The exponential growth of data centers

Their history goes back to the 1940s and the early military and government-owned computer systems in the US. They flourished with the rapid spread of the internet, email, video streaming and conferencing. Another growth spurt came with the rise of social media and, hence, cloud computing. And they are now inextricably linked with the exponential growth in artificial intelligence, or AI.

The total amount of data created and consumed globally is expected to triple to more than 180 zettabytes in the five years to 2025, according to Statista. One zettabyte is equivalent to a trillion gigabytes, or the storage capacity of 100 billion top-end iPhones. Other forecasts suggest similarly mind-boggling growth.

In economic terms, this is a $400 billion global market, reckons Statista - though other estimates are lower - expanding at $100-150 billion in new capital investment per year, so at more than 30%. And it is not slowing down, judging by the sector’s financing needs.

Valtin Gallani, Group Head of Societe Generale’s TMT Finance & Advisory Americas, says: “Our pipeline is growing rapidly, and the rate of growth is not decelerating. Currently, there is no sign that it will.”

The rate of expansion in the Asia-Pacific region, at 20-30% a year matches that of North America, adds Eugene Tan, Head of TMT Finance & Advisory at Societe Generale in Asia Pacific, and shows no signs of slowing either.

From hype to hyperscale

Broadly speaking, three types of company construct and operate data centers. Telecoms companies or smaller ‘Mom & Pop’ providers have for many years built so-called ‘co-location’ facilities that allow a number of enterprises, financial institutions and other organizations to house and run their servers offsite under one roof. 

While this is a steady business, virtually all of the market’s recent growth has been in larger (and increasingly, much larger) data centers for the ‘hyperscalers’ that dominate cloud computing and are now investing heavily in AI: Amazon, Google, Meta, Microsoft and Oracle from the US, but also China’s Alibaba and Tencent and, lately, Nvidia and OpenAI. 

Perhaps 50% of the time, the hyperscalers build their own data centers, taking advantage of substantial cash piles and a competitive cost of financing. But construction is not their core business.
This has led to the emergence of a group of specialist data center builders and operators: companies like Vantage, CloudHQ, AirTrunk, QTS, Equinix and others, who do this on behalf of their giant customers and who, like them, are largely from the US, which currently dominates the sector.

The specialists find suitable plots of land, handle the permitting and install the latest power and cooling technology, and sometimes the fiber optic connections as well. Within two years, if everything goes smoothly, they hand their customer a finished ‘shell’ (building) with anti-septically clean data halls (the ‘white’ space) and everything else (the ‘grey’ space) fully operational. 
The hyperscalers then install their computing equipment into the white space, often spending multiples the cost of the shell. 

Bigger and bigger

It is hard to convey just how large and expensive these pieces of infrastructure are becoming, says Mr Gallani. His team used to primary finance and advise wholesale and enterprise co-location facilities that cost an average of $50-60m to build; over the last 2-3 years the team has been primarily arranging financing for data center projects that cost well over $800m to build, and sometimes north of $3.0 billion. These new data center facilities are built-to-suit for larger hyperscaler tenants like Amazon and Microsoft that AI and Cloud computing workloads. The team these days is structuring loans that are over $2.0 billion in size, financing data centers that encompass multiple floors and sometimes multiple buildings – but are being leased by a single hyperscaler tenant. 

The market is also evolving in other ways: training AI’s large language models does not require the same latency as, say, high-speed trading, so operators are able to move further away from prime city centers and into secondary markets where land and power are cheaper.

Beyond its cost, the sheer availability of power is becoming perhaps the sector’s key issue given the steep increase in demand that data centers are projected to generate - and the creaking grid infrastructure in many countries. The big hyperscalers have also set stringent net zero targets, considering the environmental impact of such infrastructure. So they need not just lots of electricity but renewable electricity. It is noteworthy that Microsoft, Google and Amazon have all struck deals this year to source nuclear power, which they value as both clean and stable.
Big capital needs

The ability to find suitable sites and arrange speedy grid connections is a core competency of the specialist operators, notes Mr Gallani, and explains why they are growing so quickly. In turn, it means they have vast capital requirements, which is why Societe Generale works closely with many of these firms to structure their own financings, which increasingly include green and sustainable loans or bonds.

Since these operators are investing heavily now but only reap returns over the course of 10- or even 20-year leases, a large proportion has traditionally been owned by private equity and more and more are now moving under the umbrella of large infrastructure funds. The $16 billion acquisition of Australia’s AirTrunk by Blackstone’s infrastructure arm is a case in point.

Societe Generale provides financing & advice on such M&A deals as it seeks to grow its already meaningful market share in what is one of the world’s fastest growing sectors.

More on the same topic:
•    Success Story: Supporting the development of a data centers leader in Europe
•    Video: Unleashing the Power of Data and Connectivity - A Snapshot of Some of our 2024 Deals