#b2b

Pinecone: the MongoDB of AI

Featuring

Pinecone San Mateo, CA

Pinecone builds vector databases for AI-driven search.

Listed In

#b2b

#machine-learning

Jan-Erik Asplund

Published Feb 23rd, 2023

TL;DR: ChatGPT hit 100M users in 2 months, spawning hundreds of startups and projects built on a combination of OpenAI’s APIs and vector databases like Pinecone. Now, Pinecone will have to fend off AWS and Google as they look to build a lasting, standalone AI infrastructure company. Check out our interview with Edo Liberty, the founder and CEO of Pinecone, for more.

ChatGPT was OpenAI’s iPhone moment—a consumer craze paired with an unleashing of developer creativity to build for its elegant UI (iPhone’s multi-touch screen vs. GPT’s natural language) using powerful new sources of data and context (GPS and accelerometer vs vector embeddings compressing the entire body of the internet). ChatGPT took 2 months to reach 100M global monthly active users (MAUs), making it the fastest-ever to that milestone ahead of TikTok (9 months), Instagram (30 months) and Pinterest (41 months). (link)
Just as Angry Birds stole headlines in the early years of the iPhone with 200 million downloads, app-layer successes like Jasper ($75M revenue) appear more immediately, while infrastructure layer stories—like Twilio (NYSE: TWLO) and Plaid ($250M revenue in 2021) in mobile—become visible over time through the innovation they spark. Startup apps like Uber (NYSE: UBER) and Venmo, enabled by fundamental infrastructure like Twilio, pull new infrastructure from the market that they need like Plaid, fueling further waves of new companies at the application layer. (link)
Because large language models (LLMs) like GPT operate on the core primitive of the vector embedding, the ability to store and retrieve vector embeddings is minimally needed to create apps that do more than just make API calls to GPT—with one company taking the early lead, Pinecone. Built to power semantic search and recommendations for ecommerce, Pinecone has hit an inflection point in its growth with social developer signals showing a 20x increase in usage thanks to ChatGPT and the rise of generative AI, all of which operate on vector embeddings. (link)

With usage-based pricing, Pinecone is indexed on the upside of rapidly increasing AI workloads, and with the high switching costs and high net dollar retention characteristic of database businesses, Pinecone has the potential to build a reinvestment moat where it rides the tailwinds of high sales efficiency and revenue expansion to pour investment into R&D. Snowflake (NYSE: SNOW) with its top decile 174% net dollar retention and MongoDB (NASDAQ: MDB) with its top quartile 120% exemplify the stickiness and revenue expansion of database businesses. (link)
Pinecone’s success has attracted fast following cloud giants AWS and Google Cloud executing on their playbook with a free, open source vector database clone while monetizing a managed offering bundled into their infrastructure platform. AWS and Google Cloud have existing distribution into 3.5M+ business customers that they can sell AWS OpenSearch and Google Vertex Matching Engine into. (link)
Counter-positioned against the cloud giants’ offerings as cloud agnostic vendors, companies like MongoDB, Cockroach Labs ($30M revenue in 2021), and Neo4j ($100M revenue) have built standalone companies by becoming category leading brands with the upside potential as large as what Snowflake means to the data warehouse. MongoDB alone has built an $880M/year revenue business dominating NoSQL despite free open-source alternatives and the threat of AWS, Google Cloud and Azure’s own NoSQL databases. (link)
As mobile soared, producing huge winners like Uber ($68B), Twilio grew from a telephony API to a suite of products for communication—similarly, Pinecone has $10B+ opportunity predicated on bringing AI into every app armed with proprietary data to serve specific customer needs. While currently focused on their core database product, Pinecone has future potential upside around building products complementary to that database—for search, marketplace rankings, cybersecurity, and other operations on top of vector embeddings. (link)