How Many MW Does a Data Center Use?

A comprehensive guide to data center power consumption at every scale — from a single cabinet to a gigawatt campus.

Data centers are the engines of the digital economy, and they run on electricity. But the answer to "how many megawatts does a data center use?" is not a single number — it spans three orders of magnitude. A small colocation facility might draw 1–5 MW, while a hyperscale campus can consume 100+ MW, and emerging AI clusters are pushing toward 1 GW and beyond.

Understanding these scales is essential for grid planners, utility executives, data center developers, equipment suppliers, and investors who need to anticipate where demand is heading. This guide breaks down every tier of data center power consumption with real-world examples.

Small Colocation: 1–5 MW

Small colocation facilities — sometimes called "retail colo" — lease space by the cabinet, cage, or private suite. These facilities typically house tens to a few hundred cabinets and draw between 1 and 5 megawatts of IT load. They are often located in urban or suburban areas close to customers.

Typical profile: 1–5 MW IT load, 30–50 W/ft² density, single or dual utility feed, UPS and backup generator support, N+1 redundancy.

Real example: Many Equinix International Business Exchange (IBX) facilities in non-major markets operate in the 2–5 MW range. A typical CyrusOne or CoreSite colocation facility in a secondary market like Columbus, Ohio, or Nashville, Tennessee, draws around 3–4 MW at full build-out.

Enterprise Data Centers: 5–20 MW

Enterprise data centers are built by large corporations — banks, insurance companies, retailers, and government agencies — to host their own IT operations. These facilities are larger than retail colo but moderate by hyperscale standards.

Typical profile: 5–20 MW IT load, 50–80 W/ft² density, dual utility feeds, N+1 or 2N redundancy, on-site generators with 24–72 hours of fuel storage.

Real example: JPMorgan Chase operates several enterprise data centers in the 10–15 MW range across the US. A large hospital system's primary data center might draw 5–8 MW. These facilities represent the "middle market" of data center power — significant but not transformational for the grid.

Hyperscale Data Centers: 20–100+ MW

Hyperscale facilities are the backbone of cloud computing. Built by Amazon Web Services, Microsoft Azure, Google Cloud, Meta, and Oracle, these facilities are designed from the ground up for massive scale. A single hyperscale data center campus typically houses 20–100+ MW of IT load across multiple buildings.

Typical profile: 20–100+ MW per campus, 100–200+ W/ft² density, dedicated high-voltage substations, multiple transmission feeds, on-site substations with 138 kV or 230 kV service.

Real examples:

• AWS Northern Virginia: AWS operates multiple data centers in the Ashburn/Chantilly area with an estimated total load exceeding 1,200 MW across its entire Virginia fleet. Individual campus phases typically run 25–50 MW each.
• Google Council Bluffs, Iowa: One of Google's largest data center complexes, estimated at 80–100 MW across multiple buildings, powered by wind and solar PPAs.
• Meta Altoona, Iowa: Meta's Altoona data center campus draws approximately 80 MW at full capacity, primarily serving Facebook and Instagram workloads.

AI Clusters: 100+ MW to 1 GW

AI training clusters represent a new category of power demand that barely existed five years ago. Unlike traditional cloud workloads, AI training runs thousands of GPUs at near-maximum utilization for days or weeks at a time. The power draw is relentless.

Typical profile: 100+ MW per cluster, 700W+ per GPU (Nvidia H100/B200), 40–60 kW per rack density, requiring liquid cooling, dedicated on-site substation with 230 kV+ service.

Real examples:

• Microsoft's AI infrastructure: Microsoft has disclosed plans for AI data centers exceeding 100 MW each. The company's global AI infrastructure investment exceeds $50 billion.
• xAI Colossus in Memphis: Built in record time in 2024–2025, the Colossus cluster in Memphis, Tennessee, draws approximately 50–100 MW for 100,000 H100 GPUs. xAI has since announced expansion plans that would double its footprint.
• Texas AI Corridor (800 MW planned): A proposed AI-focused campus near Fort Worth, Texas, is planned for 800 MW — a project we are actively tracking on our projects page.

The Scale Comparison

Why This Matters for the Grid

The data center industry's power demand is growing at an unprecedented rate. The Electric Power Research Institute (EPRI) estimates that data centers could consume up to 9% of total US electricity generation by 2030, up from roughly 2–3% today. This growth is concentrated in regions like Northern Virginia (PJM), Texas (ERCOT), and California (CAISO), where interconnection queues are already strained.

Our platform tracks major projects like Project Titan (1,200 MW planned) and the proposed Texas AI Corridor (800 MW) to help stakeholders monitor this transformation.

Key Takeaways

• Data center power spans 1 MW to 1,000+ MW — a 1,000x range
• AI clusters are creating a new "megawatt class" of 100+ MW per facility
• Grid interconnection is the primary bottleneck, not data center construction
• Transformer shortages (2+ year lead times) are delaying energization dates
• Understanding these scales helps stakeholders plan capacity investments