Object Storage

Object Storage

S3-compatible object storage for model artefacts, dataset lakes and MLOps pipelines across our GPU platform.

Overview

Hot, cool and archive tiers under a single S3 API. Used by AI Token Foundry to store token routing metadata, by Managed Slurm for dataset lake staging, and by Kubernetes Service for inference model artefact registries. Move data between tiers with lifecycle rules — no re-ingestion required.

Performance

Hot tier latency
<10 ms first byte
Hot tier throughput
Multi-GB/s per prefix
Durability
99.999999999% (11 nines)
Max object size
5 TB

Capabilities

S3 compatible — drop-in for any SDK

Works with boto3, MinIO client, and any S3 SDK. Multipart upload, presigned URLs and event streams all supported — TKS pods and Slurm jobs access it natively.

MLOps streaming for training pipelines

Used by Managed Slurm to stream training datasets from the object store to GPU nodes, and by Kubernetes Service for model artefact registries served to KServe inference endpoints.

Integrated with AI Token Foundry

Token routing logs, API usage billing data and model call distribution metrics from the Token Foundry platform are stored as S3 objects with lifecycle tiering.

Strong consistency for checkpoint writes

Read-after-write consistency across all regions — Slurm checkpoint writes from Bare Metal MI350X or H200 nodes are immediately visible to downstream jobs.

Pricing

Hot tier£0.015 / GB-month
Cool tier£0.008 / GB-month
Archive tier£0.002 / GB-month
Egress£0.06 / GB

Ideal for

AI Token Foundry token routing and billing data archive
Managed Slurm dataset lake staging for large training runs
Kubernetes KServe model artefact registry for inference endpoints
Bioinformatics Computing molecular structure and docking result storage

Ready to put Technova to work?

Talk to our team about a custom GPU cluster, managed Slurm or one of our vertical AI solutions.