E2E Networks Limited (E2E) Earnings Call Transcript & Summary

Ladies and gentlemen, good day, and welcome to E2E Networks Limited Q3 FY '25 Earnings Conference Call, hosted by Go India Advisors, LLP. [Operator Instructions] I now hand the conference over to Ms. Soumya from Go India Advisors. Thank you, and over to you.

Thank you, Steve. Good evening everyone. We welcome you to E2E Networks Limited Q3 FY '25 results con-call. We have with us on the call today, Mr. Tarun Dua, Managing Director; Ms. Megha Raheja, the Whole-Time Director and Chief Financial Officer; and Mr. Ronit Gaba, the Company Secretary. I must remind you that the discussion on today's call may include certain forward-looking statements and must be viewed in conjunction with the risks that the company may face. I now request you, Mr. Tarun Dua, to take us through the company's business and financial highlights subsequent to which we'll open the floor for Q&A. Thank you, and over to you, sir.

Thank you, Soumya. Good afternoon, everyone. Thank you for joining us for the E2E Networks Q3 FY '25 Earnings Call. I hope you are all doing very well. We are happy to share our journey and provide an update on the steps that we are taking to sustain our growth and strengthen our position in the cloud GPU and AI/ML workload space. As you are all aware last quarter we achieved a strong operational momentum driven by E-advancements deployed in our cloud infrastructure. Our cumulative deployment have now grown to nearly 700 H100 GPUs, 256 H200 GPUs, and around 700 non-H100/H200 GPUs. And we also successfully raised another round of more than INR 1000 crores from L&T for expanding our accelerated cloud infrastructure and focusing on the next-generation cloud GPUs and GPU clusters. This quarter, one of the most important developments for us has been deepening of our strategic partnership with L&T, a collaboration that marks a significant milestone in our journey towards revolutionizing AI and cloud infrastructure in India. With L&T's deep expertise in data center management and our advanced AI compute infrastructure and cloud experience, both the companies can jointly offer robust and scalable solutions to the enterprise, government and organizations. This collaboration will open new revenue streams, drive growth in high-demand AI services, and we will also improve our operational efficiency leading to enhanced profitability and revenues. Together, we will leverage each other's strength to capture a larger market share and create a more powerful presence in the AI and cloud sector. Each quarter, we continue to stay ahead of the curve by embracing latest advancements in cutting-edge GPU technology. In line with our growth plan, we are also expanding our data center capacity, from 4.2 megawatts to nearly 10.2 megawatts, which is a move which is supported by our recent fund raise. We continue to innovate at a rapid pace to expand our better tested scalable, high-performance cloud infrastructure for education, research, enterprises, government and start-ups. The surge in demand for the cloud and AI services has helped us achieve strong growth and strengthen -- we have strengthened our position in the market. By leveraging our advanced AI cloud platform, we are committed to driving India's digital transformation, empowering government and enterprise initiatives and positioning India as a global leader in AI and cloud innovation. So our platform supports wide range of cloud native services today, including CPU and GPU environment, virtual machines, native containers, serverless architecture mode. We offer flexible high-performance storage solutions across object storage, across block storage, parallel file systems. And we also offer like advanced networking, load balancing, firewalls and relational database services, NoSQL database services, VectorDB services. We have continued to develop our AI/ML platform called TIR, which is designed for data scientists and developers to streamline their AI/ML workload like training inference and model endpoint deployment. With an early mover advantage in the AI/ML space since 2020, we continue to offer a superior price-to-performance ratio, helping customers scale without any long-term commitments to our cloud. As the AI market, especially Generative AI rapidly expands, we are positioned to help businesses capitalize on AI's transformative potential, filling the demand supply gap in India and supporting country's digital evolution. Now I would like to hand over our call to CFO, Megha, who will briefly touch upon the financial and operational highlights of the quarter under review. Over to you, Megha.

Thank you, Tarun, and good afternoon, everyone. Let me first start by giving you some of the key financial highlights. I will summarize the performance of Q3 FY '25. For Q3 FY '25, the total revenue stood at INR 416 million, which witnessed a substantial growth of 73.7% on year-on-year basis. EBITDA for the quarter is INR 246 million which further provides a growth of around 119% on year-on-year basis. EBITDA margin for the current quarter is 59%, which demonstrates a growth of 1223 bps year-on-year. PAT is reported at INR 116 million, which demonstrate growth of 108% on a year-on-year basis. PAT margin for December quarter is 27.8% and Diluted EPS is INR 7.03 for the quarter, which is around 86.5% on year-on-year increase. Now we can do a quick comparison with last quarter, that is September '24. We have witnessed total revenue decline of 12.6% on a quarter-on-quarter basis from INR 476 million to INR 416 million in the current quarter. EBITDA for the current quarter is INR 246 million as against INR 314 million on the previous quarter. As a result of this, PAT for the current quarter is INR 116 million, exhibiting quarter-on-quarter decline of 4.3% from Q2 FY '25. While our outlook -- overall outlook on AI compute infrastructure in all forms remains very strong. There has been some dip in the revenue numbers as compared to previous quarter. So the decline in revenue and growth in the current quarter is mainly due to churn and downscaling of training deployments, which are, as you know, bursty in nature. And the revenue factor reprovisioning has tended to be concentrated during this quarter due to concentration and due to a smaller scale. We anticipate the effect of burst training workloads would eventually become muted at a larger scale, and then we intend to build cloud GPU computing infrastructure. We anticipate a resurgence in demand for advanced AI solutions over medium term. Now important update as well. During 2024, we had raised a total of INR 14,849 million through preferential issue of equity shares. Out of this, we had utilized INR 1,508 million till Q3 '25, and we have balance of INR 13,349 million as on December 31. That concludes the update for the quarter. And now we can open the floor for question-and-answer session.

[Operator Instructions] The first question is from the line of Astha from PKJ Advisors.

So the first question is that what was the CapEx deployed in Q3 FY '25? And how much CapEx we plan to do ahead in Q1, Q2, Q3 of FY '26?

So we have deployed INR 2,017 million till December. That is [indiscernible].

And for the next year ma'am?

So we are kind of like very agile in terms of like how we deploy the CapEx. So we are anticipating that like majorly based on the pipeline, we will evaluate. And based on that, we will do just-in-time deployment as much as possible over next several quarters. And based on that, the outlook is not based on a plan, but practically by looking it at the end of the quarter like what was the CapEx that was done. So it would be based on the demand outlook that we come up with.

My second question is, I understand the answer was given earlier, but I missed it. I wanted to understand what was the reason for a drop in quarter-on-quarter revenue?

Sure, sure. We currently have a fairly small scale. And compared to that, like some of our larger customers like their deployments were quite large. Now our goal obviously is to build a much larger infrastructure, which is spread across much larger footprint of larger customers. Now training workloads by their very nature are bursty. So which means that like typically, although you don't expect them like, okay, at what time does a training get over or to make a particular data science group besides that, okay, the trading needs are over for now. So unfortunately, all of these got concentrated during the Q3. We have also seen somewhat muted demand for the end of this quarter, December typically is a slow month, where we have not been able to recover back that unutilized infrastructure to a higher utilization rate. So that is the current status today. So obviously, in the medium term, we expect that as the footprint of customers grows larger, then the effect of 1 or 2 large customers declining their training needs, should not like impact us, so that is subject to growth of our overall scale.

The next question is from the line of Amar Mourya from Lucky Investments.

A couple of questions from my side. Firstly, if you can help us understand this L&T tie-up. And then how it is going to help us in long run in terms of gaining some large enterprise clients, that is number one. Number two is in terms of our current -- I think we have something around 4,200 megawatts. So what is the peak revenue potential of this capacity? And by then, we will be reaching that kind of peak utilization levels, and in that how much of the percentage would be enterprise versus the retail? These are 2 questions.

These are 3 questions actually, Amar. So your first question is about the L&T tie-up, how it would help us in the long run? See, obviously, a small company when they tie-up with a big company like you get the advantage of the size and scale and connections that a bigger company brings in, and also their own capabilities in terms of technology as well as infrastructure. So over here, like we obviously intend to leverage all of those things. It's quite early days. I think it's been like practically, I think, less than few weeks. So basically, where we have started interlocking, started working together and jointly exploring opportunities with the customers. So in the long run, I believe that the interlock in terms of jointly exploring opportunities would obviously help both the businesses who bring their own strength to the table. And we'll obviously be pursuing a lot more what we internally call as like long cycle business opportunities compared to working with SMEs or smaller companies, the long cycle business opportunities are obviously long term in nature and more rewarding in terms of lifetime value. And smaller company obviously make the decisions fast. That has been the mainstay of our businesses, our business till now. I think that natural change will come over like medium term where we start working with larger customers in conjunction with our partners, including L&T. And I think that is one of the major changes that we are going to see, where we are pursuing more enterprise business apart from what we continue to pursue in the SME and start-up and other organizations that we have continued to do. Now in terms of megawatt capacity like that's the measure of the data center capacity like how much IT load you can put in. Typically, these are fairly well-known numbers in the industry which are quite in time numbers like that 1 megawatt of what it pertains to today would vary between, like, say, $25 million to $50 million over a spectrum of trade a smaller player and like the large hyperscaler, so each megawatt would represent different revenue levels for those, so it could vary anywhere between $10 million to $50 million and it is also dependent on the kind of workloads. Now regarding percentage of enterprise versus retail, I think it would be very hard to predict today, but my assumption is that like mostly wherever in the industry, there are large revenues, large profit pools, obviously they are the ones who invest in amount of money into compute. So what I believe is that there would be a shift from smaller companies to bigger companies in terms of overall concentration of our revenue over the medium and long term. Now it would be very hard to say how the percentages would vary. But the kind of capabilities we are building today are very focused on meeting the requirements of the larger enterprises.

The next question is from the line of Ashwin Kedia from Alchemy Capital Management.

Just one question was, how has your revenue percentage changed from training to inference quarter-on-quarter? And how do you expect that to change moving forward for the next year?

See, the major concentration of the revenue obviously has been training, will continue to be training. I think training will continue to play a very major role for the foreseeable future. That being said, of course, the impact of large bursty training workloads what we have seen in this quarter and -- but as we scale up, I think like that effect would be less and less pronounced. So I think it's the function of us being in our very early days of our journey as of today, which is what we are seeing. I think that is probably the question you're trying to ask, right? So the other part being that like, I believe that ultimately, it's going to be somewhere the ratio eventually over the medium and long term, I think it should become more or like 50-50 or maybe 60-40 in favor of like training.

And what is your inference revenue today percentage of your total revenue?

We don't really track it today very separately because there is like -- it's a very nebulous line between building training infrastructure for foundational models being built from the ground up versus foundational models being fine-tuned at what level. So like the graduation from training to inference is like -- kind of like there is a whole range. So you can't really classify what constitutes training versus inference. So inference and production would continue to remain smaller even in the global context, so ultimately, it's the training -- so today, we are like all -- the AI is like in its early phase. So for significant future, like training is going to constitute a majority of revenue for -- not just for us, but overall in the cloud ecosystem, like training would constitute the majority of the revenue.

Next question comes from the line of Garvit Goyal from Nvest Analytics Advisory.

Can you help me understand our geographical concentration, particularly in data center business like North America is the biggest market for the data center. And if interest rates remain elevated, hyperscalers like Meta, Microsoft may scale down the CapEx, which in return will impact our business. So can you put some color on it? Like is that going to happen or otherwise, what's your take on it?

So I can't really comment on what large global players are doing, like from the view from whatever we are, like, obviously, all of us are hearing in the industry news is that like the investment into AI is definitely going up. Now regarding data centers themselves, like we rely on data centers only in India. So we have one facility in the North, in the Delhi NCR region. And we are in the process of establishing the second facility near in the Chennai -- Greater Chennai area. And so geographically, from an infrastructure point of view, we are concentrated in India. From a customer view, primarily the customers we tend to serve are in India. Now over the medium and long term, obviously, we will continue to look for opportunities to expand out of India into other markets as well, and this may necessarily not be on the public cloud, but these opportunities could be in the form of software and services and support as well.

The next question is from the line of Ketan Kapasi from [indiscernible] Investments.

What would be the impact on the company of this new GPU rule proposed by the U.S. government?

See in the short term I don't think there would be any immediate impact. So one is, of course, there is like about like more than a period of a quarter before these regulations come into effect. Secondly, based on what India is doing today in terms of overall volume. See currently, we are not really hitting the limits that are being placed. Now -- and there are certain exceptions which we have not fully studied, but there are exceptions around certain end users who are consuming up to 1,700 GPUs then not being counted in the overall limit. There would be companies based in the U.S., it's still when they bring their infrastructure to India potentially, they're not being counted. So overall, I think that impact assessment would be more like -- our belief is that it should be assets like in about like a 2-year time frame rather than on an immediate basis.

The next question is from the line of Keshav from [indiscernible].

So as I can see that your non-H100s and H200s have seen an increase of 100 GPUs. So is it like that the demand for the H200s and H100s are currently not much. So that is why you have deployed...

So the demand environment continues to remain strong. So the pipeline is there, obviously, for both H100s as well as H200s. What we are potentially kind of like missing is like -- immediate and closures that we were seeing in the past. So I think the sales cycles have grown, I don't think demand has gone away. I think the sales cycles have become longer. I think that is what we are seeing.

Okay. So like the H100s and H200s, are they completely utilized or they are underutilized as of now?

No, no, they are underutilized as of today.

So how much of that would be the percentage, if you have that number?

It's a fluctuating number because a lot of workloads in the cloud, it's very, very hard to measure like a point in time very, very easily. So -- but we have reasonable capacity on both sides, both being consumed as well as being made available to the customer.

Okay. And while we are sitting on a pile of cash. So we see like the utilization has only been INR 150 crores. So how much time do you think that the funds is going to take time to utilize? Or like how is the CapEx cycle, if you can comment on that...

See very, very hard to commit to kind of like spend that money in a market where the infrastructure like the high-tech equipment like GPUs and CPUs, they get like upgraded version. So it is very, very important not to build a lot of inventory or what would potentially kind of like will get superseded by a newer version. Although these are all long cycle products, they will sell for 7, 8 years. But it is also important to have dry powder for the latest version. So that has always been our strategy that like always be able to invest for every version. So that's where we will continue to evaluate the demand and continue to try and keep the utilization high, but not have so little capacity idle like we are not able to service new demand. So it's a balance that we have to draw out, I think that balance will get established. And I think those -- future CapEx cycle they will obviously continue to get established, and when we look back we'll be able to say where certain considered okay, the CapEx that was spent, as opposed to trying to predict today. So we are reactive, not predictive. So that is what we'll continue to do. Overall outlook for both medium term and long term, we continue to see that the AI transformative potential and its adoption is definitely picking up.

My last question is NVIDIA simply launched bids, I guess their new product, I'm recollecting the name correct, so is that a play to the cloud GPU since the devices...

In the compute market, you have to understand that there would be multiple approaches to solve various problems. So obviously, they would have -- NVIDIA would have seen gaps in terms of what can be serviced from the cloud or like micro unit. See, the edge devices have always existed. So if you look at edge compute, that has always existed. Now they built a slightly more powerful edge machine. It does not take away the need for data center in anyway. I think this would be like one more thing to the ecosystem. I don't think like it's a zero-sum game that whatever you were able to do on the edge, there would be still further more workload for the cloud.

So technically, what is the difference if you could expect a bit?

See, like there's a whole range of devices that are kind of like do a bit of AI. This is a slightly more powerful edge. So when you talk about what you can do in the cloud, like it's a lot more number of things compared to what you can do on the edge. See, it's not just about like being able to do an inference for a training in isolation, you also need access to VectorDBs. You also need access to relational databases, you also need access to a lot more storage, you also need redundancy, you also need reliability, you also need security. So there are certain things that you can do on your laptop, certain things you can do on your mobile, certain things that you can do on work stations, certain things that you will be able to do on edge device which is very powerful required for one particular organization or one particular business unit. And there would be a lot more things that you will continue to do on the cloud. So it's not like really comparative in terms of like a straight away movement of workloads from here to there.

The next question is from the line of Akshay from CD Integrated Services.

My first question is what is our sourcing strategy for the GPUs? Do we source GPU from NVIDIA and then we do add some value like we make our architecture or we do source it from the companies like HP or Network Technologies?

That's a good question. So GPUs do not work in isolation obviously. So I think like before the hopper architecture during the MPR architecture, the A100s and A30s and A40s, et cetera? Like there was a very large prevalence of GPUs being deployed PPI form factor where you typically bought the GPUs and deployed them in compatibles that were -- which were typically on the hardware compatibility list of major vendors like NVIDIA or AMD or Intel. So that needs to be the approach. Today, it's a lot more integrated approach where a vendor like Dell or HP or NetWare newer or Super Micro, I think there are like, I think, 10 to 15 such vendors who build GPU servers. So now today the approach of -- kind of like buy a complete integrated hardware, which is a design -- which is done by a principle like an NVIDIA or an AMD or an Intel, and the major of materials also comes from an NVIDIA or Intel or AMD. Currently, of course, we are very majorly focused on NVIDIA. That does not preclude us to work with other players in the future. And we buy these integrated servers. Typically, that is called HGX line, that's the major one. Some of the other GPUs like L40s, L4s, they continue to follow the older model of being able to deploy [ GPI ] cards into boxes, which typically come without GPUs, so we do that as well, but these are less powerful GPUs typically, the more powerful GPUs come in their pre-built integrated boxes from the major GPU server vendor.

Very well understood. And sir, my second question is what is the competitive intensity in our kind of business like do we face intense competition from the companies like -- multinational companies like Amazon web services and Google Cloud or...

Yes, yes, so as AI has become very, very mainstream since 2019, '20 since we have been working on AI, the competition has become far more intense today, hence competition in the compute and technology world has always been intense. So that has definitely increased. So typically, there are a lot more players today who are doing the cloud GPU infrastructure. As the market has expanded, there is space for even more number of players. I think that competitive intensity and the market both will continue to expand.

So what is our competitive edge? Like if any player -- if any company would have to choose from the other company like ours then what is our competitive edge and why they will choose us?

See, like everyone will build their niche solutions, which are targeting particular solutions for data science team, particular industry solutions and everyone will eventually find their niche to succeed in. So again, like I said, these are early days. We continue to learn from our customers. We continue to learn from the ecosystem. And we have an R&D team, which has the ability to move very fast and build. And we have build a lot of intellectual property that is integrated into our cloud. And this will continue to play a big role in terms of the transition.

The next question is from the line of Abhishek Shindadkar from InCred Capital.

Congrats for the good 3Q. My first question is, did 3Q play out as you had anticipated? Or were there any positive/negative surprises?

Sorry, I didn't understand the question.

So did the 3Q in terms of both revenue margins and in terms of ramp-up, did it play out as we anticipated or there was a negative surprise? And if yes, it was towards the end of the quarter or towards the mid of the quarter? Any color in terms of how the revenue paid out for us...

In terms of when you are a cloud operator like all these things basically like you -- like we have maintained, I think, over the last many calls, that we react to every situation. We don't try to predict any situation. So basically, nothing surprises us basically. We operate on the principle that we have the ability to react to anything.

And regarding -- you mentioned about these bursty workloads. Was it part of -- the change was among our top large customers?

So obviously, these are the larger customers that's why the impact is visible. If it was smaller customers than we wouldn't have seen the impact. Now our whole philosophy over here is that eventually, as we scale up our infrastructure as we scale up our customer base with larger customers and larger deals, so eventually, each single large customer will not impact us that much, and that is definitely the path that we continue to follow.

Just last one data point. In terms of depreciation, we saw a significant jump in the depreciation. How should I read this in the context of the hardware, especially the GPU numbers that you've shared between the last quarter and this quarter, where the increase is only in the A100s, V100s, so how should I read that depreciation numbers?