Rumors regarding a $2B valuation for Pinecone contradict the fact that the company raised $138M in funding at a $750M valuation. This funding round included Andreessen Horowitz, ICONIQ Growth, Menlo Ventures, and Wing Venture Capital. Ash Ashutosh leads the company as CEO to expand market leadership, while Edo Liberty acts as Chief Scientist to drive AI innovations. Pinecone is a purpose-built vector database founded in 2019. The company aims to make AI knowledgeable. It serves customers like Gong and Zapier. The company operates in San Francisco, New York, and Tel Aviv. The Slab architecture uses a log-structured indexing scheme. Writes land in L0 slabs. This process guarantees immediate freshness. Background compaction prevents queries from scanning fragmented data files. Immutable data blocks allow for easy distribution across machines. The transition from legacy pod-based architecture to the current serverless model in 2025 allows Pinecone to decouple storage from compute and scale resources elastically in response to actual query volume peaks throughout the day. Does the shift to serverless change how engineers plan for long-term storage costs?
Latency and Throughput Benchmarks
Latency and throughput metrics differ between managed providers at the 1M vector scale with 1536 dimensions and cosine similarity. Pinecone Serverless delivers 18 ms p99 latency, 850 queries per second, and 0.991 recall at top-10. Weaviate Cloud is 22 ms p99 latency, 720 queries per second, and 0.987 recall at top-10. Qdrant Cloud reaches 12 ms p99 latency, 1,100 queries per second, and 0.994 recall at top-10. For pgvector using HNSW, p99 latency is 45 ms and throughput is 380 queries per second.
| Database | p99 Latency | QPS (p50) | Recall@10 | Filtered QPS |
|---|---|---|---|---|
| Pinecone Serverless | 18 ms | 850 | 0.991 | 780 |
| Weaviate Cloud | 22 ms | 720 | 0.987 | 580 |
| Qdrant Cloud | 12 ms | 1,100 | 0.994 | 1,050 |
| pgvector (HNSW) | 45 ms | 380 | 0.982 | 290 |
You must evaluate these benchmarks against your specific datasets and hardware. Weaviate is an open-source system with HNSW and product quantization that provides a hybrid search mechanism. Qdrant is a Rust-based system with a custom HNSW implementation that includes indexed payload filtering. Pinecone is a managed service with a proprietary distributed vector index. Chroma is a lightweight option for Python and JavaScript developers. It is an embedded mode in single-node deployments or a client-server mode via HTTP.
Pricing and Use Case Suitability
Costs vary based on whether a user selects a consumption-based or a resource-based model. Pinecone Serverless is $0.33 per GB of storage per month, $8.25 per 1M Read Units, and $2.00 per 1M Write Units. Weaviate Cloud is approximately $0.095 per 1M vector dimensions. For a workload of 10 million vectors with 1536 dimensions and 50 GB of metadata, Pinecone Serverless is $64 per month, while Weaviate Cloud is $85 per month. Qdrant Cloud is $130 per month for the same 10 million vectors. pgvector on an RDS instance is $200 to $250 per month for that volume. Self-hosting Qdrant on an AWS EC2 instance with an EBS volume is $160 per month, but you must add the cost of engineering time. Chroma is a lightweight option for developers who want to run a database locally in minutes. This open-source tool is good for local RAG prototyping. It is an embedded mode in single-node deployments or a client-server mode via HTTP. Since it is a Python-native design, developers find it to be the most ergonomic choice for Python-based AI applications.
