Hello, everyone. My name is Mark Seneca. I'm a partner at Orrick and I'm the head of Orrick’s sell-side M&A practice. I'm also the co-head of Orrick’s global M&A and private equity group. I've been practicing technology M&A in Silicon Valley for almost three decades and have had the opportunity to help sell more than 200 companies, including Instagram to Facebook, Nest to Google, and most recently, two important deals in the AI space representing Mosaic in its $1.3 billion sale to Databricks and Casetext in its $650 million sale to Thomson Reuters. I'm joined today by Naveen Rao, the CEO and founder of Mosaic. I've been fortunate to have helped Naveen sell his first company, Nervana, to Intel in 2016 and most recently on the closing of the Mosaic sale to Databricks that occurred earlier this month. Naveen is one of the most successful entrepreneurs in the space AI space having founded Nervana, then having been the head of AI at Intel for several years, and most recently founding and selling Mosaic. Welcome, Naveen. It's been great to work with you on Nervana and Mosaic and congratulations on the recent closing of the Mosaic sale to Databricks.

Thanks. You guys have been a wonderful partner from inception of both companies actually. It's not just the M&A process, it's all the way from literally the beginning of each one—formation of a company, all the advice along the way, and yeah, I can't think of a better partner. And in doing it again, you know.

The audience is very interested in your story. And, personally the experience that we've had in two sales and you being officially a serial entrepreneur has been one of the most rewarding of my career in terms of watching the speed of success and the dexterity of execution, and how you've built your companies. Let's start at the beginning of your work with AI and machine learning. What inspired you to move into this space, both in terms of your education and then your entrepreneurship?

Yeah, I mean, this is a, you know, it's one of those things like, yes, some of the sales have happened quick, but this has been a long, long, time in the making. I'll date myself here. But, you know, I graduated undergrad in ’97, and actually I did research in this field back in the nineties and there were two things that actually were extremely fascinating to me, of course, being kind of a geek. You know, I love sci fi. And, you know, AI was always going to be a big theme as part of that. But I went in and took a computer science course in 1995 called AI and really it was a bit unsatisfying because it was sort of just creative algorithms. There was nothing really truly intelligent. It was sort of like, ‘Well, how do I, you know, do inference for different kinds of problems and constrained spaces?’ And then I actually took a computational biology course after that and did a research project around neuromorphic circuits. And one thing that just fascinated me from the start was the fact that our brains run on about 20 watts of energy. And that singular fact is probably the thing that led me to do everything else after that. I mean, you know, after undergrad I came out to Silicon Valley and did the startup thing for a number of years. After ten years, I actually decided it was time to go back and understand how brains maybe work or what we know about brains and actually quit my job to get a Ph.D. in computational neuroscience. And so, all of this kind of went back to that initial thing. And, you know, then it became clear to me that like, research world is great… …but the way to really influence how technology works is through companies, through products.

Yeah, it's fascinating how you progressed in terms of finding your way to solving these types of problems, or at least getting the educational background to do so. As we've talked about, you've founded two AI companies, and each company that has started, you're trying to solve a problem and create a commercial product, and as you said, really change the world through technology. What was your vision for each of the companies that you've started?

Yeah, I think it's all about like trying to uncover a problem that, if solved, unlocks a whole new market. When you unlock a whole new market you get a whole bunch of new users, you get much more influence in technology. And, founding Nervana was…I finished my Ph.D. and I was in a research group at Qualcomm looking at neuromorphic architectures, which was something that I was interested in from undergrad even. And, you know, it became clear that something had shifted. Some of the problems of how we actually sort through very large search spaces for problems have somewhat been solved with large scale neural networks. And so, then it became clear to me, it's like, well, if that's true, we need computing architectures that really enable cost and time reduction. We need to think about AI or our neural network computing and back propagation learning as a first class workload that that needs to be optimized all the way to the hardware level. And really, that fundamentally changes what a computer looks like. And so, if you do that, you make these capabilities much more available to many more people. It's very similar to what happened with PCs, right? Once PCs got cheap enough, everyone could have it in their home. People start to learn to program all of this. And so, it's very clear to me that hardware was the unlock for that. And I actually pitched this internally at Qualcomm and was turned down to do it, and then went and talked to a few investors and lo and behold they said, ‘Yes, let's go.’ Ali Partovi was our first one. He's actually the one who put me in touch with Orrick and Don Keller at that time who was running the startup practice. And, yeah, they helped us, you know, structure the company in the right way to get going. And we were off to the races.

Yeah, so Nervana was much a hardware company as a software company and in trying to get the right chip architecture. To be able to support these types of applications.

That's right. Yeah. It's really, you know, for an entrepreneurs out there, what, what I think is really interesting is that people talk about AI as being some sort of like, you know, magical capability or something. Really, it's mapping some kind of algorithm onto hardware at large scale. That's really what it comes down to in the essence of it. And it sort of brought Silicon Valley back in a sense of like, hardware became very important—very low level software and optimization became very important again. These things kind of went away for a while, and they're going through this application space of like SaaS software and all of that. Now it came down to really hard-core engineering concepts became what differentiate companies. Every one of them that has done well is really good at systems engineering.

So after your time at Intel, what was the new problem that you had identified and were interested in solving as you thought about launching your second company?

Yeah, so even at Nervana, we, we anticipated this idea of scale—that you need to have many processors working in parallel to solve these problems. And it really lends itself to a neural network as a neural network it is sort of a parallel implementation. And we knew that bigger was going to be better. You know, in 2014 when we founded Nervana we didn't really know that transformers were going to exist and auto regressive language models were going to do what they do. But we did know that concept was going to be important, and so we were actually building towards that then. And after I left Intel in 2020, the next thing was really how do I make large scale neural networks accessible to more people? The hardware was kind of being solved. Nvidia had really, you know, really won most of it and is continuing to do so. Other players were coming with new solutions that were interesting as well. So, I was confident that side of things was going to get solved. Now, the next piece of it was, how do I make all this stuff work together from a software and algorithms perspective? And then, how do I bring a cost down to make it tractable? It's really the same thing. What can I unlock to open up more people developing? And really it came down to, we thought we could engineer the algorithms to utilize the hardware better and bring the cost down from millions of dollars for training these large models, down to hundreds of thousands. And we were able to do that. And I think that's what has happened now is that, all of a sudden, many developers can access this stuff. Many enterprises can access this stuff. And we're just seeing an explosion.

Is also part of the idea of Mosaic that in addition to bringing the costs down, that you're allowing enterprises to utilize AI in more customized ways on their own proprietary datasets?

Yeah, absolutely. I mean, part of this is there are sort of two major blockers to accessibility for general enterprise. And when I say general enterprise, I mean not the open AI’s and the Googles of the world, who have highly specialized talent to going to do this. I mean general enterprises who have some talent, they have software capabilities, but they're not going to be, you know, the ones who are doing the cutting-edge research. They really—cost is a big one. Being able to budget $10 million for a project that doesn't really have a clear ROI is not going to happen even in large organizations. They need to be able to take smaller bites and prove value over time. That's one. The other one is just sheer complexity. Being able to train on a thousand GPUs is a hard thing. It's complicated. And there really wasn't a canonical software stack that could make that easy—easy to use. It was very complicated and required kind of bespoke expertise. So, we went off to solve those two things. And really that enables these enterprises to now take these methods, take these capabilities and apply them to their own data.

Now, if one looks at your track record in building companies and selling them, one might conclude it's an easy process. But I'm sure that that is not the case. I’m sure our listeners would be interested in hearing about what you learned through building companies and selling both, and what other founders could learn from how you've generated opportunities. In my experience, oftentimes, entrepreneurs are building, and the opportunities present themselves rather than, you know, if you've built a good product and you know the opportunities present themselves as opposed to a more intentional decision to now sell. And let's try to drum up interest. What can you tell about the trajectory or the experience of learning about building companies and successfully selling them?

Yeah, I mean, you're absolutely right. I never have set out to build a company to sell. That has never been my intention. In fact, I was I was somewhat allergic to that whole concept with Mosaic. I can walk through a little bit about that, the process there. But yeah, it's executing to build something amazing. When you build something amazing, you get interest around this, and really it's always been about assessing does this acquisition help me accomplish something that I was trying to accomplish? The framework I've used actually in both acquisitions is really—does this help me have greater impact in the world? If the if the answer is clearly yes, then it makes sense to go and figure out if the acquisition can work. If it's not clear I mean, it's not just about selling it for an exit or something like that. To me, anyway, these—these are deeply personal decisions, as most founders will identify—your company is kind of an extension of yourself and it shouldn't be taken lightly. I think there are people who do build companies to sell them and get a return. That's never been how I worked and how why I've started a company. It's really a part of me, feel like it's a it's a child or something like that. You know, it's really just how I make my team and my companies have been they feel like family. So these decisions have never been taken lightly. With Nervana, you know, I saw what I perceived a lot of headwinds coming where, you know, we were going to need a lot more capital. This is back in the old days, quote unquote, where, you know, a seed round was less than $1,000,000. You know, we'd raised $24 million building chips. It's expensive. It was getting more expensive. And, you know, this is 2016 when we sold the company and we'd been on a bull run from a macroeconomic perspective for about five years. And so I was thinking, well, you know, are we going are we headed for a downturn where are things like I didn't contemplate zero interest rates and how that was going to impact funding. But really, these are the things weighing on me. I was like, well, do the, risks of staying independent outweigh the benefits? And it seemed like it was a hard thing to justify, you know, if I'd known, of course, how the funding environment was going to explode and all of that around AI, I probably would have thought differently. With Mosaic… …it was actually a similar thing—less about fear of macro, but much more about how can I go faster? I actually had offers from multiple large companies and it was very clearly no. I didn't feel that we would be able to execute faster, better, and capture a market. Whereas at Databricks, you know, they're in some ways—they're like Mosaic five, seven years out right there. They're still very founder wide. They're like a big startup. I got on really well with the founder founding team and it was clear to me that they had built the trust of enterprises, which are our customers. And so, what it seemed like very clearly was that their go to market function was something we could leverage to just go faster and really, you know, make a much bigger dent in the universe in a much shorter amount of time. Really… …that's what we were optimizing around.

And that deal being predominantly a stock deal, you're not only expected but motivated to really continue to build the value of the combined company and not really view it as necessarily an immediate exit, which sounds like it's consistent with your philosophy and probably had some underpinnings as to why you thought the Databricks deal was so, so compelling.

Yeah, actually my first conversation with our corporate development team I was very clear, like, if we do this, it's a stock deal because everyone has to be tightly aligned. That equity ownership creates that incentive—that aligned incentive. And I don't know if it's made public anywhere else, but—you know—I didn't take any money out, not a single dime. Everything I roll into Databricks is stock. So, I'm here to make the company combined entity great and grow it faster than it could have been separately.

Awesome. So now you've been with our group at Orrick through a couple of sale processes, and from an entrepreneur standpoint I can appreciate that selling a company is kind of an inconvenience or the process of selling it is an inconvenience. You want to get it over with and get back to work. And each process is a little bit different. There's always a bit of chaos and frustration through the process. What was it, if anything, that the second time around, you know, still surprised you or was something that you would advise other entrepreneurs going, you know, approaching the first-time sale process to be aware of?

Well, I mean, there's a lot of things. I think I made a number of mistakes in the first deal I mean, just from how we structure it, what you negotiate upfront: things like organizational structure of the combined entity, what you're going to do, what the functionality is, what the budgets look like going forward. These are things you can discuss with the acquiring company ahead of time, and I think it's very good to have some alignment on that ahead of time. I didn't do that as well as I should have in the first deal. With this Databricks acquisition that was actually the first set of conversations kind of started there and then worked backwards, right? Economics came second and really the economics were like, this is what needs to happen to get the deal done versus being the first thing you talk about. I think one thing it was very surprising to me was the complexity of a stock deal—just the mechanics of it versus a cash deal. Nervana was a cash deal. It's just a lot simpler, right? There's so many fewer corner cases of conversions and this and that. And of course, you know, Orrick, you guys helped help us a lot, kind of thinking through it. But I think, every, every one of these is probably different, right, is my guess. And there's so many little nooks and crannies that to think about and they can have large implications. Right. It's just not obvious when you go through it. And that was that was the thing. I was like, ‘oh my God,’ stock deal was way more complicated to get right.

Yeah, a lot of times you don't have a choice. Sometimes it's dictated by buyers and, you know, the circumstances of the market. In this case it was pushed, it sounds like, more by you as the seller than the buyer or just as much as the buyer.

That's right.

How important has the luck of your investors, really your partners in building the company as well as your outside advisors been, in as you look back on your entrepreneurship and building and selling companies?

Yeah. I mean, Ali Partovi who is now principal at Neo VC, he was our advisor—my main advisor at Nervana. He was also our first investor. He wrote the quote unquote terms, right, for the for the seed round and wrote the first check. And he was extremely important to me. We're jumping into this thing. I'd never, you know, been the CEO of a company before. I never really dealt with investors before. He helped us navigate that. I would say anybody who's going through this would definitely have somebody like him who's been through it, who can at least give you the norms of-of how things should look and how they shouldn't work. What's good, what's bad, who the players are that you should trust and all this kind of thing. It's just, it's really hard to navigate otherwise. And, he was really important to us and Nervana. You know, through all of that and through the acquisition, I mean, I built a pretty big network of investors and advisors that I could pull from. I think one of the things that really is important from the investor side of things is, trust—trust between, from the entrepreneur to the investors. Right, that they're going to have your back—that they're bought into the vision. They see this as a long-term bet. Those alignments need to be pretty strong and something you should really spend a lot of time on. I know a lot of early entrepreneurs, they get enamored with certain names, right? You got to look at a partner. It matters who's on your board, who you're interacting with, not the name of the firm many times. This is actually much more important. And again, it's hard to convince people this. They're like, ‘Oh, you know, I see this big name VC and they're interested in me.’ That's wonderful: really, it comes down to who you're working with and that trust that you build in that direction. On the other side, like those investors need to believe in you as the entrepreneur and give you the room to do what you need to do. I mean, it's never a straight line. There are many, many times that the decision process is completely murky. And in fact, that's probably more common than decisions being clear. You know, what direction do we do we take? Do I spend money on this or do I spend money on that? The investor is not the person who should be making those decisions—they're not the expert. They're investing in you as an entrepreneur, and it's up to you to make the right choices here and call in whatever advice you can. But ultimately, those investors need to trust in you. And I think those that trust by directional trust, that trust is super important. I would say that's probably the most important thing. I've heard of boards where, you know, people on the board from different investors are dictating what should be done in a company. That, things have gone off the rails if… …that's happening—in my opinion.

Yeah, absolutely. You talk about you know, needing the passion, treating your startup as something that you're committed to and not in for a short-term exit. And, you know, being committed to get through all the hard stuff. Like I said, you've made it look easy. But I know it's been hard every step of the way, and building companies is hard. And for sure, having a mentor and trusted advisers obviously is key, key to success. Let's zoom out a bit to the broader AI landscape, and I'm sure our listeners would be interested how you personally think about the various AI and AI related segments of the industry. And what applications or technology are most interesting and exciting? And how do you see different areas that are burgeoning? How do you think that they may or may not require consolidation?

Well, I think from a technology standpoint there's a singular pattern that I see emerging that is probably going to define how AI is used, at least in the next 3 to 5 years. And it’s sort of this kind of co-pilot for X pattern, building tools that augments people in the field, whether that be law, whether it be finance—you know—whether it be medicine, whatever it is. This narrative about full replacement of humans is a bit science fiction where we're not quite there. We haven't built models that truly reason why people do and you know, really are as good at causal inference and things like that as, as humans. But we have built tools that can really augment humans from a retrieval standpoint. What kind of knowledge do you have access to? I mean, this isn't a new concept. I mean, really, this is what Google did with search to some degree. We've just made the interface way more natural for humans. I mean, imagine where I'm being asked a set of questions and I'm asked to give an explanation on something. I might know something about that topic, but if I can type it into like a chat bot that knows about that topic, I can probably give a much clearer answer and then add in my own anecdotes and things like that as a human. So, I see, I see that pattern as being probably the defining one for the next few years. Now, what are the things that need to happen to make to enable that that new world? It's not going to be one general bot in my opinion. There's a lot of expertise out there that needs to be encapsulated correctly into a model. And sometimes you need very, very high precision of recall and sometimes you actually don't want high precision of recall. When we're talking about, say, creative writing or image generation where you're doing something new and interesting, you actually don't want retrieval, you actually want creativity. Those two things are a little bit at odds with one another. But if I'm a medical doctor or a nurse and I'm trying to like understand how I can interpret what a human is saying, like a patient is presenting, I actually need a bit more accuracy and precision in that recall. So, I think building bots that understand how to do those things differently for those different applications is important. Maintaining security and privacy in the different scenarios is extremely important. Finance simply cannot exist in a world where everything's open. It just doesn't work that way. There's too many rules around it, which is good. They're there to protect us, and I think we need to have solutions that get these technologies into the hands of practitioners but maintain those kind of the security boundaries that are required.

Fascinating stuff. We're all interested in the ride over the next two years. Now that the generative AI bots have been released on the world and interested to see how things develop. A self-interested question for my group and other lawyers out there that may be interested in your perspectives. What do you think lawyers need to understand about AI technology to be effective advisors to AI entrepreneurs and executives? Or, is it stay out of our stay out of our way, let us handle the technical side…

No, I think there's a—I was actually going to say it goes the opposite direction, really. I think entrepreneurs need to understand a little bit more about the legal side of things. And I say that slightly begrudgingly, you know, because it's always it's always a blocker for me. Right? And that's how most entrepreneurs think about it is like, this is time. The laws are telling me what I can't do. Right. But I've actually come to the…I've come to the conclusion that actually these things enable us to do more. When we, when we build tools that respect data privacy, it actually enables us to proliferate technologies. Right? I'll give a great example and I've kind of been using this a little bit more recently, is that I think we're a bit in the Napster moment of AI. Napster came out, didn't respect any ownership or privacy or copyrights. But what it did do is demonstrate that there was a definite desire from consumers to have access to their music in a streamed way to, you know, kind of freely be able to flip through large collections of music. Yeah, it was shut down, right? Because of this there were a lot of interested parties whose IP was being violated. Right. And if by understanding that up front, you might have actually been able to build a better solution. What happened later was Apple, by understanding those rights, and carving out the right kind of contracts and agreements, was able to make something that actually proliferated and it was built upon that demand for the same thing. I think we're in a similar moment right now where it's a little bit of the Wild West. We throw everything we can at these models, data wise—you know, IP ownership be damned to some degree. And understanding a little bit more about that, which I'm trying to get smarter on understanding licenses and different ways that we can actually permission data, or at least trace where data came from that went into a model, might actually unlock more use cases. So I think that's actually pretty surprising to me. And, you know, I'm learning every day, you know.

Yeah, and firms like ours and others are in the midst of getting ahead of all of those issues. The three issues—privacy issues, IP protection. There's a whole slew of issues ancillary to this new technology that is, as you say, is the Wild West. It's being it's being architected as we speak and will evolve over the coming years. So it's an exciting next phase in in technology in Silicon Valley and we're seeing a lot of the repeat cycles that we've experienced in other contexts, social media and others. Do you think that there will be a third time? I know you got your hands full for the meantime in your efforts with Databricks and Mosaic on a combined basis, but what's next for you in the in the more distant future?

I mean, I still love technology. I think there's a long way for us to go to build this sort of sci fi future that we've all dreamt of as kids. So I'll say, never say never. I mean, I got a lot, a lot of work to do, and I'm very focused on that. I think I have to identify one of those unlock points again. Right. I think a lot of people talk about like, oh, this person had a great vision, but it was too early. In my view, that wasn't a great vision. Being timed right for when technology's ready and when markets are ready, consumers are ready, or you know, customers are ready is actually extremely important and intrinsic to vision. Otherwise it doesn't happen, right? All of those things need to be true. And so building the right thing at the right time is what I would need to think about. And that includes understanding that technology, is it really ready? And understanding markets, are they really ready to take on this technology? I think there are a few ideas I have. I mean, really, one thing that we don't do very well yet is this idea of multimodal kind of information… …right? We have multiple senses as organisms. And we actually learn about the world through all of them simultaneously. This starts looking a lot like robotics. Are we ready to start solving some of those problems? We're getting there. So, this is this is kind of the next thing I'm thinking about.

Well, like we said at the top of the podcast, we've been thrilled to have our relationship with you. I'm predicting there will be a third child sometime in the future, and we're excited to watch what problem you decide needs to be solved next. Very exciting stuff and thank you so much for taking the time to join us. I'm sure that as I mentioned, a lot of folks will appreciate and be very interested in the perspectives you shared. Thank you Naveen.

Absolutely. Thanks for having me on.