That's from research conducted by David Sarokin at Uclue, an online pay-for-answers service. The figures cover computers and monitors (roughly two-thirds of the total), data centers (one-eight) as well as networking and transmission equipment. They do not cover the energy that goes into producing and distributing computers and equipment, nor that powering printers and other non-communicating devices. Also, left out is the fast-growing non-computer set of Internet-enabled devices, such as PDAs, smartphones, etc. Sarokin has published the details his calculations, and for what I can judge they look generally accurate -- although of course a generous margin of error should be considered, given the difficulties inherent with such a calculation but also specific elements such as, for example, the relative growth of laptops (they consume less than desktops) etc..

Read more here.