Obsidian Peaks
Pilot

Your own RTX 5090. Same machine every month.

For one workload that runs all month, on a machine nobody else uses during your agreement.

Our RTX 5090s are owned by us and run in the Stockholm region, Sweden. Public traffic exits through our server in Helsinki (EU).

What it is for

One workload, running all month.

A dedicated box pays off when it works most of the month. Three jobs it fits:

  • A production vLLM endpoint

    Your model on an OpenAI-compatible URL behind your own key, up around the clock, with no per-token bill.

  • A ComfyUI or WAN video queue

    A render queue that runs through the night, with models, workflows and outputs kept on the disk between jobs.

  • A multi-day fine-tune

    Training that runs for days without a stop or a move, with checkpoints on local NVMe.

The machines

Each box, as it is.

One card per machine you can reserve, from the live feed. Each is a whole machine; you get one.

Loading the machines

Arriving

RTX PRO 6000 Blackwell from 1 December.

For models and jobs that need more memory than an RTX 5090 has. Reserve now and we reply with a price and a start date.

Cards
4. Rent one, several or all of them.
Memory
96 GB per card
Start
From 1 December
Price
On request
Location
Stockholm region, Sweden
Reserve a PRO 6000

Included

In the monthly price.

  • The whole machine: GPU, CPU, RAM and disk
  • Electricity
  • Traffic
  • A fixed hostname and ports
  • Root access
  • Uptime monitoring
  • Reboot and terminal from the Console
  • A signed agreement
  • A named contact: Nicholas Muradov

Dedicated inference box

Your model, your OpenAI-compatible URL.

Tell us the model, any open model on Hugging Face, and we serve it on an OpenAI-compatible endpoint (vLLM) on your box, behind your own key. Root access stays yours.

  • OpenAI-compatible endpoint
  • Any open model
  • VS Code, Cursor, Claude Code

Questions

Questions and answers.

What is a dedicated box?
One whole machine with an RTX 5090, owned by us and run in the Stockholm region, Sweden, yours by the month under a signed agreement: the same machine every month, root access, reboot and terminal from the Console, uptime monitoring, and storage and traffic included.
What does it cost?
8 000 SEK/month ex VAT on a monthly term, or 7 200 SEK/month on a 12-month term. The offer is for companies. Storage, traffic and electricity are included in the monthly price.
What is the dedicated inference box?
The same machine with a model served on an OpenAI-compatible endpoint (vLLM) behind your own key: tell us the model, any open model on Hugging Face, and we set it up on your box. Root access stays yours.
Does anyone else use the machine?
No. During your agreement the machine is yours alone. Our operations key is used only at your request, to restore a service, for urgent security maintenance, to investigate an incident, or when the law requires it; every use is logged.
Is there an uptime guarantee?
We run the service with best effort and do not publish an uptime percentage yet. If a dedicated box is unavailable for more than an hour in a row for a reason on our side, and we have confirmed it, we credit that day at the daily rate, the monthly price divided by 30, against the next invoice.
How do I reserve one?
Send the reservation form, no account needed, or book a call. We agree the box and the start date with you and set it up under a signed agreement.
What happens when the agreement ends?
Access closes and the disk is kept for 14 days so you can copy your data off. Then it is securely erased and we confirm the erasure to you in writing. We keep no backups at any point, so keep your own copies while the agreement runs.
Where is the machine, and where does my data go?
The machine is ours and runs in the Stockholm region, Sweden. Your workload data stays on the box and transits our relay server in Helsinki (EU). Account, billing and support data go to our subprocessors: Stripe, Cloudflare, Resend and Google (US, under the EU Standard Contractual Clauses or the EU–US Data Privacy Framework), Hetzner (Finland) and Telegram (the operator’s support inbox).

Not a fit

When a dedicated 5090 is the wrong buy.

Do you have machines with more than one GPU?
No. Every box has one RTX 5090. If your job needs several GPUs in one machine, a provider with multi-GPU servers is the better buy.
My model needs more than 32 GB of VRAM.
Then this is not the box for it. An RTX 5090 has 32 GB; a model that does not fit in that, even quantised, needs a larger GPU.
My jobs run a few hours at a time, now and then.
Then rent by the hour somewhere else. A dedicated box pays off when it works most of the month; for bursty jobs under a day, an hourly cloud costs less.
We need SOC 2, ISO 27001 or a Tier-3 datacenter.
We have none of those. The boxes are our own hardware at our own site in the Stockholm region, not in a certified datacenter. If your customers or auditors require a certification, choose a provider that has it.

Book a call.

Tell us the workload and when you want to start. You talk to the person who runs the machines.