Inside Africa's Data Centers Revolution | Bloomberg Next Africa

On this edition of Next Africa, the massive build-out of data centers is underway with billions committed to powering African economies. We unpack where the continent fits into the growing ecosystem. Plus, Google expands its network above and underground. What that means for the continent's digital infrastructure, cloud ambitions, and the future of AI. We're trying to build AI infrastructure so that it's close to where the users are and where the demand is going to be. And in the case of Africa in particular, we're always at a dual goal, which is infrastructure for Africa in our minds is both connectivity and compute. And what's power got to do with it? Why the data center power debate may play out differently than we see in other markets. Welcome to Next Africa. I'm Jennifer Zabasanja in Johannesburg. The AI boom is driving a global race to build more chips, memory, and data centers. And Africa wants a place in that expansion. As Bloomberg's Tiwa Adebayo explains, one of the biggest infrastructure investment opportunities in decades isn't without its challenges. The rapid adoption of artificial intelligence is fueling an unprecedented wave of investment in data centers and the digital infrastructure that powers them. McKinsey estimates the industry could require up to 6.7 trillion in cumulative investment by 2030, making data centers one of the largest infrastructure build-outs of the decade. And the ripple effects extend well beyond the tech sector. Utilities, construction firms, cooling technology providers, semiconductor companies, and data center rates are all positioned to benefit. However, the challenge isn't building the facilities, it's powering them. Goldman Sachs expects data center electricity demand to increase 165% by 2030, putting utilities and electric grids under growing pressure worldwide. Now let's take a closer look at Africa, where the data center build-out is gaining momentum. Investors are expanding capacity as demand for AI, cloud computing, and digital services continues to grow across the continent. For more, I'm now joined by Loni Prince-Lew, Bloomberg senior tech reporter here in the studio. So Loni, thanks for being with us. Tiwa was just explaining the global outlook for some of the build-out we're seeing. What does it look like across the continent, where you look? Basically, it's a mix about independent builders, and then you have your hyperscalers coming in. You have your telecoms companies like an MTN coming into the continent. So there is a good mix. We have Equinix currently from the U.S. saying that they will invest $400 million over the next two years in Cape Town and South Africa. We have TerraCode, which is your biggest independent builder, continuing to invest. There are some smaller players like Araxio. They recently raised a bit of money to build. More underserved areas like Angola, Mozambique. Yeah, so there's definitely some action on the continent. Demand is picking up with the AI and the big language models and so on being used. Yeah, we're seeing some action. And it seems like each of these companies has a different strategy for where they're focused. When we compare, though, investment flows from some of these companies and the hyperscalers, how does it compare to what we're seeing elsewhere in the world? So it's picking up, but it's not quite comparable yet. McKenzie is expecting that we will have demand of about just over two gigawatts in Africa in the next four years. That is compared to our 400 megawatts that we have installed currently. Last year, we gathered about $3.5 billion in investment. We're expecting that that will grow to $9 billion in the next four years. That's not comparable to the hundreds of billions of dollars that we're seeing in the US, but it's definitely picking up. Demand is definitely picking up. There are people there trying to build out and pick up that demand. We'll need about $20 billion in the next four years of investment to get to over that two gigawatts that we need. South Africa's Terraco has become one of the continent's largest data center providers, serving cloud giants, including Microsoft, Amazon Web Services, and Google. And while AI is driving a global boom in massive training data centers, Terraco's CEO, Jan Nijsdu, believes Africa is unlikely to host multi-gigawatt AI campuses because of the continent's power constraints. In African countries, I think, edge data centers, smaller sites like we operate in Durban, those are starting to proliferate. It makes a lot of sense because you want the content to be as close to the end user. So they have an amazing experience. You wanna be able to access content, video, and get it straight away and consume it. I think from a cloud perspective, though, that becomes a bit more challenged in the rest of Africa, just by way of economies of scale, it's large-scale infrastructure. And I think that that's gonna develop in hubs. So at the moment, the major hub is South Africa. And then the two other hubs that are starting to merge is Nigeria and Kenya. Which is interesting because there are, as we see with data centers elsewhere, power is a big debate and water is a big debate. Why is that not happening as these hubs are forming? Yeah, two important debates. And actually becoming even more topical as everyone becomes more aware of sustainability. Actually, there's a lot of myths as well. So maybe the easy one to address is water because water use comes up quite often. Actually, we use technology. So firstly, we operate in a water-scarce environment, South Africa's water-scarce. We have large facilities in Cape Town where it can close to day zero. So we incorporate technology. It's called a mixture of free air cooling and a closed-loop system. So we actually don't use water in the ongoing process. So actually, if you take all our sites and aggregate our water use, we use probably 1 20th of the average golf course in America over the course of a year. So a bit of a myth with regards to water use. Renewable energy, that's probably more challenged. And there, we're a little bit in the hands of our national utility. You know, they've done really well. So we now moved into a power surplus position. But we've got a renewable energy target. We want to get to 100% renewable energy by 2035. How do you get there? So again, we're in the hands a little bit of ESCOM. They obviously, we're an emerging market, 80% fossil fuel. They have a path to improve that, but we want to accelerate that. So we've invested $125 million in our own utility scale solar plant. It's 120 megawatts. It's located about 100 kilometers outside of Johannesburg. And we'll be wheeling that power across ESCOM's grid to our sites in Cape Town and Johannesburg. Is that, do you see that being the models that we need to see in some of these other hubs, as you call it? I mean, is it a combination of private and public coordination? Yeah, I see that happening more and more. I mean, if you take Terraco, we don't just connect ourselves onto the network. We're actually investing in the grid as well. So we're building substations, whether it's for the municipality or for ESCOM. We'll end up building the substations as well as the connections to our site. So we're investing in the grid alongside them. And yeah, to the extent that they aren't able to provide us with an accelerated renewal path, we'll do it ourselves. Where does AI and higher computes come into the conversation because we've also been talking to quite a lot of people about AI data centers and more and more of them that are being rolled out. Is that something that Terraco is thinking about? And do you think maybe the African continent is a bit early in that development? So I think I'd differentiate our adoption to AI data centers. I don't think the two need to necessarily go hand in hand and I'll come back to that. But these large AI data centers that you're reading about, it's almost like weekly coming up in the news. Large sites, hundreds of megawatts or gigawatts being built in the US, in the Middle East and in China. Cost of electricity is obviously a huge factor. So the cost of electricity is like between six to eight US cents. I don't see these big sites being developed in South Africa or actually on the African continent. I think it becomes a little bit more challenged to do that. What I do see and what we'd like to see coming into our facilities is the inference model. So those big gigawatt sites is the training models, the learning models. Once that's been refined into an inference model, let's call it the production model, that needs to be, again, like content close to the end user because you're throwing a whole lot of information at this model and then you want to get an answer as quickly as possible and then act on it. It could be a prompt, it could be interactive with a bot. You want to get an answer super quickly. So we'd want to get those models into our sites. And if you add that up and aggregate it, potentially it could be sizable demand. Why not AI data centers though? So our data centers are AI enabled. I do see AI coming. I just don't see these huge gigawatt factories coming to us. The demand or the supply? I just see, from a power perspective, it'll be challenging to do a gigawatt site on the African continent at this point in time. So then what is needed to enable some of these to get to where you see as maximum operation? So a function of two things. Power infrastructure, gigawatts, a lot of power. And the second one is the price around us. So again, you're competing against these markets. We're the Middle East, very low cost of electricity. Some of the U.S. markets, six to eight U.S. cents. You'll have to get the cost of electricity down to that price. And coming up, the battle to build Africa's AI backbone from data centers to subsea cables. My conversation with Google's James Manjika is next. Welcome back to Next Africa. As the race to build AI infrastructure gathers pace, Google is expanding its footprint across Africa, investing not only in data centers and cloud computing, but also in the subsea cables to support the next generation of AI services. I spoke with Google's Senior Vice President James Manjika about why the company sees Africa as a long-term AI infrastructure play. So we're trying to build AI infrastructure so that it's close to where the users are and where the demand is gonna be. And in the case of Africa in particular, we're always at a dual goal, which is infrastructure for Africa in our minds is both connectivity and compute. So we wanted to make sure we had compute infrastructure close to the users who are gonna need that compute capacity and infrastructure in Africa to either deploy models, transform their businesses and build on top of it. So this is part of that whole long-term strategy that we've had as we think about Africa. I keep hearing you say going to be. Is the demand not here yet in your mind in terms of AI demand? How do you see it developing over the next few? Oh, it's only gonna grow. I think we've seen a huge surge in demand for compute for AI, both on the one hand for training models, but on the other part for actually serving models. So when people are using, whether it's chatbots or any of the other capabilities in the AI systems, they need to be able to have inference capacity to be able to use these systems. And that demand is only going up. That's why you've seen as a company continue to make bigger and bigger investments in compute infrastructure is to try to meet this moment and the demand that we see in all our markets. If it's gonna grow, where is it at then right now, would you say? It's growing. It's very fast growing. I mean, that's why, as I said, that's why we're making these big compute investments because of that infrastructure and the demand for it. And the biggest part for compute for AI, the biggest part that's growing the fastest is actually what I call the inference part of this because the training of models, that's typically done by us and a few others who have training models, but inference is what you need when you're trying to serve users who are demanding to be able to use these models. So that's why we're building compute infrastructure. When you do inference, you typically wanna do that close to the markets and in the regions where the demand is. So when you see us make investments in Africa, for example, in our cloud data center, is because we actually see growing demand for that compute. Do you need local partners on the ground? What's your capacity to do it on your own and what are some of the local partners that you're working with in order to get there? Well, whenever we're doing a lot of these things, we work with local partners. So yes, we may design the data centers ourselves, but often the build out of those data centers and the ability to actually operate them, we often work with partners. In this case, we're quite glad to be working with some African partners because Cyber Technologies, for example, is one such partner that we're working with. But we're trying to work with different partners in different regions. And some of those partners sometimes are the governments themselves. Because keep in mind that when, for example, when we make, whether it's infrastructure investments or, for example, the cable and connectivity investments, it's important for local entities and governments to be able to connect to that infrastructure. So they often have to make investments themselves to connect to that infrastructure to make use of it. What are some of the challenges in that? I mean, we know there's been obviously the conflict and tension in the Middle East. How do you deal with some of the geopolitics that may be inevitably get tangled up in some of the own projects that you have? Well, we try to make sure that we're building our infrastructure to be resilient and to be available everywhere in the world. So we have redundancy in our systems. And thankfully, we haven't seen any of our infrastructure get affected by some of the conflicts we see. But we always try to make sure we have very resilient infrastructure wherever we are. And so all of the cables then around the continent have been able to withstand any of the tensions that we have? We haven't seen anything yet. In fact, we're actually, as I said, we're very excited about the Equiano cable, which goes from top to bottom, but also the recently the Umoja cable, which also connects Central Africa down to the tip of it and then connects out to Australia. Because again, data costs, I think, are so important. I always think that in the case of Africa, you kind of have two infrastructure challenges, compute and the cost of data and connectivity. Whereas in some parts of the world, you don't worry as much or think as much about the connectivity. But in Africa, you have to think of both as real challenges. Coming up, powering the AI boom. We discuss why electricity may be the biggest challenge facing data centers. Welcome back to Next Africa. As investments in data centers gather pace across the continent, operators say solving Africa's power challenge will be critical to supporting the next wave of AI and cloud computing. These machines are at the heart of Africa's digital revolution. Data centers enable everything from video streaming and AI requests to online banking and cloud computing. But keeping all that data moving isn't easy or free. It requires guaranteed power 24-7. We've also deployed backup generation ourselves and we are working to onboard other providers that will provide either gas to us or provide energy to us as well. So we've made significant progress. We've connected the data center, power is running and we are ready to go. With Africa's data center market expected to more than double to almost $8.8 billion by 2031, there's a lot riding on the need for uninterrupted electricity supplies. Nigeria's KASICloud is one company hoping to capitalize on the investment in data centers across the continent, building facilities in Lagos. Cloud and AI factories and AI data centers will be critical as Africa begins to migrate digitally. So we decided to focus on West Africa because West Africa has the population and it was an easy decision for us. Focusing on the capital to build it and build it at scale was the real challenge. But they're also faced with the question of providing constant power for their data centers. Roxio Group, which develops carrier neutral data centers across several African countries, says the continent cannot build a digital economy without local infrastructure. CEO Robert Scott says this is key and will need to include independent power producers or IPPs. What should happen, in my mind, is that you need to build an IPP structure. So IPPs, companies are more than willing to build power plants, supply the power, et cetera, to these data centers. But what we need in many countries, not so much in South Africa, but certainly in other countries, is the permission for private power to supply to private offtakers. And that is a change of regulation that does take time. But regulators are starting to take action. Nigeria's central bank will require all banks and payment companies to store transaction data locally by 2027. This could have positive knock-on effects. You can't do the digitalization without having your data local. And that's what these governments have realized and are actively pushing for. And we want it to be part of that journey. As more African governments push for digital information to stay local, data centers seem destined to become more than just storage facilities, but a resource which could shape the continent's digital economy and its fortunes for years to come. And for more, we're now joined by Bloomberg NEF's Senior Research Associate, Nelson Nsetem. Nelson, thanks for being with us. What would you say to someone who claims that the power required and the compute for data centers in Africa isn't as great of a factor as it is in the West? What is it that your research has pointed out? Yeah, I think we find that to be true. So it's true that data centers in Africa typically are less power intensive, less power demand coming from these data centers. Typically in Africa, you have more traditional cloud-based data centers rather than AI serving or training data centers that are doing much heavier computes as you'd find in the West. So power demand going to data centers is much less. If you look at South Africa, the largest data center market in Africa today, you see data centers only accounting for less than 1% of power demand. And that means over the next 10 to 15 years in our economics-driven energy transition scenario where we look at power demand growth for across a range of different drivers, we see data centers along with electric vehicles being the largest drivers of power demand growth in Western countries, the UK, the US, Germany. But in Africa, we expect power demand growth to come from more traditional sources like air conditioning, buildings as well. So the power compute is much less coming from these data centers that are cloud-based in Africa. Well, so if that's the case, Nelson, if power availability is a factor, but maybe not the driving force, what then is maybe superseding some of the discussions around where these data centers get built? Is it land? Is it taxes? Is it fiber? What have you and the team found? So earlier this year, we published research tracking five key points that tell us how suitable different markets are for data center developments. And like we've said, energy availability, reliability and cost come up high amongst those things. But the other things we consider were fiber, taxes, land availability and permitting as well, and also actually existing data center capacity. Because on that, we found that it's simply just easier to develop a data center in a market where there's an existing labor force, there are private fiber networks that allow for low latency communication with other data centers. So all of those things are quite important, but the relative importance between these things depends on, again, the type of data center being built. For a traditional cloud-based data centers, again, which was still the most common in Africa, we find that fiber access is the main priority. So you see a lot of build in clustered areas, clustered commercial areas, or by the coast where there's access to subsea cables as well, because again, most customers will be willing to pay a premium for low latency. But for AI data centers, we again see that energy and land availability are much bigger factors. Cost is a big factor there as well. So we're starting to see that there will be potentially more interest in markets either with existing grid capacity that's clean and cheap, Ethiopia comes to mind, huge hydro capacity there, or that has great solar wind resource that can actually serve off-grid data center demand as well. Regarding that final thing, the last bit of analysis that we put out this year was trying to compare how these data centers will be powered. Will they be powered by clean power or by gas or existing coal grid generation as well? We do find that solar plus batteries and wind can be competitive, but it's really, really difficult to get 100% uptime that a typical data center would require. And that's why many providers are choosing South Africa, I'm sure. Nelson, before we let you go, you mentioned some of these countries that we could see more data centers being built out over the next few years. You said Ethiopia, Egypt, I mean, what else sort of comes to mind? Because it's interesting because we're not necessarily seeing most of these providers going to the countries that you mentioned. So what does this look like? Where does it look most interesting for you? Yeah, South Africa will continue to be a big market. That's where we see most of our planned capacity by data center developers. So South Africa, Kenya, Nigeria already existing strong markets. Again, it shows the extent to which existing capacity and also demand for web services and demand for AI services potentially is driving that. So you're seeing it in the biggest economies, South Africa, Kenya, Nigeria, that's where we see a lot of demand. And some of the other countries, like I mentioned Ethiopia with cheaper power, potentially we've yet to see that level of interest, but in the longer run, those things will start to play a part. And that is Next Africa for this month, but don't forget our weekly podcast with more of what's happening across the continent, available on Apple, Spotify, and wherever you usually get your podcasts. You can also subscribe to our daily Next Africa newsletter, available on the terminal and website for even more reporting from the Bloomberg team. I'm Jennifer Zabisaja in Johannesburg. We'll see you next time.