Managed AI operations
Operate infrastructure, models, retrieval quality and continuous improvement over time.
Enterprise AI, operated in Europe
From model API to private AI factory.
Deploy production inference, dedicated GPU compute and secure AI applications on infrastructure operated by Arctur in Slovenia. Build with APIs - or let our team deliver and manage the complete solution.
EU operated
Slovenian infrastructure
API → cluster
One deployment continuum
Private by design
Dedicated and on-prem option
Applied AI
RAG, fine-tuning, analytics
Built for real production workloads.
OpenAI-compatible APIs, dedicated endpoints, GPU/HPC capacity, private networking and applied-AI engineering—under one European operating model.
ONE PLATFORM, FOUR WAYS TO BUILD
AI capacity that meets you where you are.
Start with a model endpoint or a single GPU. Expand into a private cluster, a managed knowledge assistant, or an AI system installed in your environment.
01 / Inference
Model API
OpenAI-compatible access to curated language, embedding, reranking and multimodal models.
EXPLORE Inference →
02 / COMPUTE
GPU & HPC capacity
Shared instances, dedicated GPU nodes and high-speed cluster capacity for AI and HPC workloads.
EXPLORE COMPUTE →
03 / SERVICES
Managed AI
Private RAG, copilots, data engineering, model adaptation and end-to-end AI application delivery.
EXPLORE SERVICES →
04 / HARDWARE
AI systems
Buy, rent, host or operate purpose-built AI servers and private AI appliances.
EXPLORE HARDWARE →
ARCTUR MODEL API
Production inference without public-cloud compromises.
Use a familiar API surface while retaining a path to reserved throughput, dedicated serving and private connectivity.
Qwen3.5-35B-A3B / Qwen3.6-35B-A3B
Multimodal multilingual reasoning
Qwen3.6-27B
Efficient coding and language model
Qwen3.5-122B-A10B
Complex enterprise workflows
DeepSeek-V4-Flash
1M-context high-speed inference
Custom or private model
Dedicated or private deployment
SELECTED MODEL
Input €0.10 / 1M · output €1.30 / 1M
Qwen3.5-35B-A3B / Qwen3.6-35B-A3B
Two comparable multimodal 35B-A3B model variants for multilingual chat, image understanding, knowledge assistants, structured extraction and tool-enabled workflows.
| CONTEXT
32K |
MODALITIES
Text · Image · Code · Math |
| REASONING
Strong multilingual |
TOOL CALLING
Supported |
SELECTED MODEL
Input €0.10 / 1M · output €1.50 / 1M
Qwen3.6-27B
An efficient model for production applications, particularly suitable for coding and language tasks with a long context window.
| CONTEXT
262,144 tokens |
MODALITIES
Text · Image · Code |
| REASONING
Efficient coding |
TOOL CALLING
Supported |
SELECTED MODEL
Input €0.40 / 1M · output €8.00 / 1M
Qwen3.5-122B-A10B
A higher-capability model for complex enterprise reasoning, longer workflows and advanced assistant use cases.
| CONTEXT
64K |
MODALITIES
Text · Image · Code |
| REASONING
Complex enterprise |
TOOL CALLING
Supported |
SELECTED MODEL
Input €0.20 / 1M · output €3.50 / 1M
DeepSeek-V4-Flash
A high-speed, long-context model for demanding applications that benefit from a 1 million token context window.
| CONTEXT
1m tokens |
MODALITIES
Text |
| REASONING
High-speed long context |
TOOL CALLING
Supported |
SELECTED MODEL
Dedicated endpoint, reserved capacity and private deployment available on request.
Custom or private model
Bring an approved open-weight model, deploy one from the Arctur catalogue, or design a dedicated serving environment for your workload.
| CONTEXT
Configurable |
MODALITIES
Model dependant |
| REASONING
Model dependant |
TOOL CALLING
On request |
Transparent commercial model
Inference plans that scale with your product.
Every paid plan includes its full monthly commitment as inference credit. Overage discounts apply after included credit is used.
| Plan | Monthly commitment | Included inference credit | Overage discount | API keys | Concurrency | RPM | Support / controls |
|---|---|---|---|---|---|---|---|
| Free | €0 | €0 | 0% | 1 | 5 | 30 | Community / best effort |
| Starter | €59 | €59 | 0% | 3 | 10 | 75 | Usage dashboard, hard spend cap |
| Development | €250 | €250 | 5% | 10 | 25 | 150 | Email support, API key/project budgets |
| Enterprise | €2,000 | €2,000 | 12% | 50 | 150 | 800 | Priority queue, SLA, audit logs, named contact |
| Custom | €5,000+ | 100% of commitment | 15%+ negotiated | Custom | Custom | Custom | Reserved capacity, private deployment/networking, custom SLA |
Concurrency = simultaneous active requests. RPM = requests per minute. Model token rates are quoted separately and vary by model and deployment tier. Custom plans can include dedicated capacity and private network access.
Location
Secure Slovenian infrastructure
Network
InfiniBand and high-speed Ethernet
Storage
NVMe-backed performance tiers
Operations
24/7 monitoring and support options
Arctur AI compute
Rent a GPU. Reserve a node. Build a cluster.
A clear progression from standard CPU capacity to dedicated accelerator nodes and private, high-bandwidth AI clusters.
CPU COMPUTE
Standard Compute
- 2 × Intel Xeon E5-2690 v4
- 28 CPU cores total
- 256 GB DDR4 ECC
- 25G Ethernet
From €1.80 / hour
High performance CPU
EPYC High-Perf
- 2 × AMD EPYC 9474F
- 96 CPU cores total
- 512 GB DDR5
- 200G InfiniBand
From €3.00 / hour
Dedicated GPU
RTX PRO GPU Node
- 2 × NVIDIA RTX PRO 6000 Blackwell
- 2 × Intel Xeon Gold 6458Q
- 512 GB DDR5
- 200G InfiniBand
From €5.00 / hour
AI accelerator node
MACx Node
- 8 × MACx 64 GB accelerators
- 2 × AMD EPYC 9474F
- 384 GB DDR5
- 200G InfiniBand
From €5.00 / hour
Illustrative configurations and starting rates. Availability, hardware specifications and pricing should be validated before publication.
Applied AI engineering
Move from AI ambition to an operating system.
Our team combines AI infrastructure, data engineering and application delivery so customers do not have to assemble an AI platform from disconnected vendors.
01
02
Data engineering & analytics
Data preparation, lakehouse and warehouse delivery, dashboards, forecasting and governed analytics foundations.
03
Fine-tuning & model adaptation
Dataset readiness, LoRA/QLoRA, evaluation, quantisation and production serving for domain-specific models.
04
Custom AI products
AI-enabled workflows, document intelligence, agents, vision systems and product integrations built for measurable outcomes.
A repeatable path to production
01
Architecture sprint
Prioritise the use case, evaluate data and establish deployment/security boundaries.
02
Proof of value
Build and evaluate a focused working system against customer-approved success criteria.
03
Production deployment
Integrate identity, access control, data sources, monitoring and operational processes.
04
Managed AI operations
Operate infrastructure, models, retrieval quality and continuous improvement over time.
Arctur AI systems
Hardware with an operating model attached.
Acquire an AI system, rent it in the Arctur data centre, host customer-owned hardware, or deploy it as a managed private AI environment.
Featured Configuration
MACx AI Server
A high-density AI accelerator server for private inference, RAG, fine-tuning and scalable enterprise workloads. Available as customer-owned hardware, hosted infrastructure or an Arctur-managed service.
| Accelerators | CPU | Connectivity |
|---|---|---|
| 8 × MACx AI | 2 × 128-core AMD Genoa | PCIe 5.0 / 5U |
Commercial Options
Deploy it your way.
| Managed in Arctur DC | from €3,000/mo |
|---|---|
| Purchase hardware | from €79,000 |
| Private/on-prem design | On request |
Hardware examples and commercial figures are indicative. Final configurations, lead times, installation, warranty and sales arrangements are provided in a formal quote.
Talk to Arctur AI
Design the right AI operating model for your organisation.
Tell us whether you need an API, GPU capacity, a private RAG assistant, an AI server, or an end-to-end delivery partner. We will route your request to the appropriate technical team.