Skip to content
← Back to cinfo.es
Catalogue

SERVICE · INFRASTRUCTURE

Cloud GPU and AI infrastructure

AI compute capacity with your data hosted on home soil.

Our own datacentre in A Coruña's Ciudad de las TIC, plus elastic deployment across multiple clouds.

IndustryServices

Contact cinfo about Cloud GPU and AI infrastructure Download data sheet (PDF)

We take you to cinfo's contact form with this item already filled in, so you don't have to explain what it is about.

Cloud GPU and AI infrastructure

THE CHALLENGE

Training and serving models without depending on one single cloud

AI needs GPUs, and GPUs force decisions: where the data lives, what a demand peak costs, and what happens when the provider changes prices or terms.

  • Unpredictable cost

    Training and inference peaks send the public cloud bill through the roof.

  • Data outside the perimeter

    Regulated sectors that cannot process their information in just any jurisdiction.

  • Vendor dependence

    Architectures tied to one cloud that are expensive to move.

WHAT IT IS

Our own datacentre plus elastic multi-cloud deployment

cinfo has virtualised its own AI datacentre in A Coruña's Ciudad de las TIC and runs its own elastic computing system to deploy and manage workloads on local servers or on any cloud provider.

3,328 high-grade AMD EPYC CPUs and 8 TB of RAM.

33 NVIDIA L4 GPUs with 24 GB and 6 NVIDIA A40 GPUs with 48 GB of video memory.

114 TB of disk plus 500 TB more on low-latency NAS.

Automated hybrid deployment on AWS, Azure, Google Cloud, Oracle, Akamai or on-premise.

Our own datacentre plus elastic multi-cloud deployment

HOW IT WORKS

BERT: find resources and deploy where it makes sense

cinfo built BERT, its cloud resource finder and elastic deployment manager, which locates capacity and places the workload where it works out best.

  • The workload is sized

    Analysis of compute, memory, storage and latency requirements.

  • Resources are located

    BERT finds available capacity in our own datacentre and in the contracted clouds.

  • It is deployed in containers

    Containerised deployment on Docker Swarm or Kubernetes clusters.

  • It is operated and observed

    High availability, lifecycle management and observability dashboards.

CAPABILITIES

Infrastructure built for AI and video workloads

  • Our own datacentre

    Facilities in A Coruña's Ciudad de las TIC, with the team's offices in the same building.

  • GPUs for training

    39 NVIDIA GPUs dedicated to model development, training and inference.

  • Low-latency storage

    114 TB on disk and 500 TB on low-latency NAS for large datasets.

  • Multi-cloud deployment

    Workloads placed indifferently on public cloud, private cloud or physical servers.

  • Containers and orchestration

    Docker Swarm and Kubernetes with high availability and lifecycle management.

  • Automated CI/CD

    From git through integration, test, staging and production.

  • Observability

    Monitoring with analytics dashboards and alert triggers.

  • Compliance

    Cybersecurity compliance as part of the development and operations cycle.

BENEFITS

Capacity, cost control and data sovereignty

  • Data on home soil

    Processing can stay entirely inside the A Coruña datacentre.

  • Cost under control

    Our own capacity for the baseline, public cloud only for the peaks.

  • No single-vendor dependence

    The deployment system is independent of the cloud provider.

  • Real GPU experience

    Infrastructure used daily for cinfo's own video and AI products.

SCOPE

Ways of contracting the infrastructure

The service ranges from architecture consulting to fully operating the client's platform.

  • Architecture consulting

    Sizing, cloud selection and deployment plan design.

  • Compute capacity

    Use of cinfo's GPU infrastructure for training and inference.

  • Managed deployment

    Platform go-live in containers, with high availability.

  • Continuous operations

    Monitoring, alerting, updates and lifecycle management.

  • Hosted models

    Open-source LLMs served from the cinfo cloud for your applications.

  • Migration

    Moving existing workloads to a hybrid, multi-cloud scheme.

METHODOLOGY

From sizing to operations

  • Assessment

    Analysis of current workloads, costs and regulatory requirements.

  • Architecture

    Design of the hybrid scheme and the workload split between our cloud and public clouds.

  • Deployment

    Containerisation, orchestration and continuous integration pipelines.

  • Operations

    Observability, high availability and evolution of the system in production.

The infrastructure described is cinfo's own datacentre in A Coruña's Ciudad de las TIC. Sizing for each project is defined in the architecture phase.

GALLERY

Infrastructure and deployments

The service architecture of a platform in production.
The service architecture of a platform in production.
The data architecture of a multi-cloud project.
The data architecture of a multi-cloud project.
Operations and observability dashboards.
Operations and observability dashboards.

CASE STUDIES

Infrastructure behind real services

  • RTVE PLAY

    Over 7 million videos served a day

    The AI preview service runs on cinfo infrastructure with global CDN delivery and peaks of millions of daily requests.

    RTVE PLAY

  • TIIVII

    750 events produced every week

    Automatic production, the mixer and the OTT platforms run on cinfo's elastic infrastructure, with peaks concentrated at weekends.

    TIIVII

  • TIIVII AI

    LLMs served from Galicia

    cinfo's AI platform serves open-source models from its own datacentre, guaranteeing the client's data never leaves the agreed environment.

    TIIVII AI

CINFO IN NUMBERS

The datacentre in numbers

  • 3.328high-grade AMD EPYC CPUs
  • 39NVIDIA L4 and A40 GPUs
  • 8 TBof available RAM
  • 614 TBof disk and NAS storage

FREQUENTLY ASKED QUESTIONS

Questions on infrastructure and data sovereignty

Where is the infrastructure physically?

In cinfo's own datacentre in A Coruña's Ciudad de las TIC, where the technical team's offices are as well.

Can I combine your cloud with mine?

Yes, that is the usual scenario: the elastic deployment system places workloads in our datacentre, in public cloud or on the client's servers alike.

Is it for training or only inference?

For both. The GPUs are used daily for development, training and inference of cinfo's own video and AI models.

What about demand peaks?

The usual scheme covers the baseline with our own capacity and absorbs peaks in public cloud, which keeps cost under control.

Who operates the platform?

Either cinfo as a managed service, or the client's team following knowledge transfer.

Does it meet cybersecurity requirements?

Cybersecurity compliance is part of cinfo's development and operations cycle, and is documented for each project.

WHO IT IS FOR

Who needs this capacity

  • Companies with AI projects

    Training and inference without depending on one public cloud.

  • Regulated sectors

    Cases where the data's location is a project condition.

  • Public administrations

    Digital services with sovereignty and traceability requirements.

  • Video platforms

    Workloads with concentrated peaks and heavy transcoding needs.

Related items

LET'S TALK

Let's size your workload

We analyse what you need to train or serve and propose the split between our cloud and public clouds.

Request a sizing