As a bank that’s active in the digital infrastructure sector, both in South Africa and the UK, we regularly ask ourselves what the future looks like for the sector, and what risks lie ahead when it comes to demand for digital assets (data centres, fibre, tower companies).
Specifically, there seems to be a huge expansion in data centre (DC) investment currently. Investec has been fortunate to be part of this and partner with some of the best in the industry. Massive growth in any asset class requiring finance is great news for a bank (and the economy), but any responsible financier needs to contemplate what the implications of this growth are. Could it be a bubble? Or is there a genuine sustainable need for the new investment?
Towards the end of 2022, ChatGPT launched, and with it, artificial intelligence (AI), and generative AI in particular, became the phrase on everyone’s lips. This has certainly been the year of AI, and it seems to be a story that is only just beginning. Its rapid penetration into all aspects of life points to the fact that it remains in its infancy and will continue to be applied and expanded globally.
Marc Ganzi, CEO of Digital Bridge, one of the pre-eminent investors into digital infrastructure and a shareholder in one of the DC specialists we finance, recently said that global data centre capacity will need to grow by nearly 300% to meet the incoming demand triggered by AI. According to thetechcapital.com, during the company’s second-quarter earnings call, Ganzi said: “Public cloud, which really has been building and leasing space to the data centre marketplace over the last 10 years, is at about 13 gigawatts."
“Ultimately, to drive AI and to get networks to where we think they can go, we believe the opportunity set is close to 38 gigawatts. So, we are just literally in the first innings of a potential 9-inning baseball game.”
Want to know more?
Find out more about how our Corporate Leverage Finance team can support you.
Ganzi also made the bold statement that, based on the feedback Digital Bridge is getting from its investee company CEOs and their conversations with customers about ramping up capacity requirements, “the opportunity is going to be at least as big as the public cloud market was over a decade ago”. Nvidia’s CEO, Jensen Huang, recently predicted that USD1 trillion will be spent on upgrading data centres for AI over the next four years, due to the simultaneous shift into both AI and accelerated computing.
A consideration for us here in South Africa is that generative AI workloads are notorious for their high-power consumption. The latest AI chips, produced by companies like Nvidia, AMD, and Intel, consume two to three times the power of previous generations. Large language models like ChatGPT have billions of parameters, which means they consume a lot of power.
READ MORE: AI investing 101: What to consider before jumping in
As South Africans, we are acutely aware the country’s power supply challenges, so it’s no surprise that power is one of the key constraints to new DC development in South Africa. However, thanks to the rollout of renewables, many DC operators are now procuring green power (usually PV) through wheeling arrangements. So, in addition to the investment made directly (often as foreign direct investment), these DCs are also enabling new investment into renewable energy plants, since they are the kinds of power purchase agreement off-takers that commercial and industrial energy companies need to attract finance, which in turn results in green power for other industries as well.
All the factors seem to indicate that the global demand for DCs will accelerate rapidly, not taper off, so we are likely to see further advances in DC investment in South Africa across edge, enterprise, hyperscale and AI-specific sub-categories (there are many different types of DCs being built globally that are specific to the service that each provides and engineered to that service).
While AI is getting the headlines, it’s not the only factor driving demand for DC space. In an insightful article in TechCentral, Duncan McLeod explains that Microsoft plans to move Windows fully to the cloud. This means that, instead of every laptop and PC being pre-installed with Windows when you buy it, you will in future stream Windows, using the modern broadband networks to push Windows to our PCs and laptops.
Where does the capacity to do this sit? Where will all the storage and processing capacity be situated, in order to effectively power every laptop and PC? You guessed it, in a DC. What are the advantages of this? Potentially, we won’t have to invest thousands into expensive hardware, but instead we will be renting space in the cloud, space that is used to process what we need, in a DC that is owned or leased by Microsoft.
Why would Microsoft do this? I decided to ask our in-house Investec AI engine, Zebra GPT, and this was the answer:
“Microsoft is interested in moving Windows to the cloud for several reasons. One of the main reasons is to make Windows more accessible to users on different devices and platforms. By moving Windows to the cloud, Microsoft can offer a more consistent experience across devices, as users can access their Windows environment from anywhere, using any device that has an internet connection."
“Another reason is to reduce the cost and complexity of managing Windows environments for businesses. By moving Windows to the cloud, businesses can offload the responsibility of managing and maintaining their Windows infrastructure to Microsoft, freeing up IT resources and reducing costs."
“Finally, moving Windows to the cloud also enables Microsoft to offer new services and features that are only possible in a cloud environment, such as AI-powered automation and advanced security features. Overall, moving Windows to the cloud has the potential to offer many benefits for both users and businesses.”
This seems like a very intelligent answer, and it took all of one second to generate (perhaps illustrating my earlier point about the processing power behind AI).
The positive effect of having gigabit-speed internet connections, facilitated through fibre networks that have been built in much of the world, is seemingly going to have a profound effect on the way we use our home and even work computers.
As McLeod says, this is not unlike the old days, when companies had dumb terminals all connected to a mainframe computer in the building. The more things change…
At the end of the day, it seems that the digital infrastructure space is one that will continue to feed off itself. As more applications are developed to utilise the potential of our fibre networks, the opportunities for other assets – such as DCs – multiplies, and that spurs more fibre investment. A great example of a virtuous circle.
To return to my initial question: it seems that any concern that there may be too many DCs being built is unfounded; perhaps we need even more to be built.
Browse further in