Rising Emissions From ‘Benign’ Technologies

Many people may naively assume that, in the era of climate change, the more modern the technologies, the smaller carbon footprint they will have. We have seen, however, that in one way or another, several technologies have gained currency by cleverly disguising their polluting nature behind apparently greener garb. Fracking for extraction of shale oil as in the US, or shifting to natural gas from oil for power generation and industrial uses in Europe, or promoting blue, grey or other such versions of hydrogen instead of hydrogen produced using renewable energy.

Much less is known, and even less appreciated, about the dangerous and rising emissions from what appears to the average person, as benign technologies that have transformed life in so many ways, and which look set to play an ever-greater role in the coming years.

Computing is one such area where companies, users and regulators are so focused on convenience, service delivery and lowering costs to consumers, that pollution in general and greenhouse gas emissions are scarcely even thought about.

Today’s laptops, tablets and smart phones have hundreds of times greater computing capacity as the room-size computers that took astronauts to land on the moon and brought them back. These devices take just an hour or so to fully charge and can be used for many hours after. People watch movies, ask questions and, nowadays, get answers not just from the search engine but from AI (artificial intelligence) services, personalised according to millions of bits of information on prior internet usage.

AI is becoming ubiquitous, used across the spectrum from science and technology, medicine and industry to cinema, the arts. Nobody thinks or worries about where large volumes of information users save are housed, they just access them with a simple click or swipe. Crypto-currencies are bought and sold with little thought given to just how these are created. Companies, which used to house large servers for their businesses, now use “cloud” services, where vast amounts of data are stored off-premises, invisible. It all seems like a magical, new world, without real environmental costs.

The “cloud” sounds wispy, other-worldly, even unreal. The reality is scary, dangerous and certainly not pretty.

Digital Emissions                 

AI services, crypto-currencies, processing of “big data,” such as consumer shopping behaviour, facial recognition, tracking of internet or social media messaging and so on, despite their apparently “invisible” nature, involve very real brick-and-mortar buildings housing enormous servers with huge computing capacity, all of which consume vast amounts of electricity, water to cool these systems, extra-large fans to aid in cooling and so on. More electricity means more emissions. Of course, if renewable energy is used, emissions would be less than with fossil energy, but that poses additional problems which we will deal with later.

It is hardly surprising that emissions from the digital sector have    skyrocketed in recent years, when one considers that there were one billion internet-connected devices in the world in 2010, increasing rapidly to a projected 50 billion by 2025 and 100 billion by 2030. Emissions from the digital sector, now at roughly 3.5% of global emissions, have now overtaken those from aviation which are around 2.5%.

Crypto-currency “mining,” a complex computing process to “create” a roughly realistic “currency” to mimic -real money, is again a data and energy-hungry process that accounts for roughly 0.7% of global emissions.

It is roughly estimated that about half the emissions from the digital sector arise from the production side and half from the use. It is expected that, with the rapid rise in internet usage, especially through smart phones and with galloping deployment of AI, emissions from use of digital services will multiply manifold.

Problems in Estimating emissions

A big problem with more accurately estimating emissions from digital services is that reliable data are not forthcoming from the industry itself. No standardised metrics or estimation protocols have been evolved unlike for other industrial emissions sources.

Some information is available about the processing side. By and large, it is known that data centres where cloud computing, data storage and AI processing takes place, not to mention crypto-currency “mining,” are energy hungry, accounting for around 2% of global electricity consumption and roughly 1% of emissions, both these figures dating to before the recent explosion in AI services and widespread deployment of AI software and related hardware by the big internet companies.

A recent study by the International Monetary Fund (IMF) has estimated that electricity consumption from data centre and crypto-currency would reach about 3.5% of global electricity consumption, roughly equal to that of the fifth largest country, Japan. However, accurate assessment is made difficult by the business models used by different companies, for example outsourcing data-processing and cloud storage to other companies in different parts of the world.

However, there has been little study of emissions caused at the usage end. Some recent research at MIT (Massachusetts Institute of Technology) and other places has sought to tackle this problem.

For example, creating one image using AI consumes as much electricity as fully charging a smart phone. A search using Chat GPT uses 10 times more electricity than a simple Google search which, incidentally, is getting difficult to do due to deployment of AI searches now increasingly becoming the default option on most smart phones. An AI software may consume 33 times more electricity than ordinary software. The World Economic Forum has estimated that the computer power dedicated to AI is doubling every 100 days.

The Big Five

The top five users of electricity, and largest emitters, in their data centres are, unsurprisingly, Amazon, Google, Microsoft, Apple and Meta (Facebook). Amazon is by far the largest emitter, responsible for double that of the next biggest emitter, Apple. Most analysts leave out Amazon from their studies since their data usage and computing models are quite different from the others, making comparisons difficult.

An investigative study by The Guardian newspaper estimates that the emissions of these four, excluding Amazon, could actually be 7.62 times or 662% higher than their official reports have claimed, partly due to business outsourcing practices and partly due to non-transparent “renewable energy certificates or RECs,” similar to the notorious carbon offsets in climate change emissions control regimes.

Analysts say that while Apple and Meta use such “creative accounting” practices to show less emissions than are actually taking place, Google and Microsoft may be somewhat more transparent.      

Google has set a “24/7” target, that is, has promised to run all its data centres on renewable energy 24 hours a day and seven days a week by 2030. Microsoft’s equivalent promise, called “100/100/0” is a pledge to use 100% renewable energy 100% of the time, buying-in 0 or no fossil-energy, again by 2030.

AI, Data Centres go Nuclear

But as the electricity demands rise steeply with AI deployment, the challenges are mounting for the data centres. Unlike most industrial, commercial or domestic users of electricity who have at least some day-night variations in electricity use, demand by data centres and global users remain more or less the same all day and around the year.

Pressure on the grid due this huge and mounting demand is posing serious problems. Data centres, whose power demand can be as high as 800-900 MWe, therefore, require their own reliable, round-the-clock, non-fossil electricity supply which cannot be assured by solar, wind or large hydro.

Enter nuclear power, an unanticipated scenario even five years ago.

Recently, a company called Constellation Power signed a 20-year power-purchase agreement with Microsoft for electricity supply to the latter from the infamous Three Mile Island nuclear power plant. Unit 2 in the plant suffered the US’ worst nuclear accident when it had a partial meltdown of the reactor in 1979. The current agreement brings to life Unit 1, which had been shut down in 2019 due to doubts about its viability. In March, Amazon Web Services (AWS) contracted for 960 MW power from Talen Energy’s nuclear power plant in Pennsylvania.

There is a burst of activity around small modular (nuclear) reactors or SMR which is seeing a revival of interest, technology innovation and investment.

So now the future of the internet is to depend on nuclear power? Watch this space.

The writer is with the Delhi Science Forum and All India Peoples Science Network. The views are personal.