GPU server for AI: sized for your model and load
A server sized for your load, not the top of the catalogue
We calculate how much compute your model needs: size, context, concurrent users, whether fine-tuning is planned. We pick GPU servers, check site readiness and set up model serving. Delivery times are planned honestly.
What we build
The most expensive mistake in private AI is buying hardware by eye: the cards can’t run the model you need or sit idle half the time. We start with a load profile: which model, how many concurrent users, how long the documents are, whether fine-tuning is planned. From it we pick GPUs, CPUs, memory, storage and network, check that the server room can handle the power and cooling, and set up model serving. A pilot can start on available or rented hardware, with the purchase made once the load is clear.
Load profile
Model, context length, concurrent users, inference only or fine-tuning too - everything else depends on it.
Server specification
GPUs, CPUs, memory, storage and network for your profile, with headroom for growth but no paying for idle capacity.
Site readiness
Power and cooling for GPU nodes, rack space, redundancy. Checked before purchase, not after delivery.
Serving and orchestration
Model deployment, Kubernetes with GPU support if needed, monitoring of load and queues.
What the work includes
Sizing and setup work is priced per task, estimate within 1–2 days. Hardware is quoted separately by the supplier.
- Load profile from your use cases
- GPU server specification with budget options
- Site check: power, cooling, rack, network
- Purchase plan that accounts for delivery times
- Model serving and monitoring setup
- Scaling recommendations as the load grows
How we work
- 01
Requirements
We clarify which data must not leave: personal data, trade secrets, critical infrastructure. We record your security team’s requirements and the use case behind the project.
- 02
Perimeter architecture
Where models, the knowledge base and logs live, who can access what, and how the perimeter connects to your systems. The design is agreed with security before any hardware is bought.
- 03
Pilot inside the perimeter
4–8 weeks on a real task: model, documents, roles. Acceptance criteria - share of correct answers with a source, speed, hand-offs to people - are agreed upfront.
- 04
Fine-tuning and scaling
If a base model with document search is not enough, we fine-tune it on your data. Users, use cases and capacity grow with the load.
- 05
Support
Monitoring, model and knowledge base updates, incident handling. The perimeter and the code stay yours.
Questions
Which GPUs does a private AI model need?
It depends on the model and the load: one or two cards are enough to pilot a small model, dozens of concurrent users and fine-tuning need more. We size from the load profile, not from the top model in the catalogue.
Can we start without buying servers?
Yes: the pilot runs on available hardware or a rented GPU server. The purchase comes once the real load is known.
Do you supply the hardware?
We prepare the specification and help choose a supplier, taking delivery times and card availability into account. Hardware is quoted separately.
The perimeter is live - next comes support
After launch we support the perimeter: model and knowledge base updates, monitoring, incidents and improvements under a contract or an SLA.
Other development areas
- Closed AI perimeterA perimeter design for models and data: roles, audit logs, network segments. Built around data protection law and your security rules - before any GPU purchase.
- Private LLM on your serversAn open language model inside your perimeter: a corporate AI chat over your documents, an API for your systems, roles and audit logs.
- Model fine-tuningFine-tuning an open LLM on your data: terminology, answer format, classification. Inside your perimeter, with a quality check.