S3-Compatible Storage
Built for AI at Scale.
Developer-first object storage with a clean S3 API, policy-aware data services, and direct co-location with GPU clusters, starting from $0.99/TB with no egress fees.
- $0.99per TB / month
- 50GB/s peak throughput
- 11nines durability
- 15µs metadata access
Purpose-built for the data needs of modern AI teams
Architecture
Self-healing storage fabric with sub-millisecond metadata.
Every object is synchronously replicated across three availability zones with erasure coding for space efficiency and rapid rebuilds. A memory-tier index keeps metadata accessible at 15 µs regardless of dataset size.
- Native S3 signature v4, multi-part uploads, and presigned URLs
- In-flight and at-rest AES-256-GCM with customer-managed keys
- Per-bucket versioning, object-lock compliance, configurable retention
- Event bus streaming notifications to queues, webhooks, or inference endpoints
- Direct co-location with GPU clusters at 2–4 GB/s intra-cluster transfer
BHK_API_KEY before calling
# export BHK_API_KEY=bhk_sk_live_4a2e8b1c9d7f3a5b
$ bhk s3 ls s3://training-data --show-versions
→ 2.4M objects · 847 GB · latest versionHow It Works
Storage that fits your AI pipeline.
Create, upload, configure, and stream, all through the API.
Create a Bucket
One API call sets region pinning, encryption, retention policy, and lifecycle rules before the first byte lands.
Upload Your Data
Automatic multi-part above 64 MB with checksum validation. Parallel streams saturate the 50 GB/s ingest ceiling.
Configure Policies
Declarative lifecycle, residency, and access policies in YAML. Drift detection alerts whenever buckets deviate from guardrails.
Stream to GPU Clusters
Mount buckets directly to RTX 3090 nodes. Stream training data into PyTorch dataloaders at 2–4 GB/s without staging.
Durability
Protection layers, by default.
Integrity, security, lifecycle, and governance, all built in rather than bolted on.
Continuous Data Integrity
Background scrubbing validates checksums every 24 hours and rebuilds corrupted shards automatically. Audit logs capture every repair event with full lineage.
Zero-Egress Envelope
Define zero-egress buckets where data never leaves specified regions. Permit GPU clusters or analytics services through short-lived signed policies.
Intelligent Tiering
Adapts media based on real-time access patterns. Frequently-accessed data on NVMe; cold data shifts to high-density media with zero retrieval delays.
Policy Automation
Declarative policy packs enforce retention, legal hold, and residency. Drift detection alerts whenever buckets deviate from guardrails.
Versioning & Compliance
Per-bucket versioning, object-lock immutability, and configurable retention periods. Meets WORM requirements for financial services and healthcare.
Event Bus Streaming
Stream object notifications to queues, webhooks, or inference endpoints the moment new data lands. Trigger GPU jobs automatically on ingest.
API Reference
Fully S3-compatible. Plus BHK extensions.
Drop-in replacement for any AWS S3 client. Set the endpoint and go.
export BHK_API_KEY=bhk_sk_live_4a2e8b1c9d7f3a5b ·
export BHK_ACCESS_KEY=BHK4A2E8B1C9D7F3A5B6Get your keys at ai.bhkcloud.com/dashboard →
| Operation | Endpoint | Notes | Sample CLI |
|---|---|---|---|
| Create Bucket | PUT /{bucket} | Supports --data-residency for regional pinning. |
bhk s3 mb s3://media-prod --data-residency=eu-central |
| Upload Object | PUT /{bucket}/{key} | Automatic multi-part above 64 MB with checksum validation. | bhk s3 cp ./assets s3://media-prod/ --recursive |
| List Versions | GET /{bucket}?versions | Returns retention class, immutability status, and lineage metadata. | bhk s3 ls s3://media-prod --show-versions |
| Presigned URL | POST /presign | Specify intention codes to scope downstream access to AI pipelines. | bhk s3 presign s3://media-prod/dataset.tar --intention=train |
Why BHK S3
Storage that doesn't charge for moving your own data.
Most cloud providers charge $90/TB to read data you already paid to store. We don't.
| Feature | BHK S3 | AWS S3 | Wasabi | Backblaze B2 |
|---|---|---|---|---|
| Storage price / TB/mo | $0.99 | $23 | $6.99 | $6 |
| Egress fees | None (zero egress) | $90/TB | First 1 TB free/mo | First 3 GB/day free |
| S3 API compatibility | Full S3 sig v4 | Native | Full S3 sig v4 | Partial (S3-compat) |
| GPU co-location | RTX 3090 clusters · 2–4 GB/s | EC2 only · separate billing | No compute | No compute |
| Metadata latency | 15 µs · memory-tier | ~5–50 ms | ~10–50 ms | ~10–50 ms |
| Intelligent tiering | Automatic · no retrieval fees | S3-IA · pay per retrieval | No tiering | No tiering |
| Event bus / notifications | Built-in · queues, webhooks | SNS/SQS (separate) | None | None |
Pricing sourced from public rates as of May 2026. Actual costs vary by region and usage. AWS egress based on internet transfer out.
FAQ
Object storage, answered.
Everything you need to know before uploading your first dataset.
Is BHK S3 fully compatible with the AWS S3 API?
Yes. BHK S3 implements AWS S3 Signature Version 4 authentication and the full core S3 API surface: buckets, objects, multi-part uploads, presigned URLs, versioning, and object lock. Any AWS SDK (Python boto3, AWS CLI, Go, Node.js) works by setting --endpoint-url https://s3.bhkcloud.com. We also provide BHK-specific extensions for data residency pinning and intention-scoped presigned URLs.
What does "zero egress" actually mean?
We do not charge for data transferred out of BHK S3 to GPU nodes, the internet, or other services. There are no egress fees, no free-tier limits, and no "first TB free" caveats. You pay $0.99/TB/month for what you store. Bandwidth in and out is included at no additional charge.
How fast can I stream data to GPU training jobs?
GPU nodes and BHK S3 are co-located on the same internal network. Direct intra-cluster reads achieve 2–4 GB/s per node, and parallel multi-stream uploads saturate the 50 GB/s ingest ceiling. For PyTorch, set the endpoint in your dataset loader and stream shards directly with no local staging required.
How does intelligent tiering work?
BHK S3 monitors access patterns at the object level and transparently moves data between NVMe (hot), high-density HDD (warm), and archival tiers. Unlike AWS S3 Intelligent-Tiering, there are no per-object monitoring fees and no retrieval charges when cold data is accessed. The tier transition is fully automatic based on a configurable inactivity threshold (default: 30 days).
Can I migrate from AWS S3 without changing my code?
In most cases, yes. Change your SDK's endpoint to https://s3.bhkcloud.com and update your credentials. For migrations, we provide bhk s3 sync which streams objects directly from AWS without local staging, preserving metadata, ACLs, and versioning history. Contact our team for migrations above 100 TB as we offer complimentary architecture reviews.
What compliance standards does BHK S3 meet?
BHK S3 supports WORM (Write Once Read Many) object lock, configurable retention periods, and legal hold for FINRA, SEC 17a-4, and healthcare data requirements. AES-256-GCM encryption is applied at rest and in transit. For enterprise compliance reviews, SLA guarantees, or data processing agreements, contact our team.
Need a storage architecture review?
Our specialists benchmark workloads, recommend tiering policies, and script migrations. Complimentary 60-minute readiness session.