# 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 calls - 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 **On-prem / local hosting: supported** IBM documents watsonx.ai on IBM Software Hub in its own OpenShift environment as well as explicitly on-premises. There is even a “lightweight engine” mode for local hosting and inference of AI models. **Private Cloud / Data Center: Covered** IBM Software Hub can be run on private or in-house OpenShift infrastructure; IBM also lists private, on-premises, and customer-owned clusters as operating options. **EU SaaS / Managed: Covered** For watsonx.ai Runtime and watsonx.ai Studio, EU regions on IBM Cloud are documented, including Frankfurt (eu-de). This confirms a provider-operated SaaS/managed option within the EU. **Hybrid: Partially** IBM generally positions watsonx as a hybrid cloud/OpenShift platform and documents both SaaS and customer-owned clusters. However, the pages reviewed do not describe in detail an explicit hybrid operating model specifically for watsonx.ai with a clear division of individual processing steps. **AVV / DPA: Covered** IBM publishes an IBM Data Processing Addendum and explicitly states that this applies to the processing of personal data under the GDPR. **No Training: Covered** The IBM FAQs state that submitting prompts does not train the foundation models and that IBM does not have access to customer data; furthermore, data and models are private to the account. A separate opt-out is therefore not necessary because the non-use for general training is described as such. **Open Source / Transparency Path: Covered** IBM documents open and transparent pathways via open-source frameworks such as PyTorch, TensorFlow, and scikit-learn, as well as the ability to use custom or open-source models in watsonx.ai and run them on-premises via IBM Software Hub. **Data Processing** The website describes two main deployment options: first, SaaS on IBM Cloud; second, software deployment via IBM Software Hub on Red Hat OpenShift within the customer’s own or private infrastructure. For the SaaS option, EU regions on IBM Cloud are documented, including Frankfurt for watsonx.ai Runtime and watsonx.ai Studio. For greater control, watsonx.ai can be operated on-premises or on a private cluster; in this case, the infrastructure is located at the customer’s site. IBM refers to the general DPA for data processing. No clear, product-specific list of subprocessors was found on the watsonx.ai pages reviewed; the IBM website sometimes only refers to them in general terms or provides a list upon request. **Conclusion** In an EU/EEA tool directory, IBM watsonx.ai stands out positively overall because a robust European GDPR compliance path is documented across EU regions and, in particular, via self-hosting/on-premises with IBM Software Hub. The strongest compliance path is not necessarily standard SaaS usage, but rather operation on one’s own EU/EEA infrastructure or, at a minimum, in an EU IBM cloud region. Less well documented are a clearly product-specific overview of subprocessors and a complete, centralized EU data residency page specifically for watsonx.ai. **Sources** - [https://www.ibm.com/products/watsonx-ai/pricing](https://www.ibm.com/products/watsonx-ai/pricing) - [https://www.ibm.com/trust/privacy](https://www.ibm.com/trust/privacy) - [https://cloud.ibm.com/docs/overview?topic=overview-services_region](https://cloud.ibm.com/docs/overview?topic=overview-services_region) - [https://www.ibm.com/us-en/privacy/dataprivacyframework.html](https://www.ibm.com/us-en/privacy/dataprivacyframework.html) - [https://www.ibm.com/docs/en/software-hub/5.3.x?topic=watsonxai-getting-started](https://www.ibm.com/docs/en/software-hub/5.3.x?topic=watsonxai-getting-started) - [https://www.ibm.com/docs/en/software-hub/5.1.x?topic=install-choosing-installation-mode](https://www.ibm.com/docs/en/software-hub/5.1.x?topic=install-choosing-installation-mode) - [https://www.ibm.com/docs/en/software-hub/5.3.x?topic=overview](https://www.ibm.com/docs/en/software-hub/5.3.x?topic=overview) - [https://www.ibm.com/docs/en/software-hub/5.3.x?topic=requirements-software](https://www.ibm.com/docs/en/software-hub/5.3.x?topic=requirements-software) - [https://www.ibm.com/docs/en/waasfgm?topic=overview-faq](https://www.ibm.com/docs/en/waasfgm?topic=overview-faq) - [https://cloud.ibm.com/docs/overview?topic=overview-compliance](https://cloud.ibm.com/docs/overview?topic=overview-compliance) - [https://www.ibm.com/cloud/compliance/soc-2](https://www.ibm.com/cloud/compliance/soc-2) For users in the EU/EEA, GDPR-compliant use in accordance with the best available practice documented on the IBM website is well-documented, because IBM provides a DPA for the processing of personal data and watsonx.ai can be operated not only as SaaS but also via the IBM Software Hub in a user’s own OpenShift environment or on-premises. In addition, EU regions are designated on IBM Cloud for watsonx.ai Runtime and watsonx.ai Studio. This provides a relatively straightforward path to data-sovereign usage within Europe, particularly through self-hosting or EU regions. However, a blanket statement that any standard SaaS usage is automatically fully GDPR-compliant cannot be substantiated in every detail on the website. **Positive** IBM outlines a data protection infrastructure that includes the IBM Privacy Statement and the IBM Data Processing Addendum. For watsonx.ai, EU regions on IBM Cloud are documented, specifically Frankfurt for watsonx.ai Runtime and watsonx.ai Studio. Additionally, watsonx.ai can be installed via IBM Software Hub on private, on-premises, or the customer’s own OpenShift infrastructure. In the FAQs, IBM also states that prompts do not train the foundation models and that customer data is intended to remain private to the account. **Negative** No standalone, product-specific subpage for watsonx.ai was found on the website that provides a clear, comprehensive EU/EEA data residency commitment for all pricing plans and all components. A specific, publicly accessible list of subprocessors specifically for watsonx.ai was not found; instead, IBM refers generally to DPAs, privacy information, and, in some cases, lists available upon request. An explicit opt-out mechanism is also not described separately on the website but is only indirectly addressed by the fact that, according to the FAQs, customer data is not used for model training. **Server Location** For the SaaS/IBM Cloud version, EU locations are documented on the IBM website, specifically Frankfurt (eu-de) for watsonx.ai Runtime and watsonx.ai Studio. An IBM article on watsonx on IBM Cloud also mentions secure facilities in Dallas and Frankfurt. When using the service via IBM Software Hub or on-premises, the customer determines the infrastructure location themselves. Other specific server locations for all watsonx.ai components are not consistently listed on the website. ## Hosting und Daten - **On-Prem / lokales Hosting:** abgedeckt - **Private Cloud / Rechenzentrum:** abgedeckt - **EU SaaS / Managed:** abgedeckt - **Hybrid:** teilweise / indirekt - **AVV / DPA:** abgedeckt - **Kein Training auf Kundendaten:** abgedeckt - **Open-Source / Transparenz-Pfad:** abgedeckt ## 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/