IBM watsonx

IBM watsonx is IBM's enterprise artificial intelligence and data platform, announced in May 2023 at the Think conference and made generally available in July 2023. The platform is the successor to IBM's earlier Watson and Cloud Pak for Data products and is structured around three primary components: watsonx.ai for model training, tuning, and inference; watsonx.data for storing and accessing data through an open lakehouse architecture; and watsonx.governance for monitoring, documenting, and managing AI systems against policy and regulatory requirements. The platform runs on Red Hat OpenShift, supports deployment on IBM Cloud, AWS, Azure, and on premises, and is positioned for regulated industries that require auditable, hybrid-cloud AI.

Components

watsonx.ai provides a studio for prompt engineering, model fine-tuning, and inference serving. It hosts IBM's own Granite family of foundation models — designed for enterprise tasks such as code generation, summarisation, and tool-calling — alongside open-weights models from Meta (Llama), Mistral, and others, with paid commercial indemnification for selected models. The studio supports retrieval-augmented generation, function calling, and the InstructLab community-tuning workflow that allows organisations to add new knowledge and skills to base models incrementally.

watsonx.data is an open data lakehouse built on Apache Iceberg, Presto, and Spark that runs across on-premises and multi-cloud environments. Release 2.3, announced in 2025, added serverless Spark on IBM Cloud, a FinOps dashboard for cloud-usage visibility, and a Lakehouse assistant that lets analysts query data through natural language.

watsonx.governance captures metadata about training data, model versions, evaluation metrics, and approval workflows, supports automated bias and drift monitoring, and produces documentation aligned with frameworks such as the EU AI Act, the NIST AI Risk Management Framework, and ISO 42001.

2025 additions

At Think 2025 IBM announced several extensions to the platform. AgentOps brings a structured lifecycle to AI agent development, covering build, test, deploy, and monitor stages. The Model Gateway, in public preview from 12 June 2025, gives enterprises a single control plane to call models hosted on watsonx.ai or in third-party environments with consistent access controls and audit logging. The watsonx Orchestrate agent builder allows business users to assemble custom agents in under five minutes using no-code or professional coding paths.

Positioning

IBM positions watsonx as the AI platform of choice for regulated industries — banking, insurance, healthcare, telecommunications, and the public sector — where governance, hybrid deployment, and indemnified open models matter more than absolute model leaderboard performance. The platform competes most directly with Microsoft Azure AI Foundry, Google Vertex AI, AWS Bedrock plus SageMaker, and Databricks' Mosaic AI offering, while integrating with Red Hat OpenShift for customers standardising on hybrid Kubernetes infrastructure.

Granite foundation models

IBM's Granite family includes general-purpose, code, time-series, and geospatial models released under permissive open-source licences. Granite models are positioned as transparent, with disclosed training data sources and ongoing publication of model cards, and are commonly fine-tuned through watsonx.ai for enterprise tasks where smaller, lower-cost models can match larger frontier models on narrow domains.

References

1. IBM Corporation (2023). Introducing watsonx: The future of AI for business. ibm.com.
2. IBM Corporation (2025). Unlocking the future of AI development with watsonx.ai. ibm.com.
3. IBM Corporation (2025). watsonx.data v2.3 release notes. ibm.com.
4. Futurum Group (2025). IBM Think 2025 — Watsonx Platform Fuels Agentic AI and Hybrid Cloud Value. futurumgroup.com.