Don’t refresh the page, your Internet connection is fine | Our data problems are getting harder to ignore

Data is invisible until it’s not there. On Oct 14, a data centre issue caused an outage affecting DBS Bank and Citibank customers in Singapore for over eight hours halting 2.5 million transactions.

Disruptions to data flow do not only cause immediate inconvenience, but also hint at the broader issues tied to data handling, such as energy consumption and data security.

As we rely more on advanced technologies, like artificial intelligence (AI), which uses substantial energy, the environmental footprint of our digital dependence becomes more pronounced, a topic that was discussed at this year’s COP28 held in Dubai from Nov 28 to Dec 12, 2023. But thanks to the inherent digital efficiency of IT, data can present both challenges and solutions for sustainability.

So, what’s the problem? Well, this page you’re reading right now, for example, is pushing data onto the screen from a data centre.

Every time you open a webpage, your browser requests data for items such as text, images and code. While this process can involve large amounts of data, the Internet uses techniques, including caching and compression, to reduce data transfer and load times.

Here’s an example.

The first time you visit the Straits Times homepage, your browser might request 21MB of data, but due to compression, only 10MB is actually transferred.

On subsequent visits, you may load just 600kB of data, thanks to caching, where your browser stores data temporarily, allowing for faster, more efficient page loading.

Data, from the photos on your phone to enterprise databases, require storage and processing, which is where data centres come in. Think of them as the “brains” of the IT sector, says Professor Wen Yonggang from Nanyang Technological University’s (NTU) School of Computer Science and Engineering.

Your request for data, such as a webpage, travels through a network of cables and wireless connections to a data centre. Your request is processed and the necessary information is sent to you.

Data centres vary in size and scope. Smaller, local “edge” data centres store frequently requested data, speeding up requests by reducing travel distance. Meanwhile, larger “hyperscale” data centres serve millions of users worldwide with enough servers to handle substantial demand.

Globally, data centres and transmission networks are responsible for 1 per cent of energy-related greenhouse gas emissions, according to data from the International Energy Agency (IEA).

The good news? Despite sixfold growth in Internet traffic between 2015 and 2022, energy use for data centres has grown moderately, thanks to improvements in efficiency, according to data from IEA.

However, as technology evolves, so does our behaviour. Developments in 5G, for instance, mean we can transfer more data more quickly.

Singapore’s proximity to nearby economies and existing infrastructure make it an attractive home for data centres. Regarding the reliability and resilience of data centres here, Prof Wen cites the Republic’s relative political stability and low risk of natural disasters as desirable qualities.

But its hot, tropical climate stands in the way of the greatest operating challenge for the industry: cooling.

Traditionally, data centres use air cooling to regulate server heat, and more energy is needed to chill air when temperatures are higher, such as in tropical Singapore. But liquid cooling, being a more efficient and reliable method, is steadily gaining traction as a new standard.

Foreign businesses favour Singapore’s skilled workforce and growing digital economy, according to an HSBC survey. Prof Wen, who began his research in data centres 11 years ago, says back then, the Government had identified high potential for the data and IT sector.

Eventually, the rise of data centres and concerns over energy consumption would prompt the government to impose a three-year moratorium on new facilities in 2019, which ended in 2022. Four data centre operators were granted approval in July to build 80 megawatts (MW) of new capacity that meet higher standards for efficiency and decarbonisation.

This additional data centre capacity, if utilised 24 hours every day for a month, could potentially consume as much energy as around 155,500 four-room flats in the same period of time, based on the Energy Market Authority’s 2022 figures for monthly household energy consumption in Singapore.

Data centres were responsible for about 7 per cent of the Republic’s total electricity consumption in 2020, according to the Ministry of Trade and Industry. To meet carbon emissions targets, the grid must shift away from natural gas towards renewable energy sources such as solar.

“We are in a situation where there’s a lot of unmet demand for renewable energy coming from the data centre industry,” says Professor Lee Poh Seng from the National University of Singapore’s Department of Mechanical Engineering. The challenge for Singapore is limited land, sun, water and wind — ruling out major sources of green energy production.

Tech companies running global hyperscale data centres, like Amazon, Microsoft and Google, are the top corporate purchasers of renewable energy, according to data from IEA.

These companies can purchase large quantities of renewable energy using tools such as energy attribute certificates and power purchase agreements, which facilitate the buying and trading of renewable energy. But these investments are less accessible to local operators due to limited supply, says Prof Lee. The solution to address lack of supply, he suggests, is partnering with neighbours.

As Singapore runs out of land, neighbouring Malaysia and Indonesia are seeing a rise in data centres as well.

“Singapore is a natural choice for data centres. If we have sufficient power and space, people will prefer to put their data in Singapore,” says Prof Wen. “But because we cannot provide sufficient power and space, people move to Indonesia and Malaysia for expansion. They have plenty of space, they have power supply, they can have more solar power naturally.”

Indonesia’s and Malaysia’s ample land area provide good conditions for solar energy production. Consequently, this primes Johor Bahru and Batam — just across the Johor and Singapore straits — for data centre construction.

One such upcoming data centre project is a collaboration between NeutraDC (part of Indonesian telecommunications conglomerate Telkom), Singtel and Indonesian energy company Medco Power, who will supply green energy to the building.

Mr Thomas Anthony, chief technology officer of NeutraDC, says one challenge in adopting green technologies is getting client buy-in. While green options are available, clients may specify requirements for data operations that do not always prioritise sustainability.

For example, installing motion-activated lighting may be more costly upfront but leads to savings on electricity in the long run, he explains. He believes the company has a responsibility to educate clients.

One way individual companies can potentially save on energy is by switching to cloud providers, like Amazon Web Services (AWS), Google, Alibaba Cloud and Microsoft Azure, instead of maintaining their own on-site servers. Doing so not only improves operational efficiency, but also lets businesses leverage sustainable technologies offered by these providers, says Prof Wen. AI-based solutions, as pioneered by Prof Wen’s research and NTU spin-off Red Dot Analytics, have been adopted by cloud providers such as Alibaba Cloud to improve their energy efficiency, thus making their operations more sustainable.

SPH Media uses AWS to host its multimedia graphics projects, including this page you’re reading now.

The responsibility to check energy use does not fall solely on data centres, says Mr Ian Chew, founder of Greenie Web, a start-up that helps organisations understand and minimise their digital carbon footprint. Companies can take steps to reduce their data usage and, by extension, demand for storage and processing.

Mr Chew suggests that companies consider shifts towards “low-carbon code” to be more efficient and green without compromising user experience. Similarly, Mr Fershad Irani of the Green Web Foundation advises Web developers to optimise code as much as possible and pay attention to the environmental impact of third-party services. Moreover, businesses can choose web-hosting providers that are powered by renewables as a step towards greener operations.

Our growing dependence on artificial intelligence, including platforms like ChatGPT, is expected to contribute to an increase in digital carbon emissions. Mr Chew points out a difference in reception to generative AI in this regard. Bitcoin was heavily scrutinised for its carbon and energy footprint, but similarly high levels of scrutiny have not been directed at large AI models, he notes.

A study found that training GPT-3 required 1,287 MWh, which is equivalent to powering more than 3,400 four-room flats in a month. In addition, ChatGPT could use 564 MWh per day to run queries from users, which is equivalent to powering more than 1,500 four-room flats in a month.

Making the Internet more sustainable should ultimately be equitable and just, says Mr Irani. Transitions should happen not just for the global north, but for all regions where data centres operate.

Mr Chew warns against falling into the trap of "green diversion”, where the purchase of renewable energy by companies redirects green energy currently used by existing communities.

“At the end of the day, do we hit netzero from a company perspective? Yes, we do. But do we do any good to the world? No, we don't,” he says.

What lessons can we learn from the IT sector? Its ability to curb its energy footprint despite exponential growth is to be envied by virtually all other sectors facing the challenge of meeting carbon reduction targets.

Innovation must continue as we adapt to a changing data landscape.

For example, AWS says it is actively enhancing its hardware infrastructure with a focus on energy efficiency, notably through its third-generation chip. “It’s designed to reduce the time and cost of training generative AI models. It’s able to cut training time for some models from months to hours,” says Ms Jenna Leiner, AWS sustainability business adviser for Asia-Pacific.

Liquid cooling is another example and presents a more energy-efficient alternative to air cooling. As hardware advances and air cooling reaches its physical limits, data centre operators are transitioning to liquid cooling, according to Prof Lee. And although its adoption will be gradual, Prof Lee anticipates that it will eventually become a standard for data centres.

AI can also be utilised to optimise resources by predicting the most favourable regions for renewable energy production and efficiently allocating resources to those areas. In Singapore, AI has been used to reduce energy consumption in MRT stations and improve the sustainability of buildings.

While data centre operations represent only a fraction of global emissions, they can have a sizable local impact as they strain energy grids. In Singapore alone, there are at least 70 data centres all around the country.

Though largely unnoticed, the presence of data centres will continue to evolve alongside our relationship with data. These facilities will play a pivotal role not just in how we use data, but how we do so sustainably for the environmental costs and benefits of an increasingly digital world.