Ahead In The Cloud: Managing IT in a Technology Company with Nishit Sahay

Chad Watt: Welcome to Ahead in the Cloud where business leaders share what they've learned on their cloud journey. I'm Chad Watt, Infosys Knowledge Institute researcher and writer. Today, I'm speaking with Nishit Sahay, CIO of Marvell Technologies. Marvell makes microchips used in networking, high-power computing, storage, and running the cloud. Nishit is also a software guy in a hardware company. We're going to talk about managing IT in an organization full of technology experts. Welcome, Nishit. Tell me a little bit more about your journey that led you to Marvell Semi?

Nishit Sahay: Well, software or hardware, I was always a semiconductor person, and there's a reason for it. I mean, I was always intrigued by the technology, it's extremely complex. Not just the technology, the supply chain is complex, the sales channel is complex, the market is very complex, and currently, the technology stack that the whole world is sitting on is completely powered by semiconductors. A lot of people didn't appreciate that before till last year where they couldn't buy a car because there was a chip shortage. So that's semiconductor for you. We might have recently become popular, but for folks like us who have been in semiconductor, we had known what our value addition to the technology ecosystem was. So Marvell has always been this legendary semicon company with tremendous intellectual property.

Chad Watt: Describe to me what your role and mission is at Marvell as the CIO?

Nishit Sahay: So, for me or, I would say, for any CIO, the main role is to bring in or enable modern digital technologies and services. And I keep reminding the team, the first and foremost thing is to make the company more efficient. That's what the technology is for. But at the same time, we focus on resiliency. We have to improve the employee experience, so when somebody comes to work, they should enjoy what they do. But ultimately and lately, one of the more important thing is we had to make a company secure. So these are four pillars that we actually work on as my role in CIO.

Chad Watt: Those are good pillars. Infosys recently launched our Cloud Radar 2023 report, and we found that cloud is working really well for most companies. But one area where there's room for improvement is using cloud to integrate acquisitions. First off, how does that finding strike you? And then let's talk about Marvell's experience in acquisitions and integration.

Nishit Sahay: I wasn't surprised by the report. It made sense because we think that cloud has been there for a long time, so it should be extremely mature, which is not. It's a journey, and it's on a good pace for that journey. When you think about cloud systems, again, cloud could be SaaS, PaaS, infrastructure service. So, in this case, we are talking about, let's say, platform as a service or software as a service.

For these type of systems, you have to have a good end state in mind. And they work excellent when you have well-defined end state. You can put your processes in. It automates really well. It's very efficient. But when you're talking about acquisition, there's always a transitionary phase. There are certain things which is common across acquisitions. But most things are depending on the company you're acquiring.

The five big companies that we went after, the supply chain were significantly different, technologies were significantly different. So when you say, "I have a fixed way of doing integration," that doesn't work. So that's one of the flaws, I would say, of the cloud ecosystem or one of the improvement opportunities of the cloud ecosystem.

Now, said that, if you look at the technology that it's enabling lately, the buzzword lately is generative AI, that has tremendous impact on our ability to do good integration. So while it is not fully mature, I don't think it's too far from being the platform of choice when you're doing integrations.

Your second question is about Marvell's acquisition. For us, it's both, organically develop our portfolio and compliment it with other portfolios from other companies, and we've had streams of successful acquisitions. If you think about it, the big AI play that we have made is significantly due to the last acquisition or one of the recent acquisitions we did, which is Inphi. So, as a company, we decided we'll try both ways of growing. And yeah, we got pretty good at acquiring and integrating them both culturally, system-wise, process-wise.

Chad Watt: What kind of advice would you have for someone who's looking to get better at integrating acquisitions from kind of a systems and process standpoint?

Nishit Sahay: Before I start grabbing all the credit, let me give the credit to the company. I had tried this in another company, it wasn't this grand a success. Now, let me just quantify the success, right? Inphi, the example I recently just gave, it's a $10 billion acquisition, a massive-sized company, extremely complicated, right? It was the Ferrari of semiconductor before. Now, we integrated all the ERP systems, supply chain, portal, CRM, name it, within two months of the deal close. Now, that, I don't think anybody can beat.

So the secret formula here is collaboration, and it's like a Goldilocks situation, right? I mean, you have to get everything right. It's not just IT is doing an amazing job and business is just waiting on the other end for IT to complete. We all come together, business, IT, our implementation or integration partners like Infosys, we come together, and we give our best for this to happen, not for the best for what?

I think, early in the project, we come up with very detailed logistics on what exactly we want to do, what are the gaps, and how are we going to fill it. So when we do the project at that point is pure execution. There's not a whole lot of misalignment. There is no confusion. And my role as a leader or my leader's role within the organization is to just make sure that confusion is on a daily basis eliminated from our ecosystem.

Then there is the alignment piece I talked about. So mostly, within the company, we have great alignment because we have done it so many times. The company which comes in and joins us, they, normally, with the prior experience for them, it is like, "How can you do something like this in two months? I don't think it's possible." So you have to drive that alignment. You have to show them the roadmap. If everybody knows the game plan, that'll happen.

Chad Watt: Is there something along the way that persuades them that you could do this this quickly? Can you capture a moment when that's like, "Okay, yes, this is real," where you convert those doubters?

Nishit Sahay: What we do is we do a first round of very draft data conversion to showcase how our whole process is going to work towards the end. So one of the structured plan, and when you have some big milestones like your first draft data conversion, you showcase the entire ecosystem in your system, it might not be perfect. It doesn't have to be perfect at that time. But then it gives them a lot of trust that this thing is going to work.

The other piece is when we show them the maturity of our ecosystem, because the key question is, "Am I stepping down in terms of technology when I'm entering Marvell, or am I stepping up?" And one of the goals that we have had over the last couple of years is we have to keep stepping up our tech stack, so that when somebody comes in, they don't miss what they had. They get onto it, and they actually enjoy the work, back to the employee experience pillar.

Chad Watt: Very good. Very good. What was Marvell's top-line revenue last year?

Nishit Sahay: Yeah. We went to $5.9 billion last year.

Chad Watt: $5.9 billion. And how many... What's the headcount there?

Nishit Sahay: It's around 7,000 employees, most of them engineers.

Chad Watt: 7,000 employees, mostly engineers, $5.9 billion. How do you run a company that lean to produce that much top-line revenue?

Nishit Sahay: Well, we are like Avengers, right?

Chad Watt: You're like Avengers, okay, which one are you?

Nishit Sahay: A handful of us can really handle the whole army of Thanos. And that's why we have one extra L on Marvell.

Chad Watt: Okay. Okay.

Nishit Sahay: So yeah, I mean, there's a play to our name, just we are lean, we have excellent employees, all that is fine, but then there's physical amount of work. And that's where IT department is coming. You need to push the envelope on automation. When an employee is coming in, their job should be all about critical thinking, right?

Chad Watt: Right.

Nishit Sahay: Anything which is repeatable, is manual, automated. That's been our journey. Like I said, we have pushed the envelope on how much we can. We continuously do. We don't stop it. That's one thing that's powering our growth.

Chad Watt: I want to get a little more grounding on chips and semiconductors and how fast that industry has evolved. Nishit, think back to when you first started working with semiconductors or microchips. What was the state of the art then? And how does that compare with the product that you guys are selling today?

Nishit Sahay: When I was, let's say, in grad school, we were talking about microchips. We thought that this is an amazing technology. You can think it's micro, and it had its own complication. There's the famous Moore's law on how things become smaller and more efficient or more throughput ways.

Now, at that time, we used to hear about, "Oh, there is one day that we're going to be in nanometer range." And that's it, technology can't get any better, so you got to talk about other areas like quantum compute for processors or nanotechnology or whatnot.

Now, here's the thing. Marvell now is making 5-nanometer chips, okay, 5-nanometer-node chips. We are making 3-nanometer-node chips. Soon enough, we'll be making 2-nanometer-node chips. Again, if you're a Marvel fan, we are in Quantum Realm right now. So when you're that tiny, imagine the complexity, your physics changes, right? It's not just about precision design. Your physics changes.

Now, when you're with that type of a size, and by the way, that's why you have the technology stack that you have. That's why you have all these modern cars and computers and data centers and Netflix is because we are able to do that, or cell phones, right? We are able to do that. That's where semiconductor is right now. So when you have something of that small size, you can't be designing on paper, right? You can't be testing things through simple scripts. And that's where electronic design automation comes in.

Now, basically, if I have to summarize it, it's a tool set, which is for the engineers to design and develop chips. But then, now, you layer in this complexity of Quantum Realm, let's say, or 5 nanometers and 3 nanometers of this extremely small size, then you can appreciate how complicated these things get. Talking about millions of simulations that's running, there is no room for mistake. So you have to be perfect with your design, with your testing, video verifications and whatnot.

Chad Watt: So EDA has been around for a while. But now, Marvell is in the process of moving EDA to the cloud. Why are you doing that?

Nishit Sahay: When we do it successfully, we'll probably be the first one for our size a company to do something like this. Now, this is interesting. So why are we the pioneers here, right? Cloud has been there forever. Now, the main thing is there was not a whole lot of optimization.

First of all, the technology and the landscape that I just mentioned, it is very complicated. You have to have tons of investment from all the parties, the hyperscalers, the EDA companies, semiconductor, other software developers, development companies to come together and build this platform cohesively, right? I mean, it's much easier if we just take it and just move it to infrastructure as a service. That's not what we are aiming for. We are aiming for using the true scalability of cloud to get us to where we want to get to. And I'll give you a rough order-of-magnitude idea on what the scalability means for us, right?

So there is one of the projects that we're working on. Well, I'm going to get a little technical here. So the project required around 10,000, 20,000 core computes on a regular basis, to run the jobs and design jobs and whatnot. But certain points in the project, when it got really critical, it required up to more than 70,000 cores. Imagine most companies don't even have it for the entire company. I'm talking one project, 70,000 cores, most companies don't even have it for the entire company.

Now, we're talking about couple of company-sized data centers to be enabled for just one project. That doesn't make sense for anybody to do it on prem. Technologies which modulate this. But then, again, at a given time, we have 20 projects. I'm just talking one project. So that scalability aspect of it is why we first got into this, right? We got to be scalable. And as in when you are pushing our envelope on 2 nanometers, 3 nanometers type of node size, this scaling need is just going to get more intense, more extreme. So that was the first one.

Second one was the EDA process, even though it has evolved significantly, is still, what you might call it, a heuristic process, right? Blunt force, try everything out. There's not a whole lot of machine learning that has gone into it. There's not a whole lot of parallel processing that has gone into it. So anybody who's done on data analytics side of the world, things have changed because of massive parallel processing. That's why your machine learning is so prevalent.

Now, that has not happened in the EDA world. What we hope to get... First of all, one thing is obvious, we'll get scalability. But on top of it, we want to drive a massive parallel processing, so anything which runs for eight or nine weeks. By the way, our jobs can run for nine weeks, cut it down to, let's say, a few days, right?

Chad Watt: Wow.

Nishit Sahay: And again, generative AI, I think it has a big role to play in making EDA more efficient, process more efficient, and cloud is a place for generative AI.

Chad Watt: In the 2021 during the pandemic, Marvell was the first to warn the world that we were going to face a microchip shortage. How did you all become aware of that?

Nishit Sahay: Well, we are a very data-driven organization. A lot of companies claim to be. I've been in companies which claim to be. But that, again, it's a mindset change to be data-driven. We are data-driven. So, for us, the information was right out there for us. It's hard for me to believe that others didn't see it.

Chad Watt: You had to have the courage to step out and say it though, right?

Nishit Sahay: It's the culture. We are all about transparency. In 2016, when we restructured the company, the first thing that we stood up on was integrity and transparency, and we showcased. We took ahead, by the way, but we showcased. So both our analytical acumen, I would say, but layered that with the transparency and integrity that Marvell always displays. And we got together. We went in for our people and said, "Look, we're going to have chip shortage," [inaudible 00:15:21], and then after, everybody came and said, "Oh, yeah, we're also going to get the chip shortage."

Chad Watt: Nishit, we can't talk about cloud and semiconductors without talking about the new thing that's going to boost demand for artificial intelligence and specifically generative AI. Honestly, I'm a little bit skeptic. I'm not impressed with the products that I've seen that are out there. Are you a believer in generative AI?

Nishit Sahay: So a lot of it could be that I'm a little biased here. I've invested pretty much most of my free time in the last five, six years in machine learning. And by the way, large language models have been there for a long time, and I've seen it. I got introduced to, let's say, OpenAI or ChatGPT early last year as a beta trial, and I was just, I couldn't believe it. It was amazing. Then I went into the details of how much training it has done, just to give you the idea here, right?

Earlier models used to talk in a thousand parameters. Maybe a million parameters is really fancy. We're talking trillion parameters here. So I'm a big believer on this. And before that, I have worked in large language models that's been there forever. Now, what those models are good at is to understand the conversation that we're doing and just make a syntax out of it. You can convert it to an API that a system can understand. They did a fantastic job on it.

Generative AI on top of it went ahead. They learned the entire world and came back to you, right? They crawled through the entire internet. They learned about every article that's out there. They learned every programming language that they can. It's the next level of sophistication on evolution that we're talking about.

Chad Watt: Where do you see generative AI having the greatest potential to make an impact?

Nishit Sahay: Let me preface that with what used to be the problem with AI before.

Chad Watt: Okay.

Nishit Sahay: The problem with AI before was it was an area for experts. Only few people who have done masters on it or spent a whole lot of time on it can really do it. You have to really have good statistical understanding, coding understanding, whatnot, understand complex models. And that's why it isn't adopted by so many people because you can't use it till you know it.

What generative AI has done is it has already learned what it has to learn, so you don't need to know how it works, it needs to know you. That makes a huge difference, right? If you're a regular person who has never done programming in all their life or haven't ever heard of what statistics is, you can still come up with an idea and implement that using generative AI. You don't need to know all of them.

Chad Watt: Right.

Nishit Sahay: That makes a huge difference. Now that the bench strength that we had of all these data scientists, suddenly, it went from maybe a few million at max to 7 billion people on the world. So, over next few years, you would see tremendous ideas coming from this area. You have enabled this complicated technology for the entire world to use. You have literally democratized machine learning by doing this or AI by doing this. So it's obviously going to be having a powerful impact in the future. But at this time, the technology that I've seen done, once adopted, they will be big game changers.

Chad Watt: Very good. Now, let's just bring this back to semiconductors. What does expanded AI driven by cloud mean for chip makers, including Marvell?

Nishit Sahay: Yeah, and of course, we are loving it, right? We had a big uptick in the beginning of the year, and a lot of it was driven by AI. In fact, I think Jim Cramer, a couple of weeks ago, was saying that Marvell is sitting on a gold mine due to our AI potential. So we are big suppliers to all these ecosystems, which basically power AI, whether you talk about the big names like NVIDIA or other customers which are hosting AI, which is training AI, which is the main power behind AI. So Marvell chip goes into all of them.

Now, Marvell as a company defined its strategy around data infrastructure. We realized that's the future. Data is the new oil or data is the new gold, whatever you want to call it. We invested in that. And earlier, it was about 5G automotive and still about 5G automotive. And then there was a big play in cloud. It became a major player in cloud. And that's where it is coming back and helping our growth because cloud is what is enabling generative AI. And it's not something you can do in your house, right?

Chad Watt: Yeah. You were talking about something that uses compute cores. Yeah.

Nishit Sahay: Think about it. Before this thing came into the picture, you might be thinking about the traditional AI models that might require some GPUs, some CPUs, a small rack, and you'd be done. So for a big cloud provider for one data center, you're talking over one rack sitting over there, which is posting this one application for AI. Generative AI sits on entire data centers, at times on couple of data centers. So, where AI used to be a part of the cloud, now, AI is sitting on top of the cloud being the biggest application that's all there. So that's going to drive the growth of cloud pretty significant.

Chad Watt: Nishit, is there anything that you wanted to talk to that we neglected?

Nishit Sahay: Yeah. So just to touch up on all these topics, data analytics, we talked about big systems coming in, intelligent automation. We talked about heavy technologies.

Chad Watt: Yeah.

Nishit Sahay: Problem with that is you need technologists. Now, we have been automating everybody else's work. Who automates my job?

Chad Watt: Yeah.

Nishit Sahay: And that's what I think I'm really hoping for generative AI to come in and do that for me.

Chad Watt: What?

Nishit Sahay: Automate what I do.

Chad Watt: So like a GenAI, ITSM type thing or...

Nishit Sahay: Something like that, right? So the biggest tech problem with this technology is the technology itself. It confuses people. It takes time. We have like few hundred systems. Where do you go for one data point unless until we bring it to you? Generative AI is going to enable ways for you to just sit in a chatbot and ask information. It'll figure out where to go and get data for you, right?

So the way I see this is all this technological development or stack development that we've been doing in the company to modernize the company to do digital transformation, we are missing this layer on top of it, which, normally, people think it's data analytics, and that's an important layer, but a layer which can help us do our day-to-day job more efficiently. And that's why I think this will come in. And we're pretty excited about it. And I think it has a lot of value to add when it comes to IT transformations or IT driving transformations.

Chad Watt: This interview is part of our collaboration with MIT Tech Review, in partnership with Infosys Cobalt. Visit our content hub at technologyreview.com to learn more about how businesses across the globe are moving from cloud chaos to cloud clarity. I'm Chad Watt with the Infosys Knowledge Institute. Until next time, keep learning and keep sharing.