E2E Networks Limited ($E2E)

Ladies and gentlemen, good day, and welcome to E2E Networks Limited Q4 and FY '26 Earnings Conference Call. [Operator Instructions] Please note that this conference is being recorded. I now hand the conference over to Ms. Raashi Khatri from Go India Advisors. Thank you, and over to you, Raashi.

Thank you, and good afternoon, everyone, and welcome to the Q4 and FY '26 Earnings Call of E2E Networks Ltd. We have on the call Mr. Tarun Dua, Managing Director of E2E Networks; and Mr. Nitin Jain, CFO of E2E Networks. We must remind you that the discussion on today's call may include certain forward-looking statements and must be, therefore, viewed in conjunction with the risks that the company may face. I will now request Mr. Tarun to take us through the financials and business updates subsequent to which we can open the floor for Q&A. Thank you, and over to you, sir.

Yes. Hi, everyone. Hi to all of our team members, all of our investors, and we hope that we have had like a good last financial year. And -- we'll do a brief coverage of like all the progress we have made during the last quarter and last year. And then I will hand it over to Nitin to talk about the financial highlights this quarter and the overall year. And then we'll open the floor to questions. So in this year, like we have like a lot of milestones as a company that as a team that we have achieved. So like we have been able to use our TIR platform for large-scale GPU sectors. And we have helped like LLM team like operate like trailing on large clusters over the TIR platform. So with this, like we have demonstrated a full stack capability of operating both bare metal and container base large-scale deployments for NVIDIA powered GPU infrastructure at a bigger scale. And we have like not only demonstrated that like we have the ability to operate the infrastructure, but also successfully monetized the infrastructure at a much larger scale than we had to when like last year. And this is a testament to the strength -- strengthening of our technology and talent, which has been used to build like very deep in-house capability. Across the entire 5 [ miles ] of the tax. And we have done like major software improvements to improve the reliability, performance and scalability of our GPU infrastructure in the last many quarters. And going forward, like we feel that like we have been saying this a lot that like, look, the future is all AI. This is like a decadal story. Now given how the market all across the world are reacting to the AI infrastructure usage being increased like very, very drastically. So I think like we have been validated in kind of like putting up like speculative GPU infrastructure. So where the world is divided between haves and have not. We have the haves in terms of like having access to speculative GPU capacity, where mass capacity is running out everywhere. So that allows us to kind of like continue our growth. So with regard to like where we see like medium-term and long-term kind of like growing in terms of like the world is shifting to GPUs from the CPUs from rule-driven software to AI-driven software, so we continue to explore like how we can build more infrastructure under our management. So where we have explored equity, where we have explored debt, we are also exploring like a variety of private trade, asset-light model to kind of like bring in rapid expansion to our GPU capabilities. So we have consistently met the industry benchmark for performance for like influence or training or GPU deployments under our management. And we expect our cluster 1 [ B2 ] under 1024 to go live like somewhere in the mid-May. And in a couple of months, we are expecting to kind of like be able to deploy another cluster of 1024, which has already been planned. And then we are very strongly positioned to capture significant uptake in the Blackwell generation of GPUs apart from continuing to run very, very strong on the Hopper division. Apart from this, like we continue to build a plan around like [ B-300, V300 ] and [ Berarube ] deployment. And we will continue to keep increasing the GPUs under our management and under our direct deployment, both through partnerships, through direct acquisition and by exploring like various parenting models that can help us procure more GPUs. So we continue to focus on that. We continue to focus on our end-to-end project management capabilities to deliver AI infrastructure to AI services and AI product companies, AI direct [indiscernible], enterprises in the BFSI and other sectors. So we'll continue to work towards that. So I think like that pretty much covers this that -- so we -- so this has been a good year and a good quarter. So pretty much like our strategy has been validated over the last many quarters where we significantly continue to add to our technology helps in terms of like having the tech talent in-house to be able to booster. So we gave a certain guidance in terms of like where we would be by the end of this quarter, like a couple of quarters back, I think like we have more or less like met and exceeded that target. And we hope to continue to do better in the future. So given that the market is growing very, very positively, and it's a broad trend, like there can be, again, like we have mentioned in the past that there are [indiscernible]. But ultimately, this is a market where the GPU utilization is going up, the demand for tokens through the roof, like India being a country of more than like 1 billion people on phone, like -- and of course, even if they were not on the phone, like there is like an inherent ability to treat every possible piece of information that is going in and out of people, in and out of machines, in and out of enterprise system to be described in the form of tokens. So tokens can be representing voice, text, any other form of data, whether in digital spectrum, individual spectrum, audible voice, inaudible sound waves, ultrasound or whatever. So like everything you've tokenized data for AI. And I think like India as a country [ that ] rates like a lot of data. So we continue to believe very, very strongly in the long-term industry of AI factories being built here in India becoming the AI factory of the world. Now with that, I would like to kind of like hand over for more numbers-driven financial highlights off to Nitin then we'll open it up for question and answer. Nitin, over to you.

Thank you, Tarun. Good evening, everyone. Thank you for joining us today. I will walk you through our financial performance for the fourth quarter and full year close FY '26. Q4 FY '26 again witnessed a good quarter for GPU networks. We have demonstrated our ability to scale AI infrastructure rapidly, achieve high utilization across GPU cluster and convert capacity into strong revenue growth. Most importantly, we have been able to drive our EBITDA margin in Q4. So let me start with the quarterly highlights for the Q4. Revenue stood at INR 956 million, which is up 126% year-on-year and [ 33.7% ] quarter-on-quarter. EBITDA stood at INR 581 million with EBITDA margin expanding approved 60.7%. Profit report tax turned positive at INR 86 million compared to a loss of INR 75 million in Q3. PAT stands at INR 64 million. This performance reflects INR 120 million in profitability sequentially driven by operating leverage and a strong education. EBITDA margin expanded by 413 basis points sequentially. At the same time, depreciation increased to INR 530 million in Q4 reflecting ongoing infrastructure investments. Despite this, we achieved positive EBIT and positive PBT. For the full year FY 2026, revenue stands at INR 2,456 million, which is up 50% year-on-year. EBITDA crossed INR 1,263 million, which is up 30.6%. However, we reported a PAT loss of INR 156 million, which is driven entirely by the depreciation on our GPU infrastructure investment. Our core business remains strongly cash positive, operationally profitable at EBITDA level. As utilization continues to ramp up, revenue will progressively outpace depreciation, improving profit to reported profitability. To conclude, Q4 demonstrates our infrastructure investments are translating into revenue, and we are only on a path for growth. Thank you for your time. Now I would open the floor for the questions.

[Operator Instructions] First question comes from the line of [ Maja Gandhi ] from Bajaj Alternate Investment Management Limited.

First question is regarding the asset-light model that we are looking out for with L&T. Could you throw some light how will the asset-light partners with certifying what sort of arrangement we have?

All, like, of course, we have an MOU with L&T to monetize GPU infrastructure that they are building. So now that is still in exploratory stage. So as the GPU get deployed and like we starting to -- start to work into monetization, like as we know more and more, like we will talk more and more about that. That being said, like this is not an exclusive arrangement. So we will continue to operate at an arms length with L&T, and we will continue to explore other partnership of similar or different nature as well. So there is like a long gap between like, say, pure debt versus pure equity. So there are like a number of structured profitability in terms of like figuring out that how to finance the GPU in partnership with a lot of different types of partners with different objectives. And we'll continue to explore all those partnerships. And as and when, like we kind of like reach a conclusion on any particular partnership, we'll obviously keep everyone informed.

Sure. But will it be like margin accretive or because the CapEx is going to be on other partners?

You need to say here how would the numbers to play. So obviously, it's very, very early to say like how those things would look like. Unless the structure is finalized, like -- we would like, see, obviously, we all work for corporate. So we'll not do anything which doesn't earn us the corporate. So that being said, so like how to put like numbers to that is not something we can do today.

Got it, sir. And on the MRR would you like to guide for the next year because you've been doing this CapEx from your own pocket also this year. So if you can...

We'll continue to kind of like focus on growth just the same as rest of the market is going, like we were to be very, very growth focused. So that being said, that I think it would not do just as to do an MRR guidance for like 1 year in the future. I think like the future is changing very, very rapidly, we can wait. So it was obviously impossible to give like a weakened, weak guidance for MRR and it would be better to kind of like what it quarter-on-quarter than kind of like predicting 4 quarters in advance.

If you can just throw some light on the asset turn for the CapEx that we are doing, at least some sort of understanding we can get in terms of the CapEx.

Like I think we want to get the framing away from things like asset turn. So essentially, like don't look at us as an asset monetization business. Look at us as a technology business. So like I think there was like a recent conversation like around the tokens becoming more valuable. So same set of tokens that were being generated by, say, open force AI or close force AI. Like what used to produce, let's say, x amount of value. We are increasingly seeing that like as the accuracy efficiency capabilities of AI increases, then the generated token also become more valuable for businesses who have figured out how to utilize those token their business. So this means that like kind of like setting us -- setting up ourselves for looking at the value of our outcomes based on the size of our assets doesn't do just as to the business. So ultimately, we want to get away from saying that, okay, this infrastructure produces only x amount of value. So we will continue to explore options for how to produce more value from the same infrastructure by figuring out like where the tokens are more valuable.

Our next question comes from the line of Keshav from Niveshaay.

Congrats on the great set of numbers. Sir am I audible?

Yes.

We procured the platforms like way back in the early quarter, and no, data still not [indiscernible]. What's the reason behind the -- are you facing some kind of procurement issues or something from the...

Global supply chains have been impacted somewhat. So it's always -- like the delays are always like for lot of the [indiscernible]. So kind of like we have been working very diligently towards like making sure that like everything is planned out and everything gets delivered on time. But then you've got to control every single component. And sometimes, like some components can be the most critical one from the point of view of getting the entire deployment done. So we are targeting like the first deployment to go live before mid of May to keeping our investor [indiscernible].

Got it, sir. And so the quarter that upon demand has been strong globally and people are seeing that although GPUs are sold out including the whole one. Are you thinking to spend in the domestic market? So are we changing the skin or [indiscernible] representation in India?

Okay. I think you -- I don't think I have understood your question, but let me try to answer what I have understood. So like -- see, obviously, it has turned into -- convert into a much better market than it was like a couple of quarters back especially for people who have the review infrastructure in place. That being said, that as a company, as an Indian company, we are very, very focused on India first. So wherever we get an opportunity to support Indian companies, we definitely like to do that. Now with that in mind, like basically like we always prioritize India first and then we are also happy to support the global infrastructure needs today. I hope the answer is in the right direction to what you were asking.

Just definitely later, I was just trying to ask like demand has been strong globally that the GPUs got sold out.

So there is fresh -- we are seeing like demand both India as well as globally at a strong both ways. And we definitely try to prioritize India. So -- but then like whenever the demand comes from eventually, like we are happy to fulfill it.

Got it. And then one last question from my side. Employee costs increased this quarter. So if you could give some color on that, like what's the key drivers behind the size?

See, as we grow, like we kind of like start figuring out a more and more interesting problems that need to be filled in this space. Obviously, our goal is to go toward what you call higher value tokens. And that obviously requires the application of quite a high level of talent. So like as we obviously grow the, base effect would still be there. But on the other hand, like we want to maintain a balance of not losing out on future opportunities because we didn't invest today on what was required 12 months later or 18 months later or 2 years later.

Got it. And one last thing, if you could also provide the GPU utilization model if possible like for this quarter.

See, broadly like across our entire infrastructure, not just GPUs. I think we are looking at like 80% plus utilization certainly, across the quarter, I would say, like only in the March month, we are looking at like 80% plus kind of utilization, certainly less than 85%. So there is like elasticity in terms of like basically like how we can increase the utilization for the organization.

Our next question comes from the line of Bharat Gulati from Dalal & Broacha.

Just wanted to get a breakup of -- what is the MRR breakup currently? Can you give us a split between India AI mission enterprise and also a split between what was the GPU contribution of that MRR compared to CPU?

Yes. See, I think like broadly, the way things have suggested like in another couple of quarters. I think GPU contribution would exceed closer-- would be closer to 85%, 90% over coming quarters. That being said, like all these numbers are fairly dynamic. So in the sense that like they really very, very rapidly from week to another week. So I wouldn't like to kind of like -- and the emphasis today is fairly small. So like what's the split is something that like we don't want to worry about today. So like as we grow the customer prices to a much larger number, then it would start making sense to kind of like talk about like, okay, how far you've got a couple of them black well, you've got a couple of outlet of them. That's the point at which like it would start making sense to say, okay, what's happening on Hoppers, what's happening on Blackwells. Clearly, entire universe of GPU, CPU, storage, all of the stuff that we have is like fairly, fairly small to kind of like put a very, very strong pattern of all those numbers today.

Fair enough. But could you just give an MRR split between the India AI mission and our enterprise or our SME clients I was just trying to get an idea of what kind of visibility do we have in that [ 374 million ] MRR. And also just to add on to that, before that our MRR for the previous 2 months, if you average it would come to like a [ 290 million ] of MRR. So just trying to understand the spike that came in the last month, what was the reason for that spike? And also how sustainable is this 374 increase in...

Broadly an increase in overall utilization that has led to the increase in the -- spike in the March month. From the perspective of like overall place, like, I think, like the India AI of like -- yes. So basically, like the overall government business, like across this quarter and has not exceeded more than like, say, 35%, 40%.

All right. So sir, is there some particular reason for that? Because ideally, our missions were supplied or India AI mission was supposed to start in the month of Jan where in March as of this quarter's reported numbers. So what kind of ramp-up are we seeing on that? And just any kind of time lines and what kind of revenues in terms of MRR do we...

That way, we have always looked at India AI mission is that like, look, we are trying to, obviously, like we said, that we are supporting India, we are putting India AI mission, we are supporting Indian companies. And we definitely gave a preference to all [indiscernible] but that being said, like so whatever capacity is not offtaken. So we are not trying to kind of like force anyone to offtake any capacity from us. So we are very, very happy to sell it to outside the scope of either AI mission or the governments were closed. So we have our customers across multiple segments to kind of like solely rely on one revenue [indiscernible]. But that being said, we continue to work very closely with the AI mission for the coming generations of GPUs as well. And we are hopeful that like we will continue to collaborate, continue to work together, continue to solve problems.

Sir, just trying to understand that India AI mission was almost all our Hopper and Hopper series GPUs were given to India AI mission. There is 80% utilization, but India...

We have not scaled up that to that extent. So majorly like whatever was the scale required by the [ lots ] of AI mission, we have provided that scale, but like we have not insisted on them having to increase their scale without like having the product. So we are quite okay with that.

So in terms of our revenue currently, what kind of clients are we currently catering to? And what's the pipeline? You said may we expect Blackwell to get deployed, do we have some orders?

This is a very, very small base, like it's like -- kind of like, okay, there are GPU customers like everywhere in the world, there are GPU customers in India. There are GPU customers outside India, like every possible segment that you can think of, like whether it is enterprise, whether they can [ BFSI ], whether it is education, everyone needs GPU. So it's basically like almost [ customer ] today. So like the [ meter ] and profile and putting like a stat on like, okay, this much of our business comes from here, this mature business comes from there is like -- kind of like very, very premature. I think like we need to grow the base of GPUs to a couple of tens of thousands before these numbers start to make sense. So today, at a very, very small base, like none of these numbers would make sense to kind of track and then kind of like say that, okay, what's the kind of number this quarter, it will say in next quarter.

So sir, would it be fair to say that until you don't build a huge base of GPUs are MRR will be lumpy in nature? Is that what you're trying to iterate that?

No, I think the demand has secularly changed in the GPU world. So like from the era in which like the lumpiness was there because our base was even smaller. I think we have come a long way over there. That being said, the lumpiness will continue to decrease as we continue to build more and more volume of GPUs under our management.

So out of the [ INR 374 million ], what exactly would be long-term MRR that at least we are thinking of visibility for the next 8 to 10 months or a year or so?

See, we are not very, very focused on -- so we have a balance of like basically like thinking about like, okay, what needs to be like 6 months, 1 year, month-on-month, which is preemptible. So we are doing like kind of like a whole series of them. So like it's a week on week effort to kind of like figure out that, okay, what is the best, what to sell hourly, what is it [indiscernible], what to sell weekly preemptible, what is the cost you want to sell monthly, whatever has what to sell yearly. That being said, like wherever larger [indiscernible], we try to do at least like 6 months to 1 year visibility with our customers. And increasingly, for even larger customers, we are trying to look at even longer visibility to at least like 2 years to even going up to 3 years. But that being said, like our view is that like as tokens become more expensive to generate and kind of like to generate more value for people, it is better to kind of like not put all the eggs in the long-term basket, but have like increase judicious mix in very short-term contracts and medium-term contracts and some degree of long-term contract.

Got it, sir, got it. And just one last thing, sir, in terms of the Blackwell that you expect to be deployed in May, do you -- can you give any idea in terms of who will be the customer? Will it be an enterprise, India AI mission and SME? Then would that customer's revenue start to come and...

We will announce that after closing that particular deal instead of like saying anything speculative today.

Our next question comes from the line of Nishant Joshi from Equisense Advisors Private Limited.

I have one question. As you said, companies...

I'm sorry to interrupt you, Mr. Joshi, but your voice is not audible.

Am I audible now?

Yes, please proceed.

So I want to say that, as the company is planning to [ affect ] the EBIT as well as equity and subsequently intend to go for the asset-light model also? So will this change our business model in future of providing hardware, we would be providing more of service will be offering our TIR platforms. So will this lead to a change in our revenue matters?

Think of this as an expansion of the number of business models without shutting down any of the existing business model. So we will continue to own, continue to operate GPUs.We'll continue to acquire new GPUs. We will continue to expand the partnership to kind of like bring increasing number of GPUs in our management available through our feedstock platform, which operates at multiple layers where it is the choice of the customers like what layers they want to buy. And kind of like nothing is like changing from the point of view that, okay, we are not going to do this or we're not going to do that. We are simply saying that like we are going to do additional number of things to expand the universe of our thinking to do a lot more than what we are doing today.

So that could be an ideally additional line of revenue you need to say?

Absolutely. Absolutely.

[Operator Instructions] Our next question comes from the line of Deepak Poddar from [ Spire Capital ].

Sir just wanted to understand, at the March MRR of INR 97.4 crores. So what is our capacity utilization and on what capacity?

So this is closer to overall capacity utilization of around 80% in the March month.

And on a capacity base of INR 3,900 crore?

Yes. So this is like CPU, GPU stored in all capacities altogether. So the utilization is closer to 80%.

Closer to 80%. And I think we have around [ INR 2,050 ] lined up in next 6 to 12 months, right?

[ INR 2,048 ] plus some sales that we are planning to deploy this financial year. Starting from May 1st, lot of [ INR 1,024 ] is expected to go live.

So, by FY '27 end would you be targeting around INR 6,000 of capacity?

I wouldn't want to place a limiting number over there, but like you could say that like that's a minimum number.

At least that -- okay. understood.

That part is already visible. So that's what we've already spoken about.

And you mentioned around 80%, 85% utilization by March. I mean you're talking about by March '27 of this expanded capacity. I mean...

No, this is the previous one by March.

Okay. And my second question is on your fixed assets. We have got around INR 1,500 crores order fixed assets, right? And last 2 years, CapEx has been around close to that only. So this entire is a depreciable asset. I mean, would that be a fair assumption?

I would let like Nitin put his perspective on that.

Yes. So the entire [indiscernible] is the depreciable assets, which constitute of the new [ B200 ], which is currently displayed as CWIP.

Correct. And what is the amortization schedule for this? I mean, is it 3 to 4 years?

6 years.

Okay, understand. And what would be FY '27 CapEx target?

Like we said like we are already planning to deploy 2,048 B200. So that is already in place as far as our loan plans are concerned. That being said, that we're not limiting our [ terms ] to those plans. So we continue to deploy capital under various business models judiciously to expand the GPU footprint like fairly rapidly.

Correct. So any number you have? I mean what is -- I mean I think FY '26 was...

I think a look that is more important than putting a number in prediction.

Our next question comes from the line of Varun Gandhi from [indiscernible] Avenue Growth Fund.

My question is on GPU rental prices. And are we seeing pressure on them? And I asked this from 2 aspects. The first aspect is technologies advancing rapidly. So we see the [ Ruban ] architecture has been disclosed by NVIDIA and which is much more cost efficient for inference compared to your -- even your Blackwell architecture. And number two, is the competition that's been ramping up across domestic competitors, right? So are we witnessing any sort of pressure on GPU rentals right now? Or do you foresee that in the near term?

I think like broadly, the trend today is that like there are not sort of GPUs in the world that people want to buy and deploy. So that's the current trend. Now whether this trend remains like for how long, like it's hard to predict. But the broad secular trend has been there that like basically the -- there are AI believers and they were AI nonbelievers. I think like the proportion of nonbelievers is like slowly dwindling. And it's quite clear that like basically -- so we are mainly expecting the [indiscernible] in terms of utilization of AI. Like we have looked at so many of our customers, so many of the organizations we work with, including our very own. So where we say that, okay, is it the case that like our AI utilization is to an extent that we wanted today? The answer for ourselves that we increasingly get as well, this is going to be like more like 10x, 20x, 50x of where we are today in terms of how we are utilizing AI. So AI is not coming out of the IT budget. AI is essentially coming out of your budget for the capabilities that we are building for the business. So that being said, that I don't think like the demand environment in the foreseeable future is going to be changing negatively for a long period of time. I don't think that's going to be the case from the visibility we have today. So like we don't see that there is like any negative pressure on the pricing today. So in fact, like there are, like, I think, good set of tailwinds which are slowly inching up the prices rather than like a decrease in pricing.

So from a demand standpoint, we aren't seeing any sort of negative. But from a technology advancement standpoint, let's say, if the [ Rubin ] architecture comes in live this year, and it should would that pressure...

There is a lot of workloads that will go into production where they will continue to operate on the targeted architecture. So typically, it's like a what you call like a funnel kind of a growth where the next layer of the funnel is bigger than the previous layer of the funnel. So it means that like the demand for existing GPUs will continue to remain very, very strong. So demand for even the NPL series has remained strong. Hopper series has remained strong, like..

Tarun, I understand the demand has remained strong. But let's say, 2 years ago, the GPU rental prices for Hopper H200s, are they same today or they are lower 30%, 40% lower than what they were quoting at 2 years ago?

I think like taking in the case of 2 years ago is slightly anomalous from the price of [ 2 ] that I think that was 1 particular year in which the demand went from, say, 1 to 10 to like I think like that capability-driven shortages where the capability to print that many GPUs was -- had to be [ bedded ] very rapidly. So that created like a, I think, like a local maximum of pricing. But broadly, we are seeing that like there is stability in the pricing for GPU over an extended period of time. Typically, we feel that like it is possible to kind of like definitely use and utilize the GPUs over a 7-, 8-year period. So we don't think that like the GPUs are like a 2 year, 3-year or a 3-year story and then kind of like there is something available at a much lower cost than you were able to ship. And those shipping costs are like very reasonable. So that's not the case. Today, our entire test demand cycle for doing AI on a new tech sets is like a whole other set of expenses that you don't want to do, like if your job is getting done by a GPU, which is available at a very reasonable price.

Got you. Now my second question is, at the beginning of the call, you touched upon how you've significantly improved your software architecture. Could you just give me a brief -- if you could just highlight a brief example of how you've done that. You iterated that you are now trying to capture high-value tokens. Is there some sort of example that you could showcase?

So this is broadly from the point of view is that like the focus themselves today are more valuable. It's just a point of like choosing the right set of focus that people are going to utilize. So it's not like a kind of like a strategy statement to go and capture one particular set of high value [indiscernible] today. It's a broad [ intrinsic ] outlook that like, look, ultimately we have to go and figure out like which are the high-value tokens. Now that being said, like from a platform improvement perspective, like we have been working on things like which are required by people running training on other clusters, people who are running influence how to support them, people who want to build rapidly the pipeline reliably across larger number of GPUs and understanding of like how -- what is it that the AI data scientists do from have been there done that perspective. So we have made improvements in all those areas for productization of our feedstock.

Right. So just as an extension to this question, see, we're very well positioned to capture the sovereign AI tailwind and my understanding is that BFSI companies would be a key client base over there. Is there any sort of strategic development that you could share where you partnering with some sort of SaaS companies or other software companies and trying to offer a bundle of software plus compute. Is there something you could help us with?

Yes. So as we do something, we'll definitely share that with everyone over here. So like as things happen, like we look back and say that, okay, this is what we've done.

Okay. But there's nothing that you can share in terms of your thought process over here?

See, like we are -- we continue to [ do ] an opportunity. So what we capitalized on, we will come and inform the market.

[Operator Instructions] Our next question comes from the line of Abhishek from InCred Equities.

Congrats on a great set of numbers. Sir, my first question is about the guidance commentary. You'd be summarized your comments on the demand, it appears to be really strong. We have a visibility of 2,048 GPUs. And then we said that we don't want to get fixated on the guidance. So just trying to understand that the purpose here was not to get fixated on the number rather than any worries in your mind about the current macro? Is the understanding right?

Yes, yes. We are not worried about like whatever numbers we give because we will be able to meet them or not. I think it's like we don't want to give a underwhelming numbers without first exploring like over week after week over like 52 weeks like what we are capable of doing. So it will become sort of like a limiter for us ourselves to say that, okay, we are only to meet this number. So that's why we are not putting a number today.

Perfect. That's very helpful, sir. The second question is on the pricing. So contrary to the prior question, I was -- the recent articles are suggesting that the GPU spot rates for April have gone up substantially higher. So for an H100, the prices are up anywhere by 25% to 30%. Can you just...

We are seeing that. We are looking at that. That being said, like we obviously, like we tried to capture some of that. But then that being said, like we are also trying to support the Indian companies and India as a country. So where we want to have a balance between long term and short term. So short term, of course, that we want to make more money and long term, of course, we want to work with the kind of customers we will sustain and grow on our infrastructure over a long period of time.

Perfect. Sir, just last one data point from Nitin sir. So the depreciation for the next quarter should be up by around INR 25 crores a quarter, right? Is that assumption right?

I'll let Nitin answer the questions around the depreciation. Nitin, over to you.

Sorry, could you repeat again?

Sir, the depreciation for the next quarter should be up by -- should go up incrementally by INR 25 crores a quarter, is the assumption right?

So why do you expect that?

Because of the B200 cluster.

B200 cluster would be that we are seeing the time line of [indiscernible] so that's what there would be increase in the depreciation, but not to that significant effect, that's what you are [ turning ] on.

Our next question comes from the line of Neil Munot from PICO Capital.

Sir, my first question is building on last participant's question. The capacity utilization is...

Sorry to interrupt you, Neil but your voice is breaking.

Is it better now?

Yes, please proceed.

Yes. So the first question is basically building on participant's question. Is that the 80% utilization that you're talking about, I'm assuming this is excluding the new B200 that has...

Yes, that is including the B200.

Yes. So when we talk about the 80% utilization, with respect to the remaining capacity available, have you kept the size for POCs or demand is still ramping up investment?

No. So the demand is still ramping up over there. So like -- it's like all cost of difference at port of capacity. So like they all work in different populations. So an ideal utilization number could be like -- kind of like -- it depends on like basically liquid you end up but that's not a limiting factor on the MRR from a very straight line perspective. So for instance, like the same particular piece of hardware for like under different circumstances, we build at like widely varying rates.

What is my assumption?

These are all averages like essentially. So these are all important from a look back perspective. So what -- this is our assumption on that like we are like doing what is like sort of like an 80% capability. Now 80% could be 80%, 80% could be maybe 85%, 80%, even 75%, maybe even 70%.

Okay. Understood. And sir, with respect to the demand scenario in 200, have we closed AI mission?

We have not closed any deals on the Blackwell capacity, which is going up as we are. Like obviously, once we close the deal, we will kind of inform everyone about it.

So the reason for this is, is the reason ramp-up? Or is there a reason that we're still closing POCs or in that sense?

So there is a lot of interest in the platform capacity that we are building. So kind of like we are working towards the multiple customers to figure out like what sort of problems they need us to solve for them and what is it that they are looking at from us. And we obviously want to work with the customers that are best placed to utilize what we have to offer.

Next question comes from the line of Hardik Gandhi from HPMG Share and Securities.

Congratulations on a set of good numbers. I wanted to push on a question asked by the previous participant. I think so majority of the investor group just wants to know whether the revenue uptick is due to any lumpy in nature? Or are you seeing any sustainable growth in the long run? Because in the last December -- but we suddenly dropped the numbers. We stated that the start-ups were no longer wanting our ecosystem, right? So we just want to know from a long-term perspective or at least some...

Let me try to answer that question as best as possible. So see, the lumpiness is a part of the nature of the business. So the lumpiness will keep going down as we keep increasing our GPU base. Now that being said, we are currently seeing the kind of customers who are buying GPUs have sustainable long-term workload, so these are not of the short-term matters for most part. So we do expect to have sustainability for these workloads over the medium term. So I don't think like there is going to be a massive amount of lumpiness in the near term. Medium term, of course, like as a number of GPU growth as I think the effect of lumpiness would be far more muted than it has shown up in the past in the early days of first building with the [indiscernible].

Right, right. And the second question from my side was to understand, again, on this part itself that what percentage of revenue are we allocating to a long-term MRR -- sorry, long-term contracts, so that our MRR...

It's a fair dynamic nature. So like basically like it is not like possible to predict...

Yes, we are not -- I'm just saying that from a...

We are not setting any hard lines over there, like there are, I think, like far more number of variables to consider then it would be possible to describe over here that like what to pick up then in terms of like customer interest. So like I don't think it could be -- like I would be able to give like a very fair picked answer to that question. So like we are definitely interested in having some percentage of long-term contracts. What that percentage will come out to be? Is it like something on -- we need to look at it on a look-back basis, when the size and scale of the GPU installation is sufficient. So I think in the short term, there is no [indiscernible] to do that.

Right. But just a small data point, Nitin. What percentage of revenue did come from customers outside India versus domestic demand for the last quarter?

I'll let like Nitin take a stab at the question.

So the international revenue for the last quarter is roughly around 35-odd percent, 35% to 37% is the international customer revenue.

Our next question comes from the line of Srinivas [indiscernible] from Trust Investment Advisors Private Limited.

Am I audible?

Yes, yes, loud and clear.

Yes. Sir, what percentage of Q4 revenue came from inference workloads versus training? And any -- what would be the future setup of this mix directionally sir?

So I think there is a lot of confusion about like what persists like of inference revenue versus training revenue. I think these are not like super hard lines. Sometimes the same set of GPUs are used by the customers for kind of like busting up the interest that grows and the inference of growth are high. And at a time when the instance or codes are low, they get converted to training. So the workload management is far more flexible today to kind of like got to like say, what percentage of inference -- what percentage of training today. And I think, again, like at a much larger scale, like these numbers would become clearer. So today, of course, like it's hardly possible to kind of like super depreciate between, okay, who is using that 1 GPU, 4 GPU, 8 GPU, 16 GPU cluster either for training or inference. But that being said, like majority of revenue is still like closer to training than to inference.

Okay, sir. And you said the 6,000 GPUs by FY '27, at least minimum, right? So what would be the upside, sir, could you deploy 8,000 an upside? 8,000 to 10,000 GPU?

I don't want to put any numbers right now. So like let's look at those numbers as I look back, like over coming period of time. So rather than like putting out a number today.

Our next question comes from the line of Rohan Nagpal from [ Helos ] Capital Management.

Congrats on the good set of results. My first question was also you said that there is a variety in terms of your current MRR being short-term, medium-term, slightly long term. Could you provide some -- or some like directional commentary on how much of that MRR is 6 months plus or 1 year plus?

So I think like one is like we are still on a very small base of GPUs. So kind of like we are not doing any hard if in terms of like where we want to be or where we are today. And it's fairly dynamic and rapidly changing both with new capacity coming up online. So all these numbers are subject to very rapid and very wide amount of change. So these do not add to any understanding for anyone today.

I mean if there is a 20% to 25% increase in the spot price and capacity that is contracted out for 6 months or 1 year will not be in a position to take advantage of any increase in realization, right? So I think it would be somewhat beneficial for someone looking at it from the outside to get a sense of some amount of revenue uplift or exposure to increased realization?

So like that's a very [ premier ] effect of understanding that you are taking on a very small size of the infrastructure installation base compared to the global peers. So I think like we would like to avoid this conversation for today. And as we grow in size, then we kind of like talk more about that, okay, like now these numbers actually mean something and they are making sense to someone on the street. So like today, these numbers would not make much sense.

Okay. Sure. And then second question was, so you talked about higher value tokens. So could you provide some color on which industries are sort of taking advantage of higher value tokens or from your vantage point?

Not really. I think it's more closer home than that in terms of like, okay, what's the higher value tokens. So like asking a question like, okay, what is closer to a such query is like a reasonably low value tokens, someone using the focus to solve this problem that is paying the bills for a company is like a completely different value conception for tokens. So answering a basic transactional support query is like a reasonably low value token. Being able to answer a more complicated query or being able to confirm a more complicated transaction with a customer on call is like a higher value token. So like there are like infinite number of variations of how people use AI and the key is to find those customers who are utilizing the higher value tokens. We are not worried about okay, what's my ultimate for our GPU price. Sort of like what's the cost of my -- following a particular problem integrating higher ROI. So again, we are not like directly involved all the time in identification of higher value tokens and kind of like being able to standardize in a very, very static manner that they should do and sell to this customer, they'll probably pay higher for this particular GPU. It's a broad industry trend, where we are seeing increasingly the demand coming from customers who are not price-sensitive because of their ability to generate or consume higher value token through a combination of their own software or proprietary [indiscernible] software that is available extent in the market today.

So in that case, what is E2E's role in this high-value token generation? I mean, that will just be an outcome of if the person who is renting a GPU or taking on retail capacity able to extract higher value, they'll be willing to pay more, right? So the price discovery is just take care of who is generating higher value token. I was going to clear on how E2E is sort of working towards higher value tokens.

It's how you were able to quickly get to that point of being able to generate and utilize those tokens for your business. Like so it's a fundamentally how to build on various parts of the software and hardware stack to make it available quicker. I think that's where the role is played by E2E.

Ladies and gentlemen, due to the time constraint, that was the last question for today. I would like to hand the conference over to the management for the closing remarks. Over to you, team.

Yes. Thank you to all of our investors, all of our team members, all of our customers, all our ecosystem partners for all the support you have all extended to E2E Network over last many years. It's been an overwhelming amount of love for us that we are seeing in the community and overall community and we wish you all like a great day ahead. And thank you for patiently listening to our call, and we would like to kind of like want to thank everyone, and this call.

Thank you.

Ladies and gentlemen, on behalf of E2E Networks Limited, that concludes this conference. Thank you for joining us, and you may now disconnect your lines.