GPUs
Shared vs. Dedicated GPUs for Enterprise AI
Compare dedicated GPUs, MIG, vGPU, and time-slicing for enterprise AI. Learn how isolation, performance, utilization, and workload type affect GPU allocation.
Read field notePricing
Transparent by default. Custom where it counts.
Open source means no license line-items. You pay for capacity and operations: per minute on public cloud, per node on private platforms, per token for inference. No lock-in penalty when you leave.
Scoping call within one business day. Written quote within 24 hours of that call.
Billing units
Three ways to pay. Nothing hidden in any of them.
01 / Pay as you go
Pay as you go
Public Cloud
Per-minute billing, no quote needed. Create an account and deploy in minutes. No contract, no commitment, no sales call.
View rates02 / Per node, yearly
Per node, yearly
Private Cloud · Kubernetes · Ceph · GPU Infrastructure
Flat, predictable pricing per node (per GPU on hosted GPU infrastructure), hosted in our data centers or deployed on your hardware, supported or fully managed. Quoted around your footprint, starting from listed entry points.
See starting prices03 / Per token
Per token, quoted
AI Inference
Metered like public cloud: you pay for the tokens you process. Priced like a platform: the rate is quoted, because model choice, volume, and placement change the per-token cost.
How it's pricedPublic cloud · Self-serve
Public cloud. Every rate published.
Virtual machines from $7.50/month, billed per minute. Choose x86 or Arm at the same price. Scale up or down freely. You pay for the minutes you use, nothing else.
| Instance | vCPU | Mem | Disk | BW | /hr | /mo |
|---|---|---|---|---|---|---|
| v3-starter-1 | 1 | 2 GB | 10 GB | 2 TB | $0.0106 | $7.50 |
| v3-starter-2 | 2 | 4 GB | 20 GB | 4 TB | $0.0212 | $15.00 |
| v3-starter-4 | 4 | 8 GB | 40 GB | 5 TB | $0.0424 | $30.00 |
| v3-starter-8 | 8 | 16 GB | 80 GB | 7 TB | $0.0848 | $60.00 |
| v3-starter-16 | 16 | 32 GB | 160 GB | 10 TB | $0.1696 | $120.00 |
| v3-starter-32 | 32 | 64 GB | 320 GB | 13 TB | $0.3392 | $245.00 |
| v3-starter-64 | 64 | 128 GB | 640 GB | 15 TB | $0.6784 | $495.00 |
| v3-starter-96 | 96 | 192 GB | 960 GB | 15 TB | $1.0176 | $742.00 |
Storage
Object · up to 1 TB
$0.10/GB/mo
Object · up to 50 TB
$0.09/GB/mo
Object · up to 100 TB
$0.08/GB/mo
Object · 101 TB+
Contact us
Block · NVMe
$0.15/GB/mo
NVMe snapshots
$0.15/GB/mo
File sharing
$0.12/GB/mo
+ $20.00/mo
Image service
$0.10/GB/mo
Networking
Public transfer
$0.10/GB
Private transfer
Free
Public IP
$0.0028/hr
≈ $2.00/mo
Private network
$0.0014/hr
≈ $1.00/mo
Router
$0.0014/hr
≈ $1.00/mo
Floating IP
$0.0028/hr
≈ $2.00/mo
Load Balancer
Highly available + SSL
$0.03/hr
≈ $20.00/mo
All prices USD. Private network traffic between your instances is free.
AI Inference · Quoted
AI Inference. Metered by the token, quoted to your workload.
Like public cloud, inference is usage-based: you pay for the tokens you process, nothing else. Unlike public cloud, the rate is quoted rather than published, because the model you run, your expected volume, and where it runs genuinely change the per-token cost.
AI Inference
Priced per
million tokens
Open, commercial, or custom models behind a managed endpoint, hosted or in your data center.
Most deployments run tens to hundreds of millions of tokens a month; quotes hold across that range and beyond.
- Billing unitTokens, shaped to model choice, volume, and placement.
- Price driversThe model you choose · expected volume · where it runs · data-handling requirements (zero-data-retention available).
- Always includedManaged endpoint, serving path operations, room to change models without re-architecting.
Platform products · Quoted
Platform pricing. Starting points up front, quotes built around your footprint.
Node count, hardware, location, and operating model genuinely change the number for these products. So we publish where pricing starts and quote the rest. Every quote is itemized, engineer-led, and delivered within 24 hours of your scoping call. These are list prices. Quotes reflect your footprint and commitment, and most come in below list.
Private Cloud
From
$750 /node/year · 24×7 support
$62.50 /node/mo
Dedicated OpenStack cloud, hosted in our data centers or deployed on-premises.
Typical entry: 3-node HA cluster.
- Billing unitPer node, yearly, 24×7 support included.
- Price driversNode count and hardware class · our data center or yours · supported or fully managed.
- Always included24×7 engineers, zero-downtime upgrades, monitoring, no per-VM or license fees.
Kubernetes
From
$1,200 /node/year · 24×7 support
$100 /node/mo
Pure upstream Kubernetes, on our cloud, major public clouds, or your hardware.
Typical entry: 3-node cluster.
- Billing unitPer node, yearly.
- Price driversCluster size · where it runs · supported or fully managed.
- Always includedCertified Kubernetes conformance, 24×7 experts, upgrades and security advisories. No licensing fees, ever.
Ceph Storage
From
$900 /node/year · 8×5 support
$75 /node/mo
$1,500 /node/year · 24×7 support · $125 /node/mo
Block, object, and file storage from one cluster, hosted or on your hardware.
Typical entry: 3-node storage cluster.
- Billing unitPer node, yearly, at the coverage you choose: 8×5 or 24×7. No per-TB, per-socket, or per-protocol fees.
- Price driversNode count and drive configuration · hosted or on-premises · 8×5 or 24×7 coverage · supported or fully managed.
- Always includedAll three protocols (RBD, S3, CephFS), zero egress fees, 24×7 storage engineers.
GPU Infrastructure
From
$750 /node/year
$62.50 /node/mo
Dedicated GPU capacity, in our data centers or on hardware you already own.
Typical reservations run 2 to 8 GPUs; single GPUs and multi-node clusters are quoted the same way.
- Billing unitPer node, yearly, on your hardware; hosted capacity is priced per GPU, quoted at reservation. Fully dedicated, never oversold.
- Price driversGPU class and count · hosted or on-premises · supported or fully managed.
- Always includedThe full cloud platform around the hardware: provisioning, networking, storage, lifecycle.
Operating model
Two ways to run every platform.
Supported
Your team operates. Our engineers are on call 24×7 for guidance, troubleshooting, and escalation.
Fully Managed
We operate everything: deployment, upgrades, monitoring, and incident response, around the clock.
Services
Migrations and professional services start with a fixed-scope assessment, then a quoted project. No open-ended time-and-materials surprises.
Every from-price above is the supported entry point. Support coverage also factors in: Ceph publishes both tiers, 8×5 and 24×7. Fully managed is quoted at scoping.
Our approach
Why some prices are quoted
Hardware, placement, and operating model can swing the real cost of a platform by a wide margin. A single fixed price for a private cloud would either overcharge small footprints or hide the true cost of large ones, so this page shows where pricing starts and everything beyond that is quoted in writing.
01
Engineer-led.
Your scoping call is with an engineer who will run your infrastructure, not a sales script.
02
Itemized.
Every line explained. No bundle fog.
03
Fast.
Scoping call within one business day; written quote within 24 hours of that call.
True across everything we sell
Egress
$0.10/GB on public cloud. Free on private platforms and Ceph.
No license fees
100% open source stack, no proprietary line-items.
Per-minute granularity
On public cloud.
Flat per-node pricing
On platforms: no per-socket or per-TB traps.
Leave anytime
Everything is upstream open source. Your workloads stay portable, so our pricing has to stay honest.
Trusted By Engineering Teams At
FAQ
Pricing questions, answered
Per minute, pay as you go. No contracts or commitments; you pay only for accrued usage.
Next step
Know your number by Friday.
Start on public cloud in minutes, or book a scoping call: engineer on the line within one business day, written quote 24 hours after that.
Scoping call within one business day. Written quote within 24 hours of that call.
Field notes / Latest
What production is teaching us.
Engineering notes from operating open infrastructure: the failures, design decisions, and upstream work that make open infrastructure better.
Browse all field notes
OpenStack
Why Infrastructure Flexibility Matters More Than Choosing the “Right” Cloud
Why infrastructure flexibility matters and how open cloud technologies can help organizations adapt as workloads and business needs change.
Read field note
GPUs
AI Networking: What GPU Infrastructure Really Requires
Learn how AI workloads change network design across GPU clusters, storage, east-west traffic, RDMA/RoCE, Kubernetes placement, and production inference.
Read field note