AI startup Pinecone raises $100 million as vector database market for LLMs heats up

Pine cone, the vibrant vector database company in New York City that provides long-term memory for large language models (LLMs) such as OpenAI’s GPT-4, today announced it has raised $100 million in Series B funding at a $750 million valuation. The financing round was led by Andrews Horowitz.

Pinecone introduced the vector database in 2021, a managed service that enables engineers to build fast and scalable applications using AI model embeddings and get them into production quickly. In today’s generative AI era, Pinecone helps engineers connect chatbots to their own company data to get the right answer, not hallucinate.

The emergence of ChatGPT last fall saw Pinecone boom, with the tool quickly becoming an integral part of the software stack – the memory layer – for AI applications. The company said it has seen an explosion in paying customers — inclusive — so far in 2023 Gong And Zapier — in all sectors and sizes. The vector database category has grown to include other tools such as chroma, Weviaten And Milvus.

>>Follow VentureBeat’s ongoing generative AI coverage<

Pinecone kicked off the explosive shift to generative AI

While the company was founded with the rise of LLMs in mind, the speed and explosiveness of the generative AI shift came as a surprise, Edo Liberty, founder and CEO of Pinecone (and former director of research and head of Amazon AI Labs), told VentureBeat in a Zoom interview.

“It sort of broke a collective psyche,” he said. “It grew gradually, but then it went overnight.” When ChatGPT launched, he explained, “Millions of developers around the world got excited and super creative about the kinds of things you can do with this – they started building great applications.”

He also pointed out that generative AI suddenly became a boardroom-level discussion. “It doesn’t matter if you’re an architect or a law firm or a consultancy, this is going to undermine or amplify potential and you have to figure out what to do with it,” he said. “I don’t think there’s a single company I speak to that doesn’t have something going on with regard to language and AI.”

And interest in Pinecone continues to grow among developers, who continue to explore how LLMs can be used. For example, over the past two months, the AI ​​community has been abuzz with the long-term potential of AI autonomous agents, with tools popping up including Auto-GPT and BabyAGI. “Both projects use Pinecone,” said Liberty. “Again, that drove tremendous growth; I think at one point we were getting 10,000 registrations a day.”

The long-term outlook for vector databases

Coincidentally, there was a lot of chatter on Twitter this week about a new research paper on the potential of a new architecture, the Recurrent Memory Transformer (RMT), that will allow LLMs to store information on up to 2 million tokens. Some said RMT could reduce the need for vector databases, but others said it wouldn’t because the RMT requires a much longer inference time.

But Greg Kogan, VP of marketing at Pinecone, told VentureBeat earlier this week that while the company had no comment on the specific paper, “there’s a big gap between something that works in the lab and something that works for large-scale, real-world applications.” where cost, performance, ease of use and technical overhead are important factors, that is the gap we want to bridge.” He added that chatbots are a breakthrough technology that Pinecone leaned on and “found a way to enable large-scale applications in the real world.”