Its cool to see how their server farms work. It are just containers full of motherboards. They connect power, internet and water(cooling) and the container is 'one' entity in the farm. If more then xx% (don't know the exact number) of motherboards in a container fails, they simple swap out the container and a tech can work on it. Container failure is not really a problem since they have so much redundancy. When they build the serverpark in Belgium they build a watertreatment plant to cool down the servers with water from a river nearby. After the water is cooled again in special heat condensors its dumped back in the river, cleaner then ever.
Google is currenlty working on a system where serverparks are only online at night. Since they have so much datacenters in different timezones its more ecological (and cheaper) to only have the datacenters online where it is nighttime. This way they use power that is much cheaper (night tariff rates) and ecological because at night less power is consumed but nuclear reactors still keep pumping out the same power (you can't turn a nuclear reactor off). Since it doesn't really mather on the internet where the servers are this is a very first and new green strategy being applied by a company. This means however they need twice as much servers as normal since half of them would be powered down at all times.