Aura Digital
Sofia · 42.69°N 23.32°E
Services · GPU hosting

A server earns nothing in its box

Buying the GPU server is half the job. The other half is power, cooling, network, monitoring and customers - and that is where most projects stall. We take it on, whether the machine is ours or yours.


01

First we work out whether it is worth it at all

A GPU server pays for itself only at a certain utilization and a certain electricity price. If the arithmetic does not work, you hear it before you buy hardware.

  • Which card for which workload - training, inference or rendering
  • Real rental rates on the market today, not forecasts from last year
  • Power, rack space and network - the actual monthly bill
  • When the money comes back, and at what occupancy
aura@roi: ~ ONLINE
$ aura roi --gpu=rtx-pro-6000
rental rate · live data
power · at site tariff
output · payback period

02

Colocation - the machine goes into a data centre

The server lands on a site with redundant power, climate control and staff on the floor. We handle the mounting, the cabling and the bring-up.

  • Rack mounting, power and network connection
  • The site is in the EU - GDPR applies in full, with no third-country transfer
  • Remote console access, so you are not driving out for every fault
  • Thermal test at full load before it goes into service
  • Works with hardware you bought elsewhere
aura@dc: ~ ONLINE
$ nvidia-smi --query-gpu=temp.gpu,power.draw
eight cards · responding
temperatures · within spec at full load
✓ machine ready for work

03

Listing and monetization

Spare capacity is offered through Vast.ai. Verification there is not a formality - disk speed, network and reliability are measured and have to pass.

  • Preparation for verification: drivers, containers, ports
  • Pricing against what the market is actually paying right now
  • Watching occupancy and correcting the price
  • Or a direct lease, off-marketplace, if you prefer a single tenant
aura@vast: ~ ONLINE
$ vastai show machines
status · verified
occupancy · checked daily
price · tracked to market

04

Monitoring and upkeep while it runs

A machine under load around the clock is not left alone. We watch it, and when something breaks it is our problem, not yours.

  • Monitoring of temperatures, power draw and reachability
  • Driver and container updates inside an agreed window
  • Component replacement on failure, through the warranty process
  • Monthly report: occupancy, revenue, downtime
aura@watch: ~ ONLINE
$ aura watch --node=gpu-01
availability · continuously checked
alerts · thermal, power, network
report · monthly

Who this is for

Not for everyone. Here is when it makes sense and when it does not.

  • You own a GPU server that sits idle, or sits in your office
  • You are considering buying one but do not want to babysit it
  • You need GPU capacity for a project without buying hardware
  • You are leaving the cloud because the bill now exceeds a machine
  • ⛔ Not for you if you need one GPU hour a month - the cloud is cheaper there

Frequently asked questions

Do you have your own data centre?

No. We work in colocation at third-party sites. The first GPU server is already live this way, with 8 RTX PRO 6000 Blackwell cards. We also do this as a contractor for NEOXIS AD, an early-stage GPU compute venture we co-founded. We say so plainly, because promising "our data centre" would not be true.

Where does the data sit, and does that satisfy GDPR?

The site is inside the European Union. GDPR applies in full and there is no transfer to a third country - the question that has followed US providers since Schrems II does not arise here. The contract is with a company registered in the EU. We do not own the data centre and do not claim otherwise; the terms of the specific site are named before anything is signed.

Should I rent or buy?

It depends on utilization. At a few hours a week, renting is cheaper almost every time. Under steady load, owning the machine pays back - usually inside two years at current rental rates. The calculator runs your case against live market prices rather than our assumptions.

Can you colocate a server I did not buy from you?

Yes. We take hardware bought elsewhere. There is an intake check - configuration, thermal test under load, condition of the cards - and you get told what we found before it goes live.

What happens if a GPU fails?

We go to the site, not you. The machine comes off the market, the component is replaced through the warranty process, and it returns to service after testing. Our own platforms carry a 3-year warranty; third-party hardware carries its manufacturer’s.

How long does it take to go live?

With hardware ready and space available, usually days rather than months. If the server is still being built, the build time applies. Vast.ai verification adds a little more, and that part is not entirely up to us.

Tell us what hardware you have, or plan to buy

You get back the arithmetic: live rental rates, real power cost and a date it can be live. If it does not pay off, you will hear that from us.

Book a call
Call Us