The 4Ds megatrends: Digitalization

The hype around ChatGPT and generative AI earlier in 2023 brought a sudden jolt of energy to the sector. Some digital infrastructure investors immediately became skeptical about the hype, as they have seen similar stories play out before.

Alex Leung, Head of Infrastructure Research & Strategy

The COVID-19 pandemic highlighted the importance of digital infrastructure, as high speed internet became essential with the accelerated adoption of remote work and education, and the popularization of high-definition video conferencing. Digital infrastructure deal volumes set a high-water mark of almost USD 200 billion in 2022.

Since then, deal volumes have declined significantly. Just as public market investors began scrutinizing growth-oriented technology stocks, private infrastructure investors have also become cautious about digital infrastructure investments, especially those that were underwritten using overly optimistic debt-fueled growth assumptions.

The hype around ChatGPT and generative AI earlier in 2023 brought a sudden jolt of energy to the sector. Publicly traded data center companies such as Digital Realty (DLR) and Equinix (EQIX) have seen their stocks increase by 20% this past year, after experiencing severe declines of up to 50% from their peaks in 2021.

EQIX and DLR stock performances (Indexed to 2019)

Some investors immediately became skeptical about the hype, which in this case, we were sympathetic to at first, since digital infrastructure investors have seen similar stories play out before.

For example, two years ago, cryptocurrency was supposed to accelerate data center investments due to crypto miners’ unsatiable demand. Some of those projects have since been abandoned. The metaverse was also supposed to drive the adoption of decentralized “edge” data centers and internet-of-things (IoT). That also has not happened.

At the same time, investors should not be too dismissive of this opportunity, and take a more balanced approach. Unlike crypto mining, AI is championed by large and credible hyperscale companies, who are also some of the most important data center customers. They will therefore take the lead in determining whether new infrastructure investments are actually needed or not. Based on estimates from various sources, the total spend on AI and AI related infrastructure could be enormous.

AI and data center investment forecast (USD billions)

Within digital infrastructure, AI will therefore benefit hyperscale data centers the most. The opportunity for colocation data centers is less obvious, as hyperscale’s are driving the adoption of generative AI, although this can change in the future.

Despite many large numbers being thrown around, the actual infrastructure investment opportunity will likely be in the tens of billions, as pointed out by Equinix (see above chart) and McKinsey. Keep in mind that enterprise spending such as software and equipment (servers, routers etc.) are often lumped into AI investment estimates, even though are typically spent by customers, rather than by the infrastructure provider.

Global data center construction capex excl. equipment and enterprise spend (USD billions)

The impact on other digital infrastructure assets such as fiber, towers and small cells will be longer term, as current AI learning models can be trained at data centers in remote locations without the need for high connectivity or low latency (akin to sending your child to a remote university to avoid distractions).

Technologically, AI-focused data centers use GPU chips that consume up to 4x more electricity per rack than traditional data centers that use CPUs, which raises concerns about carbon emissions. This requires a rethink of data center design, including the incorporation of cheap and cleaner energy sources, and the use of advanced cooling equipment (e.g. liquid cooling). Obsolescence therefore becomes a bigger risk for AI-focused data centers, especially if new equipment or chip designs¹ emerge.

Geographically, since AI training does not require much fiber connectivity, data centers beyond traditional markets such as Northern Virginia and FLAPD (Frankfurt, London, Amsterdam, Paris and Dublin) with access to cheap renewable electricity will become more attractive (e.g. Texas or the Nordics). This trend is already evident based on the recent investment announcements, as none of these projects are located in traditional data center markets.

Recent data center capex announcements

Although it is good to have a healthy amount of skepticism, sophisticated investors can still find opportunities as long as they can separate commercial behavior from hype, and focus on infrastructure characteristics. This includes having high quality counterparties, long-term contracts, access to clean energy, minimal technology risk, and avoiding speculative capex.

1. For example, Google is developing TPU chips that are designed for AI and could be significantly more energy efficient and cost effective