Apr 19, 2025 · 27m · latent-space

The Rise and Fall of the Vector DB category: Jo Kristian Bergum (ex-Chief Scientist, Vespa)

Jo Kristian Bergum · 17m spoken
0:00 / 0:00
▶ Watch on YouTube →

gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions

Former Vespa Chief Scientist Jo Kristian Bergum analyzes why the standalone vector database category is declining and shares practical guidance on building resilient, hybrid search architectures for AI systems.

How this conversation actually went

Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. How this is scored →

The hosts as informed peer 5.3 Guest teaching 4.0 Guest disagreement 3.3 The hosts pushing back 2.5
05100:0010:0020:001:00–6:28 · The hosts as informed peer 5/10 The Emergence and Pitfalls of Dedicated Vector Databases The host highlights his early involvement in the vector DB ecosystem by mentioning he wrote the OpenAI cookbook Chroma example. Jo Kristian details his 20-year search background, clarifying that while vector DB companies might survive by pivoting into general search engines, the standalone vector DB infrastructure category itself is dead.6:28–12:03 · The hosts as informed peer 6/10 Venture Hype, Embeddings Utility, and Database Architectures The host demonstrates domain expertise by comparing vector DB funding rounds to MongoDB's entire lifetime venture raise and questions the convergence of database storage and search indexing. Jo Kristian provides technical depth, explaining why pure semantic similarity fails without web search signals like freshness and detailing how PGVector rapidly caught up to dedicated vector engines.12:03–18:35 · The hosts as informed peer 5/10 Recommender Systems, Practical RAG Architecture, and PostgresML The host asks about the exact operational sequence for building retrieval pipelines and queries Jo Kristian on running models inside databases like PostgresML. Jo Kristian firmly rejects PostgresML's approach, arguing against embedding logic in SQL and explaining the multi-stage cascade architectures used in production search and recommender systems.18:36–26:08 · The hosts as informed peer 5/10 Long Context Limits, Graph RAG, and Embedding Frontiers The host dismisses the long-context-versus-RAG narrative as engagement farming and prompts Jo Kristian on Graph RAG and vision-based embedding models. Jo Kristian nuances the long-context debate with specific token calculations and demystifies Graph RAG by emphasizing that knowledge graph extraction is the real hurdle rather than graph database storage.1:00–6:28 · Guest teaching 3/10 The Emergence and Pitfalls of Dedicated Vector Databases The host highlights his early involvement in the vector DB ecosystem by mentioning he wrote the OpenAI cookbook Chroma example. Jo Kristian details his 20-year search background, clarifying that while vector DB companies might survive by pivoting into general search engines, the standalone vector DB infrastructure category itself is dead.6:28–12:03 · Guest teaching 4/10 Venture Hype, Embeddings Utility, and Database Architectures The host demonstrates domain expertise by comparing vector DB funding rounds to MongoDB's entire lifetime venture raise and questions the convergence of database storage and search indexing. Jo Kristian provides technical depth, explaining why pure semantic similarity fails without web search signals like freshness and detailing how PGVector rapidly caught up to dedicated vector engines.12:03–18:35 · Guest teaching 5/10 Recommender Systems, Practical RAG Architecture, and PostgresML The host asks about the exact operational sequence for building retrieval pipelines and queries Jo Kristian on running models inside databases like PostgresML. Jo Kristian firmly rejects PostgresML's approach, arguing against embedding logic in SQL and explaining the multi-stage cascade architectures used in production search and recommender systems.18:36–26:08 · Guest teaching 4/10 Long Context Limits, Graph RAG, and Embedding Frontiers The host dismisses the long-context-versus-RAG narrative as engagement farming and prompts Jo Kristian on Graph RAG and vision-based embedding models. Jo Kristian nuances the long-context debate with specific token calculations and demystifies Graph RAG by emphasizing that knowledge graph extraction is the real hurdle rather than graph database storage.1:00–6:28 · Guest disagreement 3/10 The Emergence and Pitfalls of Dedicated Vector Databases The host highlights his early involvement in the vector DB ecosystem by mentioning he wrote the OpenAI cookbook Chroma example. Jo Kristian details his 20-year search background, clarifying that while vector DB companies might survive by pivoting into general search engines, the standalone vector DB infrastructure category itself is dead.6:28–12:03 · Guest disagreement 2/10 Venture Hype, Embeddings Utility, and Database Architectures The host demonstrates domain expertise by comparing vector DB funding rounds to MongoDB's entire lifetime venture raise and questions the convergence of database storage and search indexing. Jo Kristian provides technical depth, explaining why pure semantic similarity fails without web search signals like freshness and detailing how PGVector rapidly caught up to dedicated vector engines.12:03–18:35 · Guest disagreement 5/10 Recommender Systems, Practical RAG Architecture, and PostgresML The host asks about the exact operational sequence for building retrieval pipelines and queries Jo Kristian on running models inside databases like PostgresML. Jo Kristian firmly rejects PostgresML's approach, arguing against embedding logic in SQL and explaining the multi-stage cascade architectures used in production search and recommender systems.18:36–26:08 · Guest disagreement 3/10 Long Context Limits, Graph RAG, and Embedding Frontiers The host dismisses the long-context-versus-RAG narrative as engagement farming and prompts Jo Kristian on Graph RAG and vision-based embedding models. Jo Kristian nuances the long-context debate with specific token calculations and demystifies Graph RAG by emphasizing that knowledge graph extraction is the real hurdle rather than graph database storage.1:00–6:28 · The hosts pushing back 2/10 The Emergence and Pitfalls of Dedicated Vector Databases The host highlights his early involvement in the vector DB ecosystem by mentioning he wrote the OpenAI cookbook Chroma example. Jo Kristian details his 20-year search background, clarifying that while vector DB companies might survive by pivoting into general search engines, the standalone vector DB infrastructure category itself is dead.6:28–12:03 · The hosts pushing back 3/10 Venture Hype, Embeddings Utility, and Database Architectures The host demonstrates domain expertise by comparing vector DB funding rounds to MongoDB's entire lifetime venture raise and questions the convergence of database storage and search indexing. Jo Kristian provides technical depth, explaining why pure semantic similarity fails without web search signals like freshness and detailing how PGVector rapidly caught up to dedicated vector engines.12:03–18:35 · The hosts pushing back 2/10 Recommender Systems, Practical RAG Architecture, and PostgresML The host asks about the exact operational sequence for building retrieval pipelines and queries Jo Kristian on running models inside databases like PostgresML. Jo Kristian firmly rejects PostgresML's approach, arguing against embedding logic in SQL and explaining the multi-stage cascade architectures used in production search and recommender systems.18:36–26:08 · The hosts pushing back 3/10 Long Context Limits, Graph RAG, and Embedding Frontiers The host dismisses the long-context-versus-RAG narrative as engagement farming and prompts Jo Kristian on Graph RAG and vision-based embedding models. Jo Kristian nuances the long-context debate with specific token calculations and demystifies Graph RAG by emphasizing that knowledge graph extraction is the real hurdle rather than graph database storage.

speaking balance: gold is the hosts, purple is the guest (3 minute bins)

0:00 · the hosts 0% · guest 100%0:00 · the hosts 0% · guest 100%3:00 · the hosts 0% · guest 100%3:00 · the hosts 0% · guest 100%6:00 · the hosts 0% · guest 100%6:00 · the hosts 0% · guest 100%9:00 · the hosts 0% · guest 100%9:00 · the hosts 0% · guest 100%12:00 · the hosts 0% · guest 100%12:00 · the hosts 0% · guest 100%15:00 · the hosts 0% · guest 100%15:00 · the hosts 0% · guest 100%18:00 · the hosts 0% · guest 100%18:00 · the hosts 0% · guest 100%21:00 · the hosts 0% · guest 100%21:00 · the hosts 0% · guest 100%24:00 · the hosts 0% · guest 100%24:00 · the hosts 0% · guest 100%27:00 · the hosts 0% · guest 100%27:00 · the hosts 0% · guest 100%
Sharpest disagreement ▶ 17:23 Outright rejection of in-database ML inference

Jo Kristian immediately shuts down the host's question about PostgresML, declaring that executing large SQL queries to handle embeddings inside the database creates serious DX, cost, and scaling issues.

Hardest push from the hosts ▶ 20:07 Dismissing long-context hype as engagement farming

The host calls out commentators claiming long-context windows will kill RAG, labeling their arguments nonsense designed purely for social media engagement.

Biggest teaching moment ▶ 10:27 Detailing PGVector's feature leap over vector databases

Jo Kristian explains how PGVector rapidly implemented advanced capabilities like HNSW, IVFFlat, halfvec, and binary quantization, outperforming many standalone vector database startups.

The host holds their own ▶ 6:51 Benchmarking category fundraising against MongoDB

The host brings concrete market data, showing that the 230 million dollars poured into vector DB startups exceeded the entire lifetime VC raise of MongoDB, proving the category was overfunded.

the scores for every segment, with the reasoning behind each
ChapterTopicThe hosts as informed peerGuest teachingGuest disagreementThe hosts pushing backWhy
The Emergence and Pitfalls of Dedicated Vector Databases 5332 The host highlights his early involvement in the vector DB ecosystem by mentioning he wrote the OpenAI cookbook Chroma example. Jo Kristian details his 20-year search background, clarifying that while vector DB companies might survive by pivoting into general search engines, the standalone vector DB infrastructure category itself is dead.
Venture Hype, Embeddings Utility, and Database Architectures 6423 The host demonstrates domain expertise by comparing vector DB funding rounds to MongoDB's entire lifetime venture raise and questions the convergence of database storage and search indexing. Jo Kristian provides technical depth, explaining why pure semantic similarity fails without web search signals like freshness and detailing how PGVector rapidly caught up to dedicated vector engines.
Recommender Systems, Practical RAG Architecture, and PostgresML 5552 The host asks about the exact operational sequence for building retrieval pipelines and queries Jo Kristian on running models inside databases like PostgresML. Jo Kristian firmly rejects PostgresML's approach, arguing against embedding logic in SQL and explaining the multi-stage cascade architectures used in production search and recommender systems.
Long Context Limits, Graph RAG, and Embedding Frontiers 5433 The host dismisses the long-context-versus-RAG narrative as engagement farming and prompts Jo Kristian on Graph RAG and vision-based embedding models. Jo Kristian nuances the long-context debate with specific token calculations and demystifies Graph RAG by emphasizing that knowledge graph extraction is the real hurdle rather than graph database storage.

Statements from this episode (8)

Opinion
Bergum: The standalone vector database infrastructure category is dying
“I'm not saying that the companies are dying, right? I'm just saying that the separate infrastructure category is dying, right? Because you have vector search capabilities in almost any DB technology nowadays, right?”
Jo Kristian Bergum Apr 19, 2025 ▶ 4:10
Insight
Bergum: Search, not vector storage, is the natural abstraction for RAG
“I think that's a more natural abstraction for connecting AI with knowledge and all the arguments for doing rag. I think the natural concept there is search.”
Jo Kristian Bergum Apr 19, 2025 ▶ 5:25
Prediction Not checkable as stated
Bergum: Pinecone will not endure like MongoDB because it is too narrow
“So there's always like this convergence, but MongoDB kind of, it sticks, but I don't think that for Pinecoin that was originally leading that movement, it won't like stick in the same way. It's too narrow. It's too, too narrow.”
Jo Kristian Bergum Apr 19, 2025 ▶ 7:56
Opinion
Bergum: pgvector is outpacing some standalone vector DBs in search capabilities
“So actually, what you can see, PJ Vector is doing more in the capabilities of vector search than some of the real vector database players, right?”
Jo Kristian Bergum Apr 19, 2025 ▶ 11:02
Insight
Bergum: Build RAG with BM25 first, hybrid search second, re-ranking third
“I think actually that a very strong baseline is the classical BM-Five like algorithm that's been around for 30 years, right? It's keyword matching, but it offers a very useful baseline for a lot of different search use cases because it gives you that baseline,…”
Jo Kristian Bergum Apr 19, 2025 ▶ 14:55
Opinion
Bergum: In-database ML like PostgresML is the wrong architectural direction
“No, I'm not. I'm sorry. I, I'm not. I think this is, yeah, we also seen other players that tries to, you know, move a lot of the logic into the database, agentic embedding inference and whatnot. I think if the right direction is to Keep infrastructure a little…”
Jo Kristian Bergum Apr 19, 2025 ▶ 17:24
Insight
Bergum: Building knowledge graphs is the bottleneck in Graph RAG, not databases
“The core issue issue is actually to build the knowledge graph, right? The entities, the relationships. So if you say graph, graph, You know, databases or graph rag is going to kill vector rag and all that discussion. I think the first issue is to actually buil…”
Jo Kristian Bergum Apr 19, 2025 ▶ 22:19
Opinion
Bergum: Standalone embedding API startups face a difficult business model
“I think it's a difficult business model to be in, like, because you have to have an API based service and you have to do batching and you have to make up for the compute and then, you know, are people willing to pay for it? And I think maybe that's why Voyage …”
Jo Kristian Bergum Apr 19, 2025 ▶ 25:25
Made with StarZero

Turn any episode into a week of clips.

This entire site, over 200 episodes transcribed, diarized, checked and made playable, runs on the StarZero media pipeline. Drop in your own episode and the podcast clipper finds the moments worth sharing, cuts them, captions them, and reframes them for every feed.