Estimate Your Monthly Vector DB Storage Spend

Size your RAG index before the bill arrives. Enter how many embeddings you store, their dimensionality, and your provider's price per GB to see raw vector size, index overhead, and projected monthly cost.

How the calculation works

Every embedding is a list of numbers, so its raw size is simply dimensions × bytes-per-value. A 1,536-dimension float32 vector occupies 1,536 × 4 = 6,144 bytes, or about 6 KB. Multiply by your vector count to get the raw payload. The calculator then applies an index multiplier, because production vector stores rarely keep flat arrays — an HNSW graph adds neighbor lists that typically inflate footprint by 25–40%, while IVF-PQ adds far less. Metadata (source text, document IDs, filter fields) is counted separately at your stated bytes-per-vector, since RAG payloads often carry chunk text that dwarfs the vector itself.

The full formula is:

perReplica = (count × dims × bytes × indexMult) + (count × metaBytes)
totalBytes = perReplica × replicas
monthlyCost = (totalBytes / 1,073,741,824) × pricePerGB

We convert with the binary gibibyte (2³⁰ = 1,073,741,824 bytes) rather than 10⁹, because memory-resident indexes are provisioned in GiB — using decimal GB would understate RAM-backed cost by about 7%. The information gain most calculators miss is the interplay between precision and index overhead: dropping from float32 to int8 cuts raw size 4× and shrinks the absolute graph overhead proportionally, so quantization compounds. Binary embeddings push this to 32×, which is why high-volume RAG systems re-rank a binary first pass with full-precision vectors. The optional growth rate compounds the monthly footprint over 12 months so you can budget for an expanding corpus rather than today's snapshot. All math runs locally in your browser; nothing you type leaves the page.