Hyperscale Data Center

A hyperscale data center is a single campus engineered for very large, horizontally-scaled compute — typically tens of thousands to millions of servers under one operator. The label is informal but is generally applied to facilities with at least 5,000 servers and a contiguous IT load measured in the tens to hundreds of megawatts, with new AI training campuses crossing into the gigawatt range. Hyperscale operators — the term is mostly used for cloud platforms and a few large internet companies — design their own racks, networking, power distribution, and increasingly their own silicon. Standardisation across a fleet lets them push power usage effectiveness below 1.2 at the best sites, well under the industry-wide average that has hovered around 1.58 since the mid-2010s. Site selection is driven by four constraints: available transmission-level power, water or air cooling capacity, low-latency fibre, and tax or regulatory incentives. The result is dense regional clusters such as the Northern Virginia data center alley, the Dublin corridor in Ireland, and the suburbs of Phoenix, Atlanta, Dallas, and Singapore. Concentration raises local electricity grid stress even when global totals look manageable, because new transmission and substation capacity takes years to build.