Start Fast, Scale on Your Terms
Prototype with pay-per-token access to a curated model catalog, then move to dedicated GPUs once you need guaranteed capacity and predictable latency. Both run on the same platform, so there is nothing to migrate.
Run AI Inference Your Way
Dedicated GPU Endpoints or Pay-as-You-Go Access to a Curated Model Catalog
Exoscale Inference gives you two ways to put AI models into production. Reserve dedicated NVIDIA GPUs for full control and predictable performance, or call a pay-per-token API against a curated catalog of ready-to-use models. Both share the same OpenAI-compatible API and run exclusively on sovereign European infrastructure.
Why Run Inference on Exoscale?
European Data Sovereignty
Run inference in Exoscale's European and Swiss locations, with control over where your workloads and data are hosted. This helps you meet your data residency and GDPR obligations.
Built for Production
Both inference options are engineered for stable, predictable operation, with 99,95% SLA. Open interfaces and an OpenAI-compatible API keep your workloads portable and your integrations simple.
Inference Options
Choose the inference option that matches how much control you need and how your traffic looks. Dedicated and On-demand Inference share the same sovereign infrastructure and OpenAI-compatible API, with Batch Inference coming soon.
Dedicated Inference
Reserve dedicated NVIDIA GPUs for any open-source or custom AI model, public, gated or private, from Hugging Face. Get an OpenAI-compatible endpoint with full isolation, per-second billing, and the option to scale to zero when idle. Zero infrastructure operations required.
Discover
On-Demand Inference
Call a curated catalog of production-ready models through a pay-per-token API. No GPU sizing, no capacity planning, no infrastructure to manage. Ideal for rapid prototyping, variable traffic, and teams who want AI capabilities without running dedicated GPUs.
Discover
Batch Inference
Process large volumes of AI requests asynchronously for workloads that do not require real-time responses. Ideal for data enrichment, document processing, evaluations, and scheduled jobs. Coming soon.
Contact usWhat You Get With Either Option
-
European Sovereignty & Sustainability
Your data remains in Europe, ensuring GDPR compliance and sovereignty. Our data centers use renewable power and efficient cooling to reduce environmental impact significantly. Built in Europe, for Europe.
-
Predictable, Transparent Pricing
Pay per second for the GPU time you reserve with Dedicated Inference, or per token for what you consume with On-demand Inference. Transparent billing, no hidden fees.
-
OpenAI API Compatibility
Both Dedicated and On-demand Inference integrate easily with your existing AI tools. No vendor lock-in and a simple transition to our sovereign cloud infrastructure.
-
Dedicated Hardware & Privacy
When you need it, Dedicated Inference runs on fully isolated infrastructure, guaranteeing maximum privacy and performance for your workload.
-
Local, Expert, Premium Support
Get 24/7 support with a 30-minute response time, handled directly by Exoscale engineers in Europe.
-
A Complete, Integrated Platform
Run your entire AI stack on Exoscale: GPUs, vector databases, inference, compute, storage, networking, and more, all fully integrated, cloud-native, and sovereign.
Complete Your AI Stack
Pair Inference with the rest of Exoscale’s AI infrastructure and cloud services to build a complete, sovereign platform.
High-Performance GPUs
Accelerate AI training, fine-tuning, and volume inference using powerful NVIDIA GPUs. Scale from a single GPU to multi-GPU setups without complex orchestration. Perfect for machine learning, data processing, 3D rendering, inference, and scientific computing.
Discover
Managed Vector Databases
Essential tools for modern AI, powering Retrieval-Augmented Generation (RAG) and semantic search workloads. Fully managed PostgreSQL with pgvector and OpenSearch-based vector search.
Discover
Support Plans
Get the help you need to run your infrastructure with confidence through flexible support plans that provide expert guidance and guaranteed response times (SLA), ensuring our experts are there when you need them most.
DiscoverTrusted by Engineering Teams Across Europe
Running AI inference in production needs a dependable partner. Our engineering and support teams help organizations across Europe deploy, scale, and move between Dedicated and On-demand Inference on Exoscale’s sovereign, sustainable cloud platform.
Contact usFrequently Asked Questions
What is the difference between Dedicated Inference and On-Demand Inference?
Dedicated Inference reserves NVIDIA GPUs for you. You get full isolation, predictable latency, and per-second billing, and you can run any open-source or custom model from Hugging Face. On-demand Inference gives you pay-per-token access to a curated catalog of ready-to-use models, with no GPU sizing or capacity planning needed. Both share the same OpenAI-compatible API.
Can I move from On-Demand to Dedicated Inference as my usage grows?
Yes. Both options run on the same sovereign infrastructure and expose the same OpenAI-compatible API, so moving a workload from On-demand to a dedicated GPU deployment does not require rewriting your integration.
When Should I Choose On-Demand Inference?
On-demand Inference is ideal for rapid prototyping, applications with variable or unpredictable traffic, and teams that want to use production-ready AI models without sizing GPUs or managing infrastructure. You pay per token and can access models through an OpenAI-compatible API.
What is Batch Inference for?
Batch Inference is designed for high-volume jobs that do not need an immediate response, such as document processing, data enrichment, evaluations, and scheduled workloads. It is coming soon on Exoscale.
Can I use my own AI models on Exoscale?
Yes, with Dedicated Inference. You can deploy any public, gated, or private model from Hugging Face, including models from your own organization’s Hugging Face account, on dedicated GPU infrastructure.
How is Inference priced?
Dedicated Inference is billed per second for the GPU time you reserve. On-demand Inference is billed per token, so you pay only for what you consume, with no upfront GPU cost. Both are transparent, with no hidden fees.
Is Inference on Exoscale GDPR compliant?
Yes. All Inference workloads run in European and Swiss data centers, so your data never leaves European jurisdiction. This applies to both Dedicated and On-demand Inference.
Does Inference support open standards?
Yes. Both Dedicated and On-demand Inference expose an OpenAI-compatible API, so you can connect your existing tools and libraries without proprietary lock-in.