Sovere S4 modular AI system

AI you own.From hardware to models.

A sovereign AI data centre, installed in your building. No one outside can read it, train on it, bill it per token or switch it off.

See what it can do

Designed for
regulated work.

Whoever controls your AI
controls your business.

Most companies rent their AI. The provider holds the key, sets the price, runs the servers and makes the rules. Here is what that means for you, with the evidence.

Your own AI data centre.
The Sovere S-series.

Two modular data centres, with two or four GPUs. Sized for your workload and installed on your premises.

Sovere S2 open black aluminum cube with two NVIDIA GPUs and blue front trim

NVIDIA RTX PRO 6000 Blackwell

96 GB per GPU. Up to four isolated partitions per card.

Sovere S2

An office. A server cupboard.

A compact starting point for Code, Documents or Knowledge. Two GPUs, with installation planned around your site.

GPU memory
192 GB
People
Up to 150
GPUs
2

Complete system€142,256Includes software, setup & first-year service

Sovere S4 open black aluminum cube with four NVIDIA GPUs and blue front trim

NVIDIA RTX PRO 6000 Blackwell

96 GB per GPU. Up to four isolated partitions per card.

Sovere S4

Your existing server room.

More capacity for teams and voice workloads. Four GPUs, with installation planned around your site.

GPU memory
384 GB
People
Up to 800
GPUs
4

Complete system€201,306Includes software, setup & first-year service

Sovere learns from your code, documents and conversations. That knowledge stays on your own hardware, and only your team can use it. That is what sovereign AI means.

Connect your code, documents and recorded conversations. Sovere learns your people, your processes and the decisions behind them.

See how the software works

Your code. Your context.

A context manager for your engineers. It indexes your repositories and gives coding agents the right files and conventions for each change.

Your data and included models
Learns from
Repositories, libraries, design decisions
Stays in your building
Your source code
Open models
Qwen3.8-27B, Gemma 4 31B, GLM-4.7-Flash
Explore Code
Code
Close-up of the two NVIDIA GPUs and precision metal cooling fins inside Sovere S2
Your repositories → useful codeIllustration
Starting hardwareSovere S2
Compare benchmarks

Close to Claude Opus 4.8.
Tested on your work.

Published coding scores against Claude Opus 4.6 and 4.8. We also compare Sovere and Claude on 20 of your own tasks.

Published model scores · Higher is better

Sovere compared with Claude Opus 4.6 and Claude Opus 4.8 on public tests
BenchmarkSovere on S2Qwen3.8-27B NVFP4Claude Opus 4.6AnthropicClaude Opus 4.8Anthropic
SWE-bench ProFix real problems in large codebases61.7%53.4%69.2%
Terminal-Bench 2.1Engineering tasks in the terminal74.0%—Not published74.6%
Where it runsThe same model as above1 GPUIn your buildingCloudOn the provider’s serversCloudOn the provider’s servers
Sovere compared with Claude Opus 4.6 and Claude Opus 4.8 on public tests
Sovere on S2Qwen3.8-27B NVFP4Claude Opus 4.6AnthropicClaude Opus 4.8Anthropic
SWE-bench ProFix real problems in large codebases
61.7%53.4%69.2%
Terminal-Bench 2.1Engineering tasks in the terminal
74.0%—Not published74.6%
Where it runsThe same model as above
1 GPUIn your buildingCloudOn the provider’s serversCloudOn the provider’s servers
Sources and methodology

Sources: SWE-bench Pro published by Qwen (Qwen3.8-27B model card, full precision, August 2026). Terminal-Bench 2.1 published by NVIDIA for the NVFP4 version Sovere runs (September 2026). Claude Opus 4.6: Anthropic, 16 April 2026. Claude Opus 4.8: Anthropic, 28 May 2026. A dash means the maker published no score.

Works with the tools
you already use.

Choose your application

Pre-built models are prepared in 1–2 weeks. Custom models take up to 8 weeks.

One fixed price.
No bill per token.

Your team can use the included models without a new API charge for every prompt.

Choose a system
Sovere S2 open black aluminum cube with two NVIDIA GPUs, CPU cooling, memory, blue accents and corner docking points

Sovere S2

2 NVIDIA GPUs · 192 GB

For teams up to 150 people

First year · everything included

From€142,256

  • The hardware, in your building
  • The software and knowledge base
  • AI models set up for your company
  • 12 months of managed service

From year two€107,363a year

Per token€0for every user

Example: 150 people using AIThe first 12 months

Cloud: about €179 per person a month, inside the €134–€1,789 range reported for Uber engineers. Sovere S2: €142,256, with the first year of service included.

See what’s included

Your own sovereign AI.
Working in eight weeks.

Eight weeks from your order to a working system in your building. We agree the scope and delivery date before you order.

What happens in those 8 weeks
  1. Week 0Scope agreedQuote, tests and date fixed.
  2. Weeks 1–2PreparedHardware configured. Sources agreed.
  3. Weeks 3–6TrainedModels learn your work.
  4. Weeks 7–8LiveDelivered, installed and approved.

The usual path: MIT NANDA, The GenAI Divide, 2025, for large companies. Gartner, May 2024: 8 months from AI prototype to production.

A few things
worth knowing.

From the space you need to the people who hold the key.

  • A small, complete data centre that you own and keep in your building. Compute, power, cooling, networking and physical security are integrated. It can work without internet.

  • Two reasons. You can check where your data goes, because the models, the knowledge and the logs stay in your building. And your bill stops growing with use: included models have no per-token charges.

    How the software works
  • Yes. The price of each token keeps falling. But companies use far more AI every year, so the total bill still grows. Company spending on AI grew 3.2 times in 2025, while token prices fell.

    See the evidence
  • Ready-made systems come in two sizes: Sovere S2 and Sovere S4, each with published prices. Custom builds are designed around your own site, power and workload. Both run the same software. Custom installations, additional applications and specialist security requirements are quoted to scope.

  • Eight weeks from your order to an installed, working system in your building. We agree the scope and delivery date before you order, then configure and test the system before delivery.

  • No. Managed AI service covers keeping the AI trained, patched and measured, within the scope agreed for your system. Your first 12 months are included; annual renewal starts in year two. Your team uses it; ours looks after it.

  • We compare each Sovere model with a cloud model built for the same job, on public tests: Claude Opus 4.6 and 4.8 for code, Gemini 3 Pro for documents, OpenAI’s text-embedding-3-large for search and ElevenLabs Scribe v2 for speech. Your people also grade a set of real business tasks. We measure both models on them before and after adaptation, including the tasks we lose.

    See the scores for each model