get a written capacity answer from the utility before you commit to a site. Not a verbal "should be fine." A document.

Utility interconnection is the longest-lead item in most builds. Depending on the region, the wait for a new high-capacity connection can run longer than the entire construction schedule. That's why experienced developers now treat power as a parallel workstream that begins on day one, not as a line item for later.

You have three broad options, and most serious projects blend them:

1. Grid supply only. Simplest to operate, but you're entirely exposed to the utility's timeline and pricing.
2. Grid plus on-site generation. Backup generators are standard. Increasingly, sites add on-site solar, fuel cells or gas turbines to cover part of the base load.
3. Grid plus storage. Batteries or thermal storage shift when you draw power, which can reduce the peak demand the utility has to provision for.

The third option gets less attention than it deserves. A utility that can't supply your full peak load on a hot summer afternoon might be able to supply it if you can shave that peak.

That's a negotiation you can only have if you've designed storage in from the start.

Cooling, and why storage changes the math

Cooling is the second-largest energy draw in a typical facility after the IT load itself, and it peaks at exactly the wrong time: hot afternoons, when grid power is most expensive and most constrained.

This is the layer I know best from direct work. Earlier in my career, I delivered thermal energy storage systems to corporations, municipalities, schools, the U.S. military and the Los Angeles Department of Water and Power. The principle carries straight over to data centers. You make cooling capacity at night, when power is cheaper and the grid has slack, and you spend it during the day. Typically that means chilling water or making ice in insulated tanks, then drawing on that stored cold to carry part of the afternoon load.

The detail that matters in practice: thermal storage doesn't reduce how much cooling you need. It changes when you buy the electricity to produce it. For a site with tight grid capacity, that can be the difference between approval and a two-year wait.

A few honest tradeoffs:

• Tanks take physical space, which matters on dense urban sites.
• Very high-density AI racks often need direct liquid cooling at the chip, which changes how storage integrates with the rest of the system.
• The economics depend heavily on your local tariff structure. If your utility doesn't price peak demand differently from off-peak, the financial case weakens.

For operating temperature and humidity ranges, the reference most engineers use is the guidance published by ASHRAE Technical Committee 9.9. Widening the allowable inlet temperature, within those ranges, is one of the cheapest efficiency gains available.

Resilience tiers, without the sales pitch

Decide how much downtime you can tolerate before you choose a tier. Then pay for that and no more.

The most widely used framework is the Uptime Institute's Tier Classification System, which grades facilities from Tier I (basic capacity) to Tier IV (fault tolerant). Each step up adds redundant paths for power and cooling, and each step adds cost.

A common mistake is specifying Tier IV for a workload that could tolerate a few hours of downtime a year. Public agencies do this often, usually out of caution. It's understandable, but the redundancy costs real money in both capital and ongoing energy use.

One more point. A vendor saying their design is "Tier III equivalent" is not the same as a certified Tier III facility. Ask which one they mean.

Measuring efficiency honestly

The standard efficiency metric is Power Usage Effectiveness, or PUE, developed by The Green Grid. It's total facility energy divided by IT equipment energy. A PUE of 1.0 would mean every watt goes to computing; anything above that is overhead, mostly cooling and power conversion losses.

PUE is useful. It's also easy to game.

A facility measured in winter will report a better number than the same facility in August. A design PUE on a vendor's slide is a prediction, not a measurement. And PUE says nothing about water consumption or the carbon intensity of the power you're buying.

When you evaluate a proposal, ask for:

• Annualized PUE, measured over a full year, not a best month
• Water usage per unit of IT energy
• The source mix of the electricity, not just the efficiency of using it

I'll be honest about a limit here. There isn't yet one universally accepted metric that captures power, water and carbon together in a way buyers can compare across sites. You'll need to look at all three separately.

A 12-week planning sequence

This is the sequence I'd run for a corporate or public-sector buyer starting from scratch. It won't finish a project in 12 weeks. It gets you to the point where you can make a confident go or no-go decision on a site and a delivery model.

Weeks 1 to 2: Define the load. Write down the IT load you need at opening and in five years, in megawatts. Specify the workload type (general enterprise, AI training, public records, edge). AI training racks run far denser than standard enterprise racks, and that changes the cooling design entirely.

Weeks 3 to 4: Talk to the utility. Submit a formal capacity inquiry for two or three candidate sites. Ask for the timeline, required upgrades, and who pays for them.

Weeks 5 to 6: Model the energy mix. Run scenarios for grid-only, grid plus generation, and grid plus storage. Include tariff structures, especially demand charges.

Weeks 7 to 8: Set the resilience target. Pick your Uptime tier based on the actual cost of downtime for your workload, not the default assumption.

Weeks 9 to 10: Shortlist delivery models. Build and own, build-to-suit lease, colocation, or a joint venture with an energy partner. Each shifts risk differently.

Weeks 11 to 12: Stress-test and decide. Put your shortlisted options through the developer questions in the next section. Kill any option that can't answer them in writing.

Questions to put to any developer

Use these verbatim in your first serious meeting. A credible partner will answer them directly. Evasion on any of the first three is a red flag.

• "What is the confirmed utility capacity for this site, and can you show me the utility's written response?"
• "What is your plan if the grid connection is delayed by twelve months?"
• "What's the annualized PUE of your most recent operating facility, not the design figure?"
• "How much water does the cooling system consume per year at full load?"
• "Which components in the power path are single points of failure?"
• "Is this design Uptime-certified, or described as equivalent?"
• "How does the cooling design change if we move to liquid-cooled AI racks in year three?"
• "Who owns the on-site generation and storage assets, and who carries the operating risk?"

Site selection checklist

Before you sign a land option or a letter of intent, confirm each item:

• Written utility capacity confirmation, with timeline
• At least two physically diverse fiber routes
• Water source identified, with allocation confirmed if using evaporative cooling
• Zoning permits data center use, or a realistic path to rezoning
• Space for on-site generation, fuel storage, and thermal or battery storage
• Flood, seismic and heat-risk profile reviewed
• Local community and council engagement plan in place
• Clear tax and incentive position in writing

The community item gets skipped more than any other. Opposition over noise, water and grid strain has delayed projects that had everything else lined up. If you're a municipality, you already know this. If you're a corporation entering a new town, budget time for it.

Where projects go wrong

Most failures I've seen come from sequencing, not engineering. Teams lock in a site, then discover the power isn't there. Or they design for today's rack density and find, two years in, that their tenants want liquid-cooled AI hardware the building can't support.

Cross-border projects add another layer. Over the years I've worked across China, the Gulf States, Europe, Mexico, Southeast Asia, Africa and North America, and the pattern repeats: the technical design travels well, but regulatory approvals, grid rules and partnership structures don't. A cooling design that works in Arizona works in Bahrain. The utility negotiation and the ownership structure won't look anything alike.

This guide also has limits. If you're a small business needing a few racks, you don't need any of this. A colocation contract will serve you better than a build. And if your workload is purely cloud-native, the infrastructure question belongs to your cloud provider, not you.

When energy and compute are planned together

The direction of travel is clear: data center infrastructure and energy infrastructure are converging. The sites that get built fastest are the ones where power generation, storage and compute are designed as one system.

That's the model behind Volarix Global's Bahrain platform, announced in 2026 through Volarix Global W.L.L. It brings together advanced energy, data-center infrastructure, digital infrastructure, AI, EV charging and energy-backed digital finance under one platform, so the power plan and the compute plan aren't handled by separate parties who meet for the first time at the substation.

For a government or public agency, that structure can simplify a cross-border project. For a corporation, it can mean one partner accountable for both the energy supply and the facility it feeds. Volarix Global works through founder-led partnerships and joint ventures, which suits buyers who want shared ownership of the risk rather than a pure vendor relationship.

If you're weighing how electric mobility or EV charging fits alongside a data center site, I'll cover that in a separate guide on energy-backed infrastructure.

Take the next step

If you're planning a data center and the power question is still open, that's exactly the conversation worth having now, before a site or contract locks you in. Volarix Global works with corporations, municipalities, public agencies and governments on energy-first infrastructure, from storage-backed cooling to cross-border joint ventures. Bring your load estimate and your candidate sites, and we'll talk through what's realistic. Reach out through dwaynecorbitt.com to start that conversation.

Frequently asked questions

What are the main components of data center infrastructure?

The core layers are power supply, power distribution, cooling, the physical structure, connectivity, and monitoring and control systems. Power and cooling carry most of the cost and schedule risk. Connectivity and controls are well understood and widely available.

Why is power the biggest constraint for new data centers?

Utility interconnection for high-capacity sites can take longer than construction itself, and many grids are short on spare capacity. Designing in on-site generation or storage can reduce the peak load you ask the utility for, which may shorten the wait.

How does thermal energy storage help a data center?

It produces cooling capacity at night, usually as chilled water or ice, and uses it during the afternoon peak. That shifts when you buy electricity rather than how much cooling you need. It works best where tariffs charge more for peak demand.

Which Uptime Institute tier does my facility need?

Choose based on what downtime actually costs your workload. Tier III, which allows maintenance without shutdown, suits most enterprise and public services. Tier IV is generally justified only for workloads that can't tolerate any single failure.

Is PUE a reliable way to compare data centers?

It's useful but incomplete. Ask for annualized PUE measured over a full year, and look separately at water consumption and the carbon intensity of the power supply, since PUE captures neither.