Power Capacity: The Foundation

Introduction Picture this: A colocation sales rep tells you they have "10 MW available" at their facility.

You're ready to sign a contract for 2 MW to support your company's cloud expansion.

But here's the question that separates novices from experienced operators-are they talking about total facility power or available IT load capacity? That distinction could mean the difference between getting what you need and finding yourself with 40% less usable power than you planned for.

Power capacity represents the absolute foundation of every data center decision.

Measured in megawatts (MW) or kilowatts (kW), it determines what you can build, what you can sell, and ultimately what revenue you can generate.

A facility's power capacity directly constrains its ability to house servers, storage arrays, and networking equipment-the actual infrastructure that generates business value.

Without adequate power capacity, even the most sophisticated cooling systems and redundant network connections become irrelevant.

You simply cannot operate what you cannot power.

This lesson breaks down the fundamentals of power capacity measurement, the critical distinction between facility power and IT load, and the direct relationship between watts and revenue that drives every data center investment decision.

You'll gain the framework to evaluate capacity claims, calculate actual usable power, and understand why operators obsess over every kilowatt.

Understanding Megawatts and Kilowatts Power capacity uses two primary units of measurement in data center operations.

One kilowatt (kW) equals 1,000 watts of power.

One megawatt (MW) equals 1,000 kilowatts, or 1,000,000 watts.

These aren't abstract numbers-they represent the actual electrical capacity flowing into and through a facility at any given moment.

A single server rack typically draws between 5 kW and 20 kW of power, depending on the density of equipment installed.

Modern high-performance computing (HPC) racks or AI training clusters can push 50-100 kW per rack.

When Equinix talks about a "10 MW data hall," they're describing a space designed to support approximately 500 to 2,000 racks, depending on power density assumptions.

Scale matters in this industry.

A typical enterprise server room might operate at 100-500 kW.

Regional colocation facilities often range from 5-20 MW.

Hyperscale campuses operate at a completely different magnitude.

Meta's Prineville, Oregon campus delivers over 300 MW of capacity across multiple buildings.

AWS operates facilities in Northern Virginia that individually exceed 100 MW.

Microsoft's Chicago campus reaches similar scales.

Think about it this way: every MW of capacity represents roughly $1-2 million in annual revenue potential for colocation providers, assuming typical pricing and utilization rates.

That's why capacity planning drives every major financial decision in this industry.

Total Facility Power vs.

IT Load Power Here's where many beginners make their first critical mistake.

When someone mentions a facility's power capacity, they might be referring to three different measurements: Total Facility Power represents the maximum electrical capacity delivered to the building from the utility company.

This is the raw power coming into the facility before any distribution or conversion losses. Critical Load includes all power used by data center infrastructure-IT equipment plus all mechanical and electrical systems supporting that equipment.

This encompasses cooling systems, UPS units, power distribution equipment, lighting, and all other operational systems. IT Load Power (also called white space power) represents the actual power available for servers, storage, networking equipment, and other revenue-generating IT infrastructure.

This is what customers actually use and pay for.

The relationship between these measurements determines a facility's Power Usage Effectiveness (PUE).

A facility with 10 MW total utility power and a PUE of 1.5 delivers approximately 6.67 MW of IT load power.

The remaining 3.33 MW goes to cooling, power conditioning, lighting, and infrastructure overhead.

Facility Type	Total Power	Typical PUE	IT Load Power	Infrastructure Overhead
Legacy Enterprise	10 MW	2.0	5 MW	5 MW (50%)
Modern Colocation	10 MW	1.5	6.67 MW	3.33 MW (33%)
Hyperscale Optimized	10 MW	1.3	7.69 MW	2.31 MW (23%)
Best-in-Class	10 MW	1.15	8.70 MW	1.30 MW (13%)

Digital Realty's newer builds typically target 1.3-1.4 PUE.

Older facilities, particularly those built before 2010, often operate at 1.8-2.2 PUE.

When evaluating capacity, always clarify which measurement you're discussing.

A provider advertising "20 MW available" might mean 20 MW of utility capacity with only 13-14 MW available as IT load.

That distinction represents millions of dollars in revenue potential and directly impacts your deployment plans.

The Power-Revenue Relationship Power capacity doesn't just enable operations-it directly generates revenue.

Every kilowatt of IT load power represents billable capacity that customers will pay to consume.

Colocation providers typically price power in three models: Flat-rate allocation charges customers for a committed power capacity regardless of actual usage.

A customer might pay $125-175 per kW per month for allocated capacity in major markets like Northern Virginia or Silicon Valley.

At these rates, a fully-utilized 1 MW data hall generates $1.5-2.1 million in annual power-related revenue alone. Metered usage charges customers for actual power consumption, typically at $0.08-0.15 per kWh plus a base facility fee.

This model shifts utilization risk to the customer but requires sophisticated metering infrastructure. Hybrid models combine a base allocation with metered overages, providing predictable revenue while capturing additional value from high-density deployments.

QTS Data Centers has reported average power pricing around $140 per kW per month in their premium markets.

CyrusOne typically generates 35-45% of total revenue directly from power-related charges.

Switch emphasizes high-density deployments at their Las Vegas campus, supporting customers with 20-30 kW per rack requirements.

Hyperscalers approach this calculation differently.

AWS, Azure, and Google Cloud don't sell power directly-they sell compute, storage, and services.

But their unit economics depend entirely on power efficiency.

A 1% improvement in PUE at a 100 MW facility saves approximately $800,000-1.2 million annually at typical utility rates of $0.08-0.12 per kWh.

Consider the math: 100 MW operating continuously equals 876,000 MWh per year.

At $0.10 per kWh, that's $87.6 million in annual power costs.

Improving PUE from 1.5 to 1.4 reduces power consumption by approximately 58,400 MWh annually, saving $5.84 million.

These economics explain why Google, Microsoft, and Meta invest heavily in power efficiency technologies.

Capacity Planning and Utilization Building power capacity costs between $8-15 million per MW for traditional construction, depending on location, utility infrastructure requirements, and design specifications.

That capital intensity makes capacity planning one of the most critical decisions in data center development.

Most operators don't build entire campuses at once.

They construct facilities in phases, installing utility capacity and shell infrastructure first, then building out power distribution and cooling in stages as customer demand materializes.

CoreSite's Santa Clara campus exemplifies this approach-initial site development included utility feeds capable of supporting 30+ MW, but individual suites were built and powered incrementally based on leasing velocity.

Utilization rates vary dramatically by business model.

Colocation providers typically target 80-90% IT load utilization at steady state.

They maintain 10-20% reserve capacity for existing customer expansion, new sales opportunities, and maintenance flexibility.

Hyperscalers often operate at higher utilization (85-95%) because they control both supply and demand, allowing more precise capacity planning.

The relationship between committed capacity and actual consumption creates interesting dynamics.

A typical enterprise customer might deploy infrastructure using only 40-60% of their allocated power initially, planning for growth over 3-5 years.

Colocation providers must balance this reality against their need to maximize revenue per MW.

Digital Realty manages this through flexible power density offerings.

Some customers need 3-5 kW per rack for standard enterprise workloads.

Others require 15-25 kW per rack for high-performance computing or AI training.

The same physical space might generate vastly different revenue depending on power density and utilization patterns.

Constraints and Dependencies Power capacity doesn't exist in isolation-it creates cascading requirements throughout facility design.

Every MW of IT load requires corresponding cooling capacity, typically measured in tons of refrigeration.

Using standard efficiency assumptions, 1 MW of IT load requires approximately 400-600 tons of cooling capacity, depending on facility design and climate.

Power distribution infrastructure must handle the full capacity path from utility interconnection through rack-level PDUs.

A 20 MW facility might include:

• Dual utility feeds with automatic transfer capability
• 20+ MW of UPS capacity for power conditioning and backup
• 15-20 MW of generator capacity for extended outages (sized for critical load)
• Switchgear, transformers, and distribution panels throughout the facility
• Thousands of PDUs delivering power to individual racks Each component in this chain introduces potential bottlenecks.

CyrusOne discovered this during a facility expansion in Phoenix-while the building had adequate generator capacity, the switchgear infrastructure couldn't distribute the load effectively.

They needed to invest $2+ million in electrical infrastructure upgrades before selling the remaining capacity.

Space constraints also matter.

Generator sets require significant real estate.

A 2 MW diesel generator occupies roughly 200-250 square feet including clearances.

A 20 MW facility might need 10+ generator sets for N+1 redundancy, consuming 2,500+ square feet of valuable land.

Utility infrastructure represents another potential constraint.

Bringing 50+ MW of power to a greenfield site can require new substations, transmission line upgrades, or dedicated utility feeds.

These projects often take 18-36 months and cost $15-30 million.

Northern Virginia's "Data Center Alley" has well-developed utility infrastructure because Dominion Energy has invested billions in capacity specifically to support the data center industry.

Emerging markets lack this infrastructure, making capacity delivery more expensive and time-consuming.

Practical Example: Evaluating a Colocation Proposal You're evaluating two colocation proposals for a 500 kW deployment supporting your company's regional cloud presence.

Both providers claim "adequate capacity" but structure their offerings differently. Provider A (Regional Operator) offers 500 kW in a Tier III facility at $150 per kW per month.

They quote their facility as "10 MW total capacity" with current utilization at 65%.

Your monthly cost: $75,000.

Their 10 MW facility specification refers to total utility capacity.

At a PUE of 1.5, they actually have 6.67 MW of IT load capacity.

With 65% utilization, they're currently using 4.33 MW, leaving 2.34 MW available.

Your 500 kW deployment represents 21% of their remaining capacity-a significant commitment that should provide negotiating leverage. Provider B (Tier 1 Colocation Brand) offers 500 kW at $175 per kW per month in a similar market.

Monthly cost: $87,500.

They describe their facility as "30 MW IT load capacity" currently at 80% utilization.

This provider is clearly specifying IT load power, not total facility capacity.

At 80% utilization, they have 6 MW remaining.

Your deployment represents only 8% of available capacity-you're a smaller customer with less negotiating power.

Beyond pricing, consider density capabilities.

Provider A's facility supports maximum 12 kW per rack.

Provider B handles up to 30 kW per rack.

If your application roadmap includes high-density compute or AI workloads, Provider B offers more expansion flexibility despite higher costs.

The capacity availability numbers also indicate different risk profiles.

Provider A has less remaining capacity cushion, suggesting they might struggle to support your growth without building additional infrastructure.

Provider B has more headroom but operates at higher utilization, indicating strong market demand and potentially less flexibility for custom requirements.

Practical Example: Calculating Deployable Capacity A hyperscaler is evaluating a build-to-suit opportunity.

The site can support 50 MW of utility capacity.

They need to determine actual IT load capacity and resulting server deployment potential.

Using their standard design achieving 1.25 PUE:

• Total utility capacity: 50 MW
• IT load capacity: 50 MW ÷ 1.25 = 40 MW
• Infrastructure overhead: 10 MW (20% of total) Their standard server configuration draws 350W per server at average utilization.

Calculating deployable servers:

• 40 MW IT load = 40,000 kW
• 40,000 kW ÷ 0.35 kW per server = 114,285 servers These servers generate revenue through compute instances.

Assuming average revenue of $45 per server per month:

• 114,285 servers × $45/month = $5,142,825 monthly revenue
• Annual revenue potential: $61.7 million Building this facility costs approximately $450-600 million at $9-12 million per MW for total facility capacity.

At $61.7 million annual revenue, the project achieves 10-14% annual return before operating costs-acceptable returns for hyperscale economics.

Now consider a PUE improvement to 1.15:

• IT load capacity increases to 43.48 MW
• Deployable servers increase to 124,228
• Annual revenue increases to $67.1 million
• Additional revenue: $5.4 million annually That PUE improvement, achieved through advanced cooling design, adds $5.4 million in annual revenue without increasing utility capacity.

The ROI on efficiency investment becomes clear.

Common Misconceptions "More power capacity is always better." Capacity beyond what you can monetize represents stranded capital.

Building 50 MW when market demand supports only 30 MW means you've invested $160-240 million ($8-12M per MW × 20 MW excess) generating zero return until demand materializes.

Phased development approaches match capital deployment to revenue opportunities.

Equinix rarely builds more than 18-24 months ahead of projected demand, preferring to add capacity incrementally as customer commitments materialize. "IT load power and critical load are the same thing." Critical load includes IT load plus all infrastructure supporting that load-cooling systems, UPS units, power distribution, lighting, and building systems.

A facility with 10 MW critical load might only have 6-7 MW available as IT load power.

When providers reference "critical load capacity," you're not hearing about usable customer power.

This terminology confusion leads to serious miscalculations during capacity planning.

Always clarify whether capacity figures represent total facility power, critical load, or IT load power.

Summary & Key Takeaways

• Power capacity measured in kilowatts (kW) and megawatts (MW) represents the fundamental constraint on data center operations-1 kW equals 1,000 watts, 1 MW equals 1,000 kW, and typical facilities range from 5-300+ MW depending on scale and operator type
• Total facility power differs significantly from IT load power-PUE determines the relationship between these measurements, with modern facilities achieving 1.3-1.5 PUE (meaning 67-77% of total power available as IT load) and best-in-class facilities reaching 1.1-1.15 PUE
• Power capacity directly drives revenue potential-colocation providers generate $1.5-2.5 million in annual revenue per MW of IT load at typical market rates, making capacity planning one of the most critical financial decisions in facility development
• Building power capacity costs $8-15 million per MW-this capital intensity drives phased development approaches where operators build shell capacity first and deploy power/cooling infrastructure incrementally based on customer demand
• Capacity constraints extend beyond power itself-every MW of IT load requires corresponding cooling capacity (400-600 tons), space for generators and electrical infrastructure, and utility delivery capability that can take 18-36 months to develop in greenfield markets

Next Steps Continue to the "Power Distribution Architecture" lesson to understand how facilities deliver power from utility interconnection through rack-level PDUs.

You should also explore "Understanding PUE and Efficiency Metrics" to deepen your knowledge of the relationship between total facility power and IT load capacity.

The "Cooling Fundamentals" lesson provides essential context on how power capacity decisions drive cooling infrastructure requirements.