The Carbon Footprint of Your Online Habit

February 11th, 2010

A 2-minute talk given at TED-U
Friday 2/12/2010.
Neil Hunt


Is your online habit cooking the planet?

We read scare stories about how new large datacenters and cloud computing facilities are going to overtax our electricity grid, but is there a prize to changing your online behavior?
Before you can decide how to change things, you should quantify to make sure you are fixing the right problems.

Your take aways should be:
1. The expensive part is all the stuff that isn’t shared with millions of other users – the stuff in your home, and shutting it down (or suspending) when not in use gains almost all the benefit.
2. Almost any physical media, communications, or travel is going to be magnitudes more expensive in carbon than online equivalents.
3. Moore’s law means that things will only get better (less CO2) in the future.


The average US consumer emits 20 Tonnes per year of CO2 (Source: Google, Wikipedia).

That’s about 5x the world average. It’s about 1kg each half hour. What IT can you get for that much emissions?


1GB of emails or web-pages, blog posts, or tweets costs about 1kg of CO2.

Source: personal knowledge of Netflix datacenters: without revealing actual numbers, the ratio is 100kW of power and cooling generates 75GBytes of data out. Using the US average 600g CO2 per kWh gives roughly 1GB -> 1kg CO2. (Sources: US DOE and DOE). In practice many datacenters are built where hydro or natural gas power is available, significantly reducing this figure — natural gas is about 300g per kWh, and pure hydro is arguably near 0.

Using 100kBytes per email (seems large, but that’s the average size of email in my current folder), or 50kBytes per web page gives 10,000 emails or 20,000 web pages per kg CO2.

And that’s the expensive, personalized data that’s generated or processed individually and not shared with millions of other users, driven by large stored databases of personal preferences and history.


Generic data – such as images, audio, photo and video assets are much cheaper – about 500GB per kg CO2.

Source: personal knowledge of Netflix CDNs built in 2008. A single rack of 10 blades and some switch gear is provisioned with 30A x 220V, but runs at a fraction of that after the power-up surge. Estimating the steady state at half surge peak, and cooling as the same as the power input, we get 6kW or 4kg CO2 per hour. This rack has a network capacity of 20Gbps, a measured peak output of about 15Gbps, on a daily cycle the average is about half the peak, and derating for operating growth, we operated these CDNs server racks at about 5Gbps average, or a little over 2GBytes per hour, for 2g CO2/GByte or 500GBytes/kg CO2.

You’d have to watch 20 Netflix videos at once to emit CO2 as fast as you breathe it out.  (20 x 2Mbps = 40Mbps = 18GBytes/hour = 432 GBytes/day -> 1kg CO2 per day).  Source for human respiration emission 1kg CO2/day: EPA.


The cost of transit is nearly negligible, in part because the power-hungry gear is shared between millions of users, and is kept at high utilization, deploying new equipment only when existing is near full capacity.

If you doubt that the costs are negligible, imagine the few pennies it costs a business to purchase 1GB of transit, and imagine all that as purchasing electric power, and you’ll see that it’s way smaller than 1kg CO2. Without revealing our actual costs, if we take a very high estimate of $.05/GB were spent on power for transit, it would purchase about 500Wh, or about 300g of CO2. The actual numbers are quite a bit lower.

However, the last mile is not shared with millions of others, and it’s always on, whether you use it or not. The components I’m counting here are are include a cable headend servicing 400k customers, a hub covering 30k customers, and a node shared by 800 customers, for perhaps 4W total per customer.  Most of this equipment equipment and power consumption is shared with transmitting the broadcast TV signal, and it is unclear how much power is incremental to support the over-the-top data.  For a rough guess, I split responsibility in halves for TV and data.

Thus I am estimating roughly 1kg per month of use, whether you use use 1Byte or 1Terabyte.


And the equipment in your home isn’t shared at all!  A 5W cable modem and 10W WiFi costs 7kg per month!

Perhaps you are beginning to spot a trend…
The less shared the equipment, the bigger it’s footprint.


If you have a nice efficient laptop (60W) that’s always on, that’s 30kg/month or 2% of a typical person’s total!

If you just close your laptop lid at night, you’ll save half your total IT CO2 emissions


When I sum up my rather extensive online habit, I get about 20kg CO2 per month, or about 1% of my overall CO2 emissions. That assumes I suspend or hibernate the PC overnight (10 hours per day duty cycle), and ignores other in-home switches, external monitors, printers, and external hard-drives, most of which are either shut down or suspended except when in use. In particular, I have set my PC to power-down the monitor after a few minutes of inactivity, which cuts the monitor power consumption from 100W to 1-2W.


How could I offset 20kg per month in other areas?


I could hold my breath for 3 weeks. Unlikely to be a sustainable strategy.


Or I could cut out meat for 4 meals a month…

American food is an extremely inefficient energy delivery medium. Energy goes into agricultural equipment, fertilizers and pesticides, harvesting, processing, storage, and transport.  The beef-cow itself yields only a fraction of the calories it consumes, so only a tiny fraction of the total input is recovered by the eater.  Furthermore, cattle emit considerable quantities of methane, especially in the common case of cattle fed on corn in a feed lot. Methane is considered to be 20 times more potent as a greenhouse gas the CO2. Many articles have contemplated the energy content of meat. This one appears well researched and pegs the CO2 equivalent of a burger between 3 and 6kg.

Vegetarians enjoy about a magnitude better food growing efficiency than meat eaters, since they don’t pay the energy loss of growing the animals.


Not flying 30 miles, or equivalently, leaving behind 30 pounds of luggage for my trip to TED, would save me 20kg.

Calculation: flying is tricky to quantify, since such a large amount of energy is consumed to get the plane airborne for zero distance travelled, and since the fuel itself is a large fraction of the takeoff weight, necessitating extra fuel to carry the fuel. This leads to some fuel consumption figures that are exponential in payload weight and distance. I used 10kg/km for level flight, 8000kg/takeoff and climb to altitude (covering 250km of distance), assumed 350 passengers and 1000km flight duration averages to get my 30 mile figure. Since my personal journey to TED was about 300 miles, I could extrapolate wildly and assume that not taking 30 pounds of baggage (which is 30/300 of my personal payload of 200 pounds of me and baggage + 100 pounds of seat, cabin staff, drinks, etc.) would equate to 20kg of CO2.


Or I could stay at home from work 2 days a month – and work online…


But if I choose to ride my bike to work instead, and then fuel 300 calories per hour per 15 miles with one burger it’s only slightly more efficient in carbon than driving. Wow – that’s stunning.

On reflection, perhaps it’s not so surprising. Perhaps there’s an order of magnitude more energy goes into the vegetable food I eat than I get out of it, and animal food implies another order of magnitude inefficiency. Feeding meat to a human is an incredibly inefficient way to fuel physical work.

But if I fuel vegetarian, I get back one of those magnitudes, and get 50 miles per kg instead of 5.

Also, 300 calories per hour is rather conservative – since I pound the pedals and climb two significant hills on my commute, I suspect my supplemental energy consumption is more in the 500-800 calorie range for the 15 mile trip, which would, if meat-fueled, potentially put cycling in a less-efficient zone than driving.


Our online habit replaces a lot of physical product – especially media. For example, manufacturing a single DVD emits 1kg – and you still have to power the DVD player and TV to watch it!
That’s from an unpublished internal study commissioned by a studio. It covers manufacture, packaging, and distribution into the channel (e.g. Walmart). I’m ignoring your cost to get it home, because I’m assuming you purchased it at Walmart serendipitously while buying necessities, or more favorably, because you ordered it from Netflix, and the incremental cost of shipping it via USPS to your doorstep is already covered by the cost of delivering 5 pounds of junk mail to your doorstep.

You should compare this figure with the 500 hours of streaming video that can be delivered for 1kg of CO2.


Overall, your habit probably costs 1-2% of your total CO2 emissions, but probably saves much more than that in physical products, communications, and travel.

My recommendations:

  • Stay online.
  • Suspend your computer when not in use – at least overnight.
  • Skip physical media.
  • Ride your bike (the health benefits will make it worthwhile) – but go vegetarian at least 4 meals a month!

Also, Moore’s law works in our favor. It’s very likely that the all the technology components will get twice as efficient every 18 months on average. Some of them, like your CPE and computer won’t be replaced as fast as every 18 months, but when you replace after 3 years, they may be 4 times as efficient. Some of that will be used up for newer, richer, more personalized services (or higher resolution video streams). But for sure the energy cost per GB served will come down too.

Another favorable development is the migration of more and more of the server technology to cloud-based servers. Sharing the physical hardware on an hour-by-hour basis between different uses allows cloud datacenters to achieve utilization factors up to 40%, vs. 15-20% as the pinnacle of utilization in single-user datacenters. (From this post, personal communication with AWS staff, and personal knowledge of Netflix datacenter utilization factors).