Sustainability: Balancing growth with green innovations
This article first appeared in Digital Edge, The Edge Malaysia Weekly on September 9, 2024 - September 15, 2024
Malaysia’s emergence as a data centre hub brings significant environmental challenges, particularly since the country relies on electricity generated from fossil fuels.
“Data centres are significant consumers of energy. Managing this consumption efficiently is vital to both operational costs and environmental sustainability,” says Tham Chee Aun, group CEO of Ditrolic Energy Holdings Sdn Bhd.
As Malaysia seeks to solidify its position as a key player in the global digital economy, balancing this rapid development with environmental sustainability is crucial, especially since it is rather easy to set up a data centre in the country.
“As long as they get the necessary approvals for the utilities infrastructure, they are generally allowed to set up a data centre anywhere in Malaysia,” says Tang Chee Khoay, chair of the working committee for the development of a data centre tool by the Green Building Index (GBI).
The power usage effectiveness (PUE) metric is crucial in assessing a data centre’s energy efficiency. Historically, a PUE of 1.6 was considered the “gold standard”, but new data centres now aim for a PUE of 1.4, which signifies a highly energy-efficient cooling system. Older data centres often have PUE values exceeding 1.6, with some even surpassing 2.0 — an indication of energy inefficiency.
“Poorer energy efficiency for older data centres is largely due to infrastructure availability around the data centres. In an urban environment, water is not enough for a water-cooled air-conditioning system to be used. Therefore, air-cooled air conditioning is used instead, lowering energy efficiency,” says Tang.
“In addition, existing data centres are less concerned about energy efficiency and carbon emissions. Most of them are waiting for their existing clientele to ask for less carbon-intensive data centres before they will do anything. If their clients are not asking for it, there is no reason for them to do anything.”
The belief that data centres are already as efficient as possible, which he says has exacerbated the perception that improvements will require a substantial financial investment and consequentially slow the adoption of advanced energy-efficient technologies.
While new data centres are often certified by green rating systems like the Leadership in Energy and Environmental Design (LEED) in the US or the Green Building Index (GBI) in Malaysia, these tools were originally designed for people-centric buildings and do not fully address the unique environmental concerns of energy and water consumption in equipment-centric environments. As a result, GBI is developing a tool specifically for data centres, says Tang.
After the IT infrastructure, cooling systems are the second-largest energy consumers in data centres, accounting for 20% to 40% of total energy use in less densely utilised facilities, says Adrian Koh, head of secure power division for Malaysia and Brunei at Schneider Electric.
Tang illustrates, “Some data centres in Malaysia are targeting hundreds of megawatts in power requirements. For a rough comparison, 300mw is enough to power a small town such as Muar, while a 2,000mw data centre with water-cooled chillers will match the water consumption of Muar.”
Water shortages are increasingly a concern in the country; therefore, water use in data centres is critical since they consume such large quantities. To mitigate this, operators are adopting technologies like dry coolers with adiabatic evaporation and liquid cooling to reduce water usage and investing in water replenishment programmes, says Koh. Despite Malaysia’s tropical climate, climate change still poses the risk of drought.
“During such a period, should water be prioritised for data centres or for the people? We must think about this and perhaps come up with regulations to determine what data centres and other high-water-consuming industries need to do in a drought year,” says Tang.
Indirect evaporative cooling has shown promise in Malaysia, potentially cutting chiller usage by two-thirds and water consumption by 50 to 60%. However, advanced liquid cooling, including direct-to-chip and immersion cooling, can drastically reduce energy consumption, bringing the PUE close to 1.0 and nearly eliminating the need for sophisticated air-conditioning systems. It also reduces water consumption to zero by removing the need for cooling towers, he says.
Koh concurs, adding that it can offer benefits such as higher energy efficiency, a smaller footprint, a lower TCO (total cost of ownership), enhanced server reliability and reduced noise levels at similar server densities than current values.
While there seems to be a preference for direct-to-chip liquid cooling, Tang says its adoption in Malaysia is slow due to the lack of supporting infrastructure and concerns about the risk of liquids near circuit boards, despite non-conductive liquids being used.
It is important to note that “as the demand for AI processing power grows and thermal loads increase, liquid cooling has become a critical element in data centre design”, according to Koh.
Although a large number of data centres in Malaysia already utilise energy-efficient cooling systems and advanced IT infrastructure to manage energy use, there is still room for improvement in adopting renewable energy sources, says Tham.
Harnessing renewable energy
Local data centres have several options when it comes to procuring renewable energy to power their operations, says Ditrolic Energy’s Tham. They can generate solar energy on-site under the Solar Self-Consumption scheme, directly using it to run their equipment. Alternatively, they can purchase Renewable Energy Credits (RECs) from qualified renewable energy plants, such as solar or hydropower, to verify the origin of the power they use.
“Another increasingly popular option is purchasing renewable energy from an off-site renewable energy plant in large volumes. This can be done through a Virtual Power Purchase Agreement (VPPA) under the Corporate Power Purchase Agreement (CPPA) or the upcoming Third-Party Access (TPA) scheme, which essentially allows data centres to buy renewable energy directly from a solar power plant,” he says.
“Through optimisation of on-site solar generation and building off-site solar plants with co-location of a battery energy storage system (BESS), data centres are able to enhance their energy usage efficiency and increase the renewable energy mix in their operations. One fine example is the solar-powered NCS data centre in Singapore, which was built by Ditrolic Energy.”
Renewable energy may help reduce a facility’s overall carbon footprint and environmental impact but the generation of renewable energy sources (RES) is characterised by variability, affected by the changing intensity of sun and wind, which can impact power generation.
“To effectively manage this and enable the integration of intermittent renewable energy, grid-scale energy storage becomes essential. The implementation of such storage solutions supports the development of resilient grids that can accommodate 70% or more RES, as demonstrated by MIT research,” says Koh.
“Additionally, long-duration energy storage (LDES) capabilities, coupled with micro-grid designs, are critical for large energy users like data centres. Facilities with significant critical power needs can also benefit from new technologies that allow for substantial deployments of batteries, such as lithium-ion batteries in intelligent uninterruptible power systems (UPS). These technologies are crucial for increasing storage capacity and facilitating greater RES adoption.”
But Tang disagrees, stressing that on-site renewable energy is limited, as the data centre’s energy consumption is simply too high for its roof to be able to contribute much at all. “Usually, it is less than 0.5% contribution even if the whole roof is used. Furthermore, the roofs are full of air-conditioning equipment rejecting heat into the air, with very limited space available for solar PV on them.”
AI’s dual impact
The growing use of generative artificial intelligence (AI) complicates the sustainability of data centres. While it can optimise energy usage, it also raises concerns about increased energy consumption, says Koh.
“We predict that AI workloads will grow two to three times faster than legacy data centre workloads and represent 15% to 20% of all data centre capacity by 2028. More workloads will also start moving closer to the users at the edge,” he adds.
“However, it is crucial to carefully evaluate AI’s broader impact on energy consumption and the environment. In fact, Gartner Inc reveals that 80% of CIOs (chief information officers) will have performance metrics tied to the sustainability of the IT organisation by 2027.”
The rapid growth of AI demands a significant increase in data centre power — a figure that is projected to reach 18GW by 2028. This far outpaces the current growth rates and presents substantial capacity and sustainability challenges.
AI requires a fundamentally different data centre architecture, which includes specialised IT infrastructure, power and cooling systems, particularly for training large language models (LLM) with high-density GPU clusters.
Currently, most data centres can only support peak rack power densities of about 10kW to 20kW and face challenges in supporting AI clusters with rack densities ranging from 25kW to 120kW. Nevertheless, some industry experts foresee that accelerated computing may enhance efficiency.
“When aiming to decrease energy consumption and carbon dioxide emissions in IT and data centres, companies need to establish a factual baseline and have access to real-time and historical data. Software can be used to measure and report data centre performance based on historical data and trends analysis,” says Koh.
Schneider recently partnered with Nvidia to revolutionise its data centre infrastructure, launching three retrofit designs for adding AI clusters to existing facilities and a scalable new-build design tailored for Nvidia’s computing clusters. These designs, optimised for AI applications like data processing and generative AI, enhance performance, scalability and sustainability.
Overcoming challenges to achieve net zero
The advantages aside, the growth of data centres undoubtedly complicates Malaysia’s efforts to achieve its net zero goals outlined in the National Energy Transition Roadmap (NETR).
“Malaysia must be more aggressive in energy efficiency and renewable energy to match this growth of data centres here. Currently, GBI intends to address this demand via a new GBI Data Centre Tool, which is designed to address the huge consumption of energy and water in a data centre. One factor that is being discussed now is the water consumption of a data centre during a drought,” says Tang.
Tham agrees, saying: “It is essential to integrate renewable energy sources and improve energy efficiency in these facilities. It is essential that data centres transition to greener energy solutions, such as solar power, which can significantly reduce their carbon footprint.
“Additionally, having the option to purchase energy from renewable sources through multiple energy suppliers would greatly contribute to the sustainability of the data centre industry in Malaysia, preventing a net increase in carbon emissions despite rising energy consumption.”
The good news? Malaysian data centres are increasingly looking to integrate renewable energy sources to reduce their carbon footprint. “However, they face obstacles such as the intermittency of renewable energy, the high initial cost of installation and the need for substantial infrastructure investment to support large-scale renewable energy adoption,” he points out.
While it is encouraging to see proactive governmental measures and roadmaps like NETR, it is also important to recognise the importance of including Scope 3 emissions as a standard prerequisite across all regions, says Koh.
Rather than just country-specific comparisons, it is crucial to make Scope 3 emissions a standard focus across regions, he adds. These emissions, spanning the entire value chain from mining to manufacturing, are challenging for data centre operators to address but are critical to decarbonising data centre products. Reducing them requires selecting the lowest-carbon products possible.
“By integrating such digital solutions, data centres can advance their energy efficiency, contributing to the decarbonisation of the sector and supporting Malaysia’s broader climate goals,” says Koh.
AI posing power conundrum for data centres
Stories by Vanessa Gomes
Data centres around the world have seen a lot of scrutiny on sustainability over the last 13 years and collectively, the industry has started to hold itself to a higher standard. The delicate balance between the need for these facilities, the commerce they drive and the sustainability implications they have on countries and the environment is a tricky one, especially from the design, build and operate perspective.
Nevertheless, be it greenfield or brownfield data centres, there is an intent to do better, and energy efficiency is one area that has been challenged over the years and has since seen a plateau after years of successful exploration.
Shaheen Meeran, vice-president of business development for the secure power division at Schneider Electric, says sustainability is now a key tenet in the industry, where the focus is on power usage effectiveness (PUE).
Not only are service providers declaring their PUE in an effort to be more competitive, but even enterprises with their own data centres are volunteering information on PUE for greater transparency on sustainability.
“Energy efficiency has matured very well, both from the perspective of products that have performance characteristics that contribute to good energy efficiency and tools to understand operations and gaps better for further improvement,” says Shaheen.
Inroads are being made to source green power and understand the carbon footprint of the current energy use. There is also a growing focus on how deployed products work together for a more sustainable data centre.
Shaheen acknowledges progress in these areas but notes that more is needed. Large data centre operators are starting to structure procurement tenders to encourage vendors to provide more sustainable products and disclose embodied carbon.
“But I don’t think it has filtered down to every data centre user and enterprise in the world. So, I think there is still a journey there and much more needs to be done,” he says.
The industry is ripe for developing countries to seize opportunities and learn from mistakes made in developed countries. Opportunities are greater now, says Shaheen, especially since it is now possible to operate a data centre at a remote location, away from where the actual information is needed.
This means data can flow between countries with the help of a colocation service provider. “We’ll see an emerging breed of cloud providers that specialise in infrastructure that is AI-ready that need not be in the same location as the company that decides to run a training model on its premises,” he says.
“With this remote model, emerging countries can potentially start looking at these as revenue streams and build new data centres. The fact that they are built new can potentially mean they can be designed and engineered in a way that is very sustainable,” he points out.
“This can become a unique proposition for investment in the country and my take is that developing nations could look at emerging AI opportunities as something that can become a revenue generator.”
Modular data centres to support customers’ needs
Schneider has a set of prefabricated data centres. The company also does engineer-to-order (ETO) data centres that are designed to meet a customer’s specific needs and requirements.
The advantage of modular data centres, unlike data centres that have been built from the ground up, is that they are built in factory-controlled environments and can be deployed in eight weeks. A large data centre will see a lot of human intervention when it is being installed on-site, which creates an opportunity for human error.
“Something that is prebuilt and pre-engineered in a factory is tested before it gets shipped out and that creates much less opportunity for error, making it more reliable,” says Shaheen.
“These prefabricated data centres can be deployed quickly. So, if there is a business opportunity a company needs to address, it can be done by deploying this new small data centre. These systems can also be designed to support AI server loads, and the real make or break would be the ability to deploy liquid or air cooling.”
Prefabricated data centres are built for immediate need, so if a company requires one to be built in the next six months, it can be done. If more capacity is needed later on, it can be built as the need arises.
The advantage of modular data centres is that a company will always utilise the infrastructure at maximum capacity. “High utilisation means high efficiency, which means it is much greener,” says Shaheen.
“Let’s say 15 years down the line, the company decides that it wants to vacate a space. The prefabricated data centre can be carried with the company, making it reusable and very sustainable.”
In December last year, Schneider released a white paper on the total cost of ownership (TCO) analysis of a traditional data centre versus a scalable, containerised data centre. It revealed that standardised, pre-assembled and integrated data centre facility power and cooling modules provided TCO savings of 30% compared with traditional, built-out data centre power and cooling infrastructure.
Schneider has not deployed a prefabricated data centre for an AI workload to date, but she believes that if there is a need from a customer, it should be able to do an ETO feasibility study for the project. “It could work very well for the heated AI market that we have today,” she says.
The challenge would be figuring out how to deliver the large quantum of power demanded by AI, not just internally in the data centre but externally as well.
“How do we create that capacity in the grid to be able to provide that quantum of power to the data centre? That challenge, I think, goes beyond the industry. Governments will have to identify ways to either upgrade the grid or find alternative pockets of power generation, which could be very close to these data centres or far away and just feeding into the grid,” says Shaheen.
“That level of collaboration will definitely require some amount of political will and, to some extent, a legislative framework. Deep collaboration between the data centre constructor, industry and government.”
Infrastructure needs to adapt to changing times
The rapid growth of the data centre market in Malaysia is a clear indication of the country’s strategic position in the digital economy. In just a few years, data centre investments significantly increased, securing a whopping RM114.7 billion from 2021 to 2023, according to Minister of Investment, Trade and Industry Tengku Datuk Seri Zafrul Abdul Aziz.
This aligns with the growth of AI, presenting data centres in Malaysia with numerous opportunities to innovate and expand their services to support AI-driven applications and enterprises.
Data centres are crucial to enable AI adoption across various sectors, but they must upgrade infrastructure and operations to meet demand, leading to challenges like higher power consumption, Schneider Electric Malaysia’s country president Eugene Quah tells Digital Edge.
“To maintain their trajectory and support Malaysia’s AI ambitions, data centre companies will need to focus on scaling their operations sustainably and efficiently, ensuring that they can meet the burgeoning energy demands of AI technologies without compromising on environmental or operational sustainability,” he says.
Traditionally, companies prioritised cost, quality and speed to market, often at the expense of sustainability, as the impression is that it is capital-intensive. This led to resistance from companies to adopt sustainable solutions, says Quah.
“AI can potentially be the key to balancing cost and sustainability. This is because AI can be used to streamline operations through predictive analytics, optimise performance and enhance efficiency, as well as aid in resource optimisation, thus minimising energy consumption. Overall, this means long-term savings and balancing costs in the process,” he adds.
There is definitely a pressing need to expedite efforts to establish a robust ecosystem that fosters the development and adoption of AI technology. However, the primary challenge lies in the lack of education and awareness of the need to implement AI or digitalise operations.
“To navigate these complexities and fully harness the power of AI, companies need to explore collaboration opportunities with experts who specialise in this field,” says Quah.
Being prepared for greater data processing
The development of large language models (LLMs) requires significant computational resources, often involving thousands of graphics processing units (GPUs). Quah says this demand influences the design and power density of data centres, with large AI clusters needing substantial energy, ranging from 1mw to 2mw for each cluster, and rack densities varying significantly based on the GPU model and quantity.
Currently, most data centres can only support peak rack power densities of about 10kW to 20kW. Deploying tens or hundreds of racks, with each exceeding 20kW, in an AI cluster will present substantial energy and infrastructure challenges for data centres. Quah emphasises that it is crucial to carefully evaluate AI’s broader impact on energy consumption and the environment.
The Malaysia Digital Economy Corporation (MDEC) has collaborated with the International Data Center Authority (IDCA) to develop a green data centre and digital hub blueprint to minimise the industry’s carbon footprint. This strategic development not only attracts global tech giants and investors but also supports the backbone necessary for efficient and sustainable AI operations in the country, says Quah.
“When we focus on the sustainable development of Malaysia’s AI infrastructure, we pave the way for smarter, faster and more sustainable decision-making processes powered by AI, benefiting both the nation and the broader region,” he points out.
Increasing AI use through education
Despite AI being a popular topic, a full understanding of the technology requires piloting multiple projects and learning from failures, which is resource-intensive and often impractical in tough economic conditions.
Many companies are eager to adopt AI but struggle to focus on impactful use cases to achieve a substantial return on investment (ROI), says Quah. Ready-to-use AI solutions, like chatbot software, can accelerate the return of investment by saving time and money on customer service.
However, he notes that AI solutions need to be customised to address specific challenges effectively, especially in industrial environments. These solutions, developed with domain experts, can integrate more seamlessly into existing systems, minimising disruptions and maximising ROI.
“For instance, in consumer-packaged goods manufacturing, a customised AI algorithm can optimise energy efficiency by adjusting machine settings based on high-frequency production data. This approach is crucial in complex industrial environments as it allows AI to address specific needs, enhancing efficiency and sustainability,” Quah explains.
The data centre industry comes together
The Infrastructure Masons (iMasons) is a non-profit professional association of technology and business leaders, representing more than US$150 billion in infrastructure projects in over 130 countries. On April 25, 2022, Christian Belady (Microsoft) and Dean Nelson (iMasons) announced the official start of the iMasons Climate Accord (iCA), focused on uniting on carbon reduction in digital infrastructure.
iMasons has four strategic industry priorities: increase awareness, enhance education, champion diversity and inclusion, and inspire sustainability. A critical tool in enabling the measuring and reporting of embodied emissions is the environmental product declaration (EPD), which enables the measuring of digital infrastructure’s embodied carbon.
In an open letter dated July 16, the governing body of the iCA called on all suppliers serving data centres to support greater transparency in Scope 3 emissions as part of broader efforts to reduce the industry’s carbon footprint. Amazon Web Services, Digital Realty, Google, Meta, Microsoft and Schneider Electric made up the governing body.
Hyperscalers have implemented strategies to reduce and/or mitigate Scope 1 and 2 emissions. As they seek to reach net zero carbon emissions in the coming years, solving the next piece of the sustainability puzzle lies in reducing Scope 3 emissions, which can represent anywhere from 38% to 69% of data centres’ total carbon footprint. Scope 3 emissions are not produced by the company itself but comprise the indirect emissions produced along the value chain.
“A challenge facing the data centre industry (among other industries) is that there are simply not enough vendors producing EPDs and making them available for data centre owners and operators. This hampers the industry’s ability to procure lower-carbon materials and equipment, as well as to reflect vendor emissions reductions in our reporting to key stakeholders (including regulators) which can jeopardise our contracts with some customer groups,” the letter reads.
Save by subscribing to us for your print and/or digital copy.
P/S: The Edge is also available on Apple's App Store and Android's Google Play.
The content is a snapshot from Publisher. Refer to the original content for accurate info. Contact us for any changes.
Comments