# IBM watsonx.ai

## Kurzbeschreibung
**“A one-stop, integrated AI developer studio for end-to-end development of AI applications.”**



IBM watsonx.ai is a development platform for building, customizing, testing, and deploying generative AI, traditional machine learning models, and AI agents. It combines foundation models, prompt and agent tools, RAG, AutoAI, APIs, notebooks, and production runtimes in a single environment. The platform can be operated as IBM Cloud SaaS or via IBM Software Hub in your own OpenShift environments.

## Claim
A comprehensive, all-in-one AI development studio for end-to-end development of AI applications.

## Geeignet für
- API Integration
- Automation / Workflows
- Data Analysis
- Data Extraction / Document Analysis
- Customer Service & Chatbots
- Productivity & Planning
- Programming / Software Development
- Research
- Texts / Content
- Sales / Sales
- Knowledge Management / Internal Search

## Kernfunktionen
- Data Analysis
- Embeddings
- Function Calling
- AI Agents
- LLM API
- Model Training
- Open Source Model
- Language Model

## Preismodell
- **free:** **Toolbox Playground / Free **Free plan with a limited monthly foundation model quota, limited machine learning compute hours, and limited text extraction. Suitable for testing, playground use, and initial prototyping.



**Lite Runtime Plan** Free runtime plan with limited capacity for testing machine learning and inference. Foundation model tuning is not supported on the Lite plan.
- **subscription:** **Essentials – Pay-as-you-go** For production deployments with no fixed minimum usage; usage-based billing for models, machine learning, text extraction, model tuning, and other resources.



**Standard – Pay-as-you-go** Enterprise plan with a larger included compute quota and additional usage-based billing; designed for extensive production and enterprise workloads.



**HIPAA-Ready **Specialized plan for generative AI and machine learning under HIPAA security and privacy requirements; officially available only in the IBM Cloud Dallas region.
- **other:** **IBM Foundation Models** Use IBM Granite models via pay-as-you-go inference or dedicated on-demand hosting, depending on the model.



**Third-party models** Access to models from Meta, Google, DeepSeek, Mistral, and others; usage-based inference or dedicated on-demand deployment, depending on the model.



**Embedding and Reranking Models** Usage-based billing for IBM and third-party embedding models, as well as reranking capabilities for semantic search and RAG.



**Deploy on Demand** Dedicated hosting of selected foundation models with hourly runtime billing; availability depends on the model and region.


**Machine Learning / AutoAI** Usage-based billing via Capacity Unit Hours for training, AutoAI, model deployment, and scoring.



**LoRA/QLoRA Fine-Tuning** Customization of supported foundation models using your own data; according to IBM, QLoRA is available in Frankfurt and Dallas, among other locations.


**Custom Foundation Models** Import, hosting, and deployment of customer-owned or customized foundation models on dedicated infrastructure.



**IBM Software Hub / On-Premise** Self-operated watsonx.ai installation on Red Hat OpenShift, including a private registry and optional air-gapped environment; customized software, infrastructure, and support agreement.



**Enterprise Support** Extended support and SLA options in addition to the included basic support.

## DSGVO und Datenschutz
**Gesamteinschätzung:** Yes

**Overall assessment:**

Watsonx.ai can process prompts, responses, documents, datasets, embeddings, training and tuning data, machine learning models, custom foundation models, RAG indices, project metadata, and technical logs. The platform supports Prompt Lab, Agent Lab, RAG, synthetic data, text extraction, classic machine learning functions, LoRA/QLoRA fine-tuning, customer-owned foundation models, and on-demand deployments.


Training on customer data: According to official security documentation, IBM does not use uploaded content or generated outputs to further train or improve foundation models. However, customers can intentionally use their data for their own models, tuning procedures, or RAG systems. These customer-specific processes are distinct from general IBM model training.



Data residency: On IBM Cloud, projects, catalogs, and data are tied to the selected region. Frankfurt, London, Dallas, and Tokyo have documented private runtime endpoints. Availability may vary for other regions and specific features.



Deletion and Retention: IBM documents the secure deletion of personal data from watsonx.ai Runtime. Specific retention periods depend on the service used, data type, plan, and the associated Data Processing and Protection Data Sheet. There is no blanket retention period for all watsonx.ai data.


**Conclusion:**

Watsonx.ai is one of the most flexible platforms for enterprises with high hosting, security, and compliance requirements. The Frankfurt IBM Cloud region is suitable for standard projects; for trade secrets, critical infrastructure, or particularly sensitive data, on-premises, private cloud, or air-gapped environments are the stronger options.


[Security policies and responsibilities in IBM Cloud](https://www.ibm.com/docs/en/watsonx/saas?topic=cloud-security-policies-responsibilities-in) [Privacy Statement](https://www.ibm.com/us-en/privacy)

**Overall assessment:**
****
Well-suited for GDPR compliance, provided that a European region or a controlled on-premises deployment is selected and the IBM contractual documents are properly incorporated. IBM provides a Data Processing Addendum for watsonx.ai. According to official documentation, the IBM DPA and the associated Data Processing and Protection Data Sheets apply when personal data as defined by the GDPR is processed in customer content. The data sheets describe product-specific permissible content, processing activities, data protection features, retention, and data return. Changes to subprocessors and opt-out options are governed by the DPA.



**A positive aspect** is the explicit training rule: IBM states that uploaded content and the outputs generated by foundation models will not be used to further train or improve IBM models or other models. This means that watsonx.ai is generally suitable for business data, internal knowledge, and confidential RAG applications. However, this statement does not relieve companies of the responsibility to review the third-party, open-source, or customer-owned models they select.


**Server location: **IBM watsonx.ai is available on IBM Cloud in locations including Frankfurt, London, Dallas, and Tokyo. Projects, catalogs, and data are region-bound; a project stored in Frankfurt cannot simply be opened via an instance in Dallas. For German companies, the Frankfurt region or eu-de is therefore particularly relevant. IBM also supports private service endpoints in these regions, allowing watsonx.ai Runtime to be operated without public internet access.



**A drawback**—or rather, a point requiring review—is that individual models and features are not available in every region. Watsonx.ai also includes models from IBM as well as third-party providers such as Meta, Google, DeepSeek, and Mistral. Model licensing, deployment type, regional availability, and any additional terms must therefore be reviewed on a model-by-model basis.



IBM also notes that personal data should not be written to training log files, as these may be accessible to other users within the customer’s organization and, if necessary, to IBM Support. For personal data in watsonx.ai Runtime, IBM describes a process for secure deletion.



**Conclusion:** Watsonx.ai is comparatively well-suited for GDPR-critical organizations, particularly when using the Frankfurt region, private endpoints, or an installation on their own OpenShift infrastructure. A DPA and data sheet, a documented region selection, role and permission concepts, retention periods, logging rules, and an individual review of the foundation models and data sources used remain required.


[Security policies and responsibilities in IBM Cloud](https://www.ibm.com/docs/en/watsonx/saas?topic=cloud-security-policies-responsibilities-in) [Privacy Statement](https://www.ibm.com/us-en/privacy)

## Hosting und Daten
- **On-Prem / lokales Hosting:** abgedeckt
- **Private Cloud / Rechenzentrum:** abgedeckt
- **EU SaaS / Managed:** abgedeckt
- **Hybrid:** abgedeckt
- **AVV / DPA:** abgedeckt
- **Kein Training auf Kundendaten:** abgedeckt
- **Open-Source / Transparenz-Pfad:** teilweise / indirekt

## Standort
**Land:** USA

**Taxonomie:** USA

International Business Machines Corporation, 1 New Orchard Road, Armonk, New York 10504-1722, USA. German subsidiary: IBM Deutschland GmbH, IBM-Allee 1, 71139 Ehningen, Germany

## Vorteile
- Comprehensive AI lifecycle, from experimentation to production
- Generative AI and traditional ML on a single platform
- Wide selection of models and a "bring-your-own-model" approach
- Powerful RAG, agent, and document processing
- Access via web interface, notebook, SDK, or API
- Available in the Frankfurt IBM Cloud region
- Data encryption during transmission and storage
- On-premises and air-gap-capable deployment options
- IBM states that it will not use customer data, customer models, or model outputs for its own models.

## Nachteile
- High technical and organizational complexity
- Often more extensive than necessary for small, standalone applications
- Costs are incurred across multiple units, such as tokens, resource units, compute hours, GPU runtime, and document pages.
- Model and feature availability varies by data center region.
- Third-party models are subject to their own licenses and terms.
- Foundation models are regularly replaced or discontinued, which can result in migration efforts.
- The complete governance solution is a separate watsonx product.
- On-premises operation requires OpenShift, storage, and, in some cases, significant GPU infrastructure.
- AI outputs must be reviewed by humans due to potential errors, biases, and hallucinations.

## Quellen
- Offizielle Website: https://www.ibm.com/de-de/products/watsonx-ai/pricing

## Letzter Datenstand
2026-06-10

## Originalseite
https://kifox.ai/en/ki-tools/ibm-watsonx-ai-en/