Data Center Sustainability: Making the Internet Green
When it comes to business’ carbon footprints, we often hear about the importance of flying less, offering more sustainable commute options, and eliminating single use plastics in the office. But rarely is data center sustainability ever mentioned.
Data centers are the engines that drive the internet, and as more organizations around the world embark on digital transformation, data centers risk causing serious harm to our environment. That is, unless concerted efforts are made to make data centers more sustainable. This article explores the environmental impact of data centers and what organizations are doing to lessen that impact.
Table of contents
The climate impact of data centers
Carbon emissions are probably not top of mind when most people jump on a Zoom or Teams call. But every time you stream a movie, send an email, or load a web page, somewhere a server works to compute the request. That requires power—lots of it.
In 2021, data centers around the world used between 220 and 320 Terawatt hours (TWh) of electricity. For context, the United Kingdom used 294.4 TWh of electricity in the same year. It’s estimated that data centers currently account for one percent of global carbon emissions.
Beyond power, data centers also require significant amounts of water for cooling, carbon-intensive raw materials like steel and concrete for construction, large tracts of land for all those server racks, and, of course, new equipment and replacement parts.
We’ll get to these other impacts in a moment, but first let’s take a look at where data centers use electricity.
How do data centers use electricity?
Data centers are big operations that require a lot of electricity to maintain uptime. Though it’s important to note that most data centers are not gigantic buildings, we’re talking specifically about hyperscale data centers—that is, facilities ranging in size from 10,000 ft2 up to over 400,000 ft2. Despite accounting for a small fraction of the market, hyperscales process the vast majority of data transmitted in the cloud.
In general, data center energy consumption can be split into four sectors:
- Infrastructure. Sometimes called overhead, infrastructure refers to the basic functions of a data center building, like cooling, humidity control, and lighting.
- Network. The hardware that facilitates the transfer of data within the data center network and out to the internet. This includes switches, modems, and routers.
- Storage. The hardware used to store data, both for backup and recovery and for long term storage. These can be tape drives or hard drives.
- Servers. A core component of data centers, servers are the heavy duty computers that perform compute functions and communicate with each other through networking.
Of these four sectors, infrastructure and servers consume the most energy by far. Each represents nearly one half of total energy consumption, with network and storage accounting for much less.
For example, energy consumption might be split as follows: 44 percent for infrastructure, 44 percent for servers, 10 percent for storage, and 2 percent for network.
Water usage at data centers
In addition to electricity, data centers also use a lot of water for cooling.
According to a peer-reviewed paper from Imperial College London in Nature, data centers in the US use 1.7 billion liters of water per day. A medium-sized data center uses as much water as three hospitals and more than a standard golf course.
Servers emit a lot of heat, especially during periods of increased workload. To prevent servers from overheating, data centers must be cooled, usually in one of three ways:
- Chillers. Heat generated by servers is transferred to water through heat exchange, which is then cooled and recirculated throughout the data center to move heat from inside the building to somewhere else, usually outside.
- Cooling towers. Hot air is exposed to water—usually by blowing the air across a damp filter or mat—wicking away heat as it evaporates, similar to the way sweating helps people cool down.
- Adiabatic economizers. Water is sprayed onto heat exchanger surfaces or directly into the air entering the data center, cooling the air inside.
These are effective ways to lower the temperature inside a data center, but they use a lot of water. What’s more, many data centers use potable water, which can put a serious strain on the local supply of drinking water and water for crops, especially in dry areas.
Land use of data centers
As previously mentioned, hyperscale data centers are large operations. According to Virginia Mercury, data centers require at least 30 acres of land, while an article in Data Center Dynamics says a 1:1 acre to megawatt (MW) ratio is a good rule of thumb for estimating data center plot size.
For example, Microsoft is building a new $400 million data center on a 67-acre plot of land in Loudon County, Virginia—one of two counties in Northern Virginia that constitute Data Center Alley. Data centers are built in geographically strategic areas, with availability zones consisting of one or more data centers to cover a region, like the Eastern US, for instance. Smaller data centers called edge locations are located in more densely populated areas to reduce latency as much as possible.
While some data centers (especially edge locations) can use existing buildings, the larger data centers used for availability zones usually require new construction. This entails all the normal environmental impacts that come with large construction projects, including clearing land, operating heavy machinery powered by gas or diesel engines, and using carbon-intensive building materials like concrete and steel.
E-waste from data centers
E-waste is one of the fastest growing streams of landfill waste, with global e-waste generation projected to reach 120 million metric tons annually by 2050, according to a report from the World Economic Forum. This form of toxic waste is mostly generated by Europe and North America, with IT and temperature exchange equipment making up a sizable portion of total e-waste generated each year.
Data center and cloud computing providers have come under scrutiny in recent years for their contribution to these startling figures, and in response, many of them have made efforts to transition to a circular economy. More on that in a bit.
How data centers can be more sustainable
So what should cloud providers and data center operators do about their climate impact? Let’s break it down by the impact areas we covered above.
First, data centers should decarbonize. That means using electricity generated by renewable energy sources like wind, solar, and hydroelectric instead of fossil fuels. While data centers can generate some electricity onsite through sources like rooftop solar, they’re largely at the mercy of their electric utility provider when it comes to deciding how electricity is generated.
That being said, data centers can offset the dirty electricity they consume by purchasing renewable energy certificates (RECs), and they can increase the supply of renewable energy feeding into the grid by signing power purchase agreements (PPAs). Doing so can help a data center achieve 100 percent renewable energy with the ultimate goal of running on 24/7 carbon-free energy (CFE).
Decarbonization is already a huge challenge, one that will be much more difficult with inefficient data centers. Data centers measure energy efficiency using a metric called power usage effectiveness (PUE), which is a ratio of total facility power consumption to IT equipment consumption. The closer that ratio gets to 1, the more electricity goes to powering servers instead of building operations.
There are many ways to boost PUE, from the physical architecture of data center buildings to using new and improved servers. But in recent years, virtualization has been a major driving force in bringing average global data center PUE closer to 1.
By effectively breaking up servers into multiple virtual instances, data centers get much more compute power out of a single server than they would if they ran everything on bare metal. This significantly reduces the need to buy as many servers, lowering energy consumption and increasing processing power per square foot.
Slash water consumption
As previously mentioned, data centers use a lot of water. Cooling accounts for the majority of water consumption at data centers, but operators should also consider how much water data centers use indirectly through power usage.
Fossil fuel and nuclear power plants use a lot of water to generate electricity, and that all factors into a data center’s environmental impact. This makes decarbonization all the more important, but that alone won’t meaningfully reduce how much water data centers use. To make a difference here, data centers should focus on using non-potable water (like rainwater) and implementing less water-intensive cooling methods.
Joe Capes is the CEO of LiquidStack, a Massachusetts-based company that offers a cooling method called 2-phase liquid immersion. LiquidStack submerges servers in airtight data tanks containing non-conductive fluid. The servers generate heat, causing the fluid to boil. The resulting gas vapor then condenses on chiller coils in the tanks to reject that heat before dripping back into the tank to be boiled again.
“Because we’re basically rejecting the heat from the servers as a vapor gas inside the tank and then recondensing that gas into a liquid…we can use water from the chiller plant at a much higher temperature,” Joe says. “As a result…you can essentially free-cool or reject that heat without any adiabatic assist, any pre-wetting of the chiller coil.”
Using 2-phase liquid immersion cooling, Joe says hyperscale data centers can reduce water usage by up to 50 percent, reject 21 times more heat per server rack, and reduce electricity consumption by as much as 41 percent. These kinds of efficiency gains are crucial as the climate and ecological crisis drives a global shortage of freshwater.
Reduce land usage
Hyperscale data centers take up a lot of space. That means clearing large tracts of land, a process that damages ecosystems at a time when animal populations have experienced a 70 percent average decline since 1970.
Data center and cloud providers can minimize this impact through compaction and better facility design, but Joe says the industry is also seeing a trend towards smaller regional and local edge locations. These are smaller data centers that can fit into the built environment, reducing the need to clear virgin land.
“A typical edge data center might be in the range of 250 kilowatts to maybe 2 megawatts, versus a traditional hyperscale data center, which is 40 megawatts to 480 megawatts,” Joe says. “That also means we’re going to see data centers in places where we’re not used to seeing them—in urban environments, street corners, building facades. That means compaction becomes really, really important.”
This is especially true in developing countries, where Joe says data center providers are leapfrogging over centralized hyperscale data centers to build smaller edge locations. Smaller, decentralized data centers could offer a more sustainable future for the industry, especially ones that can be housed within existing buildings.
Shift to the circular economy
Last but not least, data centers should switch from the linear economy to the circular economy—an ideally closed-loop system where resources are reused, recycled, refurbished, or resold instead of sent to landfill.
Data centers can switch to the circular economy by composting food scraps from facilities, buying hardware made from recycled materials, reusing old equipment for different applications, recycling equipment that can’t be reused, and by refurbishing old equipment for resale to other organizations who can still use them.
This extends beyond IT equipment. Some data centers collect rainwater to help cool their servers, while others are exploring recycling waste heat from servers for district heating, greenhouse farming, and even generating electricity, in the case of Montreal-based startup NovoPower.
The most sustainable cloud providers
The cloud provider you choose can make a big difference in your organization’s overall carbon footprint. Migrating to the cloud is already a good way to reduce emissions, and choosing a cloud partner that prioritizes sustainability at its data centers can pay dividends for emission reduction efforts.
When evaluating a cloud provider’s environmental impact, look for these stats:
- Data center power usage effectiveness (PUE). PUE is a metric of how energy-efficient a data center is, expressed as a ratio of total facility energy usage to IT equipment energy usage. The closer to 1, the more efficient the data center.
- 100% renewable energy. This is a similar concept to carbon neutrality. The name can be a bit confusing, but this stat does not mean that a cloud provider didn’t use any electricity generated from fossil fuels. Rather, providers who achieve this milestone offset their dirty energy usage with power purchase agreements (PPAs) and renewable energy certificates (RECs) for renewable energy like wind or solar.
- 24/7 carbon-free energy (CFE) targets. Hitting a 100% renewable energy goal is laudable, but it’s even better to achieve 24/7 carbon-free energy. Unlike 100% renewable energy, 24/7 CFE matches electricity usage every hour with zero-carbon electricity generation. Achieving this is quite a feat, since renewable energy is naturally intermittent.
- Water usage targets. More cloud providers are setting goals to be water-positive by a certain year—that is, to put more water back into the communities where they operate than they consume.
- Zero waste commitments. Data centers can be huge contributors to the growing e-waste crisis. Zero waste commitments aim to reverse that trend by reducing the need for additional hardware, selling used or refurbished hardware, or recycling old hardware when possible.
With that in mind, here’s how the Big Three cloud providers stack up on environmental impact based on their most recent sustainability reports.
|Average data center PUE (compared to industry average of 1.57)1||1.07-1.15||1.23||1.10|
|100% renewable energy||By 2025||By 2025||Carbon neutral since 2007, 100% renewable energy since 2017|
|24/7 carbon-free energy (CFE)||No commitments at time of publication||By 20302||By 2030|
|Water positive||By 2030||By 2030||By 2030|
|Reducing waste3||Amazon is a member of the Ellen MacArthur Foundation, which works to speed up the transition to a circular economy. Other than that, language surrounding circular economy efforts at AWS are vague and point mostly to recycling.||Commitment to zero waste by 2030 in addition to other circular economy initiatives like reusing servers and components, making fully recyclable products and packaging, and investing in companies driving circularity||In 2016, Google committed to achieve zero waste in all of its global data centers, and it's been a member of the Ellen MacArthur Foundation since 2015. It has a number of circular economy initiatives, including a virtual accelerator for startups supporting a circular economy.|
1Reporting methods for this figure vary across the Big Three cloud providers. AWS reports an average enterprise data center range; Azure reports both a PUE design goal and actual PUE, the latter of which I used to calculate an average global actual PUE of 1.23; GCP reports a comprehensive trailing twelve-month (TTM) PUE.
2Microsoft calls this its “100/100/0 commitment,” meaning “100 percent of Microsoft’s energy supply, 100 percent of the time” will come from zero-carbon energy sources
3More information on zero waste and circular economy initiatives can be found in each organization’s sustainability report, which can be found here for Amazon (AWS), here for Microsoft (Azure), and here for Google (GCP).
And the most sustainable Big Three cloud provider is…
Google Cloud Platform (GCP) is probably the most sustainable choice of cloud providers among the Big Three, though it’s worth noting that all three companies are making real progress towards some very impressive goals. Google is widely recognized as a sustainability leader in cloud computing, and they’ve generally been ahead of the curve in regards to circularity, water conservation, and 24/7 carbon-free energy.
In second place is Microsoft Azure, with AWS bringing up the rear. Both are behind GCP in terms of renewable energy commitments (although Amazon as a whole is the largest corporate buyer of renewable energy in the world). AWS was the last of the three to announce a water-positive commitment, and it also lags in data center circular economy commitments, as far as we can tell.
Ultimately, this is a hard call to make in terms of which provider delivers the greatest carbon reductions for its customers. All three providers offer some kind of carbon tracker tool for cloud customers to see the carbon footprint of their cloud usage, so those are worth exploring no matter which provider you choose.
Cloud still has the best track record for data center sustainability
Data centers are highly complex operations, the largest of which consume more electricity than some cities. For most organizations operating their own data centers, the most sustainable option is to migrate their workloads to the cloud.
GCP, Azure, and AWS operate the greenest data centers in the world among hyperscale providers, owing in no small part to their massive purchasing power and technical resources. For organizations working hard to hit climate targets, migrating to one or more of these three cloud environments is a positive step.
Scalable cloud talent, on demand
Hourly, contract, contract-to-hire, full-time—we've got you covered. Explore our cloud talent solutions today to get the help you need.
Whether you’re a developer looking to expand your programming literacy or a recruiter writing a job post or prepping for an interview, understanding the differences between C++ and Java is important. Read on to learn more about what makes these programming languages...
If given the chance, would you want to live in a synchronous world? “Nope,” said Dr. Werner Vogels, VP and CTO of Amazon.com, in a The Matrix-inspired opening video to his Thursday keynote address at AWS re:Invent 2022. What follows is an overview of Dr. Vogels’...
With the amount of data organizations collect paired with tools to store, analyze, and make predictions from that data, it’s tempting to think of data functioning in an organization as it would in the brain. That’s the goal in many ways, but we aren’t quite there yet,...