The energy and water consumption habits of datacentres often see them painted as the villain in discussions about climate change prevention, but the situation is not quite so cut and dried as that, says Dermot O’Driscoll, vice president of product solutions within the infrastructure line of business at chipmaker Arm.
Datacentres are a paradox. They are some of the most important buildings in the world – the epicentre for global commerce, scientific progress, and public debate. At the same time, they are anonymous: most look like a shoe warehouse surrounded by razor wire.
The same dichotomy pervades their environmental profile. A hyperscale datacentre can consume up to five million gallons of water a day and usually require megawatts (MWs) of power capacity. Don’t forget real estate: the largest datacentre today, the Citadel outside of Reno, occupies over 7 million square feet.
However, datacentres are also emerging as one of our chief means for fighting climate change. The Exponential Roadmap Initiative estimates digital technology accounts for 1.4% of global emissions but holds the potential to reduce overall emissions by 15% by 2030, with datacentres serving as a nerve centre for efficiency efforts.
Continuing to make datacentres more efficient, while bringing these new applications for reducing carbon elsewhere, is essential for laying the foundation for a more sustainable world.
1. Paving the Way for Virtual Power Plants
Utilities have little visibility into how much power you are consuming at any given moment and even less control over how much you use. To ensure against blackouts, they invest in so-called peaker plants that cost several million dollars, spew emissions and only get used a fraction of the time.
Cloud platforms from companies such as AutoGrid and Voltus effectively predict power demand and take actions to reduce loads at factories or office buildings. Everybody wins: customers avoid hefty demand charges, utilities can reduce their capital equipment footprint and communities get cleaner air.
Datacentres help in two ways. First, they make managing power in the cloud realistic. Historically, the core technology for managing power demand was the phone: a utility would call big commercial customers and ask them to dim their lights or move energy-intensive activities to night-time.
Today, the cloud provides scale and automation, tapping into more customers and more appliances to save power more often.
Second, datacentres serve as a lab for these applications. Google shifts workloads to evening hours when power is cheaper or cranks up servers when renewables are producing at their max. The company is also looking at ways to roll applications to different regions. Their experiences will percolate to buildings, factories, and other facilities.
2. Replacing real power plants
Bloomberg New Energy Finance estimates that uninterruptible power supplies (UPS)— the big battery packs that guarantee an even flow of power—in European datacentres alone could contribute between 3.8GW and 16.9GW to the grid by 2030. That’s the equivalent of having two to eight nuclear power plants on reserve.
UPS-on-demand could also serve as a source of revenue for mid-sized cloud providers or companies that still want to operate in-house datacentres too. In this scenario, utilities will pay substantial sums for power on demand. Battery demand, of course, would also lower prices and further encourage battery systems at universities, office buildings, and downtown business districts.
Smart battery systems would also help reduce that other scourge of datacentres: backup diesel generators. A single 3MW backup diesel generator can emit 1.6 million tons of greenhouse gases per year and cost over $1 million to operate. There is an estimated 20GW of backup generator capacity parked worldwide at datacentres.
3. Jumpstarting water innovation
Over 50% of the world will live in water-stressed areas by 2050. Water also has a deep carbon footprint: 30% of the cost of water revolves around electricity.
Datacentres are paving the way for one of the most promising techniques for solving this problem: water recycling. E.g. cleaning wastewater and injecting it back into the system. Singapore already gets up to 40% of its water from its NEWater recycling program and even though most recycled water doesn’t get rerouted for human consumption, the concept still gives pause to many people.
Hyperscalers are already taking advantage of pools of old industrial wastewater in the state of Georgia, Ireland, and elsewhere to cool their facilities. Korea’s Tomorrow Water wants to build datacentres at sewage treatment plants. It would cut water processing costs, help fund new infrastructure, and the land next door is often cheap.
Additionally, we’re seeing a lot of efforts to improve the Water Usage Effectiveness (WUE) of datacentres, which is a measure of water conservation for server farms, where a lower number is better. Facebook has a WUE of 0.3 liters per kilowatt-hour. The industry average is closer to 1.7.
4. Optimising natural resources
Food production will need to double by 2050 to avoid food scarcity and famines. Yet up to 35% of the grain harvested each year can be lost to spoilage before it gets to consumers.
Telesense, a startup out of Berkeley, California, uses a sensor (containing a processor from Nordic Semiconductor) that links to the cloud to actively monitor carbon dioxide, moisture, and other parameters to help grain distributors curb losses. Cargill, ADM, and other large granaries are experimenting.
Reducing losses minimises the need for new cropland and recovers the value of embedded energy and water that would have been lost. There are similar cloud-based efforts in mining, livestock, shipping, and other fields.
5. Cutting the cost of renewables
Xcel Energy increased the overall productivity of its wind farms in Colorado by $46 million in part through better real-time forecasts for wind and power demand.
Similarly, solar producers can minimise repair costs with predictive analytics. Cutting half of the maintenance from a single, moderately-sized 50MW solar power plant can mean $560,000 of additional profit a year, according to Steve Hanawalt, CEO of PowerFactors, and that’s only one plant. The United States alone has close to 21GW (21,000MW) of solar and 118GW of wind.
And in your home, companies like Enphase, Span and SolarEdge use smart devices linked to the cloud for modulating power in your home (and riding out blackouts in style). The bottom line: software and silicon help eliminate cost and other issues with renewables.
6. Decarbonising compute
Datacentre energy consumption grew dramatically between 2000 and 2010, then leveled off for a decade. Experts now debate whether energy can stay flat in the face of rising demand and new technologies like artificial intelligence or whether it will start to rise.
Although much of the low-hanging fruit might be gone, new ideas are already showing how more data traffic and larger workloads can be accommodated without necessarily increasing power. Hyperscale cloud providers are already showing how specialized processing— i.e. building datacentres with Data Processing Units (DPUs) and other cloud-optimized processors instead of traditional chips—can increase operations per watt by 50% or more.
Likewise, carriers are experimenting with new processors and computing architectures to keep 5G energy efficient while still delivering higher speeds and data rates. More performance won’t have to mean more energy.