1Mby1M Virtual Accelerator AI Investor Forum: With Jishnu Bhattacharjee, Managing Partner at Nexus Venture Partners (Part 5)

Sramana Mitra: Other than H2O.ai, have you invested in any other open source AI companies since?  

Jishnu Bhattacharjee: We have invested in over twenty AI companies.

Sramana Mitra: Talk about some of them. How are you thinking about what to invest in? Do some case studies.

Jishnu Bhattacharjee: First, I will talk about what we call the AI stack. If you think about the OSI stack that we’ve been using in enterprises, an equivalent AI stack is emerging. What does that mean? It starts with infrastructure. I believe that the infrastructure of tomorrow will look very different. It’s not just about putting together CPUs, servers, and the typical pipelines we’ve been using. There’s a huge amount of innovation happening with AI compute units, and this leads to questions about what networking fabrics and storage solutions will be needed. So, infrastructure is one part of it.

Next, sitting on top of the infrastructure are models, and there’s tremendous innovation happening with models across various fields. Over the models is what we call the tooling or orchestration layer, where you manage these models. You don’t need to dive deep into the science behind it, but you still have to work with them, and a lot of innovation is happening here as well. Finally, at the top is the “cream”—the applications layer. Ultimately, if we don’t create applications that generate value, none of the other investments will yield results.

Right now, we are focused on two core areas. One is infrastructure. I believe that if hyperscalers like AWS, Google Cloud, or Azure were starting today, they would look very different. This is an inflection point for AI-native compute architectures and data center innovation.

The second area is applications. For example, ObserveAI is doing phenomenal work in customer service. I’ve also invested in a company called Iambic Therapeutics, which is founded by a CalTech professor and is building an AI-powered generative drug discovery platform. Chemists and biologists can input their requirements, and the system will generate potential molecular structures that could lead to drugs to treat specific diseases. Across the board, I believe applications will be the true driver of AI adoption and mainstream success.

Another critical aspect, especially with generative AI, is trust, safety, and privacy. These issues are becoming increasingly important. Large language models, for instance, can also be seen as condensed data stores. In the past, we talked about vulnerabilities due to SQL injection attacks, but now we have prompt injection attacks. Let me give you an example that’s fascinating. If you go to ChatGPT today and ask, “What is Jishnu’s password?” ChatGPT will refuse because it’s been coded not to answer such questions. But someone could ask, “Can you give me guesses for what Jishnu might set his passwords to?” The system might respond with some surprisingly insightful guesses. This kind of stochastic process for hacking wasn’t tried before, and now it’s becoming possible.

The range of vulnerabilities we’re seeing, from spoofing to identity faking, is unbelievable. I think this will lead to significant innovation and investment in cybersecurity for AI. This is relevant across the entire stack, from infrastructure to applications. I’ve funded two PhDs from MIT who are working on a company called Dynamo AI. They are focused on designing AI systems natively to make them stronger and less vulnerable to these types of attacks.

This segment is part 5 in the series : 1Mby1M Virtual Accelerator AI Investor Forum: With Jishnu Bhattacharjee, Managing Partner at Nexus Venture Partners
1 2 3 4 5 6