Growth

Years without a published figure are absent rather than interpolated. LBNL states these particular years as numbers; the gaps between them are not filled in here, because a straight line drawn through two real points is not a third real point.

It is tempting to divide reported national electricity by the mapped facility count and call the result the power of an average data centre. That number would be meaningless. The numerator covers every data centre in the United States; the denominator covers only the 1,853 that OpenStreetMap happens to know about. Their ratio measures mapping coverage at least as much as it measures anything physical.

The same caution applies to the per-region megawatt figures elsewhere on this site. Those allocate the national total across mapped facilities by building footprint, so they are shares of a reported quantity — and upper bounds, since the facilities nobody has mapped have their consumption handed to the ones that have been.