UltiHash
High-throughput, S3-compatible object storage built specifically to feed GPU/TPU pipelines for AI training, inference and RAG workloads.
What it does
UltiHash is object storage designed for AI workloads, offered either as a managed serverless service in the EU or as a self-hosted, Kubernetes-native cluster deployable on-prem or in any cloud. It targets high-throughput reads for GPU/TPU training and inference, and reduces storage costs through binary-level deduplication.
Core features
S3-compatible API for drop-in integration
High-throughput parallel reads to keep GPUs/TPUs fed
Binary-level deduplication to cut storage costs
Kubernetes-native self-hosted deployment option
Managed serverless option hosted in EU datacenters
Support for structured and unstructured data in one store (lakehouse use)
Best for
→Feeding large-scale model training pipelines without I/O bottlenecks
→Storing and versioning outputs from generative AI systems
→Serving embeddings and RAG source documents with low latency
→Unifying analytics and AI data on one storage layer
Pricing
Test license
free
Pay-as-you-go: ~
$10.40/TiB
Subscription
$7.20/TiB
Serverless Hetzner 1TB: €0.15/GB/month
Serverless AWS 1TB: €0.29/GB/month
Toolspool rankingby monthly traffic