Turbopuffer, a company that sells a vector database, just published a blog post arguing vector databases are finished.
The post, titled 'RIP, vector database,' went up on turbopuffer's own blog on October 1, 2026, and quickly picked up 85 points and 18 comments in online discussion. The irony is hard to miss: turbopuffer itself is a vector database, built on top of object storage, competing with players like Pinecone and Weaviate. The company's framing suggests vector search is heading toward becoming a feature bolted onto existing data infrastructure rather than a standalone product category.
If the company selling the thing is calling time on its own category, that is a signal worth more than any analyst report. Vector databases had their moment during the generative AI boom, when every app needed somewhere to store embeddings; now general-purpose databases and cloud object stores are absorbing that function. For developers, it is a reminder not to over-invest in single-purpose infrastructure before the market decides whether it needs a separate product at all.
Turbopuffer still sells a vector database, for now, so the eulogy comes with an asterisk.