Three platforms on one engine, plus two products around it. The platforms are not tiers and not alternatives — they are different surfaces on the same Cozystack foundation, and they combine.
Choose by who consumes the capacity
| If the capacity goes to… | You need | Platform |
|---|---|---|
| Customers who are not you | Billing, payments, customer portal, tenant suspension, resale | Public Cloud Platform |
| Your own business units, under regulation | DORA / NIS2-aligned architecture, encryption, audit logging | Private Cloud Platform |
| Inference, fine-tuning, RAG on your own GPUs | GPU tenancy, fractional GPU sharing, model serving, vector databases | AI Platform |
They combine — that is the design, not a concession
One control plane, one API, one operations team, one upgrade path underneath all three. So the honest answer to “which one, though?” is often “two of them, and that costs you a configuration change rather than a second programme.” In practice:
- AI Platform with Private Cloud controls. The usual regulated pairing. GPU workloads sit inside the same tenant boundary the auditor already reviewed. You do not build a second compliance story for the AI estate.
- Public Cloud with GPU-as-a-Service. A provider running VMs and managed databases switches on GPU tenancy and sells it on hardware it already owns: GPU usage is measured per tenant, and charging happens in the billing system it already runs (WHMCS or its own).
- Public and Private together. A telco or bank selling a sovereign cloud product while running its own regulated internal estate operates both on one platform, with one team, rather than maintaining two stacks that happen to look similar.
Nothing in the line is a dead end. Starting on the price list at provider scale and growing into a multi-region national build is a switch, not a replatform.
Ænix Public Cloud Platform
For everyone who sells cloud — hosting providers, MSPs and regional clouds at one end; telcos, national operators and banks running a commercial cloud at the other.
A complete public-cloud product for hosting providers: full billing back-end and front-end, WHMCS integration, a customer portal with your branding, payment processing, tenant lock and suspension, and service-creation wizards for VMs, Kubernetes, managed databases, S3 and GPU. Multi-region, and it runs alongside an existing VMware or OpenStack estate while you migrate.
From $1,250 per 10 nodes per month at provider scale, live in weeks once the hardware is ready; multi-region operator programmes quoted per RFP.
Ænix Private Cloud Platform
For regulated organizations running cloud for themselves — banks, insurance carriers, public administration, telco, healthcare and regulated industry.
One Kubernetes-native platform that runs alongside VMware, OpenNebula and OpenShift while you migrate. DORA- and NIS2-aligned architecture, volume encryption, configurable audit-log retention, multi-site designs, and evidence for your own ISO 27001 work. The developer self-service layer — golden paths, GitLab CI/CD, Argo CD GitOps, self-service APIs — is part of this platform rather than a separate product.
Quoted per RFP: a 14- or 28-day assessment, then a 3-12 month build depending on scope.
Ænix AI Platform
For teams running AI on their own hardware — AI-native organizations at scale, regulated AI deployments, GPU-heavy product companies, and providers selling GPU-as-a-Service.
Multi-tenant GPU scheduling, model serving, vector databases, object storage and service APIs, with air-gapped deployment available. NVIDIA data-centre GPUs through the NVIDIA GPU Operator: passthrough of whole GPUs to VMs, NVIDIA vGPU for VMs (requires your NVIDIA vGPU licence) and, inside tenant Kubernetes clusters, MIG partitions on MIG-capable cards plus time-sliced sharing via HAMi. Cozystack is accepted into the CNCF Kubernetes AI Conformance program.
Quoted per RFP: a 14- or 28-day assessment, then a 3-12 month build depending on scope.
Enterprise support for Cozystack
For teams running open-source Cozystack themselves and wanting the engineers who maintain it on call.
SLA-backed support from the maintainers on the published tiers, from $1,250 per 10 physical nodes per month. It is the same subscription as Public Cloud Platform: every tier includes the proprietary Ænix commercial modules (billing system and WHMCS integration), which a self-run team can simply leave unused. The common entry point for teams on Hetzner, OVH or leased bare metal.
Enterprise support for Cozystack →
WHMCS integration
For hosters already running WHMCS. A proprietary Ænix module that adds Cozystack services — Kubernetes clusters, managed databases, VMs, message brokers, object storage and GPU — to the panel you already operate, with usage metering and billing wired through. Included in every Public Cloud Platform tier.
The engine underneath
All of the above runs on Cozystack — the open-source cloud platform Ænix created and maintains with maintainers from other companies, and a CNCF project (Sandbox since February 2025; Incubating application in due diligence). Apache 2.0, no per-CPU or per-core fees.
That matters commercially, not just philosophically: the engine is not ours to withdraw. If the commercial relationship ends you keep running the open-source platform on the same hardware; the proprietary Ænix modules and support stop.