Enterprise slowly adopting AI, in future it will want custom solutions, says MongoDB CEO

MongoDB soaring almost 13% today after a strong earnings beat yesterday. The company boosted guidance on growing confidence in its Atlas database. The firm also raising share buyback plan by $800 million. Stock's more than doubled over the past month. And joining us now exclusively is MongoDB CEO David Acharya.

Dave, good to see you again.

So strong quarter, stock up nicely. You got, you got a number of questions on the call about database competition from Snowflake and Databricks and the broader embrace of Postgres as a threat to Atlas. Outline for me how you think the technology that you have in Atlas is better positioned, particularly for this AI era.

Well, John, first of all, thank you for having me. What I would say is that people need to understand that Postgres is built on a technology called relational or SQL databases that really came out almost 50 years ago, in fact, over 50 years ago. And in that time, data was very organized and very uniform. and very predictable. Well, fast forward 50 years later, and data represents the modern world, which is very complex, very messy, changes all the time, and constantly evolves. And so trying to force fit today's world into a tabular structure just doesn't make any sense. And people struggle with doing that. They've tried to fix that problem with providing some support, but MongoDB is designed out of the box to be able to model the real world today's complex world where data, 70% of data is unstructured, where the data constantly changes and evolves and is highly distributed to support elastic scale. And we feel like we're really well positioned for this emerging, not just for today's applications, but for the emerging applications of tomorrow.

And so how are we seeing that reflected in the size of the customers that you're engaging with? In the quarter, you seem to see growth in self-service, which I tend to think of as those smaller customers, maybe A lot of AI there, but you've also been focusing strategically on selling into larger enterprise accounts, right?

Exactly. So just on AI, we have thousands of AI apps running on MongoDB today, and a lot of the small customers tend to predominantly build AI applications. Some, you know, most of them are small, but some of them are hitting escape velocity with hypergrowth. And those include customers like Cursor, one of the most popular codegen tools in the marketplace. But we also are serving the large end of the high end of the market. We have very sophisticated— in fact, 75% of the Fortune 100 and 50% of the Fortune 500 are customers today. So we have a wide range of customers across various geographies, industries, and, and, and using us for a wide variety of use cases. And so we serve both markets and it speaks to how big our market is and how popular MongoDB is.

I want to go back to something you just said, and that is emerging applications of tomorrow. What do you see those as being?

What we're seeing today is that enterprises starting slowly with the adoption of AI. A lot of them are what I call low-stakes use cases, conversational AI, chatbots, data summarization, data extraction, data classification. But what I think is going to happen, much like we saw in the cloud and much like we saw with the internet, over time, people want to build much more sophisticated custom solutions that truly transform someone's business, either to seize new opportunities, to respond to new threats, to drive more efficiency. And we know that that's coming because we've seen this movie before, and we think we're well positioned to do that. But today, the adoption's early for 2 reasons. One, there's limited experience with AI technologies. And so people are kind of getting up to speed on some of these new technologies, as well as there's also a risk of what's called hallucinations. AI systems are probabilistic, so you can't always guarantee the output of AI systems. And if you're a financial services customer or healthcare customer, You get very nervous when you don't know what those systems are going to produce. And so you want to be very careful and guarded about how you deploy these technologies.

If we stick with the topic of data for a moment here, I was interviewing Alex Wang from Scale AI at a conference down in Washington yesterday. And one of the things he said is keeping him up at night is data poisoning. Is that something that you are thinking about as well, too, in this new, I guess, era of AI and data being the future gold rush?

It's not a thing that we see. But one of the things that we've done is we acquired a company called Voyage AI, which helps people basically bridge their private data to the LLMs, right? The LLMs have already trained themselves on public internet data. What they don't have access to is the private data of a bank, a telco, a tech company, or any other company. And embedding models essentially enable you to convert your private data into essentially machine-readable numbers that capture the semantic meaning and context of that data. So it's a bridge between your private data and the LLM so you can guarantee the output and ensure that the output is of high quality. And Voyage produces the best embeddings in the industry. It's incredibly popular. And, and what it does is really improve the efficacy and quality of the outputs of these AI applications. And we think will really help people drive the use of AI broadly.