when pine cones A vector database for data scientists is coming out in 2021, which might be ahead of its time. But as use cases started to take shape last year, the company started pushing AI-powered semantic search. With the rise of LLM in the public consciousness, companies began to see more value in vector databases.
Investors apparently agree. Today, the company announced a $100 million Series B investment at a post valuation of $750 million. Pinecone CEO and founder Edo Liberty said those numbers are hard to come by in a conservative investment environment, but the company is growing fast and investors see an opportunity to grab the market leader.
“We were clearly the creators and leaders of the category. When we launched the vector database category, no one knew what the hell we were talking about. Now, of course, it’s a well-structured market with different players in the category , such as incumbents and cloud and so on, we’re clearly ahead. So it’s easy to bet on the leaders of the categories that have formed,” Liberty told TechCrunch.
That first-to-market advantage has helped them grow from a handful of customers last year to 1,500 today, a growth rate that Liberty says is more of a consumer tool than a highly technical database. The company is seeing interest from businesses of all sizes, including tech companies like Shopify, Gong and Zapier.
“It’s like consumer-based adoption of B2B deep tech. I’ve never seen anything like it. So you have to accelerate those capabilities, which is very expensive and very difficult,” he said.
Interest in LLMs has driven interest in vector databases, but that’s a different proposition, he said. While both take large amounts of data and let you search them, LLM bakes the data into the model and thus has less flexibility, while vector databases are built for semantic search but have the flexibility of a database.
“The whole knowledge management solution ended up being more flexible, more efficient, and easier to operate (using the vector database),” he said. He cited GDPR compliance as an example. If you have to delete a record, it’s fairly easy to do in the database, but it’s much harder to delete bad data from the model because of the way the data is structured.
Peter Levine, who led today’s investment in Andreessen Horowitz and will join Pinecone’s board, sees vector databases as a key part of the AI stack. “We believe that this vector database, and Pinecone in particular, has the opportunity to really be a fundamental part of the new AI data stack. So we really, really feel strongly that putting resources into the company will help it realize Edo’s ultimate vision,” Levine told TechCrunch.
What’s more, Levine argues that vector databases working with LLMs can act as a source of truth, which may reduce the hallucination problem we see in LLMs. “Well, they work together. I mean, think of the LLM as an application that sits almost on top of that database, and what the database will do is save the information and feed it into the LLM for more precise answers, for long-term storage of the results ,” he said. Liberty also sees it as the database that acts as the LLM’s long-term memory.
With a $100 million runway, the company will be hiring. The current headcount is around 100, and Liberty expects it could reach 150 or 200 by the end of the year.
Since Pinecone and players Qdrant, Zilliz and Chromahave recently been raising funds.
Today’s investment was led by Andreessen Horowitz, with participation from ICONIQ Growth and previous investors Menlo Ventures and Wing Venture Capital. The company has now raised $138 million, including a $28 million Series A last year and a $10 million seed round in 2021.