Deedy Das: AI’s Cyber Threat, India’s IT Jobs and Why OpenAI Is Undervalued
Corner Office Conversation
Menlo Ventures partner Deedy Das joins the podcast to dissect AI's rapid pivot from consumer apps to enterprise infrastructure, arguing that frontier labs like OpenAI and
Key takeaways
- Menlo Ventures broke its own investment rules to back Anthropic in 2023, betting on 'constitutional AI' for enterprise reliability over OpenAI's consumer-first approach.
- The Cloud Code moment flipped API adoption patterns, making the enterprise market significantly larger than the consumer market and forcing OpenAI to pivot strategies.
Main topics
- Enterprise vs. Consumer AI markets
- AI cybersecurity and autonomous agent risks
Notable quotes
Cybersecurity is probably the most concerning thing with AI today in my opinion.
Conclusion
Das provides a forward-looking framework for navigating AI's enterprise pivot, cybersecurity
Transcript preview
Speaker 2 (0:00) Cybersecurity is probably the most concerning thing with AI today in my opinion. Speaker 1 (0:09) When labs like OpenAI, Anthropic, Moonshot say that their agents were able to escape testing environments, what are the kind of risks that we're talking about in other enterprise environments? Speaker 2 (0:25) And so you can just point at a website and say, hack it. Speaker 1 (0:28) Do you think AI as a technology really has the potential to shift the global balance of power among nations? Absolutely. Speaker 2 (0:38) Right. Like I think what you make is you make capital fungible with intelligence. You couldn't just throw money at a wall and be like, give me smart people. Now you can do that. One of India's primary advantages. was human capital. The human capital advantage India had is no longer that useful. This has always been a mystery for me when people say these companies are overvalued because I think they're actually extremely undervalued and I'll explain why. Speaker 1 (1:05) If you've spent any time following AI on X or in the Silicon Valley, chances are that you've come across D.D. Das. a partner at Menlo Ventures and a former Google engineer. Didi is one among the most influential voices from the Valley decoding AI. At Menlo, he leads the 100 million anthology fund alongside Anthropic. He has a front row seat to the ideas and companies reshaping the next decade of technology. It's Thursday, the 1st of October. I'm Himanshi Lochab and you're listening to Corner Office Conversation. Speaker 1 (1:58) Hi, Didi. Welcome to the Economic Times World Leaders Forum. How's the energy been so far? Speaker 2 (2:03) It's been fantastic. Thank you so much for having me here, Imanshi. Thank Speaker 1 (2:07) you so much for joining us all the way from the US. To start with, I want to take you back to 2023, you know, when Menlo decided to back Anthropic, right? We remember Matt Murphy talked about how you really broke your own rules to participate in such a large round. And it was also just a lab then, you know, Claude was not launched commercially. What gave you the confidence back then? Speaker 2 (2:32) I think when we look back at that moment, it was really something really special about that team. that sort of stood out. It's really hard to put into words and it sort of sounds pretty vague and cliche. But on one hand, you had these people who, you know, if you remember, they'd left OpenAI to do this. So the first question you start to ask is like, okay, well, if everything at OpenAI is so fantastic, why would you leave to do this? And that was pretty compelling. Like the reason and the way they wanted to approach the problem. So one, their approach of having what they... at the time called constitutional AI and the importance they put on how specifically models should behave was compelling. Secondly, it sort of landed itself or lended itself to a world where enterprise deployments would would need, in fact, constitutional AI to run correctly on their systems in order to have any reliability in production. Something that OpenAI could do, but it just wasn't focused on. If you remember at the time, OpenAI was squarely focused on consumer. Everyone thought consumer was where the big dollars would be. for all the right reasons. But Anthropic at the time said, look, we really care about how these models behave. And we think that might apply well to enterprises. They sort of were in early discussions with Google and Amazon at the time to have distribution deals through them. If you recall, OpenAI was partnering with Microsoft. All of that combined, even though it was a really high valuation and they had no real business traction, we saw the early shoots of like, hey, let's break our rules, do something exceptional here because the market's just insanely big. And we think, you know. It wasn't one that was opening I was going after. That's why we decided to do it. Speaker 1 (4:13) So has the story changed since then? You brought about the comparison between consumer market versus an enterprise. We're seeing both labs are reporting, I mean, tremendous revenue growth. But it seems that enterprise business is sort of winning there. Everything Speaker 2 (4:27) has changed, right? So if we track the history of what really happened, Chachi PT came out. It goes viral. The primary business model on ChatGPT is subscriptions. And then at the time, and even now, they discussed running ads. Now ChatGPT has a billion users. Google went and Gemini went to compete on the consumer world. The real big shift was really the Cloud Code moment, right? What happened in Cloud Code was it flipped the API business. API businesses are usually really bad. And it's really bad because you don't call an API that many times for anything. Speaker 2 (5:02) When Cloud Code came out, the idea of having an agent that calls an LLM again and again and again and again to do one task is what drove insane amount of dollar adoption for software engineers and now a wider audience. And that's what really unlocked the enterprise play in a big, big way. And then when that started to happen, you see the dollar volume in that market just exceed any expectations that people had on... consumer. And that's when the shift changed. And OpenAI was all of a sudden put on the back foot because they were the market leader in the entirety of AI. They'd captured what people thought was the bigger pie, consumer, but then enterprise became a bigger pie. And now you see their shift in strategy too. They've cut down a bunch of divisions. They've cut down Sora. They've cut down sciences. And now they're squarely focused on chasing anthropic for enterprise because the pie is just so big. Speaker 1 (5:55) Right. But would you say that this is enough, I mean, for the kind of investments that these companies are putting in research? I mean, enterprise adoption has been slow. The revenue numbers, as compared to the investments that have been put in, there's a huge gap. Where is the next top line lift coming for these labs? Because now they're going for IPO. The markets are going to be obsessed with numbers going forward. OpenAI said that they want to do ads. How's that? This Speaker 2 (6:22) has always been a mystery for me when people say these companies are overvalued because I think they're actually extremely undervalued. And I'll explain why. It's a pretty simple take. I'm just going to use some of the reported numbers that are out there. Think of, say, an OpenAI or an Anthropic number. If you are going to be at, say, $50 to $100 billion in run rate and you have 10x'd that in the last year, say... you look at a forward projection of that 10x, you're probably going to say, okay, I'm probably not going to predict a 10x in the next year. Maybe I'm very conservative and I predict a 2x. At