Pinecone: New vector database architecture a ‘breakthrough’ to curb AI hallucinations

The vector database market was hot in 2023, since these databases help provide context and long-term memory to large language models and boost the efficiency and accuracy of RAG techniques — all in the name of reducing AI hallucinations. And no vector database company was hotter than New York City-based startup Pinecone, which raised $100 million last April and led the way in a competitive landscape.

Now, Pinecone has announced what it calls a ‘revolutionary’ serverless vector database architecture that lets companies build AI applications that are even more knowledgeable and cost-efficient. A press release claimed Pinecone serverless will deliver up to 50x cost reductions and ‘eliminate infrastructure hassles, allowing companies to bring remarkably better gen AI applications to market faster.’

The company noted key innovations including separation of reads, writes and storage, which reduces workload costs; an industry-first architecture with vector clustering on top of blob storage to provide low-latency, low-cost, fresh vector search over nearly unlimited data sizes; indexing and retrieval algorithms built from scratch; and a multi-tenant compute layer for on-demand retrieval for thousands of users.

New serverless architecture is ‘significant’ for the industry

Pinecone CEO Edo Liberty says he believes the new serverless architecture is “significant” for the industry. “I’m not saying this lightly,” he told VentureBeat in an interview. “We’ve been working on it very hard for a year and a half now — this has been our most ambitious project.”

That project’s mission, he pointed out, isn’t just to build the best vector database. “Our mission is to really enable a whole new generation of applications and capabilities in generative AI that just not possible before that,” he said, adding that he is “certain” that Pinecone can make significant progress on reducing the hallucinations that, so far, have kept large enterprises from being able to offer customer-facing gen AI applications.

Companies like Notion, Blackstone, Canva, Domo and Gong have already been working with Pinecone serverless. Liberty said that the new product now has the ‘heavy machinery’ behind the scenes that makes it easy and cheap enough for that level of customer — that has to index billions of vectors from tens of thousands, or hundreds of thousands of users, and provide RAG and knowledge over that content at scale.

“Not only can they do it now, but it’s actually much easier than ever before and it costs 10 to 100x less than it used to with any other system,” said Liberty.

A sign the generative AI tech stack is maturing

Overall, Pinecone serverless is a sign that the generative AI ecosystem and tech stack is maturing, said Liberty. The product launch includes integrations with other top AI companies in the stack, including Anthropic, Anyscale, Cohere, Confluent, Langchain, Pulumi and Vercel.

“These are the other players in leading solutions in their respective areas,” he explained. “The fact that we as companies and as CEOs go out there and say, hey, the stack is maturing, you can go build amazing products with it and they will work better together, is also another layer of the new wave of products coming up.”