Data Surge Dilemma: We Dive into APAC’s Data Centre Markets
Artwork by Dall-E
The inexorable rise of cloud computing has been accompanied by an explosion in demand by people for new services such as OTT applications and cryptocurrencies, driving much of the growth in data centre markets over the last decade.
And in Asia Pacific (APAC) all of this is overlayed with pressures created by a growing population with high digital adoption rates. According to Cushman and Wakefield data, while data centre growth in the US is moving at a healthy clip of 14 per cent, that will be significantly exceeded in APAC where growth is expected to gallop along at 25 per cent until 2028.
Now an explosive new accelerant has been added to the mix – growing demand from people and businesses for artificial intelligence (AI) and in particular generative AI and the large language models that underpin it.
Data, as the saying goes, is the new oil, and the world is thirsty. According to global commercial real estate outfit JLL, thanks to AI, global data storage capacity grew from 10.1 zettabytes (ZB) last year to 21.0ZB in 2027. That’s a 5-year CAGR of 18.5 per cent.
And of course, in this contemporary mining boom, it’s often the people selling picks and shovels – in this case, hardware manufacturers and data centre owners – who stand to clean up.
Demand is soaring for the hard drives that contain 90 per cent of all the data in hyperscale cloud environments, according to Future CIO.tech which also noted that skyrocketing data growth in Asia demands denser storage. (Rack density – the amount of storage you can stuff onto a single rack in a data centre will likely grow by 50 per cent over the next three years).
It is not just hard drives.
The experts tell us that some of the new AI data centres used for training models will be very large, something like 500MW. Then smaller (but still large!) data centres will be used for applying the models, for inference - more like around 100MW. And then there will be edge data centres, from a few hundred KW to a few MW (see below). That’s the current hypothesis.
There is of course also the extraordinary story of Nvidia whose market capitalisation has grown from $323bn at the end of 2020 to more than $2.1tn today, thanks to its exceptional ability to process all that data held in the hard drives with its graphics processing units (GPUs).
As we noted last year, central processing units (CPUs) can be used for some simpler AI tasks, but they are becoming less and less useful as AI advances, according to the Centre for Security and Emerging Technology.
Where the action is in APAC
Knight Frank reported capacity of 13.4GW across nine cities across the region in Q1 2023: Tokyo, Shanghai, Sydney, Singapore, Hong Kong, Mumbai, Seoul, Kuala Lumpur and Bangkok. Skip forward a year, and Cushman and Wakefield estimated that five countries now accounted for 80% of regional capacity: Mainland China (3.9GW), Japan (1.3GW), Australia (1.2GW), India (1.1GW) & Singapore (~1GW).
Translating this into dollars, in its recent APAC Data Centre Market report, advisory firm Arizton said that the region’s data centre market would reach $93bn in size by 2027. That forecast was made before ChatGPT burst into public and business consciousness in late 2022, and generative AI emerged as a key market accelerant…
Mining for Gold
As you can imagine, there has been plenty of data centre action in both the public and private markets, reflecting the huge amount of blue sky ahead.
Our index of DC-related stocks is up 40% over the past four years.
M&A activity has been brisk over the same period. In 2022, 187 data centre deals were inked with an aggregate value of US$48bn, following 2021’s record year of 215 deals amounting to nearly $50bn in value (up from $35bn in 2020).
Private equity investors have been dominating the action in recent years with several notable public-to-private transactions. In 2022, KKR and Global Infrastructure Partners snapped up data centre developer CyrusOne for a mere US$15bn, while DigitalBridge and IFM Investors completed their $11bn acquisition of Switch.
These days there’s scarcely a week without an announcement – for example in the last few weeks Chinese data centre operator GDS has raised $US587mn for international growth, along with reports from Reuters that Macquarie and CPPIB are planning to sell Airtrunk for an estimated $US15bn.
This isn’t going to slow down in a hurry. If anything, it’s set to accelerate.
Will this be an ESG Fail?
But if data is the new oil of growth, in sustainability terms, it has also been called the new asbestos (perhaps too emotive a term but it caught your attention, didn’t it!), with JLL saying generative AI’s hunger for energy will demand greater energy efficiency - not to mention water for cooling.
An analysis by Alex de Vries, published in the journal Joules last year estimated that the AI industry alone could consume between 85 and 134 terawatt hours each year – equivalent to the consumption of The Netherlands (from where, coincidently de Vries hails).
That’s why the giant tech platforms and data centre operators are all looking for greener ways to play their role in this great game of thrones.
Power Struggle
Jonathan Kinsey, JLL’s Global Chair of Data Centre Solutions, is quoted in the firm’s 2024 Global Outlook for Data Centres report, saying: “In many cases, existing grid infrastructure will struggle to support the global shift to electrification and the expansion of critical digital infrastructure, making it increasingly important for real estate professionals and developers to work hand in hand with partners to secure adequate future power.”
“As the data centre industry grapples with power challenges and the urgent need for sustainable energy, strategic site selection becomes paramount in ensuring operational scalability and meeting environmental goals.”
JLL further notes that the rapid growth in data centres is ratcheting up the pressure on the limited energy resources in many countries, and it calls out Singapore specifically where the government had to introduce a moratorium temporarily halting construction in certain areas while it assessed the impact on the nation’s sustainability goals.
Governments are now caught in a bind. Businesses need capacity, while societies (which of course includes businesses) are trying to reorganise themselves around ESG outcomes.
The demand for data centres can’t be reversed, so alternatives need to be found. The use of green energy is core to this, a point not lost on the hyper scalers. There are many examples:
Amazon, the owner of AWS, says it will achieve net-zero carbon emissions across its operations by 2040 and that its cloud business will reach this goal by next year.
Google says its goal is to match its energy consumption with carbon-free energy, every hour and in every region by 2030.
Microsoft says its Azure operation has been net zero since 2012 although that hasn’t stopped the company from innovating around new solutions such as undersea data centres via its Project Natick to keep ahead of the curve.
Facebook’s facility in Luleå, Sweden, and Google’s centre in Hamina, Finland, exploit the colder local temperatures for efficient cooling.
In New Zealand, which benefits from an abundance of renewable energy resources and a cool climate, companies are planning to build data centres at the very south of the country, which is also the closest geographic point between New Zealand and its larger neighbour Australia (could Invercargill become the green solution for Australasian data?)
Finally, the nature of data centres will likely start to change as edge computing becomes more prevalent. The theory is that a more decentralised approach to computing that pushes data processing closer to the point of data processing should reduce energy consumption, lower carbon emissions, and enhance efficiency.
When it all goes wrong
Get the balance wrong and people notice quickly as Singapore’s DBS Bank and Citibank discovered last year, when 2.5mn payment and ATM transactions couldn’t be completed due to data outages caused by temperature regulation systems failures at an Equinix data centre. It wasn’t the first time that DBS - the country’s largest lender - had experienced service outages.
And it wasn’t just bank consumers who felt the pinch – the MAS slapped penalties on the bank and Piyush Gupta, the bank’s CEO, took a 27% cut to his pay packet as a result of the outage.
Outages are going to get worse before they get better. According to a paper by the Uptime Institute based on research conducted in 2022 and 2023:
41 per cent of organisations encounter at least one substantial outage monthly.
In 2022, a staggering 70 per cent of outages incurred costs exceeding $100,000, compared to 40 per cent in 2019. This upward trend is projected to persist with the growing reliance on digital services.
Many service providers fall short of service level agreements (SLAs), emphasizing that SLAs are not reliable indicators of future service provider availability.
Cloud, SaaS and digital services contributed to 80 per cent of public cloud outages in 2022, rising from 66 per cent in 2016.
And reinforcing how these increasingly complex environments are subject to human frailty, the report found that software or configuration errors accounted for 65 per cent of outages in cloud services. “Additionally, over 87 per cent of outages were attributed to human error and/or management failures, underscoring the need for enhanced tools and processes to mitigate such errors.”
Outages are not really an option in a digital economy, so it is little wonder that there is such a drive to build additional capacity on top of the already significant data centre investment.
Strap in
Last year researchers from Hugging Face and Carnegie Mellon University revealed that it takes almost as much energy to generate a single image using Stable Diffusion's open source XL as it does to power an iPhone.
And one thing the world can be certain of in the years ahead is the ongoing demand for iPhones, and an insatiable thirst by businesses and consumers for billions of AI-generated images.
As APAC's data centre markets expand, driven by the relentless demand for cloud computing, OTT services, cryptocurrencies, and now artificial intelligence, the industry faces a pivotal moment in its evolution. Reconciling this growth with society’s demands for sustainability and emissions reduction will be a major challenge.
We close this article with a thoughtful summary from Simon McFadden, Aurecon’s Industry Director, Data & Telecommunications: “I see digital infrastructure now as one large network, wrapping the globe, with data centres connected by fibre and wireless networks to humans and machines. This interconnected network has continuous growth in connections and demand for data capture, transmission and storage. And this growth is exponential. Data centres are but a part of the network, but at its core, and will enable the productive technologies we are going to use in the coming years, AI, robotics and 5G, to enable breakthroughs in how we live, work and I believe, prosper as we transition to a lower carbon economy. If we can enable this growth and therefore the breakthroughs in a sustainable way, we will succeed”.