Hopsworks
Stockholm feature store and AI lakehouse with model registry and serving, available as a free tier or on-premises
Quick Overview
| Company | Hopsworks AB |
|---|---|
| Category | MLOps Platforms |
| Headquarters | Stockholm, Sweden |
| Founded | Not stated |
| EU Presence | Yes - Sweden (EU) |
| Data Location | Hopsworks SaaS, your Kubernetes cluster, or on-premises and air-gapped |
| Open Source | AGPL-3.0 (Hopsworks repository on GitHub) |
| GDPR Compliant | Yes |
| Self-Hosting | Yes - Kubernetes installer for EKS, GKE, AKS and OVHcloud, plus on-premises |
| Main Features | Feature store, model registry, model serving, experiment tracking, Iceberg, Delta and Hudi tables, GPU scheduling, Ray training, KServe and vLLM serving |
| Pricing | Free tier (1 project, no card); SaaS pay as you go; Enterprise on request |
| Best For | Teams building real-time or batch ML systems around shared features |
| Replaces | Databricks Feature Store, SageMaker Feature Store, Tecton |
Detailed Review
Hopsworks starts from the data side of MLOps. Its feature store keeps feature values in one place so training and serving read the same numbers, and the platform adds a model registry and model serving on top. The site describes sub-millisecond feature retrieval using RonDB.
The terms name Hopsworks AB, Asogatan 119, Stockholm, as the provider and apply Swedish law. The footer lists the company with Swedish and EU flags, and the page highlights air-gapped and on-premises operation for sovereign AI.
Feature store, then models
The three layers the home page names are the feature store, an AI lakehouse that works directly with Iceberg, Delta and Hudi tables, and an MLOps platform with experiment tracking, a model registry and deployment pipelines.
Zalando is quoted as using it for real-time personalisation across 25 countries, which indicates production scale rather than a laboratory tool.
Free tier and usage pricing
The Free plan includes one project with the feature store and model registry and needs no credit card. The SaaS plan is usage-based with unlimited projects and adds Model Serving, a platform SLA and community support.
Enterprise adds on-premises and air-gapped deployment and a dedicated support team. No unit prices are published, so costs depend on usage.
Self-hosting
A Kubernetes installer is published on GitHub for AWS EKS, Google GKE, Azure AKS and OVHcloud, with a recommended minimum of four to five nodes and Kubernetes 1.27 or higher.
The Hopsworks repository carries an AGPL-3.0 licence, which some legal teams review before embedding or modifying the code.
Limitations to Consider
Hopsworks is strongest on features and serving, while experiment tracking is mentioned as one item in the MLOps section rather than as the main product, so a team that wants Weights & Biases-style run charts will likely add DagsHub or MLflow. SaaS prices are usage-based with no published rates. The self-hosted path wants a four to five node Kubernetes cluster. The AGPL-3.0 licence needs a legal read before modification.
Who Should Use Hopsworks
Data and ML teams in Europe that need a governed feature store with serving, from a Swedish vendor that also supports on-premises.
Pros and Cons
Pros
- Free tier with no credit card
- Feature store, registry and serving in one platform
- Runs on-premises or air-gapped
- Swedish company, Swedish governing law
- Works with Iceberg, Delta and Hudi tables
Cons
- No published unit prices for the SaaS plan
- Experiment tracking is a smaller part of the product
- Self-hosting suggests a four to five node cluster
- AGPL-3.0 licence needs review
Alternatives to Hopsworks
Other European mlops platforms worth comparing before you decide:
ZenML
Germany · Munich open source pipeline orchestration with artifact and model tracking
Valohai
Finland · Turku platform for versioned ML runs, pipelines and GPU scheduling
DagsHub
Israel · Experiment tracking, data versioning and annotation on one platform
UbiOps
Netherlands · Hague platform to deploy and orchestrate models on any infrastructure
See all European mlops platforms or compare them directly against Databricks Feature Store on our alternatives page.
Frequently Asked Questions
Yes. The Free plan includes one project with the feature store and model registry and needs no credit card; the SaaS plan charges by usage.
Yes. Enterprise lists on-premises and air-gapped deployment, and a Kubernetes installer supports EKS, GKE, AKS and OVHcloud.
It lists experiment tracking inside its MLOps platform, but its main product is the feature store, so many teams combine it with a dedicated tracking tool.