The Ciena Insights Podcast

1 00:00:09,750 --> 00:00:11,670 Hello and welcome everyone to yet 2 00:00:11,670 --> 00:00:14,470 another episode of Ciena Insights Podcast. 3 00:00:14,550 --> 00:00:17,230 I'm your guest host, Gautam Billa, and I'm 4 00:00:17,230 --> 00:00:19,450 the CTO and sales engineering leader for 5 00:00:19,450 --> 00:00:22,330 Ciena's Asia Pacific, Japan, and India business. 6 00:00:22,350 --> 00:00:24,270 We're really excited today to 7 00:00:24,270 --> 00:00:26,650 invite on our podcast, Vinay. 8 00:00:26,770 --> 00:00:30,230 Vinay is the CTPO of Lightstorm, one of the fastest 9 00:00:30,230 --> 00:00:32,610 growing connectivity providers in the world today. 10 00:00:32,820 --> 00:00:36,140 Now, we are here to give you a glimpse of Lightstorm's 11 00:00:36,140 --> 00:00:39,570 phenomenal journey starting from India and expanding 12 00:00:39,570 --> 00:00:42,610 into Asia and the rest of the world. Vinay thank you so 13 00:00:42,610 --> 00:00:45,950 much for joining me today. Why don't you give us a quick 14 00:00:45,950 --> 00:00:48,330 introduction about yourself, tell us a bit about 15 00:00:48,330 --> 00:00:51,930 yourself and your role at Lightstorm. Thanks Gautam. Great 16 00:00:51,930 --> 00:00:56,370 to be here talking to you. So I lead the technology 17 00:00:56,370 --> 00:01:00,190 planning and operations for Lightstorm across Asia Pacific. 18 00:01:00,190 --> 00:01:05,110 In my role, my day job is figuring out how and what 19 00:01:05,110 --> 00:01:07,550 kind of network we need to build for the next five 20 00:01:07,550 --> 00:01:11,090 years, if not 10 years, and then making sure we actually 21 00:01:11,090 --> 00:01:14,270 build it. So Lightstorm is a relatively young company. 22 00:01:14,270 --> 00:01:16,330 As you said, it's also one of the fastest growing 23 00:01:16,330 --> 00:01:19,370 companies in the region. It's about seven years old, 24 00:01:19,370 --> 00:01:23,570 and it started with a simple thesis that we need to put, 25 00:01:23,570 --> 00:01:26,830 especially in India, purpose -built digital infrastructure 26 00:01:26,830 --> 00:01:30,030 not really repurposed telecom networks from the 27 00:01:30,030 --> 00:01:34,130 eras of 90s. And that thesis has now also expanded to 28 00:01:34,130 --> 00:01:37,950 across the Asia Pacific. So in my role what really gets 29 00:01:37,950 --> 00:01:41,410 me excited is we are at a inflection point where AI is 30 00:01:41,410 --> 00:01:43,990 fundamentally changing what networks need to do. I've 31 00:01:43,990 --> 00:01:46,890 been fortunate enough to go through the internet era, 32 00:01:46,890 --> 00:01:50,130 then the mobile era, and the cloud era. And so we are also 33 00:01:50,130 --> 00:01:52,150 lucky to have another revolution that's happening 34 00:01:52,150 --> 00:01:55,770 as an AI era, I would call it as. So it's about moving 35 00:01:55,770 --> 00:02:00,060 the AI and moving the network with a very specific 36 00:02:00,060 --> 00:02:03,390 requirement that's tailor-made for ai. And that's attributes 37 00:02:03,390 --> 00:02:06,500 like very specific latency, jitter, and all those 38 00:02:06,500 --> 00:02:09,530 characteristics. And how we can solve that problem as well is 39 00:02:09,530 --> 00:02:12,230 something which keeps me excited and coming to the work 40 00:02:12,230 --> 00:02:15,790 every day. Thanks, Vinay. And you guys are a relatively 41 00:02:15,790 --> 00:02:18,670 young company, India-born company, but you guys 42 00:02:18,670 --> 00:02:22,290 have quickly expanded to multiple countries now in seven 43 00:02:22,290 --> 00:02:25,090 plus markets. You guys touch 100 plus data centers 44 00:02:25,090 --> 00:02:28,130 globally that's quite amazing. Can you walk us through 45 00:02:28,130 --> 00:02:31,370 your journey and what are the types of networks that you 46 00:02:31,370 --> 00:02:34,530 have built across India and the Asia Pacific markets? 47 00:02:34,810 --> 00:02:39,090 Sure, so Lightstorm really started in India around 2019 48 00:02:39,090 --> 00:02:43,190 and one of the big bets the company had was building a 49 00:02:43,190 --> 00:02:45,850 national long distance backbone, we call it as SmartNet. 50 00:02:45,850 --> 00:02:48,970 So, essentially when we really started building it 51 00:02:48,970 --> 00:02:52,070 it was more of a deep trench fiber that would really 52 00:02:52,070 --> 00:02:55,450 connect major data centers in various hubs like Mumbai, 53 00:02:55,450 --> 00:02:59,690 Pune, Hyderabad, Chennai, and Bangalore, and Delhi and 54 00:02:59,690 --> 00:03:04,850 today that corridor is over 12000 terabits per second on 55 00:03:04,850 --> 00:03:07,430 the design capacity that's 12 petabits of the design. 56 00:03:07,750 --> 00:03:11,110 So while we build this next level of the network 57 00:03:11,110 --> 00:03:14,030 connecting these hubs, we also realize that it's 58 00:03:14,030 --> 00:03:16,130 not just having the fiber in the ground. That's 59 00:03:16,130 --> 00:03:19,200 not good enough, and you need to build a smart 60 00:03:19,200 --> 00:03:22,320 infrastructure on top of it. So we built Polarin. 61 00:03:22,320 --> 00:03:24,890 That's our orchestration platform. We put it as 62 00:03:24,890 --> 00:03:26,850 an abstract layer over the network that we built. 63 00:03:26,970 --> 00:03:30,410 And Polarin gives its customers an ability to provision 64 00:03:30,410 --> 00:03:33,190 and also manage the capacity on demand. We wanted 65 00:03:33,190 --> 00:03:35,930 to keep that single pane of glass to the customer and 66 00:03:35,930 --> 00:03:39,150 they see it real time what we are seeing in the network 67 00:03:39,150 --> 00:03:42,770 that we run. So that process was the customer need 68 00:03:42,770 --> 00:03:45,230 not wait for weeks to provision the circuit but he 69 00:03:45,230 --> 00:03:48,430 can go instantly and provision those circuits. And the 70 00:03:48,430 --> 00:03:51,970 same thesis we expanded to southeast asia and now we 71 00:03:51,970 --> 00:03:55,050 connect seven plus markets as you mentioned also like 72 00:03:55,050 --> 00:03:58,480 over 100 data centers and the idea is same everywhere. 73 00:03:58,480 --> 00:04:02,150 So we built the network seven years ago for cloud 74 00:04:02,150 --> 00:04:05,910 purpose built data networks, now we are building purpose 75 00:04:05,910 --> 00:04:08,850 -built for optical infrastructure for AI networks. 76 00:04:08,850 --> 00:04:10,190 And again, we put the intelligence 77 00:04:10,190 --> 00:04:12,690 on the top and let customers consume 78 00:04:12,690 --> 00:04:14,750 it the way they would consume the cloud. 79 00:04:14,890 --> 00:04:15,910 So simple, 80 00:04:16,170 --> 00:04:18,670 on-demand, programmable network. That's 81 00:04:18,670 --> 00:04:20,390 what customers would like to experience. 82 00:04:20,910 --> 00:04:23,770 Awesome. Vinay, let's keep focus on the India 83 00:04:23,770 --> 00:04:25,790 market for a second here. It's such a 84 00:04:25,790 --> 00:04:28,690 complex, it's such a diverse market. Scale is 85 00:04:28,690 --> 00:04:31,090 absolutely necessary when we talk about India. 86 00:04:31,090 --> 00:04:33,750 1.4 billion people. It has its sets of 87 00:04:33,750 --> 00:04:35,470 challenges when you build infrastructure. 88 00:04:35,650 --> 00:04:38,170 So we often hear the term, if you solve for 89 00:04:38,170 --> 00:04:39,830 India, you can actually build for the world. So 90 00:04:39,830 --> 00:04:41,750 what's your take on that? What are the learnings 91 00:04:41,750 --> 00:04:44,350 from the India build that you're actually taking 92 00:04:44,350 --> 00:04:46,210 out into the Asia and the rest of the world? 93 00:04:46,370 --> 00:04:47,930 True, I think that has been the 94 00:04:47,930 --> 00:04:49,770 case for anything that we build. 95 00:04:49,770 --> 00:04:51,530 So actually, I believe that. 96 00:04:51,530 --> 00:04:53,090 And it's not just as a slogan. 97 00:04:53,090 --> 00:04:55,730 India, for various reasons, is one of the 98 00:04:55,730 --> 00:04:58,110 hardest markets to build infrastructure in. 99 00:04:58,130 --> 00:05:00,250 One, you're dealing with diverse terrain. 100 00:05:00,250 --> 00:05:03,470 Two, you're dealing with complex relations. 101 00:05:03,670 --> 00:05:06,030 Three, those extreme weather conditions. 102 00:05:06,170 --> 00:05:09,470 And then the customers, and customers are king, who 103 00:05:09,470 --> 00:05:13,190 really expect world-class quality at a very competitive 104 00:05:13,190 --> 00:05:16,030 pricing, the best of both worlds to come in. So when 105 00:05:16,030 --> 00:05:18,850 you really engineer a network that works reliably 106 00:05:18,850 --> 00:05:22,980 across India, you end up with something that is inherently 107 00:05:22,980 --> 00:05:26,850 robust. And we stress test all the factors, the 108 00:05:26,850 --> 00:05:29,910 engineering standards, operational playbooks, and including 109 00:05:29,910 --> 00:05:32,450 the cost structures. All this gets just tested in 110 00:05:32,450 --> 00:05:35,070 India. And also like as you mentioned the other thing is 111 00:05:35,070 --> 00:05:39,210 scale. India is the second largest AI user base globally 112 00:05:39,210 --> 00:05:43,150 it's over 100 million active GenAI users in the 113 00:05:43,150 --> 00:05:45,970 country, and when you're really solving for that kind of 114 00:05:45,970 --> 00:05:48,570 a demand and that kind of a scale, you're not building 115 00:05:48,570 --> 00:05:51,690 a niche solution. You're actually building something 116 00:05:51,690 --> 00:05:55,030 that translates to anywhere in the region. So essentially 117 00:05:55,030 --> 00:05:57,490 if you are really going to solve that puzzle for 118 00:05:57,490 --> 00:05:59,970 India I think we would have solved the puzzle for 119 00:05:59,970 --> 00:06:02,630 most of the world and any of the regions to come with. 120 00:06:02,630 --> 00:06:05,750 Absolutely. Truly believe in that. You guys are actually 121 00:06:05,750 --> 00:06:08,490 building physical infrastructure on the ground to 122 00:06:08,490 --> 00:06:10,930 enable digital connectivity. These are actual fibers 123 00:06:10,930 --> 00:06:14,010 on the ground, equipment, dealing with subsea cables. 124 00:06:14,170 --> 00:06:16,110 Now, all this is an extremely 125 00:06:16,110 --> 00:06:18,670 capex intensive industry to be in. 126 00:06:18,750 --> 00:06:21,730 And how do you scale something so capital intensive 127 00:06:21,730 --> 00:06:24,910 to a global scale? And what are some of the challenges 128 00:06:24,910 --> 00:06:27,630 that you guys have come across in doing so? Yeah 129 00:06:27,630 --> 00:06:30,640 true. Building a network is definitely a capex intensive 130 00:06:30,640 --> 00:06:32,970 program that anybody would take around right. So 131 00:06:32,970 --> 00:06:36,150 the honest answer is having partnerships and being 132 00:06:36,150 --> 00:06:39,390 smart about where you're going to build organically 133 00:06:39,390 --> 00:06:42,990 versus where you're going to acquire the networks so we 134 00:06:42,990 --> 00:06:46,250 do not try to do everything ourselves. In India we did 135 00:06:46,250 --> 00:06:48,830 build organically because we needed to have the 136 00:06:48,830 --> 00:06:51,770 control over the fiber quality that would go in the root 137 00:06:51,770 --> 00:06:53,980 design and the engineering we wanted to control 138 00:06:53,980 --> 00:06:57,390 because we wanted to run the programmable layer on it. 139 00:06:57,450 --> 00:07:00,690 And then these qualities were always non-negotiable. 140 00:07:00,690 --> 00:07:02,910 And hence we went and built the network on 141 00:07:02,910 --> 00:07:06,210 our own. But as we expanded into Asia Pacific, 142 00:07:06,210 --> 00:07:07,950 we did acquire the Subsea Cable 143 00:07:07,950 --> 00:07:09,550 assets where it made sense. 144 00:07:09,550 --> 00:07:12,290 That's because it takes years of 145 00:07:12,290 --> 00:07:14,530 lead time to build a Subsea network. 146 00:07:14,750 --> 00:07:18,250 But acquiring this asset, what it gave us was 147 00:07:18,250 --> 00:07:20,890 immediate reach without waiting for years for the 148 00:07:20,890 --> 00:07:23,870 new builds. And then we partner with local operators 149 00:07:23,870 --> 00:07:27,070 and they would extend that to a local metro 150 00:07:27,070 --> 00:07:30,750 or in-region access. On the capital side, having 151 00:07:30,750 --> 00:07:34,070 an investor who is infrastructure focused, who 152 00:07:34,070 --> 00:07:36,990 understands long cycle assets has been critical. 153 00:07:37,050 --> 00:07:39,810 It has been an advantage for us. They are not looking 154 00:07:39,810 --> 00:07:42,830 for quick returns. They understand that fiber and 155 00:07:42,830 --> 00:07:46,310 subsea assets appreciate over decades. And they know 156 00:07:46,310 --> 00:07:48,730 the dynamics and nature of the business very well. 157 00:07:48,730 --> 00:07:51,590 So the challenge has been always regulatory. And 158 00:07:51,590 --> 00:07:54,410 every country has its own different rules and about 159 00:07:54,410 --> 00:07:58,030 landing rights or spectrum, right of way. And this is 160 00:07:58,030 --> 00:08:00,910 where all the local partnerships become essential. 161 00:08:01,030 --> 00:08:03,970 And overall, I think we had best of the experience 162 00:08:03,970 --> 00:08:06,470 whether we did build organically or acquire. 163 00:08:06,470 --> 00:08:08,730 I think that has been our focus. We want to 164 00:08:08,730 --> 00:08:12,450 control and build organically where we know there is 165 00:08:12,450 --> 00:08:15,210 a better value for the position we have. And we 166 00:08:15,210 --> 00:08:18,070 want to acquire where our customers can reach, 167 00:08:18,070 --> 00:08:20,770 scale themselves into the network. I think it's 168 00:08:20,770 --> 00:08:23,790 been quite a journey in India and you guys have been 169 00:08:23,790 --> 00:08:26,370 able to achieve a lot in a short time. What I 170 00:08:26,370 --> 00:08:29,330 hear is, Vinay, the next focus area seems to be 171 00:08:29,330 --> 00:08:32,790 Southeast Asia. What makes this region so strategically 172 00:08:32,790 --> 00:08:35,610 critical right now? And can you dissect for 173 00:08:35,610 --> 00:08:37,690 us some of the opportunities that are happening, 174 00:08:37,690 --> 00:08:40,790 let's say in Singapore, Malaysia, Thailand, Indonesia, 175 00:08:40,790 --> 00:08:42,850 that kind of corridor? What's it that really 176 00:08:42,850 --> 00:08:45,190 drives investment in Southeast Asia for you guys? 177 00:08:45,410 --> 00:08:48,390 Yeah, Southeast Asia has been a very big focus for us. 178 00:08:48,390 --> 00:08:50,970 As you mentioned, there are two things we have converging. 179 00:08:50,970 --> 00:08:53,830 First is we have massive data centers that's getting 180 00:08:53,830 --> 00:08:56,530 built out by the hyperscalers across the region and 181 00:08:56,530 --> 00:08:58,730 the neoscalers and the AI programs that are building 182 00:08:58,730 --> 00:09:01,730 those massive data centers, whether it's in Singapore, 183 00:09:01,730 --> 00:09:04,910 Johor Bahru in Malaysia, Bangkok. They are all seeing huge 184 00:09:04,910 --> 00:09:07,890 investments at a scale which we have not seen before. 185 00:09:08,090 --> 00:09:09,870 And second, it's AI adoption. 186 00:09:09,870 --> 00:09:12,300 AI adoption in the region is accelerating. 187 00:09:12,300 --> 00:09:14,670 And this is where most of the critical mass 188 00:09:14,670 --> 00:09:16,470 of the planet would live in and reside. 189 00:09:16,470 --> 00:09:18,310 And the governments are setting 190 00:09:18,310 --> 00:09:20,590 up the national AI strategies. 191 00:09:20,770 --> 00:09:22,990 And hence, enterprises are deploying all 192 00:09:22,990 --> 00:09:25,170 those inference workloads locally so that 193 00:09:25,170 --> 00:09:27,390 they can get the advantage of latency and 194 00:09:27,390 --> 00:09:29,670 also the reasons being the data sovereignty. 195 00:09:29,670 --> 00:09:32,110 But there's a gap in this. The connectivity 196 00:09:32,110 --> 00:09:34,970 between these data center hubs has not kept up. 197 00:09:35,030 --> 00:09:38,450 Most of this connectivity is still running in legacy 198 00:09:38,450 --> 00:09:41,370 submarine cables, which were all designed for the 199 00:09:41,370 --> 00:09:43,960 internet traffic not even for the cloud traffic eventually, 200 00:09:43,960 --> 00:09:48,170 and not for the kind of low latency. So we need for 201 00:09:48,170 --> 00:09:52,150 AI deterministic performance so that AI workloads and 202 00:09:52,150 --> 00:09:54,770 AI workloads demand that kind of a performance when 203 00:09:54,770 --> 00:09:57,550 we really build a network. So that has been the 204 00:09:57,550 --> 00:10:00,110 opportunity for Lightstorm. We are building that connectivity 205 00:10:00,110 --> 00:10:03,450 tissue between these AI hubs and we are extending 206 00:10:03,450 --> 00:10:05,890 the thesis that we have been started from the day one. 207 00:10:05,890 --> 00:10:08,850 And with the right network architecture from the 208 00:10:08,850 --> 00:10:12,170 day one it has made our journey easier and it's also 209 00:10:12,170 --> 00:10:16,150 able to get to scale better. Now Vinay, Lightstorm has 210 00:10:16,150 --> 00:10:19,470 talked about something called an APAC wide fabric 211 00:10:19,470 --> 00:10:21,770 now i know you might have answered a part of that 212 00:10:21,770 --> 00:10:24,770 question already but if you want to add something on the 213 00:10:24,770 --> 00:10:27,630 word fabric and what does it really mean to you? What 214 00:10:27,630 --> 00:10:30,250 does it look like in practice from your perspective? 215 00:10:30,610 --> 00:10:33,770 Fabric essentially means stitching together 216 00:10:33,770 --> 00:10:37,610 a terrestrial asset and a subsea asset into 217 00:10:37,990 --> 00:10:41,950 one programmable network. So what I mean is, say 218 00:10:41,950 --> 00:10:45,330 when a customer in Mumbai who needs to reach a 219 00:10:45,330 --> 00:10:48,270 host in a data center somewhere in Singapore, he 220 00:10:48,270 --> 00:10:50,990 should be able to provision that end-to-end 221 00:10:50,990 --> 00:10:53,390 through our Polarin, which is our orchestration layer, 222 00:10:53,490 --> 00:10:55,630 without calling three different cloud carriers. 223 00:10:55,630 --> 00:10:57,890 So when he decides, okay, he needs to connect Mumbai 224 00:10:57,890 --> 00:11:00,070 to Singapore, I should be able to do it. That's been 225 00:11:00,070 --> 00:11:02,630 the pace we want to set it. And the fabric as a 226 00:11:02,630 --> 00:11:05,090 concept is very important because AI workloads, 227 00:11:05,090 --> 00:11:07,950 especially the AI workloads, do not respect the national 228 00:11:07,950 --> 00:11:11,450 boundaries. A model provider might train in one 229 00:11:11,450 --> 00:11:14,950 country run the inference in another country and serve 230 00:11:14,950 --> 00:11:18,470 users across five, six markets at the same time. But 231 00:11:18,470 --> 00:11:20,710 when you do train, when you do inference, when you do 232 00:11:20,710 --> 00:11:23,810 so, but all of these AI workloads need a consistent 233 00:11:23,810 --> 00:11:26,770 network behavior and across all the three topologies 234 00:11:26,770 --> 00:11:29,920 or the anatomy that they are going to experience. So 235 00:11:29,920 --> 00:11:32,710 this would be like same latency profile, the same jitter 236 00:11:32,710 --> 00:11:35,350 characteristics, and same programmability. That's 237 00:11:35,350 --> 00:11:38,430 what we need to achieve in this scenario. So this is 238 00:11:38,430 --> 00:11:40,870 essentially what we mean by fabric. Giving the same 239 00:11:40,870 --> 00:11:43,850 experience across the geography, whether you're in 240 00:11:43,850 --> 00:11:46,430 terrestrial you're in the subsea or you're stretching between 241 00:11:46,430 --> 00:11:49,150 the geographies. We want to give that consistent 242 00:11:49,150 --> 00:11:52,910 software defined optical layer that spans across. 243 00:11:53,190 --> 00:11:56,950 Well, sounds like a fairly simple concept, but not too 244 00:11:56,950 --> 00:11:58,850 many operators have been actually able to achieve 245 00:11:58,850 --> 00:12:01,610 it over the last many years. So I think Polarin is 246 00:12:01,610 --> 00:12:03,970 something that we really look forward to in terms of being 247 00:12:03,970 --> 00:12:06,390 able to stitch services end-to-end. And that's 248 00:12:06,390 --> 00:12:09,250 fantastic. Now, just switching the topic a little bit, 249 00:12:09,250 --> 00:12:12,090 Ciena has been a fairly big proponent of what we call 250 00:12:12,090 --> 00:12:15,590 MOFN, Manage Optical Fiber Networks. It's a business 251 00:12:15,590 --> 00:12:18,850 model where service providers build dedicated networks 252 00:12:18,850 --> 00:12:21,470 tailored to hyperscaler needs. It's quite a unique 253 00:12:21,470 --> 00:12:23,790 architecture in terms of technology. What are your 254 00:12:23,790 --> 00:12:26,430 thoughts around MOFN and the business model itself? 255 00:12:26,550 --> 00:12:29,070 Yes, I think MOFN model validates 256 00:12:29,070 --> 00:12:31,090 exactly what Lightstorm has been doing. 257 00:12:31,350 --> 00:12:34,730 So the idea, all the hyperscalers and the AI 258 00:12:34,730 --> 00:12:37,350 companies and the neoscalers, they're all looking 259 00:12:37,350 --> 00:12:39,970 for a dedicated optical infrastructure, that's 260 00:12:39,970 --> 00:12:41,950 tailor made for them on the wavelength basis. 261 00:12:42,250 --> 00:12:44,890 And they have specific requirements and how the network 262 00:12:44,890 --> 00:12:47,630 needs to be deployed. And that has been our core 263 00:12:47,630 --> 00:12:50,490 thesis as well. We have been building this from the 264 00:12:50,490 --> 00:12:54,390 day one and our network is not shared it's not one 265 00:12:54,390 --> 00:12:58,090 size fits all kind of a solution basically. So customers 266 00:12:58,090 --> 00:13:01,570 get the tailor-made ask and we deliver it. So they 267 00:13:01,570 --> 00:13:04,150 ask for dedicated wavelengths and they want to get the 268 00:13:04,150 --> 00:13:06,910 control over the topology and also we provide the 269 00:13:06,910 --> 00:13:09,430 programmability through the Polarin. That's essentially 270 00:13:09,430 --> 00:13:12,110 what MOFN describes and we are already delivering 271 00:13:12,110 --> 00:13:14,990 it. Where it really gets interesting is when you 272 00:13:14,990 --> 00:13:17,950 combine that pluggable optics, and also how to disaggregate 273 00:13:17,950 --> 00:13:20,190 the network in ways that gives the customer even 274 00:13:20,190 --> 00:13:22,870 more flexibility. And that has been a direction we 275 00:13:22,870 --> 00:13:25,610 are also actively exploring, and that has also been 276 00:13:25,610 --> 00:13:27,550 the direction where the industry is heading towards. 277 00:13:27,910 --> 00:13:30,410 Super. Let me switch the topic a little bit, Vinay, 278 00:13:30,410 --> 00:13:33,170 here. You talked about AI inferencing earlier 279 00:13:33,170 --> 00:13:36,370 and AI moving into monetization. From your 280 00:13:36,370 --> 00:13:38,250 perspective, what's really happening in the world right 281 00:13:38,250 --> 00:13:40,730 now is on one hand, the world is shrinking 282 00:13:40,730 --> 00:13:43,510 because we are building this massive scale infrastructure 283 00:13:43,510 --> 00:13:46,190 and everything is available on fingertips. 284 00:13:46,190 --> 00:13:49,410 But on the other hand, there's so many geopolitical 285 00:13:49,410 --> 00:13:51,590 scenarios happening in the world today, which 286 00:13:51,590 --> 00:13:53,570 essentially create a little bit of a divide. 287 00:13:53,610 --> 00:13:56,310 So in your mind, is geography becoming 288 00:13:56,310 --> 00:13:58,770 irrelevant in infrastructure or is 289 00:13:58,770 --> 00:14:00,430 it becoming more important than ever? 290 00:14:00,650 --> 00:14:03,710 I think this surprises people because earlier in 291 00:14:03,710 --> 00:14:06,510 the cloud narrative or almost decade now, this 292 00:14:06,510 --> 00:14:08,870 was about abstraction, about making locations 293 00:14:08,870 --> 00:14:12,190 invisible about and giving a seamless connectivity without 294 00:14:12,190 --> 00:14:14,390 allocation info or something like that. However 295 00:14:14,390 --> 00:14:18,210 AI has totally flipped that. Now AI has various ways 296 00:14:18,210 --> 00:14:20,550 of creating the network, one of the asks from the 297 00:14:20,550 --> 00:14:23,730 AI would be training. Okay I can tolerate some level 298 00:14:23,730 --> 00:14:27,010 of latency and distance wouldn't matter so those 299 00:14:27,010 --> 00:14:29,670 jobs are long running so they can handle higher 300 00:14:29,670 --> 00:14:32,250 latency. But when it comes to inference which is where 301 00:14:32,250 --> 00:14:35,030 the market is heading, inference is incredibly 302 00:14:35,030 --> 00:14:38,190 sensitive to the latency. And When a user asks the 303 00:14:38,190 --> 00:14:41,270 question and he expects the answer under a second, 304 00:14:41,390 --> 00:14:44,010 every millisecond in the network matters and the 305 00:14:44,010 --> 00:14:46,950 inference is totally a different beast. So now the 306 00:14:46,950 --> 00:14:50,170 geography dictates where you need to place your computer, 307 00:14:50,170 --> 00:14:52,430 where you're going to build data center and 308 00:14:52,430 --> 00:14:55,790 critically, what kind of network is going to connect them. 309 00:14:55,790 --> 00:14:59,010 You cannot run a high quality inference service over 310 00:14:59,010 --> 00:15:01,850 a congested IP network, which is multi-vendor network. 311 00:15:01,850 --> 00:15:05,250 So you need a dedicated optical intelligent network 312 00:15:05,250 --> 00:15:08,630 with a deterministic performance, and that will be 313 00:15:08,630 --> 00:15:10,550 the new reality. And that is already a new reality. 314 00:15:10,870 --> 00:15:13,230 Absolutely. 100% agree with you on that. There 315 00:15:13,230 --> 00:15:15,150 is a technology aspect to it, and then of course 316 00:15:15,150 --> 00:15:17,190 there's the sovereignty aspect to it as well, 317 00:15:17,190 --> 00:15:19,830 which allows us to put data in some places and 318 00:15:19,830 --> 00:15:22,850 it limits us to put data in some other places. 319 00:15:22,850 --> 00:15:25,870 Now, of course, the entire paradigm of AI 320 00:15:25,870 --> 00:15:28,710 and cloud is completely reshaping the 321 00:15:28,710 --> 00:15:30,790 network demands and the traffic patterns. 322 00:15:30,810 --> 00:15:33,310 How does that influence your global strategy? 323 00:15:33,590 --> 00:15:36,330 Yeah, it influences everything. There has 324 00:15:36,330 --> 00:15:38,190 been a logarithmic scale that has been a 325 00:15:38,190 --> 00:15:40,450 parabolic scale now. It is called AI scale. 326 00:15:40,450 --> 00:15:43,450 So we do see AI workloads are 327 00:15:43,450 --> 00:15:45,330 increasing more than a parabolic pace. 328 00:15:45,330 --> 00:15:47,670 Most of the reports suggest that it would 329 00:15:47,670 --> 00:15:51,170 be 35% or more annually. But in my personal 330 00:15:51,170 --> 00:15:53,810 view, it's more than just 35% easily. 331 00:15:53,830 --> 00:15:56,890 And inference is also expected to overtake the training 332 00:15:56,890 --> 00:16:00,410 by 2028 in terms of how much computer power it's 333 00:16:00,410 --> 00:16:03,810 going to consume. That is a fundamental shift because 334 00:16:03,810 --> 00:16:07,810 inference workloads need to be close to the users which 335 00:16:07,810 --> 00:16:10,950 means they need metro and regional networks not just 336 00:16:10,950 --> 00:16:14,950 big fat pipes, basically. So our strategy is essentially 337 00:16:14,950 --> 00:16:18,450 built around this. In India we are evolving our SmartNet 338 00:16:18,450 --> 00:16:22,490 platform into next generation overlay that is 339 00:16:22,490 --> 00:16:25,450 purpose built for the AI traffic. It is an intelligent 340 00:16:25,450 --> 00:16:28,470 optical AI approach we are taking and it comes up 341 00:16:28,470 --> 00:16:32,370 with a very tight latency and jitter specs and also 342 00:16:32,370 --> 00:16:35,070 gives a programmable topology and software-defined problem. 343 00:16:35,170 --> 00:16:38,250 And globally, again, extending the same thesis and 344 00:16:38,250 --> 00:16:41,250 the same logic applies. As the AI adoption grows 345 00:16:41,250 --> 00:16:44,250 across Asia Pacific, the demand for this kind of 346 00:16:44,250 --> 00:16:47,830 infrastructure would follow and we want to be the provider 347 00:16:47,830 --> 00:16:52,750 that enterprises and AI companies trust and adapt for 348 00:16:52,750 --> 00:16:56,010 most of their latency-sensitive workloads. Sure, Vinay. 349 00:16:56,010 --> 00:16:58,850 You talked about SmartNet, and I believe Ciena has 350 00:16:58,850 --> 00:17:01,270 been quite an integral part of the growth story with 351 00:17:01,270 --> 00:17:05,130 SmartNet, and now the JGA cable as well. Can you 352 00:17:05,130 --> 00:17:08,070 share a little more about our collaboration, and how can 353 00:17:08,070 --> 00:17:11,170 vendors like Ciena help support your future plans? 354 00:17:11,370 --> 00:17:13,380 Ciena has been a foundational 355 00:17:13,380 --> 00:17:15,270 partner for us from the day one. 356 00:17:15,330 --> 00:17:18,910 SmartNet, which is our national backbone in 357 00:17:18,910 --> 00:17:21,490 India, that runs on Ciena optical equipment. 358 00:17:21,490 --> 00:17:24,130 The platform has been serving enterprise 359 00:17:24,130 --> 00:17:27,310 and carrier customers reliably for years now. 360 00:17:27,490 --> 00:17:30,430 When we really expanded into the subsea, 361 00:17:30,430 --> 00:17:32,450 as you mentioned about the JGA cable, 362 00:17:32,590 --> 00:17:34,670 Ciena was a natural choice because we 363 00:17:34,670 --> 00:17:36,690 already had the operational familiarity, 364 00:17:36,690 --> 00:17:39,410 the tooling, the engineering capability the team brought 365 00:17:39,410 --> 00:17:42,590 along, and the way you built around your platforms. 366 00:17:42,590 --> 00:17:45,370 And not only when we really got the JGA app and the 367 00:17:45,370 --> 00:17:48,390 recent technology upgrade to the 400 Gig, I think very 368 00:17:48,390 --> 00:17:51,150 few of the subsea cables have end to end for any 369 00:17:51,150 --> 00:17:53,870 technology and Ciena has been an incredible partner in getting 370 00:17:53,870 --> 00:17:56,370 that technology upgrade as well. So what I really 371 00:17:56,370 --> 00:17:59,310 value about partnership with Ciena is their technology 372 00:17:59,310 --> 00:18:02,930 depth. When we come to Ciena with a problem whether it's 373 00:18:02,930 --> 00:18:06,000 about a current optical switch or protection switching 374 00:18:06,010 --> 00:18:10,210 or an SDN provisioning, the engineer engagement is super 375 00:18:10,210 --> 00:18:13,950 real. It's not just sales conversation and that matters 376 00:18:13,950 --> 00:18:16,130 a lot. That matters when you're really building 377 00:18:16,130 --> 00:18:19,890 networks at a performance level where the AI is demanding. 378 00:18:19,890 --> 00:18:23,130 So we are building the next chapter together and I 379 00:18:23,130 --> 00:18:26,710 am optimistic about where it goes and this has been a 380 00:18:26,710 --> 00:18:29,510 great journey. Thank you Vinay. Thanks for the nice words 381 00:18:29,510 --> 00:18:33,230 and one final easy question for you. What really keeps 382 00:18:33,230 --> 00:18:35,910 you up at night? I wish that was an easier question 383 00:18:35,910 --> 00:18:40,810 to answer. I would say the pace. We talked about software 384 00:18:40,810 --> 00:18:44,510 pace and before that it was the telco pace and now it 385 00:18:44,510 --> 00:18:47,310 is the AI pace. I think the AI pace is incredible, it 386 00:18:47,310 --> 00:18:50,850 changes from morning to evening so quickly. The rate at 387 00:18:50,850 --> 00:18:54,090 which AI demand is growing is unlike anything I've seen 388 00:18:54,090 --> 00:18:57,130 in my career. We are not talking about gradual traffic 389 00:18:57,130 --> 00:18:59,870 increases, we are talking about a step function jump 390 00:18:59,870 --> 00:19:03,410 that happens in what customers need. My concern is that 391 00:19:03,410 --> 00:19:06,390 as an industry, we are still building networks the old 392 00:19:06,390 --> 00:19:09,730 way, with longer procurement cycles and longer deployment 393 00:19:09,730 --> 00:19:14,110 cycles. But the AI companies are moving at greater speed, 394 00:19:14,110 --> 00:19:15,950 it's more than a software speed I would call it. They 395 00:19:15,950 --> 00:19:19,650 need capacity in weeks, if not days, but definitely not 396 00:19:19,650 --> 00:19:24,050 quarters. So what keeps me up is making sure Lightstorm 397 00:19:24,050 --> 00:19:27,170 stays ahead of the curve. That we have the architecture 398 00:19:27,170 --> 00:19:30,630 ready, the supply chain is logged in, and the operational 399 00:19:30,630 --> 00:19:33,870 agility to respond when a customer says, I need a 400 00:19:33,870 --> 00:19:38,070 10 terabit capacity between Mumbai to Hyderabad and 401 00:19:38,070 --> 00:19:41,130 they're ready with that right. And that's the smart network 402 00:19:41,130 --> 00:19:43,810 with the programmable capability we want to provide 403 00:19:43,810 --> 00:19:46,570 to the customer a single pane of glass, to see it real 404 00:19:46,570 --> 00:19:48,970 time and make sure that you have the best of the class 405 00:19:48,970 --> 00:19:51,890 service from what we are going to deliver. So that is 406 00:19:51,890 --> 00:19:55,130 the bar we are setting ourselves and that keeps me up. 407 00:19:55,130 --> 00:19:58,850 It's both technology economics of scale and making sure 408 00:19:58,850 --> 00:20:01,270 that we reach to the customer and provide more than 409 00:20:01,270 --> 00:20:04,290 what they could have anticipated. Absolutely and everybody 410 00:20:04,290 --> 00:20:06,970 needs everything yesterday these days, so pace of 411 00:20:06,970 --> 00:20:09,910 change is definitely something we've never seen before. But, Vinay 412 00:20:09,910 --> 00:20:12,590 with that, it's been such an insightful conversation 413 00:20:12,590 --> 00:20:16,470 I'd like to say a big thank you to you. To take 414 00:20:16,470 --> 00:20:18,470 this time with us today, sharing your insights with 415 00:20:18,470 --> 00:20:22,010 our audience. And to our listeners, you can subscribe to 416 00:20:22,010 --> 00:20:26,410 our Ciena insights podcast on Apple podcast or on Radio 417 00:20:26,410 --> 00:20:29,350 Public app for Android users, and now even on Spotify. 418 00:20:29,350 --> 00:20:33,210 You can also find these on our website www.ciena.com/podcasts. 419 00:20:33,210 --> 00:20:37,110 So thanks everyone for listening today, and we'll 420 00:20:37,110 --> 00:20:39,510 keep coming back to you with more content. Thank you.