AWS Pelican Hill 2025, Nishit Sahay | AWS Events

Hello everyone, I'm Nishit. I'm CIO at Marvell and I'm really excited to share our journey with you. So Marvell is a leading semiconductor company. I mean apart from everything else that you see here, we are known to be the top AI player. If you think about GPU as being the brains of AI, we are the nervous system. So we're in the thick of all the innovation that's been happening. So we keep asking ourselves one question: How can we move fast? So when you're a large company like Marvell supporting a pretty sizable revenue with a very lean team, pretty significant growth and by the way very very tough competition, you have to really lean in on technology to take over some of that weight. And to unlock the potential of this technology, you have to decide on how you're going to operate. And that brings us to our three inflection points.

So I'm going to talk about three things. One is how we leverage data to make informed decisions. How are we embedding AI into our ecosystem to bring workflow automations? And of course with all these technologies that we are throwing at people, how we drive adoption.

So let's start with data. Now here's what was the case with semiconductor industry. It was very predictable. The rhythm was pretty much standard. Your problem statements in semiconductor industry for decades had not changed. Now when you have a situation like that, you don't really have a lot of need for advanced analytics. Definitely not for a fabless company like Marvell. However, we went ahead and we started investing in three different things. We invested in a very high-end data infrastructure. AWS platform with all its bells and whistles, single data warehouse on Snowflake, fancy analytics tools on top of it. That was the easy part. We invested not just in building reports and dashboards, but actually building data models around business functions. And the third one which sounds the easiest but is the hardest: we invested in building processes around this data ecosystem. You can build dashboards and reports and most of it you throw it away, but we built processes on how to use them. Now here's the thing. This is where we brought in our business system analysts to work with data engineers and data analysts to build these processes. And if you've run it, you should know BSAs hate data engineers.

So we built this super hero infrastructure, really waiting for a super villain to show up, a super villain problem statement to show up, and showed up it did. In 2021 we had the unprecedented supply crisis. I'm sure you guys tried to buy a car. Sorry about that. And that was compounded by an unprecedented demand. Well, a year later there was a demand crisis, you know, compounded by a significant increase in inventory. And right after that we had the AI revolution, another huge uptick in demand. And now with everything that's been happening, I think unpredictability has a new meaning to it, right? So all these data investments that we did, it paid off. We were able to navigate through these changing problem statements. Our problem statements are changing on a regular basis and we are able to navigate it through all of this quite effectively. And now with AI coming in, we are trying to tap into our unstructured data. As you know, more than 90% of your data is unstructured. We're trying to tap into that using data mesh architecture. But I think that's a story for some other time.

So that comes to our second inflection point: generative AI. Looks like this is one thing we are spending our time days and nights talking about now. GenAI, we started fairly early on. We know it works. The question we had for ourselves is, okay, where do we make it work, right? What is the big business impact that we can bring out of this? So like probably a lot of you, we did hackathons, innovation contests, and we got ourselves busy building it.

Now what we realize: building GenAI systems is super easy. Of course scaling is hard. I wouldn't take that away. Building is super easy. So for a company like us with 85% engineers, highly talented engineers, if it's easy for me, it's easy for them. So we went and asked them, are you building your own systems? And with a lot of pride they said, oh of course we are. So for us to avoid these shadow AI systems all throughout the company to be propping up, what we did was we actually invested in an AI development platform. So we built this platform on top of AWS with Bedrock, with LangChain, with all the bells and whistles of multi-agent framework, RAG, knowledge graph, and we invited our engineers to build on this platform. And they came in big numbers. So at this point we have around 150 developers outside IT working on these platforms, and more so 10% of the company is in some shape or form working with us on deploying these solutions.

So we were able to move fast on AI space. Now with great speed comes great responsibilities, as some said. So we have put in, like everybody else should, we have put in a great governance, robust governance framework in place. By the way, that has to evolve on a regular basis across various functions. And our goal is to not just deploy fast but, you know, responsibly with high level of security.

And that brings us to our third inflection point. You can build all these systems, if no one uses it, what's the point. So what we noticed was when we build these systems, there are users who are using it beyond the specs of the product, like we didn't build it to be used that much, and they were able to stretch it. And most of them wouldn't even log into the system. So the spectrum of technology adoption became massive because these are optional systems right now. You still can do your day job without ever logging into any AI system, right? So that became a key initiative for us. How do we narrow the spectrum? So we did quite a few things. First is we built champion networks, AI champion networks throughout the company, various sites, regions, business functions, folks who can actually sit with their co-employees and help them adopt. We have recently released an AI learning portal. The idea here is to demystify AI as well as give functional level trainings on AI systems. We did an AI event recently. We did it in person across 17 sites globally. We had 150 presentations. Again, I'm very proud to say most of it was done outside IT, and we had 55% of the company physically showed up to these sessions. Quite a bit of work, but it had a big impact. We had seen a significant uptake in adoption since we had done that.

So I don't know if I mentioned we move fast. Now how do you measure something like this? How do you know you're moving fast? There's no one key metric that you can really use to say, oh if I have this KPI that means we are moving fast. But we have some good outcomes, some good signals that we can show. The first one obviously: Marvell is known for acquisition. We don't just acquire, we actually integrate them super fast. And we have some numbers, like in five months, a $10 billion acquisition, end-to-end integration there. In the silicon design space, we are probably the first one to move our end-to-end silicon design for high-end two nanometer products all in cloud, right? And then again talking about moving fast, if you have caught up with the news, we had signed the contract pretty recently, and in five months I'll be taking one of my data centers, one of the three data centers that we have, one will be completely moved over to cloud in four to five months this year. Of course on the AI space, we are scaled up 12 production-size environments. These are production-size environments in production, and there are quite a few in progress and that should be coming up. By the way, this has a big fallout rate and I'm going to talk about that soon. But good success there.

Knowledge is great. So what have we learned from this? So I'm going to talk about three things. First is the obvious one: change management. Pretty obvious, right? Now, here's the thing. Change management should be both top down and bottoms up. The intent and need of a change has to be top down. That brings in the authority. That brings in the momentum. But the change actually happens at ground level, right? So, the change itself has to be driven bottoms up. Now, here's the hard part. You're not dealing with one type of people. You're dealing with people with different point of views, different worldviews, different approaches of doing things, different generations. This is the first time we have five generations of people. Yeah. Silent generation still is in the workplace, right? We have five generations of people working in a workplace. And you cannot use one methodology for managing change and aligning change for all of those. So you have to have different tools, different communication patterns, different feedback loops. Now this is also not a one-time thing. You can't just do it once and leave it. Throughout the change that you're trying to drive, the initiative that you're trying to deploy, you have to continuously keep doing this top down and bottoms up stuff.

But I still say this is an obvious thing and we've been talking about it. This is something that we have to learn: a private equity versus a venture capitalist mindset. Look, I'm a CIO. A good CIO should think like a private equity person. Reduce cost, manage risk, be super conservative. If you have dealt with CIOs, we are very conservative people. We reduce technologies. We don't increase technologies. Here's the thing. If a good CIO is there and if you have five ERP systems, we'll make it one. That's what we're good at. And here's the reason: your value space was fixed. What it can do, the value space was fixed. So the thing that you had to optimize was the bottom line. There was no top line for us. The thing you had to optimize was bottom line. So you were in expense optimization mode all the time. Now here's the change with AI. AI, now you're on a new value creation space. And with AI systems the failure rates are high. So the 12 that I mentioned that went live over two years, 200 failed. The six commercial applications that we have taken live production scale, over 100 have failed, and many of these companies are 20 people strong. So the traditional thought process had to be augmented, not replaced, augmented with this new VC thought process. You have to be bold. You have to take calculated risk. If it doesn't work, cut your losses. Move on. Don't try to make it happen. And if it works, scale fast.

And like I mentioned, it's not just an expense optimization space. You have to be in a value creation space. And that gets us to the last learning. And I can tell you this is probably the hardest. So we talk about attention economy. Now, here's the thing. A little while ago, every IT team or IT group would talk about doing six to seven major projects, right? These projects would take, name it, product lifecycle management, CRM project. These projects will take anywhere from six months to multi-year, and some projects never end. And in those cases, you're also dealing with different layers of technology. Server, web servers, application layer, database layer. You have to manage all of that. If you're a DBA, you spend all your time indexing database, not too far ago. Now with the current systems, with AI, a good AI system, you can actually get it implemented in few weeks. With cloud, all these layers of technology that I'm talking about, it's automated. You don't have to deal with it. You shouldn't deal with it. So the question is now you have all this time. What are you going to do with this time? Because a lot of your work is automated. So now that our value, our top line is not fixed. You can create more value. So you should do more. Now this doing more question is: can you broaden your spectrum of thinking? So we are good about thinking deep. We were always taught to think deep. Now we have to learn to think broad. If you can pay attention to it, you can actually make it happen. That's a hard thing for us to navigate through. So that's all I had. You know, tech industry is an exciting place. Marvell is an exciting place to be. I'm sure you feel the same for your industry. Well, this is the story and thanks for sharing it with me and best of luck with yours.