Tata Consultancy Services Limited (TCS.NS) Earnings Call Transcript & Summary

Good afternoon, everyone. I'm Kawaljeet Saluja, Head of Research and Lead IT Analyst at Kotak Institutional Equities. I'll be your host and moderator for this session. Friends, please ensure that your display name includes your first name, surname and company name as the system will shut your access. Today, we have with us the [ exec ] team of TCS for a fireside chat on the topic of TCS' AI data center investment and opportunity. Let me start by introducing the team. The gentleman in the center is Samir, needs no introduction. All of you have met him in the past. Our next panelist is Mangesh Sathe, who serves as the Chief Strategy Officer of TCS. He also serves as the Head of Global Consulting Practice and oversees the M&A function of the company. More recently, he was the CEO of Tata Strategic Management Group at Tata Sons, where he supported Group Chairman's office and CEO of group companies in strategy and transformation initiatives. Our third panelist for today is Deepesh Kiran Nanda, who brings with him over 3 decades of leadership across technology, energy and infrastructure. He has led large-scale power and digital initiatives from India through Japan and Australia, including senior roles at Tata Power and GE. Deepesh is widely regarded as a thought leader on sustainability, clean energy and the intersection of power and digital transformation. We'll start the fireside chat, post which we'll open up the floor for audience Q&A. Do note that this call is only to discuss TCS' data center initiatives as such. So we request the participants to restrict the questions on those slides.

Okay. Let's kick this off. The first question for you, Samir. Why data center? What prompted TCS to get into building colocation data centers? And what are the adjacencies that you have with the core services business?

Absolutely. Great question to start with, Kawal. And if you recall, during our Q2 earnings call, we announced our aspiration to become the largest AI-led technology services company. And AI and sovereign data centers are a key component to the -- overall AI value chain. And it provides TCS a unique opportunity to offer an end-to-end solution in this value chain. So it's not a stand-alone decision or an isolated decision but part of the overall -- the 5-pillar strategy, which we called out during our earnings call. Now the key driver to the investment is the demand and supply mismatch and addressing this demand creates a nice niche and an adjacency for TCS, which we believe will give us long-term committed annuity revenues. Now if you look at it from a synergy perspective, we benefit from both front-end synergies as well as back-end synergies. So that is another driver. So data center provides a unique opportunity to TCS to deepen its partnership with the various hyperscalers and AI companies. And while driving this efficiency, we also can leverage the One Tata data synergies. So all in all, if we look at TCS' dominant position in the India IT services market, -- we -- the strong partnerships with global technology companies, which we possess, the access to capabilities and most importantly, capital and the support from Tata companies, which we can look at towards connectivity, green power, infrastructure development will be the key differentiators. And then data center forms one -- the core foundation for our overall aspiration to become the world's largest AI-led services company.

Fantastic. That's a great way to start off. With the initial focus...

Sorry, Kawal, just one point I'll add. See, if you look at globally, data center market itself is also expected to see a lot of capital investments over the next decade. And this I'm not talking only about India, I'm talking across the world, especially U.S., Europe, et cetera. So one of the other benefits will also be that our expertise in working in this sector, we will be able to also then provide services related to this sector in other markets as well because that's also an important element. So because a lot of build-out is happening, which I feel technology services companies like ours should actually participate and accelerate that whole capital investment as well. So that's an added benefit to all the stuff that Samir anyways outlined.

So when you refer to the capabilities that you can take outside India, is it largely the services capability? Or would you also be open to asset-heavy business outside India?

No, I was referring to the services.

Got that. Got that. And would the investment initially be largely on colocation side? Or are you looking at integration with active infrastructure as well?

So our overall current business plan is to look at only passive data centers. We are not averse to active data centers, but it has CapEx consideration, technology refresh considerations. So any incremental considerations on the active side will be more bespoke commercial arrangements with our anchor customers rather than it being -- so as it stands, it is more passive.

Got that. And Mangesh, Samir made an important point that there's a demand-supply mismatch in the Indian markets today. But when you look at the colocation data center space, the space is heating up with multiple conglomerates announcing a gigawatt capacity plans. So how would TCS' approach differentiate itself from the competition, which also seem to be in a rush to announce the gigawatt capacity as such?

Right. So I think there are 2, 3 factors here. One, the overall demand itself. So if you see the current footprint, which is around 1.5 gigawatts, which is installed in the country. What we see from a potential perspective over the next 5 to 7 years, we see this to grow to anywhere between 10 to 12 gigawatts, right? So that's a significant, I would say, headroom that is kind of available. If I were to add all the plans that have been announced so far, I think there is still a significant headroom available, right, given today, the players that are currently in the market plus additional players who are planning to come into the market. So if I add all of that, also still there is a headroom. So I don't see any challenge from a demand perspective. Now if I just -- the second part will be, okay, now so where is this demand going to come from? If you see the Indian market specifically, I think we are going to probably look at hyperscalers, of course, driving a large part of the demand, right? And we see the same globally. I think the percentages in India would be similar to global or slightly higher but I think hyperscalers will certainly be one key component. The second interesting segment that has emerged is the AI companies. And AI companies, of course, there are a few which everyone is aware of but then there are also new companies, new model companies, new application companies in AI who are also scaling up and are also thinking of raising a lot of money globally, by the way, and they also need to invest into data centers. And then, of course, there is the private sector and the public sector as well. So now why can TCS really garner this demand? If you look at across all these 4 segments, hyperscalers, AI companies, public sector and private sector, we are working with all 4, either in the -- on the global side or on the India market side. And today, we actually have a good, I would say, front row seat visibility on what those requirements are, right? And as a company or anyone wanting a data center, they don't really want to just do a very transactional piece where they're just talking to a data center operator saying, give me a data center, then somebody else for some other services. They would ideally like somebody to come in and provide, as Samir said, the end-to-end capabilities and offer it to them. And we see that as a unique proposition that we can provide compared to some of the competitors that exist in the market.

Okay. That's fascinating, and fairly clear as well. Moving to Deepesh. Deepesh, what are the specifications of the planned data center that you have in mind? Would this be, let's say, Tier 3 or Tier 4 data centers? What kind of rack density are you looking at? What kind of cooling infrastructure? And who do you think are the competitors in the market today in India, which are operating those Tier 4 data centers?

Thank you, Kawaljeet. Taking a cue from what Mangesh mentioned earlier, we are focused on the AI side of the data center. So if I were to look at the market and I broadly split the market into non-AI and AI, our emphasis is largely going to be on the AI side of the data centers. That being said, it comes with a much higher rack density. So we are looking at anything greater than 50 kilowatt per rack all the way up to 180 kilowatt per rack and on the upper band, around 370 kilowatt per rack. We should average out at 240 kilowatt per rack. With this comes the impact on cooling. So these would be heavily skewed towards more liquid cooling. We are looking at upwards of 70% liquid cooling. And that will then reduce our PUEs to anywhere between 1.25 to 1.3 as compared to the current PUE what you see in the country, driven by air cooling of 1.45 to as high as 1.6. So more focused on the AI side, more focused on high rack densities, more focused on more liquid cooling. And hence, our PUEs will be capped between 1.25 and 1.3.

Fascinating, Deepesh. But do you think that you require those kind of high-power density, super-heated 240-kilowatt racks? Because for that, essentially, what you would require is large language model training or some, let's say, use cases around that. Otherwise, those 10 to 12 kilowatt hour -- 10 to 12-kilowatt racks can do fine. So are you basically sensing demand on AI, large language model training for you to support this kind of a capacity build-out?

Actually, that's a great question. And that's a space that we want to occupy. And to answer your question specifically, that's the kind of demand that we are seeing in the early days of the announcement that we've made. The demand is more on the AI side. And with that comes quantum load concentrated load, which is in, say, 100-megawatt multiples. So the answer to the question is yes. There is demand of that nature from international AI companies. In the Indian market currently, the AI load is in the range of 2% to 4%. Globally, it's a range of 10% to 15%. It's the right opportunity for India to get into that AI load servicing, and that's exactly what we'll be catering to in the first phase of the rollout.

And I think as far as training versus inferencing, I think each company and depending on how their data center footprint is distributed, I think they are designing basis their own considerations and latency and where their markets, demand pools, et cetera are. So I don't think we can really comment on whether training will be more or inferencing will be more. At the end of the day, whether it is inferencing or training, the GPUs will require a certain amount of infrastructure that will need to be supported. And what we are talking about is creating an infrastructure, which should be ready to support those loads irrespective of whether it's training or inferencing.

Curiosity, Mangesh, let's say your inferencing workloads, is it better supported by edge data centers? Or would your large data centers also be effective for inferencing?

It all -- see, it depends on the use cases that will be there and also who is putting up that capacity. See if it is a hyperscaler kind of a capacity, they would ideally want to put it up in a larger data center. Of course, if it is a private enterprise optimizing its own loads in -- for their specific footprint, could be a manufacturing footprint or any other footprint, they will probably design it differently. So I think each one will have a different view. From our perspective, I think our capacities are not really edge data centers. We are creating the core data centers. And that's where our focus will be right.

Right. And do you hit the ball out of the park by straightaway starting off with 1 gigawatt or Samir will tone everyone down with a smaller capacity initially?

Well, building it at 1 gigawatt would require a lot of many things, not just capital in terms of availability of infrastructure and everything also. Our plan is to phase it out. It would be in the range of 100 to 200 megawatts each part of phase, and it would be across multiple phases. So it would be over a period of time, a period of probably 5 to 7 years when we'll hit peak capacity.

Okay. And let's say, for the first 18 to 24 months, 100 to 200 megawatt, right, how much would it cost to build this data center? And how do you fund it?

So practically, as we had said in our earnings call also, each -- about this capacity will be approximately $1 billion. and it won't be completely funded through TCS' cash war chest. It would be a combination of debt and equity. And we are in discussion with -- and it could be -- the equity could be one or more partners, and we are currently in advanced discussion with a few partners and also in advanced discussion with multiple bankers on the debt structuring side of it.

And just to be clear, on the earnings call, Samir, what you essentially said that 1 gigawatt would cost around, let's say, $6 billion to $7 billion. So when you're talking about 100 to 200 megawatt, it would be in proportion, let's say, $6 billion to $7 billion, 200 megawatt. Will be $600 million to $1 billion?

Correct.

Right. And what do you think should be the optimal debt equity structure in these kind of businesses? Because I...

Currently we're in advanced stages, we'd not like to call that out, Kawal but we have a structure in mind. It would be typical of what data centers operate at. But right now, we'd not want to give out how would be that structuring. But you could -- what -- how it is otherwise funded, you could assume a similar structure.

Got that. And when you're looking at equity partnership, would you prefer that now? Or would you prefer after maybe 2, 3 years once the capacity is up and running and a possibility of better valuation improves? Or do you basically are looking for equity sourcing right upfront?

It would be a combination of -- the partners could be just the equity partner kind of it or a technology cum equity partnership.

Right.

The options are open. We could do it in each phase, we could have a different combination as well.

Right, right. And Samir, the question, which everyone has on the top of their mind is that such kind of businesses alter the ROE and ROCE profile of TCS mix of business, capital intensive, et cetera. What are your thoughts? And how would you respond to a lot of those questions and in a criticisms as well?

So one, if you look at it, given the size of the balance sheet, the strong balance sheet and the surplus funds which we have, this particular investment per se, is not going to be a significant drag on the TCS balance sheet. It allows us to smoothly navigate the investments which will require. Anyways, they are going to be phased out and not the entire investment which you talked about, you mentioned about $600 million to $1 billion is not going to be entirely a cash outflow for TCS. So given that, I don't see it as a major drag in any of the return ratios. TCS has maintained industry benchmarks in terms of margins as well as ROE. Yes, it won't get the same ROEs as you would expect from the TCS business, but we are confident we'll be able to maintain industry-leading return ratios.

Got that. And then a question for Deepesh. Let's say you start off today, right? By what time would the basic shell be ready? How much time does it take to build a power optimized design, build those racks and start getting in the revenue streams? And let's say, how much time would it take to sell that capacity. If you can just give a broad indication of the timelines of both the build part and the sale of that capacity as and when it's ready.

Thank you. So I'll ask the second question first. We're not building any speculative capacity. Capacity is built on firm demand. And to answer your first question, from the time we have land available and we can go into land, we are targeting completion of an 18-month full construction cycle. So from start date, when we press the button, land being available, the delivery will happen 18 months.

It's fascinating, Deepesh, that you say that you have the firm demand in place because when you look at hyperscalers, the way they have been working in India is a little bit different from the developed market experience. In developed markets, you end up contracting even before the groundbreaking day because the clearances are rather easy. Whereas in India, hyperscalers want to identify a spot and location, but only give a firm contract once all the approvals are in place. What you seem to indicate is that you already have that demand in place and a firm contract in place. Is that correct?

I'd say that there is active interest. There is an inbound inquiry. But the way the business planning has been designed is that there is no speculative build-out. The build-out is based against firm demand.

Okay. Okay. So that's clear. So effectively, the first revenue stream that you would have from this would be more towards FY -- end of FY '28, FY '29. Is that right, Samir?

Technically, if you take it, you should get it around the 27%, 28% mark.

Okay. Okay. Got that. And what are your thoughts on the ROE and ROCE profile of this kind of a business, right? Does it, let's say, can it generate returns, which are comfortably above the cost of capital?

Definitely, it would be -- it has to be comfortably above the cost of capital. The IRR we expect on each of the phases at the project level is in mid- to high teens.

Okay. That's interesting and fascinating as well. Okay. Another question for you, Deepesh. Basically, have you identified the land parcel and the location of the initial capacity? Of course, it would be but are you at a privy to disclose that? And typically, let's say, a data center would involve a lot of land clearances, land procurement clearances, civil infrastructure, maybe cabling, electrical substation. There are so many elements involved. Do you intend to build these capabilities in-house or leverage partners from outside or leverage the group capabilities? And if you can just give some broad thoughts on execution of the same.

Sure. So we are looking at land parcels across all the key areas where data center build-outs are happening, which includes Navi Mumbai, it includes Hyderabad, Bangalore, New Delhi, Pune as an illustrative case. In terms of how we go about construction and the contracting, so all the key items will be contracted. It's a level playing field for everyone. We, of course, have in-house strengths, which will be leveraged, but it will be available for competitive bidding across the board. And that's how we plan to go about making sure that the business gets the best that the business needs. And that will be the plan to build out these data centers.

Okay. That's clear. And initially, let's say, the contracting phase, would it the focus will be largely on hyperscalers? Or let's say, when you create a data center, it will be a mix of build-out for both hyperscalers and enterprise workloads. And I just wanted -- I mean, you just briefly mentioned about PUE, in which I guess you're using air and liquid cooling, which will differ from traditional workloads. In that case, how would the CapEx differ from your traditional data centers, which are normally based on traditional cooling mechanisms?

Okay. So to answer your first question, we are looking at large AI load, and that would come from pure-play AI players. It could also come from hyperscalers. Not to say that we won't have a play in enterprise and so on. It will be a combination of offering. In terms of the design of the data center, again, it's build-to-suit. So if an AI data center has a specific build-to-suit, there will be a specific corresponding PUE. While the costing will be appropriate to the PUE but so will be the returns equation in terms of the revenue that, that type of a data center would attract.

Just to add to it, the capacity would be led by anchor customer, which could be AI-led company or a hyperscaler. And majority of the capacity, if not all, would be led by that. And the incremental capacity, which would be available would be available for enterprise or sovereign.

Right. Got that. Now I have a question from Mangesh. Mangesh, if you look at some of the hyperscalers, they seem to be building out capacities on their own as well. I think AWS does have their capacity on -- of their own. Microsoft has some parcels. I think somewhere in Pune. I mean we don't know Google's plan. Google also have some similar plans. So how do you look at a scenario wherein, let's say, these hyperscalers are setting up their own data centers versus working with partners? And how would this dynamic play out in your...

Yes. So I think -- so today, if you see other markets also, the same dynamic is already present, where you will have hyperscalers having their own data centers plus they're working with data center companies as well. So I don't think it will be any different here. See, from a capacity perspective, I think especially the larger companies, which require multi-megawatt kind of a footprint, they will always have a scenario where they will have some in-house plus some contracted data center spaces, and the reasons could be many for doing that kind of a split. So I don't think that would change here as well. There are 2 other factors in India specifically, given, as I told you, the demand outlook that we are seeing, which is now where is this demand going to come from? Deepesh and Samir talked about the AI part. Also, if you look at the total amount of data that gets generated in the country and amount of that getting processed in the country, there is a certain ratio that one tracks. So today, India as a country will be probably somewhere closer to a lower single-digit number on that. If I take some of my Asian peers or some of the more leading emerging country players, they will probably be higher single digits. And then, of course, somebody like U.S., et cetera, will be in double digits. So from that perspective, I think data consumption in the country going higher is absolutely a given trend. And I don't think we need to really validate that. I think the -- you talked about edge cases, et cetera. So the amount of stuff that will now need to be processed and the quality of that process also, like the latency that will be involved. Cost, I think we have not covered that part but cost of, I would say, data center per unit will be absolutely a big factor for anyone to consider, right? So for hyperscalers, it's a capital allocation plus capacity allocation. For us, it is going to be a factor of what can we create uniquely in the country. So I think from both supply and demand, there is sufficient room for us to add value.

Okay. No, that's very, very helpful perspective, Mangesh. Maybe a question for Mangesh and Deepesh, both of you. Interconnect plays a major role in enhancing the stickiness of data center and boosting revenues, right? How does the company plan to address this opportunity? And would that be addressed by TCS? Or would it be group companies or someone else would capture that opportunity as such?

So in terms of the topic of interconnect, yes, ours will be a highly interconnected design, both within our own DC network and also cross DC network. In terms of how we do it, I've answered this earlier, there's a level playing field, and we'll do all that it takes to be absolutely competitive in the market such that the benefit is retained in the company and we win market share.

Okay. Okay. That's clear. Okay. Just a final question from my side. And after that, we will open this up for participants to ask questions. And actually, I can see a number of questions in the chat window as well. But just a final question for you -- for Samir. Samir, you're getting into a capital-intensive phase. There also seems to be a greater appetite for large M&A. Would it mean that there will be any changes to capital allocation policy that you have in place?

So I think, as I said earlier also, we have a strong balance sheet and substantial surplus funds, which we have been sitting on for quite a period of time. And we get beaten on both sides for spending the money or not spending the money. But coming to the capital allocation perspective, our capital allocation policy has been to return substantial free cash flows back to our shareholders. And if you look at the last 5, 6 years, we have been typically returning back 80% to 100% of our free cash flows back. And we have, at various points of time, clarified that free cash flows would be after all investments, which we typically make. Our target will be to maintain the 80% to 100% range.

Okay. That's clear. Now we'll open the floor up for participants to ask questions. [Operator Instructions] In the interim I have actually absolutely a number of questions in the chat window as well. I'll read that out in the interim for -- and in the meantime, in case anyone is interested to have a live voice conversation, feel free to do so. So the first question, which is there in my chat window, let's go with that. What's the CapEx difference between air versus liquid cooling? I guess that's a question for you, Deepesh.

Yes. So the CapEx difference between air and liquid cooling, I think it's in the range of INR 35 crores to INR 45 crore mark. It's in that range. We are more towards the INR 45 crore per megawatt mark with the AI data center focused, high on liquid cooling. There will always be a mix of liquid and air. We'll be in that bracket of 70-30. And again, the other cost aspect is on Tier 3 versus Tier 4. We'll be in that range of Tier 3 plus, which is the demand that we are seeing currently in the market.

Right. And Deepesh, would this INR 45 crores also include the interest component during the build phase of the data center or this is excluding interest costs, et cetera? Excluding...

This is capital cost. This is CapEx investment.

Capital cost. Okay. Got that. The other question is that we are seeing Indian telcos target the data center opportunity. They seem to have partnerships with hyperscalers, domestic enterprise and AI companies and willing to spend more CapEx. So why compete with them?

Yes. So I think, again, I'll go back to the same piece, right? I would have agreed with this if, let's say, the market growth was, I would say, in single digits or probably early double digits. We are talking about multiple factor growth here, right? So from that perspective, there is a lot of headroom. And given the substantial capital investments that are required in doing this, I don't think this is a market where one entity can, in a way, dominate the capacity. So I feel there is sufficient room for more players to come in. And then I think all the elements that we outlined in terms of our right to win or our ability to win in this makes us a more attractive company to participate.

Got that. The next question I have in the chat window is that what kind of challenges or pushbacks did you face from the Board and within the top management team when the idea of entering this segment was first explored.

I think as a team, we evaluated it, put it up to the Board. The Board was receptive enough in terms of -- so it doesn't stuff given, as I said, we have been sitting on surplus free cash flows. We have been considered conservative over a period of time. It's a bet, which we are taking and the Board was receptive to the same.

Yes. And I think just to add, I think if you see it goes in line with what Samir started off saying, which is the whole 5-pillar piece for AI, right? So this is not -- it's not a stand-alone investment decision that it was one amongst another proposal that we had taken to the Board. I think it comes as a very core part of our overall aspiration. And I think if we don't invest in this, we'll not be able to meet our aspirations. So I think this is a very important ingredient. So from that perspective, of course, it also ticks off all the other boxes. So it becomes certainly an attractive proposition.

The next question from the chat window is that can you elaborate on the tenure of the contracts with the hyperscalers? Is it 9 years plus 6 years of renewal? Or is it 6 plus 6 plus 6? Any insight on the tenure of the contract? And if there's any escalator -- price escalator embedded in these contracts as such?

As you said, we are in the early stages of negotiation. So there is no contracting which has firm or finalized yet. But as our discussion states, these will be long-term contracts. That's where I said we expect long-term committed annuity revenues.

Got that. What is your GTM and cross-selling strategy in the business?

Yes. So the go-to-market in this case, for us, as I had talked about various segments, I think the first 2 being on the technology partner side, which is the hyperscalers and AI companies. And then there is the enterprise side, which is on the public sector and private sector. I think on the hyperscaler and AI companies, these are more core partnership conversations where our focus is to really discuss with each of these companies to see what can be shaped together and how can we support their aspiration in this particular market or other parts of the world as well. In terms of enterprise part, we anyways are the leading system integration technology services company servicing the India geography. I think we are already in all those right boardrooms, CEO rooms, CXO rooms where these conversations are happening today. So now our ability to shape those will become better because we'll be able to provide that entire end-to-end piece. So I think on both those parts, we are well covered. And as we progress, then we'll see what additional capabilities will be needed to support this further.

And on that part, Mangesh, are the buyers of decision-makers involved in data centers will be completely different, let's say, from the hyperscalers with whom we have a partnership on the cloud side. So how do you build those cross connects and synergize so as to derive the maximum value from this venture?

So sorry, I didn't follow. So what you're saying is the...

Let's say, on the hyperscaler side, right, people who take the decision on data center, right, colo would be completely different from the relationship that you would have on the partnership and go-to-market on the services side. So there are 2 different words. So how do you make the 2 in India?

Yes. So I think all the companies that we are talking to and all the partners we are talking to, I think our partnerships are pretty deep and long term. So this element is absolutely being facilitated by the organization right from the leadership level, where they are facilitating these connects. So it's not just about a transactional conversation where I'm going to somebody who I normally deal with for a services deal. And then -- so it's being done in a very strategic way at the leadership level, our leadership as well as their leadership level. So we are not seeing any challenges on that aspect as well. In fact, they are very forthcoming on that.

Right. Okay. The next question in the chat window is that are there plans to enter the cloud segment or convert your colocation capacity into cloud into the future? Would you assume that cloud center will be a more sticky business versus colo? And will it help differentiate a business model versus other players that are also simply building colocation capacity? If so, do you have any expertise in the cloud segments and the approach going forward?

Yes. Yes. So see, we have already launched our sovereign cloud business for India and a few other parts of the world. So the short answer is, yes. Of course, I think as we scale up our sovereign cloud business, we will be leveraging the infrastructure that we are building even for that business as well. And the cloud services that we are offering, both in terms of AI as well as the non-AI part are things that we will be scaling up to both public sector as well as private sector. So yes.

Got that. Got that. And the next question is that how big in your view is the AI training and inferencing data center market in India? And will this services business of yours will also get a boost as it provide infra related to the same? That's one of the questions in the chat window. So the question is that how big is the AI inferencing and training data center market in India? How much would that grow by? And would that provide you with any edge in your AI services business as such?

Yes. So I think the -- if I look at it in 2 parts, one, if I see as of today as well as how do we see it going further. Of course, as of today, the market is small given the nature of the services that we are talking about. But what we are talking about are services that we expect to come in the future basis both the requirements that will be directly being consumed by the enterprises or by the AI companies and the hyperscaler companies. I think we talked about a few percentages earlier. So I think globally, if you see around 10% to 15% of the DC capacity overall goes with the AI, and that part is only going to keep increasing as more GPUs replace CPUs. In our case, I think we expect a similar number as we look go ahead in terms of how much capacities from direct AI load that will get consumed. See, there are 2 other factors. See one also is a factor where a lot of GPUs will also be replacing CPUs on a longer-term basis. That's another piece that we might see. That's one part that we need to factor in. And then the third part from a sovereign perspective as well, we will see AI capacities that we will need to deploy to serve the sovereign needs as well. So I think given all these 3, we feel the percentages will go up.

Okay. That's reasonably clear. The next question is that what -- does Tata Communications have any role to play in the interconnect -- or the interconnect, which you offer in the data center? And do you see them as a part of your GTM as and when the capacity comes up?

Yes. So what are the vendors in addition to the other vendors that we will release RFPs to? So the answer is that, yes, they'll be in the mix of the vendors that will give us networking solutions.

Got that. Got that. And Deepesh, Mangesh, can you just help us with the unit economics of the data center business, right? Let's say, what would our revenue per megawatt look like? How much can an EBITDA be generated per megawatt, the payback period, et cetera? And second is that, let's say, when you think about revenue streams, what are the potential revenue streams that are possible in the initial phase? So I guess the capacity revenue, power revenue, interconnect revenues but if you can just provide your perspective on it, along with unit economics, if you are ready for it, I think it's too early. But nonetheless, I think that's the question that has come up in the chat went over.

Okay. So too early. But yes, that will be the model. We're looking at a relatively high EBITDA margin business. And in terms of what will be the contracting structure, typically is that yes, there is a rental yield and then there is margin on the power supply that will be made. Networking is typically a pass-through.

Right, right. And typically, let's say, Deepesh, what will be the useful life in your view of various, let's say, assets that go into your data center. So at one end of the spectrum, you would have a building or a shell, which would have a life of 30-odd years. On the other hand, you have powers and cables, which may have a different life, a battery, UPS may have a very short life. So when I basically look at the useful life of a data center, right, what will be the duration and various components are useful life as such.

Okay. So we can get into design specifics, but I'll give a broad answer that we're looking at a 15-year point before any major overhaul happens. So it's not that life comes to an end at the 15th year because if you look at the power gear, the power gear runs for at least 25 years. These are normally governed by the number of operating hours. So the -- your terminology is a bit misleading. It's on the operating hours. The amount of redundancy that's built into these plants also make sure that equipments are run to that ratio of the redundancy and hence, the life in terms of calendar years tends to be much higher than the operating number of hours. So if you're looking at an answer in terms of years, it's upwards of 15 years.

Okay. Okay. Got that. And Deepesh, if you look at Equinix, let's say, their big point of differentiation is the points of interconnect and the software fabric they have and the more they have built around that aspect as such. So let's say, within TCS, you have mentioned in various ways, your differentiation in multiple questions. If you have to summarize differentiation for TCS overall versus, let's say, some of the global players like an Equinix or maybe a digital reality, what would those be?

I think we have a right to win in this space. We have the relationships that was very well explained by Mangesh. We are in the ecosystem. We understand power. We understand water cooling solutions. We understand networking. In India as such, all these industries are quite mature. We are in the infra business as a group. There is enough infrastructure experience that's available. And then the approach in terms of how execution would happen, you would look at safety in -- that's our #1 priority. Absolutely safe operations. Everybody goes home. Number two is in terms of quality, the benchmarks that we're looking in terms of all the process quality. We're having good discussions with most long lead item vendors in terms of how do you benchmark their process quality and so on. Third is execution in terms of delivery. Once we get the first 2 right, which is safety and quality, 70% of our focus is on execution. And last is cost. And this order is very important for us that we are going safety first, quality second, execution third and fourth is cost.

If I just add 1 or 2 points here. See, I think from anyone setting up data center capacities, I'm not talking about the physical infrastructure. I'm saying once the physical is done and I need to put in all the chips, et cetera. So it requires a lot of integration services. So -- and most of these large companies that we are talking, hyperscalers or AI companies may not have requisite capacities here, the resources here to do that. So from TCS' perspective, we can absolutely help them get the data center up and running. What you said earlier around other players, we will be able to do that very, very effectively. That's one very important part. Second part also from a differentiation perspective would be that if you look at the market, while, of course, we are talking about a certain growth rate, we still have to create that market. Market making is very important in this case, right? So we will be partnering from a TCS side, we will be partnering with these companies to also support them create that market. Now creation of that market would mean what kind of applications will be needed in a certain enterprise or in a public sector, what's the kind of offering will you need? And there, it will always be a combination of a cloud or services or some other elements of applications that we'll need to take. So I think our ability to really also shape that becomes a big differentiator when a partner is talking to us because then we will be working with them to make the market as well. So market making and then the services to accelerate the deployment also become very important capabilities to differentiate.

Right. Okay. The next question from the chat window that I have is that I think somewhere in your answers, it was mentioned that you would make money on -- you would make some margin on power. But I guess most of the contracts, the way it's structured in India, I mean, more or less power is a pass-through for hyperscalers. So Deepesh, when you said that you'll be able to make some money or some margin on power business, where does that belief come from?

Okay. So this is to do with the mix of the power that we would offer. It would be high on green energy and hence, the cost of power that you see today and the mix of power versus the cost of power and the mix that we will offer and the cost of power, there is enough margin for us to make and yet be competitive because the green power is at a much lower delta even on a round-the-clock basis as compared to the current prices on the grid.

Right. But wouldn't that be baked into the hyperscaler agreement on the power mix between green versus normal coal-fired power plants?

Yes, early days. I've just given you a kind of a directional answer that we'll be extremely competitive on the cost of electricity, and that gives us a right to earn a higher margin.

Right. Let's say, even on the renewable side, in many cases, what ends up happening is that there's a weeding charge that you have to pay to the power distributor. So does that leave enough savings if you go for renewables in the process?

Yes. So as landed costs to the data center, right, at my GIS point, all things being equal, the cost of -- the mix that we are looking at in terms of RE versus grid power will be more competitive in terms of the levelized cost of electricity. So LCOE will be lower, all things considered as compared to what we are seeing currently.

That's very helpful. Okay. I think there are like multiple questions but whatever you do is that I'll just take the last 2. The second last question is that globally, power is a major bottleneck in terms of data center build-out. Is that a bottleneck in India as well any potential bottlenecks that could delay your capacity rollouts?

I think that's a great question. And I think -- and I say that with a lot of pride being here in India because over the years, our grids have become much more reliable. And hence, when you compare with the example that was compared to, there's a lot of back of the meter happening in the Western world because the grid connectivity is not available as it is available in India. While in India, though there is much higher demand but then the situation in India on a comparative basis is much better than what's available overseas right now. And hence, that runaway market that you see in the Western markets on the back of the meter, that's not the situation in India. Number two, we are seeing the designing of the data centers. We're looking at dual connections coming in from, say, 2 different substations. In addition to that, we have captive power being made available and also newer technologies with the advent of large-scale battery energy storage systems. It gives us a much higher quality power solution at a much lower cost on an LCOE basis as compared to the Western market, which makes India as an attractive market when compared to the Western markets on a power availability, quality and LCOE basis.

That's a very fascinating point of view, Deepesh. Okay. I was not keen to ask the last question but there have been 8 versions of the same questions. So please indulge me in that final question that do you intend to get into GPU as a service? And the chat window, you can see it versions of those questions. So...

Okay. So I think this was answered earlier by Samir, that there's a passive layer. And I think the takeaway from this call is that we are all in, all hands on deck, and we are very deep in terms of understanding the MEP build-out, the timeline of that, the safety, quality delivery cost. And then we are open to looking at the active layer. We are absolutely open to looking at the active layer. But once we get the first part right in terms of making sure that execution is flawless, then the day will come when we will start graduating towards the active layer. We are open to all conversations at this point.

And just reiterating what I had said earlier, incremental active data center considerations will be dependent on bespoke commercial negotiations with our anchor customers. It won't be an incremental add-on to what we are planning on the passive data center side.

Fantastic. Samir, Mangesh, Deepesh, you have been quite patient. It's very early days but you have answered a raft of questions. I really appreciate your honest answers. Thanks a lot for that insightful session. Thanks a lot to all the participants for your interest. Until the next webinar, stay connected. Stay well, everyone. Have a good day.

Thank you, Kawal and thank you everyone.

Thank you.

Thank you everyone. Take care. Bye.