Dedicated AI servers (NVIDIA GPU powered)

A GPU that belongs
to you alone.

Dedicated NVIDIA GPU servers on HPE DL380 Gen12, run in our data centre in Hanover. No shared graphics cards and no investment in hardware and air conditioning — power and cooling are part of the monthly price.

ISO 27001 based on BSI IT-GrundschutzData centres in Germany24/7/365 operation and support
Siegel: ISO 27001 auf Basis von BSI IT-Grundschutz

Your GPU, run sovereignly in Germany

ISO 27001 based on BSI IT-Grundschutz

BSI-IGZ-0777-2026 Data centres and cloud services valid until 17.09.2029

Dedicated GPU · no shared cardsPower and cooling included in the monthly priceCertified to ISO 27001 on the basis of BSI IT-GrundschutzLocation DE01 Hanover

Why dedicated AI servers from kyberio

The GPU is yours alone

The graphics card works only for your project. No neighbour taking computing time away from you — run times stay comparable from one run to the next.

Sovereign operation in Germany

The servers are located in our own data centres in Germany, certified to ISO 27001 based on BSI IT-Grundschutz. kyberio is subject to German law, not to the US CLOUD Act.

Enterprise hardware from HPE and NVIDIA

The servers are HPE DL380 Gen12 machines with two Intel Xeon 6 CPUs, DDR5 memory and disks as SSD or NVMe. The GPU is an NVIDIA Blackwell card in the Server Edition, passively cooled. For very large workloads the B300 is added, either on its own or as a node with eight GPUs.

Which GPU suits your workload

The difference is the GPU memory. If your model fits in, it runs briskly; if it does not, a faster chip will not help either. All three are in the configurator, where you work out your own configuration.

Entry level

RTX PRO 4500 Blackwell · 32 GB

Available

Enough as long as your model fits into 32 GB of GPU memory: inference, prototypes, fine-tuning smaller models. The server comes with 64 GB of DDR5, and there are steps up to 2,048 GB.

32 GB GPU memory2 × 16 cores, 3.2 GHz64 GB DDR5 includedHPE DL380 Gen12

from €1,064.48 / month net on a 36-month term · €1,182.75 on 12 months

Work it out in the configurator

Larger models

RTX PRO 6000 Blackwell · 96 GB

Available

Three times the GPU memory. For models that no longer run in one piece in 32 GB, for longer training runs and for several services on one card. Here 128 GB of DDR5 are included.

96 GB GPU memory2 × 24 cores, 3.0 GHz128 GB DDR5 includedHPE DL380 Gen12

from €1,797.81 / month net on a 36-month term · €1,997.57 on 12 months

Work it out in the configurator

Highest performance

NVIDIA B300

Available

A single B300 or a node with eight of them. Billing is monthly; the price falls with the term, and a node can be written on up to 60 months. Memory and disks belong to the node and are not chosen separately.

Single GPU or node with 8 GPUsTerm of 12 to 60 monthsTraining across multiple GPUs

from €3,650.00 / month net per GPU in the node on a 60-month term · single from €5,657.50

Work it out in the configurator

Prices net. Setup is charged once on top: €999 on a 12-month term, €499 on 24 months, €249 from 36 months. For a B300 node it is €3,999, €1,999 and €999.

The three options

From the 32 GB card to the B300 node

Prices net, per server and month. The term discount applies to the platform price: 5 % from 24 months, 10 % from 36 months.

RTX PRO 4500 Blackwell RTX PRO 6000 Blackwell NVIDIA B300
GPU 1 × RTX PRO 4500 Blackwell Server Edition, passively cooled 1 × RTX PRO 6000 Blackwell Server Edition, passively cooled Single GPU or node with 8 GPUs
GPU memory 32 GB 96 GB –
Platform HPE DL380 Gen12 HPE DL380 Gen12 –
CPU 2 × Intel Xeon 6 6517P, 16 cores each, 3.2 GHz 2 × Intel Xeon 6 6527P, 24 cores each, 3.0 GHz –
Memory 64 GB DDR5 included, in steps up to 2,048 GB 128 GB DDR5 included, in steps up to 2,048 GB part of the node, not chosen separately
Operating system disks 2 × 300 GB SAS in RAID 1 2 × 300 GB SAS in RAID 1 –
Data disks SSD or NVMe, raw capacity freely chosen SSD or NVMe, raw capacity freely chosen not chosen separately
Network 2 × 10 GbE SFP+ · 1 Gbit/s uplink and 1 public IP included 2 × 10 GbE SFP+ · 1 Gbit/s uplink and 1 public IP included 1 Gbit/s uplink and 1 public IP included
Term 12, 24 or 36 months 12, 24 or 36 months 12 to 36 months, as a node also 48 and 60
Location DE01 Hanover DE01 Hanover DE01 Hanover
Suited to inference, prototypes, fine-tuning smaller models larger models, longer training runs, several services in parallel very large models and training across several GPUs
Price per month €1,182.75 on 12 months · €1,064.48 on 36 months €1,997.57 on 12 months · €1,797.81 on 36 months €6,935.00 single on 12 months · €3,650.00 per GPU in the node on 60 months

RTX PRO 4500 Blackwell

GPU
1 × RTX PRO 4500 Blackwell Server Edition, passively cooled
GPU memory
32 GB
Platform
HPE DL380 Gen12
CPU
2 × Intel Xeon 6 6517P, 16 cores each, 3.2 GHz
Memory
64 GB DDR5 included, in steps up to 2,048 GB
Operating system disks
2 × 300 GB SAS in RAID 1
Data disks
SSD or NVMe, raw capacity freely chosen
Network
2 × 10 GbE SFP+ · 1 Gbit/s uplink and 1 public IP included
Term
12, 24 or 36 months
Location
DE01 Hanover
Suited to
inference, prototypes, fine-tuning smaller models
Price per month
€1,182.75 on 12 months · €1,064.48 on 36 months

RTX PRO 6000 Blackwell

GPU
1 × RTX PRO 6000 Blackwell Server Edition, passively cooled
GPU memory
96 GB
Platform
HPE DL380 Gen12
CPU
2 × Intel Xeon 6 6527P, 24 cores each, 3.0 GHz
Memory
128 GB DDR5 included, in steps up to 2,048 GB
Operating system disks
2 × 300 GB SAS in RAID 1
Data disks
SSD or NVMe, raw capacity freely chosen
Network
2 × 10 GbE SFP+ · 1 Gbit/s uplink and 1 public IP included
Term
12, 24 or 36 months
Location
DE01 Hanover
Suited to
larger models, longer training runs, several services in parallel
Price per month
€1,997.57 on 12 months · €1,797.81 on 36 months

NVIDIA B300

GPU
Single GPU or node with 8 GPUs
Memory
part of the node, not chosen separately
Data disks
not chosen separately
Network
1 Gbit/s uplink and 1 public IP included
Term
12 to 36 months, as a node also 48 and 60
Location
DE01 Hanover
Suited to
very large models and training across several GPUs
Price per month
€6,935.00 single on 12 months · €3,650.00 per GPU in the node on 60 months

Work out your AI server

In the configurator you choose the GPU, memory, disks and operation — the monthly price is there straight away, and the bill of materials at the end.

The builder

What you put together yourself

The configurator asks four things: the GPU model, the resources, software and network, and finally the term. Every step shows what it costs.

For memory, the smallest size is already part of the platform price: 64 GB with the RTX PRO 4500, 128 GB with the RTX PRO 6000. Above that it goes up in steps to 2,048 GB of DDR5. For data you choose the class, SSD or NVMe, and the raw capacity; we fit the largest disks we can. The operating system sits separately on two disks in RAID 1 and is part of the server.

We set up the software stack with the managed service: Ubuntu Server 24.04 LTS with NVIDIA driver and CUDA, and on request Docker with the NVIDIA Container Toolkit or a pre-installed AI platform with a web interface. Without the managed service you get the server bare metal, without an operating system, and install it yourself.

On the network side, a 1 Gbit/s uplink and one public IP are the standard. 10 Gbit/s and further IP addresses can be added.

The four steps

  • GPU model: RTX PRO 4500, RTX PRO 6000 or B300
  • Resources: memory, data disks, number of servers
  • Software and network: managed service, uplink, public IPs
  • Term: 12, 24 or 36 months

Prices (extract)

What the build-out costs

All prices net, per server and month. For memory, the smallest size is part of the platform price; what is listed here are the surcharges above it.

Memory RTX PRO 4500 · 64 GB included

128 GB DDR5 RDIMM, 64 GB on top
€138.22 / month
256 GB DDR5 RDIMM, 192 GB on top
€341.78 / month
512 GB DDR5 RDIMM, 448 GB on top
€748.90 / month
1,024 GB DDR5 RDIMM, 960 GB on top
€1,563.14 / month
2,048 GB DDR5 RDIMM, 1,984 GB on top
€3,191.62 / month

Memory RTX PRO 6000 · 128 GB included

256 GB DDR5 RDIMM, 128 GB on top
€203.56 / month
512 GB DDR5 RDIMM, 384 GB on top
€610.68 / month
1,024 GB DDR5 RDIMM, 896 GB on top
€1,424.92 / month
2,048 GB DDR5 RDIMM, 1,920 GB on top
€3,053.40 / month

Data disks per server

SSD 1.92 TB SAS/SATA, 2.5 inch, hot-plug
€139.00 / month
SSD 3.84 TB SAS/SATA, 2.5 inch, hot-plug
€259.00 / month
SSD 7.68 TB SAS/SATA, 2.5 inch, hot-plug
€439.00 / month
SSD 15.36 TB SAS/SATA, 2.5 inch, hot-plug
€579.00 / month
NVMe 960 GB 2.5 inch, hot-plug
€44.16 / month
NVMe 1.92 TB 2.5 inch, hot-plug
€97.60 / month
NVMe 3.84 TB 2.5 inch, hot-plug
€153.88 / month
NVMe 7.68 TB 2.5 inch, hot-plug
€267.92 / month

Operation and network

Managed service patches for the operating system and NVIDIA driver, monitoring, response to incidents, pre-installation of the software stack
€339.00 / month
Uplink 10 Gbit/s instead of the included 1 Gbit/s uplink
€129.00 / month €999 setup
Public IP IPv4/IPv6, one address is included
€2.00 / month

Source: cloud configurator, Sovereign AI Server builder, as at 14.11.2026. The term discount applies to the platform price, not to memory, disks, software and network. Special equipment goes into the bill of materials as a note and is quoted on request.

HPE Private Cloud AI Enterprise AI cluster

AI at scale, on request

For companies that want to bring AI into production at scale, we offer the HPE Private Cloud AI platform. It combines NVIDIA AI Computing by HPE with AI-optimised systems — from compact turnkey kits to large cluster architectures.

Part of it is the HPE stack: Private Cloud Control Plane, Kubernetes management, observability, governance and validated frameworks and models. That shortens the way from pilot project to production, and data sovereignty along with control over security and compliance stays with you.

In practice, this means: runs take the same time whenever they start, and the monthly costs are fixed in advance — power and cooling included.

Available on request. We size the cluster with you, based on your workload.

Frequently asked questions about dedicated AI servers

Which of the three options suits my workload?

That is decided by the memory on the card. The RTX PRO 4500 Blackwell with 32 GB carries inference and the fine-tuning of smaller models. The RTX PRO 6000 Blackwell with 96 GB holds models that do not fit into 32 GB, and works faster with larger batches. For very large models and for training across several GPUs there is the B300, on its own or as a node with eight GPUs. Tell us which model should run — from that it can be calculated instead of guessed.

What does a dedicated AI server cost?

The server with the RTX PRO 4500 Blackwell costs €1,182.75 net per month on a twelve-month term and €1,064.48 on 36 months, the server with the RTX PRO 6000 Blackwell €1,997.57 and €1,797.81 respectively. A single B300 is €6,935.00 per month on twelve months; in the node with eight GPUs it is €3,650.00 per GPU on a 60-month term. Setup is charged once on top, €999 per server and €3,999 per B300 node, and less on longer terms. Memory, disks, managed service and network you add yourself in the configurator.

Are power and cooling really included?

Yes, both are in the monthly price. It is the item most often underestimated when running hardware yourself: a server GPU under load draws power continuously, and the heat has to be removed again. With us, that is priced in and does not fluctuate with the electricity price.

What do you take on, and what stays with me?

Without the managed service we provide the server, run the hardware, monitor it around the clock and replace it if something fails. You get it bare metal and install it yourself. With the managed service at €339 net per server and month, the software stack comes pre-installed, and we patch the operating system and the NVIDIA driver, monitor the server and respond to incidents. Models, data and libraries stay with you either way. What we can take on beyond that is on the managed services page.

Which terms are available?

Twelve months is the minimum. On 24 months the platform price drops by 5 %, on 36 months by 10 %; memory, disks, software and network are not affected. The one-off setup is halved on 24 months and falls to a quarter from 36 months. A bill of materials made up purely of B300 nodes can be written on 48 or 60 months, and the monthly price per GPU falls further: €3,942.00 on 48 months and €3,650.00 on 60.

Can I also use the servers by the hour?

That is what GPU as a Service is for: dedicated GPU instances on demand, without a fixed monthly commitment. The AI overview gives an overview of all the options. A dedicated server always runs on a monthly price with a fixed term, the B300 included. If you compute continuously, your own server is usually the cheaper route.

Where are the servers located?

In the DE01 data centre in Hanover, certified to ISO 27001 based on BSI IT-Grundschutz (BSI-IGZ-0777-2026). No transfer to third countries, no access under the US CLOUD Act.

Satisfied customers

IT-PHACONWOLFFAutoDoELAINE technologiesZEDALMAJA.cloudSDV SystemeAucotecPARATFP SignHyperspaceAttinaTutaElectronic Arts

Sovereignty

Paving the way for digital sovereignty

ISO 27001 based on BSI IT-Grundschutz, BSI-IGZ-0777-2026
BSI-IGZ-0777-2026 ISO 27001 based on BSI IT-Grundschutz Scope: all data centers and all cloud services Security concept
  • Owner-managed since 1997

    As an owner-managed operator with our own data centers, we make our decisions independently and are subject exclusively to German / EU law.

  • 100% made in Germany

    Our data center, product development, and customer support are all based in Germany for your digital sovereignty.

  • 24/7 Customer Support

    We are here for you: 24/7/365 customer support, in English and German

Any questions?

We look forward to hearing from you.

Florian Helbig

Senior Solution Consultant

sales@kyberio.com +49 (0)511 71260 140

Book a meeting

  • We will get back to you as soon as we can.
  • No obligation, free of charge.

For support requests please use support@kyberio.com — that way your request reaches the team on duty. More ways to reach us: Contact.

How we process your details is explained in our privacy notice.