SABER · AIR.TEK Sovereign
Your AI, on your premises.
An AI server installed at your site: the same automations, the same team, and nothing leaves the company.
SABER is the sovereign option of the AIR.TEK subscription. Three formats, from the desk to the server rack, each configured to your needs, bought or leased.

Hosting
Where do your tools run?
The subscription doesn’t tie you to any location. You choose, and you can switch as your needs change.
Your tools and the AI run with major hosting providers. The simplest way to start.
Hosting and AI with sovereign providers, for your tools as well as for the AI.
Everything runs on your server, at your site. No outbound connection.
Hosting for your tools and hosting for the AI are chosen separately.
AI server
An AI server: what for?
An enterprise AI server is a GPU-accelerated machine, installed at your site, that runs your AI models locally. Instead of sending your queries, and your data, to a provider, everything is computed on your premises. It is the foundation of a local, sovereign AI.
Your assistants and automations answer from the server, on your network, without calling any outside service.
RAG finds the exact passage in your up-to-date documents and cites its source, entirely in-house.
The model trains on your data directly on your hardware: your vocabulary, your formats, your rules.
Open-weight models, including Mistral, chosen for your case and optimized for your server.
The range
Three formats, one standard.
GPUs, memory, storage: every SABER is sized with you on two criteria, the number of simultaneous users and the size of the models to run. One server is enough to get started; we adjust it as your usage grows.
On a desk, no rack needed
- FormatTower
- GPUsNVIDIA, to your needs
- ForOne team, one site
In your server rack
- Format2U rack
- GPUsNVIDIA, to your needs
- ForSMEs with a server room
High capacity
- Format4U rack
- GPUsNVIDIA, to your needs
- ForHigh volumes, several sites
Comparison
SABER server or cloud GPU?
| Cloud GPU | SABER server | |
|---|---|---|
| Where it runs | In the provider’s data center | On your premises, on your network |
| Your data | It passes through the cloud | It stays in the company |
| Billing | Per GPU hour, variable | Bought outright, or a fixed lease |
| What you keep | Rented resources, nothing stays with you | The server and the model, if you buy them |
| Without Internet | Stops | Keeps running |
Guarantees
What only a server on your premises guarantees.
Data, results and interfaces stay yours, with no lock-in: you can take them with you.
Within your server’s capacity. No bill that climbs with consumption.
Your AI depends on no remote service that could be taken away from you.
The server contacts nothing outside: all traffic starts from your network.
Maintenance happens at your request, through secure remote access or on site.
Your data is entrusted to no one else.
Under the hood
For your IT team.
Three independent layers. You have nothing to manage; your IT manager will still know who owns what.
Screens, monitoring, integrations with your tools.
YoursOrchestration, AI engines, search across your documents, monitoring.
Maintained by the teamGPUs, memory, storage, network.
Yours, or leasedThe server connects to your existing systems, ERP, CRM, TMS, document management, and any tool with an API. The AI works on your data where it already lives, without migrating everything.
A detailed technical sheet is available for your IT manager.
The move
Start on the subscription. Move to SABER when it makes sense.
- 1You start on the subscription
Your automations run from the first weeks, in the cloud.
- 2Volume or sensitivity goes up
Together, we cost out the move to your own server.
- 3The same team migrates everything
Your automations move to SABER, with nothing to rebuild.
Questions
AI servers: your questions.
What is an enterprise AI server?
A GPU-accelerated server, installed on your premises, that runs your AI models locally: answers, search across your documents, adaptation to your business. Your data stays with you.
What is SABER?
AIR.TEK’s AI server, installed at your site, bought or leased. It is the sovereign option of the subscription: the same automations and the same team, with everything running on your premises.
Do I need a data center?
No. The SABER Tour sits on a desk; the 2U and 4U formats go in a server rack. One server is enough to get started.
How is it sized?
On two criteria: the number of simultaneous users and the size of the models to run. We define the configuration with you during scoping, then adjust it as usage grows.
Which models can it run?
Open-weight models, including Mistral, chosen for your case and optimized for your hardware. The architecture is not tied to any particular model.
Local server or cloud GPU?
A local server keeps your data on your premises, at a fixed cost, and works even without Internet. A cloud GPU is paid for by usage, and your data leaves the company. Many start on the subscription in the cloud, then move to SABER when volume or sensitivity justifies it.
How much does an AI server cost?
It depends on the configuration, chosen with you based on your volumes and usage, and on the option: purchase or lease. The server is quoted after the discovery call; the cost items of a sovereign AI are detailed on our dedicated cost page.