How AI is set to accelerate demand for data centres

We recently visited Japan, South Korea, Singapore, and Indonesia to keep pace with the electrifying growth of digital infrastructure in the region, as detailed in the embedded video above. We found that established operators such as Equinix, NTT and SingTel, are particularly well placed to capitalise on this growth opportunity, through their highly developed network infrastructure, which connects leading Global Cities. In our proprietary analysis below, we highlight an example of how Equinix has a leading position in Singapore, hosting the highest number internet exchanges and network carriers in the market, helped by their global reputation and proximity to subsea cable landings.

The sheer volume of critical data carried by these data centre users creates a gravitational pull for customers from other industries to lease space at Equinix facilities. The company then makes high margin revenue from plugging these tenants into each other, now hosting over 24,000 ‘inter-connections’ (Interconnections are fibre optic cables that run between customers within a facility, allowing them to securely exchange data between their private servers) between customers in their Singapore metro alone.

Data Centres: The modern-day shovels in the AI Gold Rush

The mass adoption of generative Artificial Intelligence (AI) (a type of AI that is capable of creating new data or content, rather than simply analysing, or processing existing data), marked by the run-away success of ChatGPT since November 2022, has sparked interest akin to the Californian Gold Rush. Investment markets have jumped back on the bandwagon to Silicon Valley, rewarding pioneering firms like NVidia, Google, and Microsoft, that have gained first mover advantage through extensive research and development efforts. Investors are understandably now looking for the next group of beneficiaries, whilst seeking to avoid high risk or even loss-making companies.

We believe that the answer is hiding in plain sight. NVidia’s first quarter 2024 earnings call was a watershed moment for the AI supply chain, highlighting the incredible recent growth in demand for NVidia’s hardware and leading analysts to upgrade their full year revenue expectations by approximately 40% (source: Refinitiv).

In explaining their upgraded outlook, the company mentioned ‘data centres’ no less than 56 times during their earnings call for investors on 24th May 2023. It is clear that their advanced Graphics Processing Units (GPUs) (GPUs accelerate the processing of large amounts of data, needed for training and applying AI models) entirely depend upon high-performance, secure, and stable data centre environments.

The evolution of data centres from communications exchanges into AI launchpads

Data centres have evolved dramatically from their origin as telecommunications hubs for hosting the internet in the late 1990s and early 2000s. Technological and policy advancements have since driven their transformation into larger, more localised, more resilient facilities, boasting state-of-the-art cooling systems, redundant power supplies and advanced security measures.

Over the past two decades in developed markets (outside of Asia), data centres were in many cases spun out of telecommunications companies, into infrastructure businesses more suited to their long-term ownership. These buildings both host the public internet and also enable businesses to privately connect to their end customers and partners. Coupled with the smartphone revolution which has led to an unprecedented surge in data generation, the demand for data storage and processing capacity continued to soar.

AI data science has been around since the 1950s, but with the affordable and widespread availability of advanced AI technology much more recently, adoption has rocketed. This is in no small part due to the cloud computing giants such as Amazon and Microsoft developing their own internal applications and beginning to make these available to businesses of all sizes.

Translating the growth potential into numbers

The potential size of the opportunity remains subject to fierce debate, but forecasters estimate that total demand for data centres, as defined by power consumption, could hit 35 gigawatts (GW) by 2030 in the US market alone, up from 17 GW in 2022. A GW is a unit of power equal to one billion watts, often used to describe the output of large-scale power plants or other energy systems.

The US currently constitutes about 40% of the data centre global market. Industry analysts estimate that the future power consumption related to AI deployments could increase from approximately 1GW in 2023 to 7GW by 2026, representing a $12 billion revenue opportunity for data centre operators and 15-20% growth versus existing total data centre capacity.

Equinix owns a portfolio of 248 multi-tenant data centres across 32 countries, that host 10,000+ companies, with 450,000+ inter-connections between them. Their services have grown beyond offering physical space, power, cooling, and connectivity to increasingly focus on network services, growing their competitive advantages. In 2023, Equinix identified a $21 billion addressable market for data centres services to support AI by 2026, based on their current operations and capacity. Whilst they will likely only take a percentage share of this opportunity, it could significantly grow their current revenue of around $8 billion per annum (source: company, June 2023).

Can our power grids sustainably keep up with AI computing demand?

Supply was already struggling to keep pace with surging demand for data centre space in major metropolitan areas, prior to the mass adoption of AI tools. Supply chain bottlenecks and a lack of distributable power have led to low vacancy rates and significant pricing power for data centre owners in these locations.

AI software is trained on enormous quantities of data to return informative responses. For advanced applications such as these, GPUs are used in place of Central Processing Units (CPUs) given their ability to more efficiently process multiple computations simultaneously. Whilst GPUs are more efficient per byte of data processed, total power consumed is likely to increase as new uses cases are introduced, as suggested in the forecasts above.

As data centres begin to host more high-power GPUs to support AI, the draw on our electrical grids will only become more acute. Some utility providers are already setting quotas that limit the construction of new facilities, making existing facilities more valuable. Singapore is case-in-point, with new data centre power quotas tightly controlled by the government and disproportionately awarded to incumbents such as SingTel, who can help to arrange novel solutions to renewable power requirements.

The AI supply chain must work hard to derive more efficiency from existing resources. Innovations around liquid cooling and optical networking technologies are advancements set to reduce the energy consumption needed to run high performance chips. What’s more, data centre operators and occupiers are working intently to add new large scale renewable energy supplies onto the grid.

AI adds to the unprecedented demand for data centres

Data centres, as the critical infrastructure behind the digital economy, are set to play a key role in delivering new AI tools to consumers and businesses.

As with any new technology, the ultimate size of the AI market opportunity remains open for debate. Data centre operators will need to carefully manage environmental and regulatory concerns. We believe that companies with long track records of innovation should see their revenue growth augmented by this new wave of applications.

Following an extensive deep-dive into key Asian cities, we believe that the region presents some of the most exciting growth opportunities. Moreover, ‘first movers’ with the most well-established digital infrastructure networks are best placed to capture the explosive demand to come.  

Sources:

McKinsey & Company (2023). Investing in the Rising Data Centre Economy. McKinsey & Company. Retrieved from source.

Evercore ISI (2023), Equity Research Note

Nvidia Q1 Results, May 2023

Equinix, 2023 Analyst Day, June 2023

Important Information

Marketing material for professional clients only

This document contains indicative terms for discussion purposes only and is not intended to provide the sole basis for evaluation of the investment solutions described.  The management of the portfolio may be/may have been delegated (including partial delegation) to another Schroders entity.  Any reference to sectors/countries/stocks/securities are for illustrative purposes only and not a recommendation to buy or sell any financial instrument/securities or adopt any investment strategy. The material is not intended to provide, and should not be relied on for, accounting, legal or tax advice, or investment recommendations.

Past Performance is not a guide to future performance and may not be repeated. The value of investments and the income from them may go down as well as up and investors may not get back the amounts originally invested. Exchange rate changes may cause the value of investments to fall as well as rise.

Schroders has expressed its own views and opinions in this document, and these may change. Information herein is believed to be reliable, but Schroders does not warrant its completeness or accuracy.

The forecasts included are not guaranteed; they are provided only as at the date of issue and should not be relied upon. Our forecasts are based on our own assumptions which may change. We accept no responsibility for any errors of fact or opinion and assume no obligation to provide you with any changes to our assumptions or forecasts. Forecasts and assumptions may be affected by external economic or other factors.

Any hypothetical results shown herein must be considered as no more than an approximate representation of the portfolios’ performance, not as indicative of how it would have performed in the past. It is the result of statistical modelling, based on a number of assumptions and there are a number of material limitations on the retroactive reconstruction of any performance results from performance records. For example, it does not take into account any dealing costs or liquidity issues which would have affected a real investment's performance.

This data is provided to you for information purposes only as at today's date and should not be relied on to predict possible future performance. There can be no assurance that this performance could have been achieved using tools and data available at the time. No representation is made that the combination of investments would have been selected at the commencement date, held for the period shown, or the performance achieved.

Any third-party data including MSCI data is owned or licensed by the data provider and may not be reproduced or extracted and used for any other purpose without the data provider's consent. Third party data is provided without any warranties of any kind. The data provider and issuer of the document shall have no liability in connection with the third-party data. The terms of the third party’s specific disclaimers, if any, are set forth in the Important Information section at www.schroders.com.

Schroders will be a data controller in respect of your personal data. For information on how Schroders might process your personal data, please view our Privacy Policy available at www.schroders.com/en/privacy-policy/ or on request should you not have access to this webpage. For your security, communications may be recorded or monitored.
Issued in the UK by Schroder Investment Management Limited, 1 London Wall Place, London EC2Y 5AU. Registered No. 4191730 England. Authorised and regulated by the Financial Conduct Authority.