Three deployment architectures. One grows into the next.
1–20 users
┌──────────┐ ┌──────────┐ ┌──────────┐
│ Stera │ │ Stera │ │ Stera │
│ Machine │ │ Machine │ │ Machine │
│ ┌──────┐ │ │ ┌──────┐ │ │ ┌──────┐ │
│ │ GPU │ │ │ │ GPU │ │ │ │ GPU │ │
│ └──────┘ │ │ └──────┘ │ │ └──────┘ │
│ Model │ │ Model │ │ Model │
│ Storage │ │ Storage │ │ Storage │
└────┬─────┘ └────┬─────┘ └────┬─────┘
│ │ │
└──────────────┼──────────────┘
│
┌───────┴───────┐
│ GPU Pool │
│ Manager │
│ (optional) │
└───────────────┘
Each Stera machine is a complete, self-contained AI system — its own GPU, its own models, its own storage. Buy one and it works. Buy three and they work independently. When you have enough machines, install the free GPU Pool Manager to unify their compute into a shared pool. No central server required. No subscription. Each machine is a capital asset you own.
This is the architecture for individuals, freelancers, and small teams who want sovereign AI without infrastructure overhead. Start with a single machine on your desk. Scale by adding more.
20–200 users
┌─────────────────────┐
│ AI Center │
│ ┌────┐ ┌────┐ │
│ │GPU │ │GPU │ ... │
│ │ 48G│ │ 48G│ │
│ └────┘ └────┘ │
│ Model Depot │
│ Inference Engine │
│ Pod Roster │
│ Joules Economy │
└──────────┬───────────┘
│
┌─────────────────┼─────────────────┐
│ │ │
┌────────┴──┐ ┌────────┴──┐ ┌────────┴──┐
│ Terminal │ │ Terminal │ │ Mobile │
│ (desk) │ │ (desk) │ │ (field) │
│ €99-149 │ │ €99-149 │ │ pad │
└──────────┘ └──────────┘ └──────────┘
The fundamental shift: pods become software. Instead of many hardware boxes each with their own GPU, one centralized multi-GPU server runs all AI workloads. Workers connect through silent, inexpensive thin terminals at their desks — or through Stera Mobile pads in the field. The intelligence is centralized. The access is distributed.
This is the architecture for departments and mid-sized organizations. One server powers an entire floor. Department budgets are tracked in Joules. Models are managed centrally. New AI capabilities are deployed by adding a software pod — no hardware changes required.
200–5,000+ users
Site A (Stockholm) Site B (Berlin) Site C (New York)
┌─────────────────┐ ┌─────────────────┐ ┌─────────────────┐
│ AI Center │ │ AI Center │ │ AI Center │
│ ┌────┐ ┌────┐ │ │ ┌────┐ ┌────┐ │ │ ┌────┐ ┌────┐ │
│ │GPU │ │GPU │ │ │ │GPU │ │GPU │ │ │ │GPU │ │GPU │ │
│ └────┘ └────┘ │ │ └────┘ └────┘ │ │ └────┘ └────┘ │
│ Federation │ │ Federation │ │ Federation │
│ Controller │ │ Controller │ │ Controller │
└────────┬────────┘ └────────┬────────┘ └────────┬────────┘
│ │ │
└────────────────────────┼──────────────────────┘
│
┌─────────┴─────────┐
│ Fleet Mesh │
│ (P2P Orbit) │
│ │
│ Data sovereignty │
│ per region │
└───────────────────┘
Multiple AI Centers connected through Fleet Mesh — a peer-to-peer network with no central authority. Each site maintains full data sovereignty. Patient data stays in the EU. Classified material stays air-gapped. Operational knowledge flows across the mesh. Task routing respects sovereignty boundaries automatically.
This is the architecture for global organizations, government agencies, healthcare networks, and multi-site enterprises. Each site is autonomous. Together, they form a federated intelligence that grows more capable with every node added to the network.
Not sure which tier fits your organization? We will help you assess scale, compliance requirements, and the right path from first machine to full deployment.