One of the areas where Energy Efficiency is in their DNA are remote sensor engineers. The Microsoft Research team who wrote this paper are responsible for sensorweb and were Xerox PARC guys.

I found this article on the Economist, and it reminds me of some people who are investigating battery operated sensors in the data center to collect power and environmental data. Why would someone go with battery power, because the easier you can install a device in data center on its own network and power, the easier it is to deploy. The costs to deploy monitoring solutions that assume they'll plug into the existing data center network infrastructure will many times be more than the cost of the equipment.

To do that, they need to float near the sea floor, since most of an earthquake's energy travels through the rock rather than the water. So a Mermaid can operate at a depth of up to 1,500 metres (about a mile). When she hears something that might be pertinent, she runs the signal through her on-board computer to decide just how significant it really is. If it does turn out to be significant, she surfaces by pumping air into a bladder and makes contact with a satellite that has been co-opted into the project. Once she has delivered her message, the air is sucked back out of the bladder and she returns to her gloomy underwater station.
The main engineering problem Dr Simons faces—apart from making something that will work reliably in the salty ocean depths—is energy conservation. When a Mermaid runs out of power, she dies. That power is provided by lithium-ion batteries and is reckoned sufficient for between 50 and 100 surfacings.
One of the ways Dr Simons saves power is in the computer. The decision to surface is made by an algorithm that depends on a mathematical function called a wavelet. This divides an earthquake wave into separate components which can be studied independently. That allows the computer to restrict energy-intensive high-resolution analyses to those sections of the waves that really need it. The other sections receive a more cursory (and thus less power-consuming) glance.