IBM opened Think 2026 in Boston on May 5 with a sharper enterprise AI pitch.
CEO Arvind Krishna framed the shift this way: running enterprise AI now requires a new operating model, with the same discipline, governance and scale that companies apply to their most critical infrastructure.
That is a different message from the earlier watsonx era, when IBM sold the stack mainly as an enterprise AI platform. This year the company put a name on the larger claim: the AI Operating Model.
The reason is simple: the deployment problem has changed.
Who governs thousands of agents?
Inside large companies, agent counts have moved from a handful to dozens across teams, platforms and models. The operational question is no longer just how to build agents, but how to know how many are running, who approved them and who is accountable when they make a bad decision.
IBM's argument is that companies are moving from deploying a few agents to governing thousands, where near-real-time oversight and auditability become the hard part.
The main product move at Think is aimed at that layer. The next generation of watsonx Orchestrate, now in private preview, is being positioned as an agentic control plane: a place to connect agents from different platforms and vendors under one policy framework. Alongside it, IBM Bob, the developer partner agent focused on security and cost awareness, is now generally available for production use.
Concert Secure Coder is being pushed directly into the development workflow, so vulnerabilities can be addressed while code is written rather than after a security handoff.
In plain terms, IBM wants AI governance to move upstream from after-the-fact audit to the point where software is created.
IBM also filled in the data layer
Agents do not matter much if they cannot reach the right operational data. IBM used Think to show a broader stack around streaming, lakehouse search, operations and sovereign infrastructure.
| Platform | Role | Status |
|---|---|---|
| IBM Confluent | Real-time streaming on Kafka and Flink foundations | Acquired and integrated |
| watsonx.data | OpenRAG, OpenSearch and GPU-accelerated Presto | Private preview |
| Concert | AI-powered operations platform | Public preview |
| Sovereign Core | Sovereign infrastructure layer | GA |
| IBM Bob | Enterprise agent development partner | GA |
One watsonx.data proof point stood out: in a Nestle proof of concept spanning data marts across 186 countries, IBM said costs fell 83% while price performance improved 30 times. IBM presented this as a production-environment result, not a synthetic benchmark.
Sovereign Core may be the more strategic product
Sovereignty is becoming a hard requirement in AI. Europe wants local data control, Middle Eastern buyers are pressing for data residency, and countries including Canada and India are debating AI sovereignty more explicitly.
IBM Sovereign Core turns compliance and operational independence at the infrastructure layer into a product. The partner list is revealing: AMD, Dell, Intel, Mistral, MongoDB, Palo Alto Networks, Red Hat, Cloudera, Elastic and Confluent.
Mistral's presence matters. It points to a model in which IBM acts as the prime contractor for sovereign AI programs, while regional model providers supply part of the stack.
The AI Divide is IBM's real theme
Krishna repeatedly used the phrase “The AI Divide”: the gap between heavy AI spending and limited business return. IBM is telling CIOs that AI should move from pilot budgets into core systems investment.
That framing plays to IBM's strengths. Compliance, disaster recovery, audit trails and long-lived enterprise infrastructure are areas where AI-native vendors often still have less operational history.
Why this move matters
In the short term, the test is whether watsonx Orchestrate and Sovereign Core can land with customers at scale. Over the longer run, IBM is making a plausible bet: governing thousands of agents may become an enterprise pain point before building stronger agents does.
Salesforce with Agentforce, Microsoft with Agent 365 and IBM with watsonx Orchestrate are all competing for that control layer. IBM's edge is its installed base and its credibility in compliance-heavy environments.
Whether the AI Divide narrows will be easier to judge when IBM reports later watsonx revenue numbers.
Sources: Think 2026: IBM Delivers the Blueprint for the AI Operating Model as the AI Divide Widens (IBM Newsroom); IBM Combines AI Operations, CocoLoop, Sovereign Infrastructure, and Quantum Drug Discovery Progress at Think 2026 (StorageReview); IBM rolls out tools to run thousands of AI agents with governance (Stocktitan)