IBM

From Watson to Granite

Founded: 1911 (AI efforts from 2000s) | HQ: Armonk, New York | Key People: Arvind Krishna (CEO), Dario Gil (SVP & Director of Research) | Valuation: ~$170B market cap (public) | Flagship Models: Granite 3.2, Granite 4.0, watsonx.ai, watsonx Orchestrate

The Origin Story

IBM's relationship with artificial intelligence predates nearly every company in this encyclopedia. The story begins in earnest with Deep Blue defeating Garry Kasparov at chess in 1997, but the chapter that defined IBM's AI identity was Watson's victory on Jeopardy! in February 2011. Named after IBM's founder Thomas J. Watson, the system defeated champions Ken Jennings and Brad Rutter using natural language processing on a scale never before demonstrated publicly. The triumph launched a thousand press releases and a billion-dollar bet on Watson as a platform, with IBM pouring over $15 billion into its Watson Health division and making bold promises about revolutionizing oncology, drug discovery, and clinical decision support. The reality fell catastrophically short. Watson for Oncology, developed in partnership with Memorial Sloan Kettering, was criticized for providing treatment recommendations that were often unsafe or irrelevant, trained primarily on synthetic cases rather than real patient data. By 2022, IBM had sold Watson Health to Francisco Partners for a reported fraction of its cumulative investment, marking one of corporate AI's most expensive cautionary tales.

Key Milestones

The Watson Health collapse forced a strategic reckoning. When Arvind Krishna took over as CEO in April 2020, he pivoted IBM decisively toward hybrid cloud and a more pragmatic AI philosophy. The launch of watsonx in May 2023 marked the formal beginning of IBM's second act in AI: a full-stack enterprise platform combining watsonx.ai for model development, watsonx.data for data management, and watsonx.governance for AI oversight. The Granite family of large language models debuted in September 2023 as IBM's answer to the generative AI wave, but with a distinctive twist — they were released as open-source models under permissive licenses, a stark departure from IBM's historically proprietary approach. Granite 3.0, released in October 2024, established the family as a serious enterprise contender, with Signal65 validation showing competitive benchmark performance against models from Meta and Google across reasoning, RAG, and safety evaluations. Granite 3.2, launched February 2025, added multimodal vision capabilities and reasoning with inference scaling. Granite 4.0, released October 2025, introduced a hybrid Mamba-2/Transformer architecture designed for efficiency, with Nano models small enough to run locally on edge devices. IBM also joined the InstructLab project to enable community-driven model improvement, signaling a commitment to open-source governance alongside commercial deployment. Gartner recognized IBM as a 2025 AI vendor leader, specifically citing the Granite-watsonx combination for domain-specific language models.

IBM Think 2026 and the AI Portfolio Expansion

IBM Think 2026, held May 4-7 in Boston, marked the most comprehensive expansion of IBM's AI portfolio in a single event. The centerpiece was the next-generation watsonx Orchestrate, repositioned as a multi-agent control plane for enterprises — a platform for coordinating multiple AI agents across business processes, with governance, observability, and orchestration built in. IBM articulated an "AI Operating Model" built on four pillars: agents (via watsonx Orchestrate), data (via the completed Confluent acquisition, integrating real-time data streams with AI agents), automation, and hybrid infrastructure (including IBM Sovereign Core for sovereign AI deployments).

The Confluent acquisition, completed ahead of the event, gives IBM a real-time data backbone for its agent platform — agents that can act on live streaming data rather than batch-processed snapshots. IBM Sovereign Core addresses the growing demand from governments and regulated enterprises to run AI infrastructure within national borders, free from foreign cloud dependencies. In partnership with Datavault AI, IBM committed $5 million and 20,000 hours to accelerate AI adoption in data-intensive industries.

Current Position

IBM now occupies a distinctive niche: the pragmatic enterprise AI vendor with a comprehensive multi-agent strategy. Unlike OpenAI and Google, which chase frontier model performance at massive cost, IBM positions Granite as “right-sized” models — efficient, auditable, and deployable on-premises or in private cloud environments where regulated industries need them. The watsonx Orchestrate platform pushes IBM beyond individual model deployment into the higher-value territory of agent orchestration, where the real enterprise spend is heading. The strategy targets financial services, government, and healthcare, sectors where IBM's existing consulting relationships and mainframe installed base provide natural distribution. IBM Consulting, with its 160,000-plus consultants, serves as the deployment engine, wrapping Granite models and watsonx Orchestrate in industry-specific solutions. Revenue from IBM's AI business is reported as part of its Software segment, making exact figures opaque, though the company claims a multi-billion-dollar AI bookings pipeline. The competitive challenge remains significant: Microsoft and Google offer enterprise AI platforms with far larger ecosystems, while open-source alternatives from Meta compete directly on Granite's open-source positioning.

What Leaders Should Know

IBM's Granite story is ultimately about risk management and trust. For organizations in regulated industries where model provenance, auditability, and intellectual property safety matter more than raw benchmark scores, IBM offers a credible alternative to hyperscaler AI. The open-source licensing of Granite models provides deployment flexibility that proprietary alternatives cannot match. The new watsonx Orchestrate multi-agent control plane addresses the next phase of enterprise AI adoption — orchestrating multiple specialized agents rather than deploying a single chatbot — and IBM's consulting-led approach gives it a deployment advantage that pure software vendors lack. The Confluent integration means IBM can now offer real-time data-driven agent workflows end-to-end. However, IBM's track record with Watson Health is a reminder that enterprise AI credibility is earned slowly and lost quickly. The question for executives is whether IBM's consulting-led approach and hybrid cloud infrastructure can convert AI promises into production deployments more effectively this time around — or whether the company remains a fast follower in a market that increasingly rewards first movers.