Pinecone launches serverless vector database on Azure, GCP
Serverless, first launched on AWS, is now available on all three major public clouds in a move aimed at enabling customers to control costs in the environment of their choice.
Pinecone on Tuesday launched its serverless vector database on Microsoft Azure and Google Cloud in a move that enables customers to use the fully managed database on the cloud of their choice.
The vendor first unveiled Pinecone Serverless in January, at which point the platform was only available in public preview on AWS. Now, after being made generally available on AWS in May, the platform's general availability on all three major public clouds is a significant step for Pinecone in terms of expanding its reach and letting users choose their deployment environment, according to Kevin Petrie, an analyst at BARC U.S.
Pinecone Serverless is a rearchitected version of Pinecone's vector database designed to remove some of the infrastructure management costs associated with cloud computing. Serverless computing platforms automatically scale up or down based on demand, which can lead to savings with Pinecone charging customers based on consumption.
Meanwhile, by expanding the platform's availability to the three major public clouds, Pinecone is now making those potential savings available to all customers.
Any viable data platform in this space should run on all three hyperscalers' infrastructure. Most cloud adopters use more than one hyperscaler, and the more they can standardize tools across clouds, the better.
Kevin PetrieAnalyst, BARC U.S.
"This is an important step to take," Petrie said. "Any viable data platform in this space should run on all three hyperscalers' infrastructure. Most cloud adopters use more than one hyperscaler, and the more they can standardize tools across clouds, the better."
In addition to making Pinecone Serverless generally available on the three major public clouds, Pinecone unveiled new features for the platform. They include enabling users to more efficiently import large amounts of data and better protect data from system failures and accidental deletes, among others.
Based in New York City, Pinecone is a vector database specialist whose tools enable customers to store and operationalize unstructured data that can be used to train analytics models and applications, including generative AI (GenAI).
The vendor raised $100 million in April 2023 as vector search emerged as a key enabler of generative AI development. To date, the 2019 startup has raised $138 million.
Cloud expansion
Vector databases are nothing new, dating back to the early 2000s. However, their popularity has surged over the past couple of years in concert with the exploding interest in generative AI.
Enterprise generative AI applications need to be trained on proprietary data to understand the company and accurately respond to queries about its operations.
While traditional structured data provides some of that needed proprietary information, it's estimated to make up less than 20% of all data. Therefore, for a generative AI application to have a full understanding of an organization and deliver the most accurate results possible, the more than 80% of its data that is unstructured -- text, images, audio files, videos -- also needs to be part of the AI's training.
Vectors, which are numerical representations of data, are a means of giving structure to unstructured data so that it can be searched and discovered to train generative AI.
Pinecone is one vector database specialist whose tools now can be used to develop the data pipelines that train and update generative AI models. Chroma and Redis are among the other vector database specialists, while data platform vendors including AWS, Databricks, Google and Oracle also provide vector database capabilities as part of their broad offerings.
Pinecone's vector database capabilities measure up well against those of its peers, according to Stephen Catanzano, an analyst at TechTarget's Enterprise Strategy Group. With Serverless now on more than one cloud, the vendor can better compete for market share.
"[Pinecone is] very innovative and at the front of what's happening in GenAI, specifically around helping companies take their enterprise data and build new GenAI apps," Catanzano said. "They're very popular for this and for building out tools to make it simple. Being on each cloud, where a customer's data for these applications lives, is an accelerator for their business."
Pinecone Serverless is available in Starter, Standard and Enterprise versions. The Starter version -- which is free to use -- is the most basic, with only community support and topping out at 2 GB storage. The Enterprise version is the most elaborate and includes enhanced support over the Standard version.
Pinecone does not publicize pricing for its Enterprise version, but Pinecone Standard costs $0.00045 per gigabyte, per hour for storage and starts at $8.25 per 1 million read units and $2 per 1 million write units.
By starting with general availability only on AWS before expanding it to Azure and Google Cloud, Pinecone was able to use its initial launch of Serverless as a learning experience to work out any problems before making the database more broadly available, according to Jeff Zhu, the vendor's director of product management.
Major architectural overhauls risk falling short of customer expectations around quality and reliability, he noted. As a result, Pinecone attempted to ensure there was no decrease in the quality and reliability of its database before making it available on all three major public clouds.
"We focused our efforts on making a single cloud 100% production-ready, and then took those learnings to accelerate the production readiness of the remaining clouds," Zhu said.
Beyond making Serverless generally available on Azure and Google Cloud, Pinecone introduced bulk imports from object storage to simplify large-scale data ingestion and backups for data stored in Pinecone Serverless. The backups, now available to Standard and Enterprise users, include protection from system failures and accidental deletes, and the ability to restore data indexes to their previous state in the event of a bad update or delete.
In addition, new role-based access control capabilities limit who within an organization can execute certain tasks within Pinecone Serverless.
While useful, the new features don't represent significant innovation, according to Petrie.
"These features are incremental improvements," he said.
Bulk imports accelerate the data migrations needed to feed generative AI applications, while access controls help allay security concerns, Petrie noted. But there's still more Pinecone could do to enable generative AI development, such as add more embedding models to transform unstructured data into vectors.
"That process is not trivial," Petrie said.
The impetus for developing the new features, meanwhile, came largely from customer feedback, according to Zhu.
With interest in developing AI applications -- including generative AI -- surging, users of all data management platforms are experiencing new challenges in their attempts to build accurate and secure tools. Among them are efficiently moving large amounts of data and protecting data once it's in position to train an application.
"These features address some of the top challenges we've heard from our customers," Zhu said.
Future plans
With Pinecone Serverless now generally available on all three major public clouds and new features in the pipeline, Pinecone aims to expand beyond its limited focus on vector databases, according to Zhu.
Developing AI applications requires more than just a vector database, so the vendor is building features such as a GenAI-powered assistant and model ranking and inference capabilities that are designed to enable better data discovery during development.
"We're working hard to provide a composable platform for developers to rapidly build, deploy and iterate on AI by providing high-quality RAG [retrieval-augmented generation] components in a single place," Zhu said.
While providing RAG components with a vector database has benefited Pinecone, expansion could provide the vendor with growth opportunities, according to Petrie.
RAG in conjunction with vectors is only one means of feeding generative AI models. Relational databases and graph databases also enable searches and can feed RAG pipelines as generative AI evolves to include more model types and increasingly benefits from diverse data formats.
"Given this convergence of model and data types, Pinecone should branch beyond just vectors," Petrie said. Knowledge graphs and SQL queries of tabular data represent still other opportunities for diversification, he added.
Catanzano, meanwhile, said Pinecone is providing innovative vector database capabilities that compare favorably with those being developed by competing vendors. Its roadmap, which could include more diversification, should also maintain its focus on being creative to retain its position relative to other vector databases, he said.
"They are doing a great job innovating and leading," Catanzano said. "I'm not sure what may be next, but [they should concentrate on] keeping up with and exceeding competitors."
Eric Avidon is a senior news writer for TechTarget Editorial and a journalist with more than 25 years of experience. He covers analytics and data management.
