Asia Centric by Bloomberg IntelligenceArtificial intelligence is fueling demand for data centers across Asia, with capacity set to double in the next four years, according to Moody’s. Projesh Banerjea, KKR director and Southeast Asia infrastructure lead, joins John Lee and Katia Dmitrieva on the Asia Centric podcast to discuss how investors can best tap into that growth – and how it can get complicated.
You're listening to Asia Centric from Bloomberg Intelligence, the podcast that explores the big ideas and trends moving money across the region. I'm John Lee in Hong Kong.
And I'm Katie Dmitrieva, also in Hong Kong.
Katya. When investors think of AI, they always think of Nvidia. That did you know there was a stock last year called Vertive Holdings that went up four times more than in Vidia.
What does the company do?
They make cooling systems for data centers.
Cooling systems, so not the data centers themselves, but the infrastructure for the data centers. Imagine there's a lot of investment opportunity there. You know, I was reading the Moody's report recently and in this region in particular, data center capacity is set to double just the next four years. So you can think about how much opportunity is there. And that's actually what we're going to be discussing today. We have a guest today, Projesh Banerjee, who's the director of Infrastructure at KKR, joining us.
Hello Projesh, thank you, thanks very much for having me.
I wanted to ask you, is there a race to build data centers right now?
Across Asia. What does that look like?
Eris is probably a strong representation. There's certainly enormous demand for data centers. There are really two major trends that are driving these volumes. One that's been playing out for a long time, and that's just the continued migration of data to the cloud from on site servers, and this trend has been a key driver of DC capacity growth over the last few years. Globally. ISSHA is playing catch up on this front as well. And the new trend, of course is AI and I'd say we're probably seeing this a little more in the West, in North America and increasingly in Europe than we are in Asia at the moment, but it's almost certain to play out here over the next few heres as well. And this is driving another step change in DEMI. So yes, there is an enormous amount of demand and activity across is in the DC space at the moment.
And PROTESSHA, can we take a step backwards? What exactly is a data center? I think people have some preconceptions, but I'd love to hear your explanation.
Yeah, what do they look like?
Well, thankfully they aren't particularly eye catching. Something's gone wrong if someone's invested a lot in the architecture and then you sort of drive by and you go, ooh, this is a tourist attraction. Hopefully these are buildings that you drive past and never really realize are there. They're just very large structures with a lot of different data holes. Just think of them as cold dark rooms, very purpose built for the most part to house a huge number of servers and digital infrastructure where data is stored, very very specifically built from a power water cooling perspective to be energy efficient or as an energy efficient dispossible and essentially maximize space and minimize power usage to the extent possible to keep the service cool. So a physical structure recently nondescript building which is basically full of goal of dark rooms which house service.
Yeah, you mentioned the water and electricity component. Is there kind of a trend and where these things are located and also does that kind of impact your investment outlook and where these data centers are actually going to get built?
Yes, there is, and it gets a little nuanced because it goes into the type of data centers and they're you know, historically say three types of data centers. So there are some that focused on interconnects, so these are more carrier hotels where there's just a lot of cross connectivity between people who house equipment in those data centers. The same there's enterprise data centers, which is a little more focused on enterprise clients. And then there's what is driving a whole lot of growth and capacity, and that is the hyper scale data centers, which is where you see this week step change in demand. That's more AI driven. Obviously from a cloud computing perspective, those are the types of clients you know we speak with as well, and from a location perspective, the hyper scale side gets quite technical around the client's network architecture as well. So if you think about how most companies architect their networks, to say, Amazon for example, has what it calls a cloud region, and when it opens cloud regions in certain countries, there are multiple availability zone so think of these is just basically zones that have multiple different data centers with their own power, cooling, physical security, connected through very low latency and redundant fiber networks as well, and cloned so that if you one data center goes down, hopefully the end customer doesn't experience an outage. So yes, absolutely. When we look at investing in data centers, we think very carefully through the type of data center and whether or not specific parcels of land from a development perspective are located within these zones and regions where the customers are building out their own network architecture. And if you do have you know the sites that are flood proof, connected to the utilities, with access to power transmission so substations and transmission capacity as well, then they can be frankly as valuable as called. They're very very attractive pieces of land at the moment.
Asia Centric is produced by Bloomberg Intelligence, where more than five hundred experienced analysts and strategists work around the clock to bring you timely, world class research. Our coverage spans two hundred market indices, currencies, commodities and industries, as well as over two thousand equities and credits. If you like what you hear, don't forget to subscribe and chirm and progress. You mentioned the hyperscalers like Amazon, Google, Microsoft, Where are the hyper scalers and your cells? Which countries are seeing the biggest demand for creating these data centers?
So honestly, there are different drivers across regions. In markets like frankly, in almost every market across Asia, there's an almost demand. India, for example, I think it's forecast to grow it over a twenty percent giga over the next five to seven years. That's true off Tokyo as well, that is true of the Philippines, true of Indonesia. These are very very high growth fits. And of course they're starting off of small bases as well. So one of the reasons we see the growth is because they're also historical under investment or there just wasn't a need for this type of capacity to be set up in these regions. But for a whole bunch of reasons, you know there is that need today. Data sovereignty is one of them. But also from a market perspective, these markets are maturing and are attractive enough for the hyperscalers themselves to be investing in a very meaningful way. And so for us as data center investors, we're investing alongside or ahead of our end clients, typically Big Deck on the hyperscale side, to ensure that we have this capacity up and running as and when they need it.
So you guys sort of pave the way you come in.
You kind of lay the groundwork and then you get the Googles and Amazons of the world on board.
It's it's a close partnership, I would say. So you do need to invest ahead of the client, but it's not quite speculative, so there's a strong sense of whether or not there is demand. So it's very collaborative in terms of working with the Googles and Amazons and Microsoft's to follow them or and often in many cases to actually have them direct the management teams to where their needs are. So these companies are very large owners and operators of data centers themselves, so these are captive data centers that they themselves manage and build. So often if we see them moving into a certain market and executing on a self build plan, then we know that there is demand there, and we know that there's probably room for a third party provider as well.
Building these data centers, especially those related to AI, must be really expensive. A lot of people are talking about how expensive it is to buy these Nvidia chips used for artificial intelligence. Now, can you give us some information who actually buys the invidio chips? Is it yourselves or is it the hyperscalers?
So that part the active equipment, the stuff that goes into the servers themselves is typically the client's responsibility, so that would be the hyperscalers themselves. The data center owners and operators typically are just responsible for the physical to more the gore shell, the civil infrastructure of the M and E, just making sure that the data holes are available, which we then lease out to the hyperscalers, you know, for them to put that equipment in. And to your point, the active equipment, so the servers and the chips going into the data centers are often a lot more expensive than the data centers themselves. So the rule of thumb is that it's called it sort of ten dollars of what for a data center itself. The equipment going in will be many times that.
There's a lot of discussion of how much computing power AI requires, and this is obviously sounds like it's driving the demand for these data centers. Can you give us some numbers if AI really takes off, is there enough capacity right now? And how much needs to be built to satis fi all that demand.
Look, there are many numbers that get thrown out, so I'll refrain from giving you numbers because they're quite country specific and client specific, just in terms of the power needs. But yes, it's not just the absolute megawatts. It's also the architecture within the data centers themselves. So the intensity of the power use is quite different for the latest generations of chips versus historical ones. And this is changing. You know, in two or three years, we're seeing enormous differences in terms of the types of power from a racked density perspective. You know, the needs of the latest generation of say the in video chips. So you know, some of the examples are just from a power demand perspective. I think globally, data centers, if historically accounted for maybe two to three percent of overall power demand, that is expected to go up there maybe about ten percent in some mats. In some markets, like say Singapore, DC is already account for ten percent of power demand, and so there's actually a moratory in some countries in terms of letting data centers continue to run unleashed, so power is being rationed out. But this is an enormous step change from a power demand perspective and what that means for associated infrastructure as well, both on the power generation side but also the transmission lines that you need the utilities that manage these electric networks is frankly an upgrading of a lot of old architecture because it is moving nations as power demands in a meaningful way.
If data center demand is rising, if it's becoming a lot more energy intensive. Are all countries prepared for that right now? I know you mentioned India, the Philippines, US with DC there. I mean, are all of these places ready or do you see the potential for the need to do a lot of upgrades.
I think you probably need to see upgrades across the board because it's also just the way some of these electrical networks, of the utilities, transmission and distribution networks were built didn't necessarily require that volume of power to certain regions and sights. So there's often bottlenecks in these systems, and it's ensuring that there is that substution capacity and redundancy, and you know there are no bottlenecks in terms of just how that transmission system was built out. So there's different degrees of upgrading that are required. That's on the utility side, but in a lot of different places. Countries that are power shot need more primary electrons in the grid anyway, So lots of markets were playing catch up. From a power perspective, there is now a whole new stack of demand that has come in, further increasing the need to have more electrons in the grid. So, yes, it's a lot to think through.
And is this extra power demand coming from renewable energy or is it just traditional sources or both both.
I think it depends on the market. It can be important powers. So Singapore is talking about multiple import projects around the region. Obviously most of Singapore's powered ninety plus percent today is gas. A lot of other countries will be in a similar position.
Yeah, you had mentioned that. You know, there's this picture of demand growing.
This is a trend that's not going away, but are there also risks to that outlook? You know, we were speaking with Deron Amoglu a few weeks ago in an episode. He's one of these AI critics. He sees this sort of as AI not having as big of an economic impact or perhaps as big impact on industry as we previously thought or as people assume. So if a lot of these data center growth is being led by AI, you know, is that one of the risks going forward that we could see maybe a pullback in that trend or do you see that as as staying.
Another great question. Look, so there are definitely different points of view here right around whether investments in AI makes sense for the people who are really buying the chips right investing behind the architecture and all the costs, and people will say, oh, this is a bubble. There have been deck bubbles in the past as well, but this is very different. So you have some of the largest, most sophisticated and well capitalized technology companies in the world investing behind this wave, as opposed to say, you know, the late nineties, early two thousand and one, it was a lot more startup. Is there some degree of hype? Frankly, you know, it's not my place to comment on what the fair price for some of these companies should be. I think a lot of people can can form their own opinions on that front. But on the data center side, these investments are typically underpinned by long term contracts to these very high quality counterparties, and so there's enormous comfort in knowing that the counterparty will honor the contract because these tend to be pretty rock solid contracts as well, and provided you're in resting in the right regions. There could be other use cases. Will it be a large language model that is ultimately trained there, or inference occurs. Could there be a cloud zone. So long as you're in the right location, there are lots of different uses that you can put power and that capacity to. So maybe the numbers won't play out to the extent that they are projected today. But one thing about the data center space is that anyone who's historically forecast growth here has typically been wrong in terms of being conservative. Demand is actually outstripped historically what almost anyone has forecast as the market demand to be. So you wager a beer at your own perle. But I my estimate is that demand's actually going to be creative in what most people expect today.
It sounds like a great business, like you make these long term contracts. Your counterparties are these massive hyperscals with a lot of cash. Everyone knows how strong they are in terms of the balance sheet. There must be a lot of competitors in this space. Is like tell us about the competitive landscape that you guys are doing.
You're right, they are great businesses if you can access the customers, if you have the right team and the right sites, and it takes a long time to build that trust and that connectivity and to have frankly invested in the space for years to be able to now reap the rewards of those early investments and those locations and that power capacity and so on. So there are lots of different folks who are trying to enter this space, and it is i would say, getting tougher and tougher to do so because access to capital is a differentiata. These are enormous gazzlers of capital these businesses, and obviously without the fundamental assets, without the land and the power and the mutility connectivity, the fib or the water, ultimately you can't build. So the people who are best positioned to capitalize on the wave are the ones who've been investing in their business for many years now and in the relationships with these hyperskillers. And the markets are fragmented. There's no player that is such a larger market here that we would say, you know, is the only one that wins. This isn't a winner take all market. This is one where frankly, many firms should be able to do well provided they execute and deliver for their clients.
Is it mostly large firms in the space or you just said it was fragmented.
Is there a lot of smaller players?
It depends on the market. So every nation will have some firms that are very large. The guys who are very large in Australia aren't necessarily very large in Singapore or India or Japan.
For example, So it's really domestic.
It's a combination. So there are some players who are multi country platforms, and there are others who are very large in one country and are now looking at building out across the board. Ultimately, our viue is that good platforms will move from country to country as their customers expanding those countries as well, So you sort of follow your customer in.
This business and protecs, who are your main investors? Are they sort of pension funds, sovereign wealth funds? Like are they rich individuals? Tell us like, who are the people that's backing you to invest in these data centers.
So our LPs are a combination of exactly those types of folks, typically a blue chip roster of the most sophisticated capital alligators in the world, and they can be pension funds, sovereign wealth funds, large insurance companies, institutional investors. It's a broad mix of both types of investors and cross geographies as well in terms of where these investors are domiciled and pratrescia.
You mentioned quite a few regions, in particular in Southeast Asia, and it seems like Southeast Asia is getting a lot of the growth, but you did not mention China. Now there's obviously a lot of geopolitical tensions between the US and China. Is this in any way impacting where hyperscalas build their data centers?
I think what I can probably say is that data is regulated, whether that's requirements to store data domestically, you know how that data is used. It gets quite technical and illegal quite quickly. Companies, I think are taking a view on where their growth is going to be, and Western hyperscalas are probably expanding across regions where they see more prolonged growth and less regulatory challenges.
Yeah, so lots of things to consider when you're trying to open a data center. Yeah, what is something that people might be missing when it comes to data centers.
I think there's a tendency to assume that all data centers are the same, or just to paint the entire sector with one brush, and that is actually not the case. So with all the changes that are happening now on the chip side and then there architecture side, the older data centers are in a lot of cases increasingly more challenged, so deployment sizes and that's just effectively megawatts are getting larger. So it's possible to have you know, a ten year old or fifteen year old site that is not quite after scratch in terms of where technology or design and architecture have moved on from a power usage effectiveness and from a scale perspective. So while there's a lot of excitement around the sector as a whole, there is a buyer beware sort of you know disclaimer as well around what type of data center you're buying and how relevant that is in terms of lease up and location for future growth as well. There's so much need for new assets to be created, there's real risk of some stuff getting outdated in the space.
What are some examples of that?
I know you can't talk about specific companies probably, but you know general, what's an example of that situation? Where is it just a data center that becomes very old and maybe even so old that.
It can be updated.
Or is it more that you know, technology is advancing so quickly that even if you start building one now, it might be outdated by the time it's done.
Now, it's more the former, I would say, because remember here, the long term contracts protect you for the duration of that contract, so at least for a while. For so long as you have that contract, you're in a pretty good place. One example I can give you is actually in our portfolio. One of the underlying PCs for our investment in Cintel's data centers, a company called Etceera for finance last year, is actually the reallocation of power to more efficient and new data centers as opposed to some of the older stuff which is the first generation. So third and fourth generation data centers today, especially in a market like Singapore, will be a lot more efficient and you can get a lot more by aggregating the power supply that goes into some of these older sites and reallocating that to a new facility that is just more purpose built. It's a great story, you know, to give the centrality team credit. They thought this through and real at least to us pioneers in the space around trying to rebuild or repower these facilities in Singapore and effectively squeeze more it megawatts out of the same power allocation, so really avoiding emissions. There's a real environmental story here around avoided emissions because the data center itself takes the same power down from the grid, but is able to sell on more of that power to its customers because it's just more efficient. So the building needs a lot less power to cool itself, among other things. So we're seeing that as a very interesting angle in certain markets that are more power constrained, and we're frankly investing behind that team as well, just just building, reallocating from the old and building new, more purpose built facilities.
Sort of what's happening with the office space now, I guess as well? A lot of empty space and then you're just reallocating it to apartments or yeah.
Yes, I so I'm more sure what the environmental angle is there, and if the apartments are more powered you know.
Oh no, just from like a space usage perspective, yeah.
Yes, repurposing and redesigning absolutely from that perspective. Yes, The interesting thing here is you're actually creating a lot more value by reinvesting in the assets, because the scarcity of power in certain markets just mean that the value ascribed to each lease is getting higher and higher. Sur pricing is going up, and so by rebuilding, oftentimes you're actually creating more value for yourselves and for your customers who are looking for larger footprints to deploy.
Across Katya data centers. A lot more technical than I originally thought.
Yeah, there's so much to think about with the data centers.
Where to put them, who's investing, how to cool them off?
Very educational. Thanks for joining us, Prajesh, thank you so much. I'm John Lee in Hong Kong.
I'm kadjd me true, but also in Hong Kong.
This podcast was produced by Clara Chen and you've been listening to the Asia Centric podcast
