Powering our future sustainably with green data centres

Sustainable approaches to powering data centres will be the way forward, especially with the launch of Singapore's Green Data Centre Roadmap in May.

Speaking at a panel titled Powering the Digital Future with Sustainable Solutions during CommunicAsia, Kavickumar Muruganathan, Sustainability & Policy Lead (APAC), Cloud Operations & Innovation, Microsoft, said that the new roadmap is heartening to see. "There is a lot of data centre capacity growth, there’s a lot of interest in harnessing technology and innovation in driving processes and innovation. We had a lot of details about how the government is going to partner industry which is very important," he observed. 

"Looking at it from a long-term perspective, how do we enhance those efficiencies? ...AI-powered data centres will have new challenges, a broader ecosystem is important. We can collaborate and innovate to find more cutting-edge solutions."

Henry Xu MBDA, CEO, Co-founder, Red Dot Analytics and the moderator of the panel, called the 300 MW target for new capacity in Singapore outlined by the roadmap both new and exciting. He said the industry will need more guidelines on using AI and digital twin technologies to optimise, automate, and manage efficiency across the data centre, especially as the typical data centre operator is unfamiliar with digital twins. 

The need for better cooling

The increase in computing power needed to support AI will mean hotter data centres. While future energy consumption estimates varied among panelists, the consensus was that sustainable new cooling methods will be needed.

Sharmel Ali, Head, Group ESG, ST Telemedia Global Data Centres said that with exponential growth in AI, traditional rack densities of 25-30 kW are likely to grow 3x to 4x whereas Xu suggested that AI GPU data centres might run up to 80-100 kW per rack.

"The challenge is that power densities are going to increase further down the road, probably to 100 kW per rack. We have to embrace new technologies like AI, host data centres should be looking at AI as well, it’s low-hanging fruit. Software-driven optimisation is more cost effective than hardware or physical changes," Xu said.

Ali shared that his company already hosts GPU clusters that power AI applications, and that innovative cooling technologies have been introduced to enhance energy efficiency. "It's part of our commitment to environmental stewardship," he said.

"The (traditional) cooling mechanism is air cooling, (but it is) insufficient to meet AI-powered density (requirements). The way forward is liquid cooling - water or others, to circulate that liquid through heat exchangers which are close to the heat-generating components of servers, or to submerge the servers in a liquid. It is very efficient in terms of heat transfers. It can reduce energy usage for cooling, enhance cooling efficiency. An additional benefit is that it ultimately results in a low carbon footprint, so we function in a more efficient way," Ali added.

Technologies like AI and digital twins could be used to predict what can happen to the physical data centre. Xu noted that harnessing AI and digital twins at a data centre will require balancing the requirements for performance, resilience and sustainability.

Xu noted that digital twins can help with AI-driven scenario planning, such as predicting what happens if another 50 kW rack is added. "(I) can ask the AI to tell me the optimal policy to give me the lowest energy consumption, we can run scenarios to give optimal or minimum energy consumption but without violating SLAs," he suggested, adding that actions can even be applied autonomously.

Sustainability

Any change has to be sustainable. Direct air capture technology*, for example, could be harnessed to add a green element to the generation of waste heat, Murganathan said. He also spoke of using sustainable alternatives to traditional fuels like hydro-treated vegetable oil (HVO) or hydrogen fuel cells, but said that the challenge is scaling the solutions globally.

Said Muruganathan: "In some of our data centres...we've completely transitioned out of diesel... we also pioneered hydrogen fuel cells for backup." 

In Singapore, servers at their end of life are broken down to the bare bones, he added, diverting them from landfills.

Public-private partnerships

An ecosystem comprised of industry peers, technology partners, and governments is crucial if common standards and regulations are to be developed to encourage sustainable practices within data centres, panelists added.

"AI-powered data centres are the next thing to come into operation in a couple of years. How can we use this to support our corporate ambitions? As a company, we have set ambitious corporate targets. So how do we enable AI data centre growth and also enable us to meet the sustainability commitments that we’ve set in tandem? I don’t think we can do it alone. We need everyone in this ecosystem to support us," Muruganathan said. 

Tew Kiat Wong, Chairman, Data Centre Chapter of Singapore Computer Society & MD of Organisation Resilience Management, touched on how the industry can be a test bed for new solutions. "We are working with key stakeholders, sharing of trends and developments with customers and technology partners, understanding what the customer requirements and needs are, understanding what their sustainability objectives are, so we not only align ourselves but also find the appropriate solutions," he said. 

"There is an opportunity for us in terms of seeing or identifying gaps in areas where we can provide solutions, collaborate to find solutions and set boundaries and specifications in terms of developing the appropriate measures and mechanisms, developing standards, collaborating with peers and the sector, coming up with common standards, working towards common goals," he said. 

AI energy consumption a fallacy?

Anthony Schiavo, Senior Director and Principal Analyst, Lux Research, said in blog post that the argument that AI will require more energy from data centres is flawed, with AI needs likely to parallel existing data centre needs. He noted that AI developers are motivated to minimise computational demands to optimise their own costs and limited GPU time.

"First, it’s worth setting a baseline on energy demand growth from data centres: You may be surprised to learn that despite the huge expansion of digital tech into every area of our lives over the last decade, data centre energy usage growth has been almost flat over that same period. This is largely due to improved energy efficiency in chips, programs, and the data centres themselves — there’s been a major shift to hyperscale centres, which are more energy efficient. 

"This lack of growth comes despite the growing number of data centres and the growing amount of computing power," he said, clarifying that cryptocurrency mining is considered separately. 

He pointed out that large language models (LLMs) are becoming more energy efficient, while searches for the same thing do not need to be repeated. "Many large-volume tasks will end up using smaller purpose-built and custom optimised models that don’t need as much energy per query," he said. 

"On top of the energy efficiency improvements in the models, there’s also a ton of real-world reasons to think energy demand growth won’t be that significant: Things like Google searches don’t need to be re-run every time, as there’s lots of common searches that are repeated. Many large-volume tasks will end up using smaller purpose-built and custom optimised models that don’t need as much energy per query," he added. 

"AI-generated photos and videos are kind of the wild card here, but our estimates show that it would take a huge amount of image generation to really move the needle on total data centre energy demand." 

*Direct air capture technologies extract CO₂ from the air using renewable energy or waste heat.