GreenTech provides some the details on how Google is able to accomplish its goal then the NYTimes - h/t mbloom1915
For Google, that meant serious work at the data centers to cut down on the electricity needed to run all those searches and YouTube clips. Those data centers now deliver 3.5 times as much computing power as they did five years ago, while using the same amount of electricity, the company said.
Google's parent company Alphabet also leveraged in-house expertise from DeepMind, its machine-learning subsidiary, to reduce energy needs. By training the computers to optimize operational factors like fans and cooling systems, DeepMind was able to cut the energy use for cooling data centers by 40 percent, yielding a 15 percent drop in overall energy overhead.
"[If] you can cover the roof of a data center with solar PV panels, maybe you cover 5 percent of the electricity needs," Touati said. "If you want to reach your...100 percent renewable target, the only option you have today is to actually source electricity from offsite projects."
That's what Google did, to the tune of 2.6 gigawatts worldwide. In some locations where renewable purchase options didn't exist, the company worked with the local utilities to create them. That process helped produce Duke Energy's Green Source Rider, which enabled Google to purchase power for its North Carolina data center from a 61-megawatt solar project. Now other companies can buy through that program as well.
Source: news.ycombinator.com