Google shrinks AI memory from 31GB to 4GB with TurboVec

Google released TurboVec, an open-source vector indexing library built on the TurboQuant algorithm, that compresses a 31GB vector dataset to roughly 4GB without sacrificing search quality. Written in Rust with Python bindings, TurboVec reduces memory usage by up to 92% by compressing high-dimensional embeddings to 2 to 4 bits per dimension. The library searches faster than FAISS, making large-scale AI applications cheaper to run and deployable on local hardware.