STRATASkip to the research
What’s Up, original XLVII plate

What’s Up

Berkeley Lab’s 2016 and 2024 reports examine electricity use across US data centers. Both distinguish historical estimates from prospective scenarios, allowing equipment efficiency and total demand to be examined separately.

Power usage effectiveness compares total facility power with computing-equipment power. A lower ratio can coexist with greater total electricity use when the computing load increases; water consumption requires a separate measurement and accounting boundary.

Historical record

  1. Efficiency becomes a publicly identified product attribute.

    The US Environmental Protection Agency launches ENERGY STAR. The program’s history offers an early institutional example of making energy performance visible to buyers, rather than treating electricity consumption as an invisible consequence of purchasing equipment.

  2. A national estimate counts data-center electricity beyond a single machine.

    Berkeley Lab’s report reconstructs historical US data-center demand and develops scenarios through 2020. Its unit of analysis is an installed population of equipment, which makes the difference between component efficiency and total consumption explicit.

  3. The accounting is rebuilt as demand and equipment change.

    Berkeley Lab releases a new US data-center energy report in December. It separates historical estimates from scenarios extending to 2028; those scenarios describe conditional futures, not measurements already collected.

A more efficient facility can still consume more electricity.

Suppose an illustrative facility supplies a steady 1 MW to its computing equipment. At a power usage effectiveness of 1.4, total facility demand is 1.4 MW and daily electricity use is 33.6 MWh. Reducing that ratio to 1.2 lowers daily use to 28.8 MWh at the same computing load. If the computing load then grows to 1.3 MW, total daily use becomes 37.44 MWh. The efficiency gain remains real, while the absolute demand exceeds the original case.

  • Display computing power, overhead power and total energy separately. A ratio alone hides the amount of work and the duration of operation.
  • Treat water consumption as a separate measured quantity with a defined boundary. It cannot be calculated from power usage effectiveness alone.
  • Review a capacity increase against a site-wide budget and its local constraints. State the workload served, the time interval and the people whose allocation changes.

Recordings and archives

Research sources

  1. Our History

    US Environmental Protection Agency / ENERGY STAR · EPA’s account of the program’s 1992 launch.

  2. United States Data Center Energy Usage Report

    Lawrence Berkeley National Laboratory · 2016 report with historical estimates and then-prospective scenarios to 2020.

  3. 2024 United States Data Center Energy Usage Report

    Lawrence Berkeley National Laboratory · December 2024 report; historical estimates and future scenarios are distinct.