# Astera Labs CEO Jitendra Mohan: Our Customers Want Everything Faster

**Channel:** iConnections
**Source:** https://www.youtube.com/watch?v=dTA7PUyKTCA
**Transcript page:** https://www.withtranscript.ai/video/dTA7PUyKTCA

## Chapters

- 0:00 — Opening Banter and Business Update
- 3:24 — Astera's Origin and Connectivity Strategy
- 10:00 — Scaling Content and Optical Expansion
- 16:48 — Hyperscaler Relationships and Customer Traction
- 18:49 — Growth Ambitions and AI's Impact

## Transcript

**[0:00] Speaker A:** Has anything changed in your business in the last 15 days since you gave guidance?

**[0:06] Speaker B:** No, fundamentally nothing has changed and our demise has been predicted twice already. Fortunately, we are here. Our company goal is to increase our revenue stand. There were two sets of symptoms and no doctor was able to put the two together and that did the trick.

**[0:20] Speaker A:** Well, are you going to disappoint me, Chhatendra, on the 30% share?

**[0:23] Speaker B:** Never doing a lot of these. So, I brought I brought a reinforcement. You don't need re reinforcements, Gavin.

**[0:31] Speaker A:** Always good to have and

**[0:32] Speaker B:** try my best.

**[0:34] Speaker A:** Um, well, look, Jatendra, thank you for being here. As we'll get into it, our companies have a bit of a long history, at least in the history of Atrades. We were early private investors in Astera Labs, and it's been pretty incredible to see you grow the the business. Um, Jatendra for background for those who don't know him, uh, is a 30-year veteran in the semiconductor space. So, between his time at National Semiconductor and then TI, he's got 30 years of experience under his belt and now obviously a co-founder and CEO of Astera. Um, Jatendra inside of Astera, little known fact, won an award as a cool as a cucumber as

**[1:17] Speaker B:** I don't know how that got out, but

**[1:20] Speaker A:** so that took some digging to find, but I can say knowing Jatendra uh and having spent a lot of time with him over the last 5 years as investors in the company, he has really navigated the AI space elegantly and taken full advantage of this opportunity. Um, and by the way, he's just a great human. super nice and very humble. So, love spending time with you. Thank you for being here.

**[1:44] Speaker B:** Yeah, man. We we did the series C. Is that when we first invested

**[1:49] Speaker A:** and that was

**[1:49] Speaker B:** 21 I think we in That's right.

**[1:51] Speaker A:** 800ish million.

**[1:53] Speaker B:** Yeah. 800 900 something like that. But I have to tell you the story.

**[1:56] Speaker A:** First time I met Gavin, my takeaway was you got to have two Red Bulls before you go talk to Gavin. He's so full of energy, so full of knowledge and so blunt with his advice. So, thank you.

**[2:09] Speaker B:** I am very direct.

**[2:10] Speaker A:** Yes.

**[2:11] Speaker B:** Everybody always knows what I think.

**[2:13] Speaker A:** I met you yesterday and he shook my hand, did not say hi and said, "This is what you should do."

**[2:20] Speaker B:** Well, you're um maybe you and Andrew Feldman could maybe you guys should have had a therapy session last night. Um but listen, man. Um, Jendra, I ordinarily would not ask this question, but given what has happened in the market over the last really two weeks, over the last month, has anything changed in your business in the last 15 days since you gave guidance? No. Fundamentally, nothing has changed. The only thing that has changed is our customers want everything faster. And I'm not even kidding. The number of meetings that we are in where people are looking to get more supply, more parts so that they can string together these racks a lot faster has accelerated if anything.

**[3:09] Speaker A:** Has anything gotten worse?

**[3:11] Speaker B:** No.

**[3:13] Speaker A:** Okay. My my news feed is burning up uh with the topic of the day, but other than that, no.

**[3:19] Speaker B:** Okay. So, business is better since earnings.

**[3:21] Speaker A:** Business is great.

**[3:22] Speaker B:** Okay.

**[3:23] Speaker A:** Good to know.

**[3:24] Speaker A:** And look, Jendra, I think there might be people in this audience who aren't as familiar with Astera as Gavin and myself. Um, so maybe give a bit of background on the company. You sit so squarely in the connectivity space.

**[3:37] Speaker B:** Yes.

**[3:37] Speaker A:** In the semiconductor world, which has really gone through a renaissance as a byproduct of AI, but give us some color on why you started the company 7 years ago. What was the impetus and what really caused you to make the jump?

**[3:50] Speaker B:** Yeah, 9 years now actually. Time flies when you're having fun. So look, um, you know, I'm a little bit of a science fiction buff. I read science fiction. I watch science fiction. And so when this AI was coming around, we

**[4:02] Speaker A:** What's your favorite book, man? Favorite sci- science fiction book?

**[4:05] Speaker B:** Isaac Asmo Foundation series. Absolutely. Um, so actually the threebody problem is very good too. But anyway, so so when kind of the AI train was coming, we me and my co-founders, we were all very convinced that AI is going to become really, really big. And in order for it to become really really big, you would need many many GPUs to work together. And there were many companies, Nvidia included, that were doing a bank job of building GPUs. But we thought that the connectivity was overlooked. And it was a little bit of a selfish realization because me and my co-founders had been in the connectivity business as you said 30 years. So it was a little bit of, you know, the problem to be solved in our backyard. So we started Asta Labs to provide connectivity solutions to to connect GPUs together connectivity for AI and cloud infrastructure and for the last 9 years that's similarly what we've been uh doing and and uh uh been here with

**[4:55] Speaker B:** the wide portfolio of devices that connect you know world's most most common XPs. So Jendra, we started off making rettimers, something really, really simple. And then you've kind of become a I would say an entire AI fabric franchise. We got switches, we got retimers, we have copper products, optical products, variety of products and now you have a entire rack scale reference design. Talk about that evolution and the importance of the new Orion rack scale platform.

**[5:29] Speaker A:** Yes, absolutely. So when we started the company me and and my co-founder Sanjay and Case we would you know talk about what is Eststera are we going to be a rettimer company or are we going to be a connectivity company and we agreed all that hey we are going to be a connectivity company we started with retimers first because that made the most business sense to do it was the right first step to take it was a product that our customers wanted to use um and we were uniquely qualified to build one. So we did that and we became so successful that for a while Eststera got known as a rettimer company. Uh but of course we had been planning for something much more than that. So since then now we've introduced you know more than four product families as you said. Uh we now have the full connectivity infrastructure at the rack level.

**[6:14] Speaker A:** When we went public the AI server was this tall. Now of course is a whole rack and with that the connectivity challenge has expanded. Um and our vision is and we provide today as part of the Orion rack the full connectivity infrastructure starting with the anchor socket of a switching device connectivity which is both copper and optical as well as addressing the memory bottleneck uh that the industry is is facing today. We have the hardware uh we have the software that goes with it. So our customers can basically deploy their GPUs and get as somebody else earlier said earlier get to the first token fastest. So, and the claim to fame that we have is if you connect a given GPU to our uh infrastructure, you can produce more tokens per second from that same GPU. Just to make it I'm I'm a simple thinker, but um Jensen has said that he thought one of Nvidia's greatest competitive advantages was invinc.

**[7:12] Speaker A:** And that is the biggest switch scaleup network franchise in the world.

**[7:17] Speaker B:** Correct. But you were the second biggest and what's your market cap? What's the market?

**[7:21] Speaker A:** 35 billion.

**[7:22] Speaker B:** 35 billion. And you have you have the second biggest switch scale up network franchise in the world. I'm a simple thinker. Foxy ask asks some smart questions.

**[7:31] Speaker A:** Yeah. So um look, Jendra, this evolution of the portfolio has obviously materialized in financial ways, right? your content per accelerator, which I think is a great way of kind of simplifying your business, has grown tremendously over the last few years. And in fact, in your earnings call, you gave a great data point that was we see our content per accelerator doubling every generation of accelerator,

**[7:57] Speaker B:** which at this point is every 18 months. So even in a world where we don't grow capex which obviously the market is worried about you can continue to compound at these high rates. So maybe talk about where you see content per accelerator going.

**[8:14] Speaker A:** Yeah absolutely I will say two things. First you know you know when you talk about financial growth there are two things that lead up to that. One is just the rising tide. The number of XPUs that are getting deployed and look we don't really have any control on how many XPUs will will get deployed. What we do have control over what is our content going to be on a per XPU basis. And so from the start of the company, that's the metric that we've been focused on. When we started out with our rettimer products, our content was sub $100 per XPU. Today, our content already exceeds $1,000 per XPU. And we said, you know, during our earnings call that we have a path to doubling that content every generation. In fact, with the Scorpio switch, uh, Gavin, that you talked about, we will have $1,000 of content per XPU just from that switch.

**[8:55] Speaker A:** And the reason this is happening is really twofold and it's very simple to understand. The GPUs are getting more and more powerful. Therefore, they create a lot more data and they want to be fed a lot more data. So, the amount of data that you need to send to the GPU is increasing. Everything is going faster. That's number one. Number two, the cluster sizes are going up. Meaning when you are given a problem to an AI problem to solve an AI query to answer more GPUs are getting involved with that which leads to a bigger cluster size which means more GPUs need to get connected in a scaleup network which is where we play. So we benefit from higher data rates that results in higher ASPs. We we benefit from having more connectivity solutions both copper where we started as well as optical where we are we are going which gives you more ASP and then finally the anchor socket which is the switch and the

**[9:42] Speaker A:** switches that we build are not any ordinary switch we build switches that are uniquely designed for AI workloads. These switches participate in the the compute that is needed for AI. Gavin as you said uh I I believe we are the only the second company where we have uh compute in the network deployed at scale.

**[10:00] Speaker A:** Absolutely. And just I mean just so everybody really understands there are three really complicated chips in an AI data center. There's the CPU. Y

**[10:09] Speaker B:** there's the XPU generally a GPU these days but not always. We got traniums we got TPUs

**[10:15] Speaker A:** um we got cerebras um and then there is the switch. This is a very complicated chip and networking has been growing faster than compute very consistently for the last several years in AI data centers and nothing suggests that's going to change but I am Jindra I uh we were at an event in Napa Valley one of these private you know um events and and I gave a talk and you were there with all your peers and I said um you know I thought you had real greatness within you and that Astera had the chance to be a really iconic company. Um, you know, much more iconic than than 35 billion. There's there's a lot more to go. You know, Broadcom's worth what 2 trillion. Um, why what is what is unique about the Astera culture? Like what would you say is is special about Astera?

**[11:08] Speaker A:** Because you guys have executed really really really well. It's not hard. A lot of people thought you were a little rettimer company that was quickly going to be killed. Why are you Why is Astera executing so well?

**[11:20] Speaker B:** Yeah, I mean that's a

**[11:21] Speaker A:** little bit of a softball, man. I'm trying to be nice for all the times I've yelled at you.

**[11:24] Speaker B:** No, no, no, no. Um, I have to give kudos to our team. You know, when people come and join Astera, somebody told me that there is an Astera factor. There is there is a vibe within the company that allows us to tackle problems that would be that would sound impossible. We focus religiously on delivering results. We are obsessed about our customers. In fact, one of the greatest achievements that we've had as a company is to earn the trust of our customers. I can tell you that there are times where our customers came and said, "There is no way, Jetendra, you and your team will be able to build this product. You have never done this before." And we said, "Okay, please give us a chance and let's do this." This happened with our Leo family of memory controllers. We built the best memory controller product out there.

**[12:07] Speaker B:** Then the second time around when it was time for switches we're like you cannot build this but because of what you've done we will teach you how to build this switch. So we have this just very very good relationship with our customers who tell us you know uh what they are building where our products fit. So we get a ringside seat of understanding what products to build. Um and then again kudos to the team they just you know just blood and sweat all day long. We work six days a week. Uh people take great pride in what they are doing. So that X factor is absolutely there. And the last thing that I'll say is we are never afraid of new opportunities. Whenever there is a new opportunity that comes, we embrace it. Whether we know how to do it or not, we figure it out from first principles.

**[12:48] Speaker B:** We make mistakes. We learn from them. Our customers tell us when they make when we make mistakes. Our advisers tell us when we make mistakes. And I think that humility is really ingrained in everybody in the company.

**[12:59] Speaker A:** Yeah. And Jendra, I think within the fabric part of your business, there was this phenomenon earlier this year where companies who were viewed as levered to copper connectivity were very much viewed as losers in the market and those levered to optical connectivity were viewed as winners. And I think your business was mistakenly put in the former bucket for some time, but you've talked a lot about how much you've built up the optical portfolio and you've been working on this for years now. So maybe talk a bit about how you've built up the optical product suite and the supply chain as well. How much are you doing there to facilitate this ramp?

**[13:34] Speaker B:** Yes. So we've been public for a little over two years and our demise has been predicted twice already and you know fortunately we are here. So uh regarding optical you know we've been in the connectivity space for such a long time that you know when optical is needed and when optical is coming. um as the the famous Andy Betosham said never bet against copper because copper just happens to be the most reliable low power no lowcost interconnect that is out there and so that's why we started with copper but we figured that optical is going to come as again as these cluster sizes grow and your reach is going from 2 m to 20 m and 50 m and whatnot so we worked on you know bringing up the talent and everything within the company a couple of years ago because we figured that we will need this and and you know some of our uh vendor s and partners will not be able to provide this technology.

**[14:23] Speaker B:** So we put this team together and just very recently at OCP just a month or so ago we announced a full portfolio of optical products including the electrical IC's that go into optics silicon photonics that goes into optics uh you know vixelbased front ends as well as co- co-ackage optics. So we are very happy to ship copper where our customers want copper and the opportunity to gain a lot more share of their wallet by shipping optical is absolutely there and we are very well positioned to capitalize on that. Let's just make this really simple and like let's just say it once. So, um to to save me from frustration when I read things that are written, but definitionally if you make a switch, either copper or optics goes into that switch, switching is now your biggest business. That's correct. And can you just talk about what happens to the dollar value of a switch when we move to CPO if that when and if that happens?

**[15:22] Speaker A:** Yeah. So I think there are two things to talk about here. First, what does the switch give gives us, right? We've I've talked about switch being our anchor socket a few times. So when a hyperscaler or an AI platform provider is putting together their rack, first they of course look at their XPU and then they look at the scaleup protocol. With the scaleup protocol they have to pick the switch vendor. So the switch vendor gets selected first right after the XPU's architecture is decided. Once the switch vendor is selected the connection between the switch and the XPU could be copper could be optical but the switch vendor has an undue level of influence on what gets selected which vendors rettimers go in there. Of course you know if it is from the same switch vendor then they will all just work very nicely together. So what we aim to do as I said earlier we provide the full connectivity infrastructure.

**[16:09] Speaker A:** We start with the switch. We go to the connectivity options whether it's copper or it's or it's optical all the way to the XPU and and one day we will do IO chiplets that go into the XPU as well. So we take full control of this. Um in terms of the time itself we said that the optical time when everything goes CPO is about $40 billion and uh we keep getting the scale up switch time wrong every single time it whenever we come up with a number is lower. But a scale scale up switch with CPO without what is the dollar differential?

**[16:40] Speaker B:** Double

**[16:41] Speaker A:** double in easily double in terms of the ASP because these optical engines sell for thousands of dollars per engine.

**[16:48] Speaker B:** Thank you.

**[16:48] Speaker B:** Thank you. And you guys are very associated with AWS. Our friend Nafa, amazing to be associated with AWS. We're very bullish on Trinidium 3, bullish on everything that they're doing. Um, we think of them as in in our venture portfolio. They're one of the hyperscalers that we most like our companies to work with because they're very innovative, but there's a lot more to Astera than the AWS relationship. So, can you talk about other hypers scale relationships, how those are developing? Yeah,

**[17:18] Speaker A:** you referenced it.

**[17:19] Speaker B:** Absolutely. So, we we are very proud to have all the hyperscalers as our customers today. uh we we got our break uh with Nvidia when when Jensen was shipping or Nvidia was shipping hopper GPUs and our rettimer was attached to each one of them. We are super thankful to them and at that point you know everybody used our rettimers. Uh since we have now gone into switching and so on our concentration is more when somebody is using a protocol that is supported by our switches. So therefore they interested from some hyperscalers and there are actually a lot of them. uh there are close to 20 opportunities that we are chasing just with PC expressbased protocol. However, our goal is to really service the entire XPU market, right? So, no matter which XPU is coming out, is it GPU or a TPU or or you know, tranium and uh Maya and MTIA, we want to service all of them.

**[18:11] Speaker B:** Therefore, we will build the switching solutions that address not only PC express and ULink and like we're doing today with Enink Fusion, but also all the other protocols. So, no matter the the the XPU, we'll be able to service that. Now specifically to your point, we're very happy with the customer traction that we are getting with our Scorpio platform. We are shipping to more than two customers uh just last quarter. We have several more uh that are going to be shipping here in Q4. Our Scorpio P series is already shipping to multiple customers. So we are all in the right track, but really uh the the big you know benefit will come when we are able to address the entire XPU market that we are very hard at work for.

**[18:49] Speaker B:** And just hitting on this TAM point to bring it kind of high level again. You've given TAM numbers that I think together between scaleup switching and optics amount to 70 billion plus. And again, you've constantly under forecasted, but we we like that as investors. Um, your Q4 guide implies you'll do about $2 billion in revenue this year. And correct me if this is wrong, but I believe that TAM excludes the NVIDIA ecosystem. So how do you bridge from here to being a meaningful slice of that 70 billion over time?

**[19:21] Speaker A:** I'm going to be disappointed under 30% share to Tendra.

**[19:24] Speaker B:** Okay. So look, personally, my goal is to go and our company goal is to increase our revenues 10x, right? When we get to eight, we'll up that goal again by another 10x and we'll see. And the reason for that, as Brad said earlier in his talk, is that if if you do everything right, how big is the price? The price can be really, really big, right? the the the capex spend from the hyperscalers exceeds a trillion dollars. You know, maybe it's going to be $2 trillion. And as you correctly pointed out, our revenue is only $2 billion. So, we have a ton of headroom to grow and we are growing with the the hyperscalers that are using our products today. And we have lot of room to grow with the other hyperscalers as we bring some of these newer optical products and and switching technologies out to the market.

**[20:04] Speaker B:** So, very excited about what the future holds,

**[20:07] Speaker A:** man. Well, are you going to disappoint me, Chindra, on the 30% share?

**[20:10] Speaker B:** Never. Good man. Um,

**[20:14] Speaker A:** two,

**[20:15] Speaker B:** we didn't even we didn't even get to CXL, which is important. But I do just want to say if you are um, you know, if you if you're in finex, if you're an investor and you're hoping to get a one-on-one with Astera, Jatindra did have a previous commitment and he has to leave immediately after this. Next year, Jatindra is going to do one-on- ones. Absolutely.

**[20:33] Speaker A:** Good man. But coming back to America, um, you know, the three questions we're asking everyone, what is AI doing to make the lives of ordinary Americans better today? Yeah, I think that's a great question. First of all, people whether they realize it or not, are already touched by AI in many, many ways that you may not see it. I mean, you put this show together and you thank everybody who did. I don't know how many agents were involved in putting together this show. Thankless. But

**[20:59] Speaker B:** can I actually just tell you something very funny? There was a uh there I thought there was a very efficient I Connections employee named Ellis Reid

**[21:08] Speaker A:** that was awake 24 hours a day and I was and I feel a little silly as an AI investor that it took me a while to realize Ellis Reed was an agent.

**[21:16] Speaker B:** Yeah.

**[21:17] Speaker A:** So yeah. So but I think just you know coming closer to how you know where I really benefit. Don't judge me. I talk to my car. I I drive a Tesla and first of all it drives itself. Secondly I'm talking to Grock and getting ready for meetings and such. But where it really hit home was somewhere somebody in our you know very close family was going through a medical situation. Um and I and I talked to to uh uh to him and said why don't you take all of your symptoms all of the test reports and just feed this into clawed or open I don't know which one he eventually ended up using and he did that and the problem was there were two sets of symptoms u and no doctor was able to put the two together. AI said hey this is happening then that's happening maybe this is the cause please go talk to your doctor about that.

**[22:00] Speaker A:** He did doctor say oh this sounds like a good idea let's stop this medication as advised by AI and that did the trick

**[22:06] Speaker B:** and now anytime I get any you know medical reports whether it's for me or my parents or my in-laws first thing they go into cloud and we figure out what it's going to to say so I think AI is going to just benefit everybody immensely uh and and if we look out 3 years as a avid science fiction reader uh I'm embarrassed I learned that on stage to tendra uh makes me feel like I need to listen more when we talk. Um but um how is AI going to make the lives of ordinary Americans better in three years and help us reclaim and revitalize the American dream?

**[22:40] Speaker A:** Yeah. So I think first of all everybody will benefit in many ways than they they even realize. But the two things that I would like to call out one is education and the other one is healthcare. I think both of these fields are going to get of course everybody knows about coding and and you you know the specialized applications but both of these fields are going to get disrupted. There is an immense amount of opportunity for the new generation to learn. You know, there is no excuse. AI will teach you whatever field you want to learn. Same thing with healthcare. The the fact that the context windows can be so large and you can hold so much information together in in an agent's memory is fundamentally going to change how healthcare and education uh shape up in the world.

**[23:19] Speaker B:** Love it. Thank you, Jendra. And safe travels to your next deployment.

**[23:23] Speaker A:** Absolutely. My pleasure. Thank you.

**[23:25] Speaker B:** Thank you. 30% baby.

**[23:26] Speaker A:** 30%.
