What Vespa.ai does
Vespa.ai is the big-data serving engine powering large-scale AI search, recommendation, and retrieval applications. The technology was built inside Yahoo over 20+ years to run search and personalization across the entire Yahoo network, then spun out in October 2023 as an independent company. Today Vespa Cloud offers the same engine as a managed service, combining vector search, lexical (BM25) text search, structured filters, and machine-learned ranking in a single low-latency serving stack.
Unlike pure vector databases, Vespa was built from day one to rank results with neural and gradient-boosted models, do hybrid retrieval, and return millisecond responses at billions of vectors. That is why companies like Spotify, Yahoo, OkCupid, Perplexity, and RavenPack's Bigdata.com run their largest production search and recommendation workloads on Vespa.
Who it's for
Vespa targets search, recommendation, and AI platform engineers at companies with very large catalogs, strict latency SLOs, and a need for hybrid ranking. Its sweet spot is teams that have outgrown a simple vector DB and need real-time ranking, filters, and personalization at scale.
Pricing
Vespa is open-source and free to self-host under the Apache 2.0 license. Vespa Cloud is offered with usage-based pricing across dev, prod, and enterprise tiers, plus dedicated managed plans.
Team & funding
Vespa.ai was incorporated in 2023 and is led by Jon Bratseth (CEO), the original chief architect of Vespa at Yahoo, alongside Frode Lundgren (CTO) and Kim O. Johansen (COO). The company is headquartered in Trondheim, Norway with a growing US presence. Vespa raised a $31M Series A in November 2023 led by Blossom Capital, with Yahoo retaining a strategic stake. Vespa reports about 60 employees and $6.3M revenue as of mid-2025.
Position vs competitors
Vespa competes with Elastic, OpenSearch, Pinecone, Weaviate, Qdrant, and Milvus. Its differentiation is depth — built-in tensor ranking, hybrid lexical-plus-vector retrieval, online learning, and decades of production hardening — at a scale most newer vector databases cannot match.