GPU on Demand for Kubernetes

Dedicated GPU power, when you need it. Extend your on-premises or hybrid Kubernetes cluster without touching your infrastructure.    

  • Native Kubernetes integration
  • Zero CAPEX, predictable costs
  • Dynamic scalability (Plug&Play)
  • Data sovereignty (100% Europe)
REQUEST SUPPORT Learn more

Extend your Kubernetes cluster with dedicated physical GPUs

Aruba Cloud’s GPU on Demand allows you to integrate dedicated NVIDIA GPUs into your existing Kubernetes cluster in minutes.
You get scalable compute power for AI/ML, 3D rendering, and analytics, with transparent monthly costs and no architectural lock-in.
Based on Aruba’s European cloud infrastructure, it guarantees consistent performance, security, and data sovereignty, with immediate activation and release.

AI acceleration

Whether it’s inference or fine tuning, Aruba Cloud’s dedicated GPUs ensure minimum latency and consistent performance. The cluster scales in minutes: experiment, test, deploy without slowing down your DevOps cycle.

Rendering

Optimized NVIDIA RTX GPU profiles for 3D rendering, visual design, and video production. Work in real time with dedicated resources, stable performance, and constant throughput. 

Machine Learning

Fine-tune on high-performance dedicated GPUs, without scalability limits. We manage the infrastructure: you focus on feature engineering and experimentation.

Data analysis

Process large volumes of data with high throughput and reduced insight times. Perfect for big data pipelines, video analytics and near real-time anomaly detection.

GPU Prices On Demand

GPU on Demand - Nvidia GeFORCE RTX 4080 Super

€ 389.90 + VAT / month Free setup if you order for 6 months*

  • CUDA Cores: NVIDIA CUDA Cores 10240
  • GPU: Nvidia GeFORCE RTX 4080 Super
  • GPU Memory: 16GB GDDR6X

GPU on Demand - Nvidia L40S

€ 1499.90 + VAT / month Free setup if you order for 6 months*

  • CUDA Cores: NVIDIA Ada Lovelace Architecture - CUDA Cores 18176
  • GPU: Nvidia L40S
  • GPU Memory: 48GB GDDR6 with ECC
REQUEST SUPPORT

Federation architecture: how it works

Secure scalability across different regions and clusters. Dynamically extend your

Kubernetes cluster by using remote GPUs as if they were local resources, without requiring application changes.

  • Integrated virtual node via LIQO
  • Dedicated and sovereign physical GPUs
  • Same APIs, no lock-in

Why choose Aruba Cloud’s GPU On Demand

No Lock-in

Extend any Kubernetes architecture — on-premises, public or hybrid — without changing your toolchain or APIs. GPUs are exposed as native cluster resources, with no platform constraints.

Sovereignty & compliance by design

Data and workloads remain in Europe, with infrastructure and processes compliant with GDPR and data residency requirements. You have total control over where your data resides and how it’s treated.

Predictable costs

Dedicated GPUs with fixed or package fees, with clear consumption monitoring. Scale or deactivate resources whenever you want, maintaining full budget visibility.

Transparent integration

Same Kubernetes APIs, same CI/CD pipelines, same tools. Integrate dedicated GPUs without changing your architecture or workflows.

Immediate scalability

Access remote GPUs on demand and scale AI workloads when needed. Maintain agility and consistent performance even during load peaks.

Fast and reversible activation

Connect dedicated GPUs in minutes and release them when they’re no longer needed. Perfect for PoC, testing, seasonal peaks, or controlled experimentation phases.

Try it with Aruba Managed Kubernetes and start building your Private AI

Dedicated GPUs + Aruba Managed Kubernetes: The fastest way to bring AI/ML to production. Activate GPUs on Demand, scale as needed, and maintain control over costs, data, and performance.

REQUEST SUPPORT Learn more

GPU on Demand: Frequently Asked Questions

  • Add extra GPU capacity in minutes for generative AI, training, 3D rendering, or analytics. Perfect for handling peaks or experimenting without changing your pipeline: connect, use, release.

     

  • The prerequisite is the use of Kubernetes.

    You need a Kubernetes cluster. You federate your cluster with a dedicated GPU server (via Liqo) and the remote GPUs become cluster resources. You schedule AI/ML pods as on any worker node.

  • Compatible with major platforms: AKS, EKS, GKE, OpenShift, Tanzu, Rancher, and vanilla distributions. No proprietary stack dependency.

  • Thanks to the on-demand model with monthly billing, you have visibility on consumption and can scale or deactivate GPUs independently. No surprises at the end of the month.

Get started with Aruba Cloud

Looking for a custom solution?

Have a chat with our solution architects.

Get in touch