Description

For the latest episode of the MLOps Weekly Podcast, join host Simba Khadder as he chats with Gautam Krishnamurthi, partner at Great Point Ventures, about the rapidly evolving world of AI and its impacts on the future of venture capital investing. They also discuss the latest trends in large language models (LLMs), venture valuations, and the impact of rising interest rates on the public markets. Gautam provides his expertise on identifying real enterprise use cases, distinguishing valuable startups amidst the noise, and the critical role of infrastructure in the machine learning landscape. Lastly, you’ll learn about the transformative power of AI, how it's reshaping industries, and what investors seek in the next wave of groundbreaking companies.
‍

Listen on Spotify

Transcript

[00:00:06.430] - Simba Khadder.
Hey, everyone. Simba Khadder here, and you are listening to the MLOps Weekly podcast. Today, I am really excited to be joined with one of our investors, and actually one of our board members, Gautam, who is an investor at GreatPoint. He joined GreatPoint in 2021. He invests in areas such as enterprise software, from AI to data infrastructure, fintech, commerce infrastructure. Prior to GreatPoint, Gautam was a partner at Green Bay Ventures, where he invested in companies like Databricks, RapidAPI, CISO. Before Green Bay, he was at Capital G, where he also invested in companies like UiPath and Robin Hood. He began his career at Goldman Sachs. Another fun fact about him is he was a member of the Varsity Football team when he was at Stanford. Gautam, so great to have you on today.

[00:00:53.050] - Gautam
Yeah, thanks, Simba. Also wanted to wish you a happy birthday.

[00:00:56.350] - Simba Khadder.
Thank you. Honestly, not much else I'd ever be doing but hanging out on a nice sunny day and talking to you about what's happening in AI.

[00:01:03.840] - Gautam
Of course, yeah, man. I appreciate you having me on. I know you guys have done quite a few of these, and so glad to get my shot on here, too.

[00:01:11.090] - Simba Khadder.
Let's kick off with a nice easy question. There's a lot going on in venture right now. There's this obviously huge transformational technology shift with LLMs and all that's happening in foundational models. Something's happening to the economy. I'm not the person to ask. Maybe we're in a recession, maybe one's going to happen, maybe one's never going to happen, who knows? Venture valuations are all over the place. It sounds like it's the have or have not. Help us understand. What's going on?

[00:01:39.180] - Gautam
I think it feels like that question was almost two parts, what's happening in the venture landscape broadly and then specifically within the machine learning, LLM world, what's happening. On the venture world broadly, I think it really, as it has with every other recession, comes down to interest rates. If you look at what the Feds have done over the past, call it three years, with major rises in interest rates, highest interest rates that we've had in the past 20 years, that's really hamstrung the public markets. When you look at the public markets as the end goal for a lot of venture backed companies, that's created a backlog or a backup in growth stage companies where growth investors are deploying less capital into these growth stage companies because they're seeing less exits on the IPO market as quickly as they had back in, call it 2020, 2021, where interest rates were extremely low.

[00:02:32.270] - Gautam
We think that as the interest rate environment starts to become more favorable, even though the Fed said something else yesterday, over time, we would expect the interest rates to start to come down over time. That's going to start to unlock the public markets, which will start to unlock the IPO markets, which will start to unlock the growth markets.

[00:02:50.190] - Gautam
That will trickle all the way back to the early stage markets and create this flowing effect or a traffic jam effect, where over time, over the next call it two or so years, even though the environment today is not great, the environment over time in the next two or so years will start to become more favorable. That being said, the LLM market and the machine learning market, for lack of a better term, has been as hot as ever. You've definitely seen over the course of 2023 that folks have been pouring a lot of money from the venture world into early-stage businesses in the LLM space, in the machine learning space, for obvious reasons.

[00:03:28.530] - Gautam
Like you said, this is a step function change in levels of productivity, what you can do with LLMs, how LLMs can affect your business and improve your businesses, how they can improve consumers' lives. From all these different regards, there's a lot to be excited about. That being said, I think over the course of 2023, there was a lot of excitement. If you think about the hype cycle, there was a lot of excitement back in 2023. I think 2024 is the year of real use cases and coming to a level of reality where there were a lot of maybe overinflated valuations back in 2023, companies that raised a lot of money on not a lot of traction.

[00:04:09.860] - Gautam
It feels like 2024, though valuations are still high, you're still going to see elevated valuations, especially for fantastic founders and fantastic companies. You're starting to see a little bit more reality, companies that are finding their footing within the enterprise, companies that are actually finding real use cases, and valuations that are coming a little bit more back to Earth. Especially from a personal investor perspective, it's nice to see prices coming a little bit more back to reality.

[00:04:37.670] - Simba Khadder.
It makes a lot of sense. I think having that background of... I think, at least for me, it's hard to think and see everything that's happening on the macro. You can only really focus on the micro. What can you do and create value and just hope other things play out. But you can always hear and feel the rumblings of the macro around you. One thing I've noticed, at least of this LLM wave, it reminds me of almost like the early mobile wave. That's one I experienced, so I remember. It was just like everyone had a startup. It was like a million startups. It was a very different environment.

[00:05:11.050] - Simba Khadder.
The barrier of entry was much lower because the transformation or the technology shift was so transformative. It feels like that again, where there's a lot of stuff coming out. There's a lot of really interesting use cases, obvious use cases where there's like 30 startups around it. There's a lot of really shiny many demos, but as we've been seeing recently, some of those demos maybe don't hold up to the flame very well. How do you parse it all? What do you really look for and how can you find in all this noise, the signal, the companies that at least you think are going to make it?

[00:05:47.690] - Gautam
That's a great question. Hopefully, it's something that I can do right, especially as a venture investor, that's in the job description. It's interesting. Like you said, there is a lot of noise. Finding the signal is getting and tougher and tougher. I think for us, really what we're looking for as an investor, as GreatPoint Ventures, really what we're looking for is those real enterprise use cases. I think especially with LLMs, especially on the infrastructure side, which I'll stay focused on right now, especially on the LLM infrastructure side, our view of the world is there's a lot of great companies who are building a lot of great infrastructure.

[00:06:24.160] - Gautam
A lot of that infrastructure is being used by startups, other companies that are very early stage to build and try out a product. The enterprise adoption still feels on the come to us. When we're looking at companies and trying to make bets on the infrastructure side, what we're really looking for is that enterprise adoption. Are large enterprises, Fortune 500s, the global 2000s, actually using the product? Are these companies able to deploy with these large companies, large enterprises, and scale up with their number of users and number of use cases, the amount of bandwidth the infrastructure takes.

[00:07:01.460] - Gautam
That to us is the most important piece because I think a lot of the use cases within smaller startups tends to come and go for obvious reasons. You're seeing that with a few of the infrastructure players on the LLM side already. Ones, like I said before, that had raised at pretty enormous valuations, but that was behind growth that came primarily from the startup ecosystem, which is great. But the real sticky use cases, the real enterprise use cases feel like they're on the come because enterprises are just generally slower to adopt technologies like this. But for us, that's really what we're looking for is when and who are the enterprises that are adopting these products.

[00:07:40.320] - Simba Khadder.
What's your sense of... Because you see a lot of different companies, and you see the ones that have been successful thus far of enterprises versus maybe the ones who haven't. Are there any pattern you see, any factors that maybe you start to see as why some companies tend to do well of enterprise and others don't?

[00:07:57.110] - Gautam
Yeah. For us, I think there's two main areas. I think it's a fantastic question, but there's two main areas that we're seeing where there's that stickiness on the enterprise side. I think first, is it a hair-on-fire problem? Enterprises have a lot of infrastructure that they're spending on, a lot of infrastructure that they're buying. Is the problem that you solve, that a company solves a hair-on-fire problem for that company? Something where either it's causing a bunch of issues with their end product, it's causing a bunch of headache and extraneous costs for the organization that are unnecessary. It poses some risk that could result in a fine or governance issue or a reporting issue.

[00:08:42.300] - Gautam
What we're looking for is companies that are solving the number 1 problem that you have. That flows into being a nice-to-have or a need-to-have. The second piece that we're looking for within solving problems for enterprises are problems that help enterprises up-level. When I say up-level, enterprises are usually, I guess, two steps behind when it comes to adopting some of these technologies. They often don't have the manpower or the expertise to be able to solve a lot of these problems.

[00:09:13.240] - Gautam
How can you give them tools to make their data science team, their ML team of five people, a team of 10, 20, 100 with the software that you provide? That scalability factor is something that we're looking for, a multiplier effect for lack of a better term.

[00:09:28.710] - Simba Khadder.
It makes a ton of sense. One thing I've been thinking about recently is there's almost two paths of value creation in LLMs, at least I'm seeing in enterprise. One I would call is democratizing LLMs. It's almost like an average person at a company… Let's say you work in marketing. You probably could do your role better and more efficiently if you were using an LLM versus equivalent, you who doesn't. Then the other style, which we see is, I would say, maybe more like hardcore use cases that look more like the traditional machine learning use cases. Like, hey, we have all these huge treasure trove of PDFs that we're going to create this whole new application around this new cutting edge model. From your perspective, is there one that you find more exciting? Is there one that you think is going to create more value creation? Do you have an opinion on that? Do you even buy that breakdown?

[00:10:20.940] - Gautam
Yeah. I think it's interesting that you say that because this is almost a question that's a little bit less on the infrastructure side, a little bit more on the application side of the LLM world. Where we've been making bets and where we've been looking is, frankly, on what you described as the harder side, where it's, like you said, next to the predictive use cases. A couple of the companies that we have invested in, a company called Diffuse Bio, which is using diffusion models to help generate proteins and protein binders based on an ingested set of proteins in a repository that they have developed over time, helps you to develop new proteins much more quickly.

[00:11:00.660] - Gautam
Huge use cases in therapeutics and biotech. The other one is a company called Nixtla, which is on the time series side where they have basically built a zero-shot inference model to help you ingest your time series data and spit out predictions on the back-end with zero training. Both of those, to me, seem a little bit more on the harder side, science-harder ML side, for a lack of a better term. That being said, we are still absolutely looking at efficiency tools.

[00:11:29.560] - Gautam
I will say one thing that's been interesting to me as an investor, especially seeing this boom of LLM companies, LLM applications, the companies on the efficiency gain side, there have been a lot more, and it's harder and harder to see the differentiation between some of those companies for two reasons. One, they're doing something that's very similar to not only their competitor set, or their peer set, but also the older generation, just giving you a big efficiency gain on top of it.

[00:12:02.340] - Gautam
But the other side of that same coin is when you're looking at these companies, the incremental efficiency gain between yourselves and, say, one of your peers is marginal. When I go back to how we invest and where our thesis in this space has been, you want to see a hair-on-fire problem. Switching me from an efficiency gain of 5X to 6X doesn't really do the trick. I want to see a gain of 5X to 50X. That side has been a little bit harder for us to invest in just because of the competition. But on the harder ML side that you mentioned, we've actually made a few bets there and are definitely excited about that space.

[00:12:41.930] - Simba Khadder.
It's interesting. It totally tracks. I also think that part of it is also on that infrastructure side, if you're building a hardcore, a GenTech or RAD system at scale as an enterprise, the problem space there is huge. There's so much parts to it. I think for lots of the more of this efficiency gain things. ChatGPT alone already does it pretty well. Then, if you put that API into Salesforce or whatever, like you said, it seems like you get a lot of the value there. It's almost like there might even be more value creation there, at least for the time being, but that value creation is going to be eating up by, it sounds like, the incumbent.

[00:13:22.300] - Gautam
Totally. That's, I think, one thing that we're struggling with as a firm is where does the value accrue in all of this. When I think about it, there's almost three layers. There's the foundational model layer, the OpenAI layer. There's the infrastructure layer where we've made bets, including with you guys, with feature form. Then there's the application layer. Frankly, I don't have an exact answer for you right now in terms of where things end up. We're trying to make bets across all of the different areas in order to hedge ourselves in a certain way to be able to play in the entire space. But it's interesting to see, like you said, especially with certain applications, it feels like the OpenAIs and the foundation model companies will start to eat up those "efficiency gain" type of applications over time.

[00:14:10.400] - Gautam
You need to build something with a stronger moat to be able to create real differentiation over time. That being said, in five years, we could look back and there could be a fantastic efficiency gain company that comes out of it. But my bet would be that efficiency gain company has somehow built their own foundation model and has their own differentiation on that side, which has created that ability in the first place.

[00:14:32.910] - Simba Khadder.
One piece that I would love to learn more about, honestly, personally, is I know you mentioned the bio-company that you invested in, and I know that GreatPoint as a firm, I know your focus is much more on software infra, but as a firm, you all do a lot of hard tech, biotech things that are maybe not things that typical software/AI engineers maybe have as much experience with. Can you explain what's happening in those spaces? You can pick one like biotech or something. Just for those of us who are maybe unfamiliar, what's happening there? How is it affecting that world?

[00:15:09.440] - Gautam
Yeah, absolutely. I know you made the division between efficiency gain and harder ML. But in the end, everything comes down to efficiency, even on the harder ML side. I think when you draw a distinctive line between... I think the distinctive line is a little bit different than efficiency gain, which seems almost horizontal to me and the harder ML problems, which are almost vertical problems. Let me just take bio as a vertical problem. If I think about what's happening there, there are companies like Diffused Bio. There's quite a few other ones that are attacking this space from different areas, but they're all trying to do similar things.

[00:15:49.810] - Gautam
How can I increase the efficiency of the healthcare system? Whether that's in the design of drugs, whether that's in the labeling of data to help me make decisions as a biotech company or a devices company. How can I be more efficient as an organization in selecting healthcare benefits, doing things on that side? It comes down to efficiency gains in the end, but I'd say the verticalized solutions is more akin to the ML problems that you were specifically talking about.

[00:16:20.100] - Simba Khadder.
I guess another thing I would love to get your take on, because you also see more late-stage companies that were built and started scaling pre-advent of at least broadly used GPT-3, 4. Is everyone becoming an AI company? Do you have to be an AI company to compete nowadays? How do you think about that?

[00:16:42.090] - Gautam
Let me break this down into two parts. Do you have to be an AI company to sell your product is one side, and do you have to be an AI company to raise more money is the other side? Let me answer the second part first because that's probably the easy answer. The easy answer on that is it makes life a lot easier. When I look companies that are able to raise their Series B, Series C, Series D companies, they might not actually be AI companies. They might not be ML companies, but they'll try and weave in some verbiage around AI and ML and large language models in order to raise money.

[00:17:18.100] - Gautam
I would recommend doing it. That's how you pique the interest of investors. That's how you get people to look at your company and for people to get excited about it. I would say have something behind that because if someone peels back the layers, and they find that it was not quite what they had expected, that could also result in some negative things. My recommendation would be actually have something behind that if you do want to say that you're an AI and ML company.

[00:17:43.320] - Gautam
But I'd say it makes life a whole lot easier to be able to fundraise, especially at the later stages. On the other side of it is just to be able to sell your product. I think it's yes and no on that question. Yes, in the sense that every large enterprise, every potential buyer has some AI initiative that they're working on, probably all since the beginning of 2023, when everything came out, I guess, at the end of 2022 with OpenAI and ChatGPT became mainstream, every enterprise had a mandate that said, We have to integrate AI into our business. We have to integrate machine learning into the business.

[00:18:23.410] - Gautam
But I think the year has come and gone. 2023 has come and gone where people are looking at those initiatives and saying, Great, we would love to have them work if possible, and our goal is to have them work. That being said, we want ROI on our infrastructure that we buy. We want ROI on the products that we buy, the software that we buy. That, I think, comes down to, again, are you solving a real problem, solving a real hair-on-fire problem? While LLMs might be hot, and they might be the new buzzword thing, they're not necessarily the thing that...

[00:18:58.720] - Gautam
Like I said before, they might a slight efficiency gain, but they're not solving the hair-on-fire problem. We're seeing a lot of enterprises still revert to, Okay, fine, but we're going to buy... LLMs are an important piece for us, but we're still going to buy the product, the software that actually creates the efficiency gain in the end for us. It'll pique people's interest, but I think in the end, solving core problems is the real need.

[00:19:24.030] - Simba Khadder.
There's two questions that I want to go through here because a ton of stuff you just said that was super interesting. One piece I want to just maybe break down a bit. You mentioned the AI story makes raising money easier. The VCs aren't dumb. They're very smart. They spend a lot of time thinking about they have to get returns. The ones who don't, don't last very long. There's something behind that that they are resonating in to. Is the perception that the next wave of IPOs are all going to be AI companies/AI-enabled companies? Is it just that that it's going to create growth? Is it just the idea of what could happen? Why is that strategy working? Why do VCs look for that?

[00:20:07.710] - Gautam
I think it comes down to for VCs, to create a sports analogy behind it, your job is to figure out not where the puck is, but where the puck is going and skate there. I think the view of not only VCs, but I think everyone, hence the interest from enterprise, is that AI is going to change the world, whether that's today, whether that's tomorrow, whether that's five years from now. As a VC, you're trying to make the prediction on who's potentially transforming enterprise today and transforming the way that they buy and sell goods, how they do business, how they do all these different things. But really, in the end, it's what are these enterprises going to be buying in five years, and what attracts them five years down the line?

[00:20:50.660] - Gautam
AI is always going to be that next step for enterprises. I think in the past two years, AI and machine learning, for that matter, were always almost like a pipe dream in the sense that I could tell you what it was, but did I use AI and ML in my daily life? No. But now my grandma is on ChatGPT.

[00:21:14.900] - Gautam
She's on there trying to find a new recipe or a new recommendation. It's become so mainstream and top of mind, I think, with everyone, that everyone can resonate with it. Hence, why folks, especially on the VC side, are saying, Okay, now it's become mainstream, it's become popular. We think that enterprises are going to move to that direction and start being buyers of LLM software, of machine learning software.

[00:21:40.750] - Simba Khadder.
It's just a transformative enabling thing that's almost like there's going to be huge winners here. It's like you're positioning yourself.

[00:21:49.180] - Gautam
Like I was saying before, I don't know where exactly those winners are. If I had a crystal ball, I probably wouldn't be doing this job. I probably wouldn't be doing this interview. I'd be off on a beach somewhere. I don't know exactly where the value will accrue over time. Will it be on the foundational model side? Maybe. Will it be on the infrastructure side? Will it be on the application layer? I don't know the exact answer, but there are going to be winners in all three of those categories. The size of those winners will all be different in all of those categories. But the need for machine learning, the need for AI has become as apparent as it's ever been, especially when it's taken up so much mind share amongst the general population acquisition.

[00:22:30.690] - Simba Khadder.
Got you. We've talked a lot about value accrual, and where's the value in the chain going to end up? One other aspect, but at least I'm seeing and curious to how you think about it, is the growth we're seeing amongst consultants. It's almost like the people making the most money are the hyperscalers and the consultants. Why is that? Do you think that's going to change? Do you have any sense of what's going to happen there?

[00:22:58.410] - Gautam
I don't disagree with that at all. You're seeing that with a lot of our portfolio companies and companies in general. The reason to me that I think that that's happening, and frankly, why I would recommend to pretty much any one of our portfolio companies or any software company in general who's selling to businesses to go and find a systems integrator, find a consultant to cozy up to, is because those guys have a lot of the core relationships with enterprises, especially to places like the center of excellence. Call it center of excellence, call it your development team, whatever you want to call it, who is doing a lot of this work that's on the forefront of AI and ML. When you get close to these systems integrators, these consultants, you're able to have very much an easier entry path to be able to get into these enterprises rather than having to break in on your own.

[00:23:49.980] - Gautam
Obviously, the economics of some of these deals changes and varies, frankly, from SI to SI and from type of company to type of company. But in the end, I think fundamentally, a SI or a consultant, especially for something that's so far-looking, far-facing as machine learning or AI, they're the best people to go to try and break down doors for you, especially in large enterprises where it takes forever to break down those doors.

[00:24:16.910] - Simba Khadder.
Yeah, that's very true. The relationships and the connections go so far. With so much noise, like we talked about, it's like you almost lean into the "who do I trust?" How about GreatPoint, and yourself, personally, what are you looking to invest in now? What startups listening to this should reach out?

[00:24:35.050] - Gautam
Yeah, that's a great question. I know it's a little self-serving, but I would love to give a little bit about us. Frankly, when I'm looking at the infrastructure world, I think about it in two buckets. There's two buckets on the buyer side. There's the data haves and the data have-nots. The data haves are the folks like the Facebooks, Googles, Amazons, Lifts, Ubers of the world, folks who can go out hire the next machine learning engineer, the next data scientist, the next data engineer to their team to help build out pipelines, build out machine learning products, do all the things that I think everyone expects from a tech company, and usually you're only seeing from a tech company.

[00:25:18.390] - Gautam
On the other side, you have the data have-nots. These are pretty much everyone else in the Fortune 500, folks where it's a lot harder to throw manpower at the problem. Like I was saying before, you have to give them products that are able to increase the efficiency of the teams significantly. Go from a team of five people on the machine learning team to help them scale and pretend that they're a team of 50 or 100 with the power of your product.

[00:25:44.640] - Gautam
We're really looking for people in the machine learning infrastructure world, in the LLM infrastructure world, who are really creating those efficiency gains and frankly are selling to those data have-nots, the folks who are looking for help that doesn't come in the form of increased human help. It comes in the form of increased software help and efficiency gains.

[00:26:05.630] - Simba Khadder.
Got you. That's awesome. GreatPoint has been an awesome partner for us. If you do have the opportunity to work with Gautam, you should definitely take it.

[00:26:12.160] - Gautam
I appreciate that.

[00:26:12.540] - Simba Khadder.
I feel like we could keep going for so much longer. We do need to cut off. I really appreciate you coming on. It was really fun to be able to chat and be able to record some of the conversations we have all the time. So thanks again for coming on.

[00:26:28.250] - Gautam
Yeah, thanks, Simba. It was great to be on here and really appreciate you and your team putting this together. Hope you get to enjoy the rest of your birthday.

[00:26:36.340] - Simba Khadder.
Thanks, man.

‍