IBM's generative AI book of business passed USD 12.5 billion, led by consulting.
Orchestrate, Bob, and Granite cover agent control, coding, and open models.
Growth is uneven, with 2026 revenue guidance now at 4% to 5%.
Arvind Krishna has led IBM since April 2020, and his view of generative artificial intelligence differs from most rival companies. He has been focusing on ensuring AI fits within daily work, such as coding, HR, and IT support.
The numbers explain why this approach matters. In January 2026, IBM said its generative AI book of business passed USD 12.5 billion. About four-fifths came from consulting, which shows clients are paying IBM to build working systems. This article looks at how Krishna's strategy shapes IBM's platforms, tools, and models.
Arvind opened IBM Think 2026 in Boston, which drew more than 5,000 leaders from over 80 countries. According to him, companies pulling ahead are redesigning how they operate, rather than just adding more AI. IBM calls this the AI operating model. This model has four parts:
Agents: coordinated AI that carries out and adapts work across the business.
Data: real-time, connected information that gives teams a shared view.
Automation: end-to-end workflows that scale across processes.
Hybrid: control over sovereignty, governance, and security wherever AI runs.
Krishna also expects generative AI models to keep leapfrogging one another and become interchangeable. In his view, value moves to the platform layer that handles governance and integration. This thinking explains why IBM builds around model choice rather than a single model.
Also Read: IBM CEO Arvind Krishna Explains Letter Behind Historic 25% Stock Plunge
Many firms now run agents built by different teams on different systems. Tracking who built each one and what data it touches gets difficult fast. IBM's answer is the next generation of watsonx Orchestrate, now in private preview. This AI platform acts as a control plane, applying consistent policies to agents from any source.
IBM Bob became generally available on April 28, 2026. It supports planning, coding, testing, deployment, and modernization. Bob began with 100 developers in June 2025 and now serves over 80,000 IBM employees.
Surveyed users report a 45% average productivity gain, though the figure is self-reported. At Blue Pearl, a Java upgrade that once took 30 days finished in three. Bob also routes tasks across frontier models and IBM's Granite models, based on cost and accuracy.
Granite 4.0 arrived in October 2025 with a hybrid Mamba and transformer design. IBM says it cuts memory needs by more than 70% and speeds up inference.
The family is open under Apache 2.0. It was also the first open model family to earn ISO 42001 certification. Regulated sectors such as banking and healthcare value these traits.
Agents perform only as well as the data they can reach. IBM's roughly USD 11 billion Confluent acquisition adds real-time streaming built on Kafka and Flink to watsonx.data.
In a Nestlé proof of concept, GPU-accelerated Presto delivered 83% cost savings on a data mart spanning 186 countries. Governance gets equal attention. IBM Sovereign Core, now generally available, embeds policy at the infrastructure level for regulated firms.
Krishna asked IBM to prove its tools internally first, under a program called Client Zero. IBM reports USD 4.5 billion in productivity gains from AI, automation, and hybrid cloud.
Its AskHR agent resolves 94% of common employee inquiries, and support tickets fell 75% from historical levels. AskIT cut calls and chats to IT support by 74% since its 2023 launch. Krishna has also said IBM's total headcount rose, as savings funded more hiring in programming and sales.
Progress has not been smooth. In the second quarter of 2026, software revenue rose 5% to USD 7.8 billion, and the Data segment grew 19%. Consulting stayed flat at USD 5.3 billion.
IBM also trimmed its 2026 outlook to 4% to 5% constant currency revenue growth. Its own CEO study found only 25% of AI initiatives delivered expected returns. Some investors also worry that generative AI could shrink the COBOL modernization work IBM has long sold.
Also Read: IBM AI Certifications: Best Courses for Beginners and Professionals
Start with one cross-functional workflow where AI can shorten cycle times.
Govern every agent from day one, with clear owners and audit trails.
Keep model choice open, since leading generative AI models change often.
Test new tools inside your own teams before selling or scaling them.
Krishna's strategy rests on one idea. Generative AI creates value only when it sits inside real workflows with clear controls. IBM's platforms, tools, and models each serve that aim, from Orchestrate's oversight to Bob's coding help and Granite's efficiency.
The next test is scale. Enterprises now want proof, and IBM's guidance shows growth is not automatic. Still, a company that trusts its own tools first has a credible story. If Krishna's model holds, generative AI will feel less like a project and more like routine practice.
1. Who is Arvind Krishna, and what is his role at IBM?
Arvind Krishna is IBM's chairman and CEO. He joined the company in 1990, became CEO in April 2020, and added the chairman role in January 2021. He now steers IBM's software-led AI strategy.
2. What is IBM's generative AI book of business?
It tracks software sales, new SaaS contract value, and consulting signings tied to generative AI offerings. IBM reported over USD 12.5 billion at the end of 2025. It is not revenue.
3. What does watsonx Orchestrate do?
It is an AI platform that manages agents across an enterprise. The next generation, now in private preview, applies consistent policies and accountability to agents from any source, helping companies govern them at scale.
4. What is IBM Bob?
Bob is an agentic development partner that became generally available on April 28, 2026. It supports planning, coding, testing, deployment, and modernization, and more than 80,000 IBM employees already use it.
5. Why do Granite generative AI models matter to enterprises?
Granite 4.0 models are open under Apache 2.0, and IBM says they need over 70% less memory than comparable models. They also hold ISO 42001 certification, which supports governance needs in regulated industries.