IBM Pitches an AI Operating Model at Think 2026 With Next-Gen watsonx Orchestrate and Sovereign Core
CEO Arvind Krishna unveiled an agentic control plane, made Sovereign Core generally available and folded the Confluent acquisition into watsonx.data as IBM argued enterprises now need an entirely new model for running AI.
IBM used its Think 2026 conference on May 5 to deliver what it called the most comprehensive expansion of its enterprise AI and hybrid cloud portfolio to date, anchored by a new operating model designed for organizations that have moved past pilots and are now trying to run AI as a critical production system. CEO Arvind Krishna framed the message bluntly: running AI in the enterprise requires a new operating model, with rigor comparable to the way companies manage their most critical infrastructure.
At the center of the announcement is the next generation of watsonx Orchestrate, now in private preview and repositioned as an agentic control plane. The product is meant to address what IBM argues is the fastest-growing pain point in enterprise AI deployments: managing hundreds of agents built by different teams, on different frameworks, with consistent policy enforcement and auditability. Crucially, the new Orchestrate is model-agnostic. IBM is explicitly supporting agents built on Anthropics Claude, OpenAIs GPT, internally trained IBM models, or any combination, with governance applied uniformly regardless of which underlying model is in play.
The companion piece is IBM Sovereign Core, which moves to general availability after months of regulated-industry trials. Built on Red Hat OpenShift and Red Hat AI, Sovereign Core embeds governance, compliance and AI execution controls directly into the infrastructure runtime, rather than bolting them on at the application layer. IBM is shipping it with an extensible catalog of vetted partner software from AMD, Intel, Dell, MongoDB and Palo Alto Networks, a deliberate signal that the platform is designed to extend across hybrid and partner environments rather than lock customers into a single stack.
Real-time data is the other leg of the framework. IBM is folding its Confluent acquisition, which brings managed Apache Kafka and Apache Flink into the watsonx fabric as IBM Confluent, and pairing it with a new context layer in watsonx.data that adds semantic meaning and explainability to the streams flowing into agents. The Concert platform, also unveiled in public preview, takes the same idea into operations, moving teams from passive monitoring to coordinated AI-driven response and including a Concert Secure Coder module that embeds security guidance directly into developer workflows.
The strategic frame Krishna offered is a four-pillar model: agents that execute and adapt across the business, data that is connected and real-time, automation that scales end-to-end, and hybrid that delivers operational independence and sovereignty. It is a deliberately unsexy pitch in a market currently transfixed by trillion-parameter models and 900-billion-dollar valuations, but it is aimed squarely at the CIOs and CISOs writing the checks. As the gap widens between enterprises that have industrialized their AI deployments and those still stuck in proof-of-concept purgatory, IBMs bet is that the winning vendor in the next phase is the one that makes governed multi-agent operations boring, repeatable and auditable.
