Hey, Hemant. Good to see you. Shall we begin?
So, I love talking to people like Hemant because he's the kind of guy who texts you the night before an event and says, 'Eric, please. No easy questions.' That's not a good start. But quite seriously, Hemant is one of the most prolific and intellectually serious voices in venture capital on the subject of artificial intelligence. He has written four books about AI. He has co-founded a nonprofit around it. He bought a hospital to test it. And in many ways, he has staked his firm's identity on AI. And Hemant, if that's not authenticity, I don't know what is. You have studied AI for years and you track its progress as closely as anyone. Just to establish some context, since the beginning of the year, the big three in the United States, Google, OpenAI, and Anthropic have released seven foundation model updates, the equivalent of one a week. And with Anthropic rolling out Claude 3.5 just today, Mistral AI, the European champion, has released two new models, and of course, just in time for China's Lunar New Year, we've seen the release of DeepSeek V4 and Alibaba's Qwen 3.5. Will it reach a plateau at some point, or will the pace of evolution in AI and in particular in these foundation models keep accelerating?
Yeah, first of all, it's a really unique time as you said. And I've been saying this for a couple years. This is all about navigating ambiguity. So, we actually don't know where the scaling laws tap out. I will say that the pace of the progress of these models actually seems to be accelerating because you have some of these recursive habits where the models are training the models themselves. You're starting to see that. So I actually think there might be another knee in the curve in how fast the progress does happen. At least a lot of what we are seeing, listening to the folks building at the frontier labs, is that it is accelerating. It's pretty exciting and it's also breathtaking, you know, if you're an investor trying to think about what is going to last and what is going to sort of, you know, in the next release not be something that's worth investing in again.
Breathtaking, exciting. Is it also at least a little bit terrifying?
It's very terrifying. I mean, I think if you go all the way back to GPT-3 when the models started becoming more visceral in terms of what they can do and the art possible, and you started to see some of these thin wrapper companies emerge, they were looking very interesting because their adoption was so fast and lots of early venture capital money and applied AI went there. And then a year later it was pretty clear that those technologies weren't really going to be enough to build a business. So I think there's a version of that that continues to happen. So thinking through durability, not just revenue growth as an investor, is very, very difficult in these times. All because it's so exciting in terms of what the technology and the models are able to do for us.
Before I left for India, I'm based in New York by the way. Before I left for India, I spoke to a CEO back in the city who runs the asset management division for one of the big insurance companies and he described his state of mind as AI-induced paralysis. The models are evolving so quickly, he said, that his management team simply can't keep up with the capabilities, let alone the disruptive power they have. How is anyone supposed to keep up with this?
It's very hard. I mean, I'll actually give you a real life example of what happened a couple weeks ago. So we have obviously a lot of companies that are building on top of these models, enterprise infrastructure to accelerate diffusion of AI. And I called one of the CTOs and I said, 'Hey, I'm going to New York also, New York, and I'd love to review the product roadmap with you.' And this was, I sort of gave him a week notice. And I showed up and he said, 'Listen, I've tried to develop this a couple of times and every time I take a shot at it, it actually changes because on a daily, weekly basis, the innovation and the progress is so fast that things you thought were the problems you were solving, the models have actually solved for you.' So the closer you are to the infrastructure side, the harder it is to be able to tell because these capabilities are expanding. If you look at the enterprises, it's the same thing, which is how do you make enterprise transformation decisions when the underlying stacks are like quicksand. We talk a lot about this inside of our firm in terms of, you know, what is the foundation—great term, quicksand—that you're building these on. So it's a tricky time. So, you know, at the same time, the worry for the CEOs is that if we don't make these investments, somebody else will and three to five years from now we're going to be rendered irrelevant by our competitors that got it right. And that's where the ambiguity is and the risky part of this is, you know, as enterprises think through their adoption.
Some people in the room might be feeling some anxiety around that ambiguity. Any tips or tricks or advice for how to manage?
A bar tonight, we can have a drink together and figure it out.
Well, I think the first thing is you can't be frozen. You have to adopt it and I think you need to create a culture where your entire team is getting their hands dirty with the models. That's probably the biggest piece of advice because the organizations need to viscerally become AI native. And once that happens, the creativity will unleash and you have to start building essentially your AI teammates in the different departments so that you are getting ready for what the organizations need to look like on the other side of where these models do start to stabilize in terms of their capabilities, or at least understanding their surface area, what they're going to cover versus what needs to be built on.
I'm going to make an observation. Most of the CEOs whom I have talked to about this very issue in the United States are following your advice. They are investing, even if they can't see, even if the future, if you will, is a bit of a hazy mirage. When I was at the World Economic Forum in Switzerland a few weeks ago, I spoke to some European CEOs and on the whole they had adopted a very different posture, much more of a wait-and-see posture. That, on the basis of what you suggest, sounds like a very dangerous thing to do.
I think it's very dangerous. And in Europe, it is a bit of a mixed bag. We see some companies and CEOs that have the courage to say we need to innovate and forge ahead and there's some that are frozen. And I think some of it is because of regulatory lack of clarity as well as to what's going to happen. Some of it is because what's the European sovereign AI? This is a place where Mistral is making significant headway. I think that's a topic here as well. As you think about strategic autonomy and critical sectors, what do we bet on in terms of models for these critical industries. So there is some of that and, you know, everybody we spend time with, we encourage them, like, you can't be frozen. You got to keep moving forward. You need to be investing using these technologies or somebody else will be better positioned in your markets.
The situation we're talking about, as you may have noticed, has finally caught up with the financial markets in the form of the so-called AI fear trade. It has been a wickedly sharp rotation away from sectors perceived as the most vulnerable to autonomous AI agents. In the United States, they're calling it the SaaS apocalypse, right? The collapse of the software as a service business model. And those of you here in India may have noticed that the Nifty IT index has cratered in the past two weeks and two examples of that are Wipro and Infosys, both down about 20% in that space of time. Hemant, you understand, I think it's fair to say, the threat that artificial intelligence, particularly agentic AI, poses to these and other sectors. In your opinion, are investors being too bearish or perhaps not bearish enough?
Yeah. So, first of all, let's talk about what is happening. So I think if you go into the social sphere today and read around what people are starting to do with Claude, a lot of what's happening in the companies is they will go and essentially replace pieces of SaaS software that they're using with Claude, building them a replacement software. And that is happening in many different use cases in enterprise. Okay. And we have companies even here, one of the CEOs we were chatting with that has gone and replaced six of their core sort of packaged SaaS companies and they're saving $8 million a year because they were able to get Claude to build what was good enough for their operations. Okay, that's what's really happening. So when you see that you start to say, boy, what is really going to happen with these SaaS companies which was fundamentally a set of workflows on top of databases and these frontier models should be able to do that pretty well. So there's an over-rotation, there's a lot of fear that these companies won't really have ability to grow as the diffusion of these models accelerates. So that's really what the SaaS collapse, you know, probably said it actually, what's really going on in the market. So my personal view is I think we have over-rotated so it's going to come back but the reality is that if a lot of this gets commoditized because the models can do this for us and frankly can build workflows that are more in line with our own secret sauces and a single business versus what was built by a company that served many of us, right, so it becomes highly tuned to our own business, if that's actually going to happen. And how do you value these companies even when you sort of get over some of the fears? Is SaaS software going to be interesting enough from a multiple standpoint? Yes. Okay. That to me is a more durable issue in what's happening. And if that becomes an issue, then think about all the companies that have been built over the last, you know, decade that haven't exited that are fundamentally good companies, are growing, but their valuations are just so far ahead of where they sit that they may never be able to grow into those values. I think that's where this will really rubber meets the road. And even last night at midnight I was on with our entire investment team reviewing our own portfolio saying what are the kinds of companies that are potentially exposed to this behavior where the market just may not think these are interestingly valued enough eventually and therefore we have to be careful about how we think about creating enterprise value in those businesses.
I'm going to come to that in just a second, but I do want to, do you think that it really is at the end of the day more about profit growth and operating margins? Or in some cases, will this be akin to what we saw when the automobile replaced the horse and buggy, when the refrigerator replaced the ice delivery man, when the spreadsheet replaced the calculator? That we'll have to re-underwrite these companies with an assumption of a terminal value of zero.
