Hosting in Alsace — Macs, RTX GPUs, Jetson — multi-fibre, solar power — first machines live in November 2026

Hosted in Alsace, connected by multi-fibre, partly powered by our own roof.

A dedicated Mac, an NVIDIA GPU server or a Jetson module to run your language models locally, without your documents leaving France; or your own server in colocation. All of it in our own building, in Centre Alsace, connected by multi-fibre, each link running at 8 Gb/s, with a photovoltaic roof covering part of the daytime consumption. No subcontracting, no anonymous data centre at the other end of Europe: a machine room twenty minutes from Colmar and forty from Strasbourg, run by the same people who answer the phone. We are twenty kilometres from the German border, within easy reach of Germany, Switzerland and Luxembourg — and your data stays in France, under European law.

Reserve a machine See the machine room

8 Gb/sper fibre: the building has a multi-fibre connection, each link reaching 8 Gb/s, with a Starlink satellite link as backup.

Solarself-consumption photovoltaic roof: part of the consumption is produced on site during the day, and the grid, the UPS and the generator cover the rest.

100 % localour building, our rack, our hands: nothing is subcontracted, and neither your models nor your data leave Alsace or France.

Dedicated Mac hosting

A Mac of your own, rented by the month

We buy the machine, install it and maintain it. You have exclusive use of it and full administrator access, as if it sat under your desk — without the desk.

Mac mini M6 and M5 Pro, Mac Studio M5 Ultra to order
Root access over SSH and remote screen over VNC
Dedicated fixed IPv4 address and IPv6, ports opened on request
iOS builds, continuous integration, remote development machine
Local inference with open AI models, up to 256 GB of unified memory
NVIDIA RTX GPU servers and Jetson modules also available
Never a shared machine: one client, one physical machine
Option to buy the machine after twenty-four months

Server colocation

Your server in our building

You have the hardware but not the place — no fibre, no UPS, no cooling, and no wish to hear it running at night. You bring it to us, we connect it.

42 U rack, from a single 1 U to a full rack
Electricity — partly solar —, UPS and cooling included
Shared link on 8 Gb/s fibre, fixed IP included
Racking and hands-on work carried out by us
Remote hands: reboot, disk, cabling, part replacement
Apple, x86 or ARM hardware — it makes no difference

Compute and local AI

Your language model, on a machine of your own.

Three families of hardware, three very different uses. We tell you which one matches your workload before renting you anything — and if it turns out a public API is enough, we tell you that too.

Apple Silicon Mac

Plenty of memory, few watts

Unified memory lets you load large models without stacking up cards: up to 256 GB on a Mac Studio. Ideal for an internal assistant, a handful of concurrent users, a development machine — and for building Apple applications.

NVIDIA RTX GPU server

Throughput, when it has to keep up

When several dozen users query the model at the same time, or a model has to be fine-tuned on your own data, the GPU is still the reference. Full CUDA ecosystem: vLLM, Ollama, Transformers, image generation, transcription.

NVIDIA Jetson

Inference that runs around the clock

A Jetson module draws a few dozen watts and runs without interruption: video analysis, sensors, agents watching a stream, small specialised models. The most frugal option when throughput is not the issue.

Compute machines — hardware supplied by RB2S, ordered for you
Machine	Memory available to the model	Power envelope	€ excl. VAT / month
Jetson Orin NX	16 GB unified	≈ 25 W	on quotation
Jetson AGX Orin	64 GB unified	≈ 60 W	on quotation
RTX 5090 server	32 GB GDDR7 · 128 GB system RAM	≈ 700 W under load	on quotation
RTX PRO 6000 server	96 GB GDDR7 · 256 GB system RAM	≈ 900 W under load	on quotation
Dual-GPU RTX PRO 6000 server	192 GB combined — large models	≈ 1 600 W under load	on quotation

Your data must not leave the region? See the Alsatian AI section. Monthly rates are given within 48 h: they depend on the purchase price of the hardware at the time of order and on actual power consumption, part of which our solar production covers. Rate card to be fixed. The machines are dedicated, never shared, with root access, a fixed IP and the ports of your choice. Supplying your own card or your own GPU server? That is colocation, and the power envelope is included in the table below.

Pricing

Published, excluding VAT, with no setup fee.

Dedicated Mac hosting — machine supplied by RB2S
Machine	Configuration	Unified memory	€ excl. VAT / month
Mac mini M6	12-core CPU · 12-core GPU · 512 GB SSD	16 GB — 170 GB/s	99 €
Mac mini M6	12-core CPU · 12-core GPU · 1 TB SSD	32 GB — 170 GB/s	139 €
Mac mini M5 Pro	18-core CPU · 20-core GPU · 1 TB SSD	64 GB — 307 GB/s	239 €
Mac Studio M5 Ultra	36-core CPU · 80-core GPU · 2 TB SSD — to order	256 GB — 1.2 TB/s	649 €

For reference, equivalent US hosting providers charge 149 to 349 $ a month for comparable configurations, with the data stored across the Atlantic. Three-month initial term, then monthly with thirty days' notice.

Colocation — machine supplied by the client
Space	Height	Power included	€ excl. VAT / month
Single machine	1 U	up to 150 W	39 €
Small server	4 U	up to 400 W	79 €
Quarter rack	10 U	up to 1 000 W	159 €
Half rack	20 U	up to 2 000 W	279 €
Full rack	42 U	up to 4 000 W	479 €

Options: guaranteed bandwidth, additional IP addresses, encrypted off-site backup, remote hands beyond the included allowance, weekend on-call. Consumption above the allowance is rebilled at cost.

The machine room

Not a giant data centre. A building in Alsace that we run ourselves.

The machines are installed in "LE 4", a new commercial building owned by the group, in the heart of the Alsace vineyards. It is local infrastructure, and we say so in plain writing rather than displaying the logo of a data centre that is not ours.

Address: 6 rue du Sommerberg, 67750 Scherwiller — Bas-Rhin
Building: Built in 2022, in service since September 2023 — 976.82 m² certified, offices, workshops and storage.
Connectivity: Multi-fibre connection, 8 Gb/s per link, Starlink satellite link as backup, RJ45 pre-cabling throughout the building.
Energy: Self-consumption photovoltaic roof: part of the daytime consumption is produced on site. UPS with batteries and a backup generator for the rest.
Security: Controlled access, intruder alarm, thirteen IP cameras, perimeter fence and gate, public fire hydrant 25 m away.
Compliance: CONSUEL certificate endorsed in 2023, annual inspection of the electrical installations by a COFRAC-accredited body.
Operations: The room is closed to the public: work on your hardware is carried out by our own technicians, seven days a week, on your written request.

Frequently asked questions

What we would rather say before you sign.

Is the hosted machine shared with other clients?

No, never. Each client gets a whole physical machine. This is not only a commercial choice: Apple's macOS licence only permits virtualisation on Apple hardware, and limits it to two virtual machines per host. We do not pool capacity and we do not oversell it.

What level of availability do you guarantee?

Let us be precise: this is not a Tier III certified data centre and we do not claim otherwise. Power is secured by a UPS and a generator, and connectivity rests on a multi-fibre connection backed by a satellite link. We commit contractually to a response within four working hours and to same-day restoration of service in the event of a hardware failure, with replacement hardware available on site.

Where is my data, and who can access it?

In Alsace, on hardware dedicated to you, so in France and in the European Union. No extraterritorial law of the Cloud Act type applies. We sign a GDPR-compliant processing agreement and we never access the contents of your machines, except at your written request under the remote hands service.

Is it really solar-powered?

Partly, and we would rather put an honest figure on it than display a green leaf. The building's photovoltaic roof covers part of the daytime consumption; the rest comes from the grid, and a compute machine running day and night obviously uses far more than the roof produces. We give the actual coverage rate to any client who asks for it.

Mac, RTX or Jetson: how do I choose?

Two questions. How many people query the model at the same time, and how large a model do you need? An internal assistant for a small business, a few concurrent requests: a Mac is enough, and it draws ten times less power. A service open to the whole company, fine-tuning on your own data, image generation or transcription at volume: an RTX server. Continuous analysis of a video or sensor stream: a Jetson, which runs on a few dozen watts. We do that calculation with you before the order, and we do it again if your usage changes.

Can I run an AI model on it?

Yes, and it is one of the uses these machines are asked for. The unified memory of Apple Silicon Macs can run serious open models — up to 256 GB on a Mac Studio M5 Ultra — at a power draw nowhere near that of a GPU server. It is the simplest answer when your data must not be submitted to any third-party service. We will tell you plainly when your need really calls for an API: for very large models or heavy concurrent loads, a dedicated machine is not the right tool.

How does provisioning work?

For a dedicated Mac: we order the machine, install it and hand over access within seven to fifteen days depending on hardware availability. For colocation: you ship or bring the server, we rack it, power it and connect it, and you receive the addressing plan and the remote access credentials.

And if I want to stop?

A three-month initial term, then thirty days' notice. Your data is returned to you and the disks are securely erased, with a certificate if you need one. For colocation, you collect your hardware by appointment.

Booking

Tell us what you want to host.

A build machine, an application server, a backup replica: describe the need and we will confirm availability and the go-live date.

contact@rb2s.com Reply within 24 working hours