IBM held its annual IBM Think 2026 conference in Boston this week with its main message: enterprise AI needs a new operating model – control.
Taking the keynote stage Tuesday, IBM Chairman and CEO Arvind Krishna made a noticeably different argument than many of his Big Tech peers including Cisco, Alphabet and Microsoft.
To win the AI race, businesses must control the enterprise stack beneath AI itself — where AI runs, how data moves, how policies are enforced, how infrastructure behaves when regulations shift and how organizations avoid becoming trapped inside a single model ecosystem.
“Every nation, every enterprise, needs AI and cloud infrastructure that they control,” Krishna said. The technology underlying AI is as important to enterprise growth and national competitiveness as either finance or defense.
His definition of control was blunt: “Nobody else can turn it off, nobody else can tamper with it” when geopolitics, regulation or infrastructure failures intervene.
His argument was that the next phase of AI competition will be won in the plumbing: hybrid cloud, governed data, sovereign infrastructure, runtime policy enforcement and the ability to move across models without surrendering control.
Not coincidentally, those are also the areas where IBM has been quietly rebuilding momentum. IBM beat Wall Street expectations last quarter, driven largely by software, Red Hat and infrastructure growth. IBM Z mainframe revenue jumped 51%, reinforcing the company’s long-running thesis that hybrid cloud, automation and enterprise infrastructure still matter in the AI era.
“AI is not helping your business,” Krishna told attendees. “It is your business model.”
During the week of keynotes and product announcements, IBM put its operating model argument into concrete form — including:
- watsonx Orchestrate (unified orchestration for planning, building, deploying, and governing AI agents at scale)
- IBM Sovereign Core (now GA; embeds governance, compliance, and control directly into the AI runtime)
- IBM Enterprise Advantage (helps enterprises build and operate their own hybrid-AI platforms, not just consume IBM’s)
The company is attempting to establish itself not as the leader of the AI model race, but as the trusted operational layer beneath enterprise AI itself — where sovereignty, governance, runtime policy enforcement, and data control increasingly matter as much as the intelligence layer sitting on top.
The contrast with other recent tech keynotes was hard to miss.
Cisco has framed AI agents as a new workforce that must be secured. “AI agents aren’t just making existing work faster; they’re a new workforce of co-workers,” Cisco President and Chief Product Officer Jeetu Patel said earlier this year, positioning the problem around trust, zero trust and agent safety.
ServiceNow has centered its messaging around “governed autonomous work,” while Google Cloud continues pushing its “agentic enterprise” vision built around Gemini, cloud data and AI-native infrastructure. Microsoft, meanwhile, keeps embedding Copilot deeper into enterprise workflows and productivity software.
Those companies are largely selling what AI can do.
IBM is selling what enterprises need underneath AI before any of that scales safely.
The question, according to IBM, is not simply whether AI agents can be trusted. It is whether enterprises can trust the infrastructure, data architecture and policy controls beneath the agents themselves.
Krishna made that point directly when discussing why many enterprise AI initiatives stall.
“They often fail because of what’s underneath,” he said, pointing to “siloed data, fragmented infrastructure, multiple clouds, with no coherent operating models.”
IBM did have an AI agent story. It just was not the whole story.
The company’s most visible agent pitch was IBM Bob, an AI development platform Krishna said is already used by 80,000 IBM developers and designed to support architecture, code generation, testing, security and compliance across the software development lifecycle. But Bob mostly served as proof of a broader argument.
IBM’s larger Think 2026 pitch was an “AI operating model” built around watsonx Orchestrate for multi-agent coordination, Confluent for real-time data, IBM Concert for intelligent operations and IBM Sovereign Core for policy enforcement and sovereign infrastructure control.
IBM has a lot to sell into this version of the AI future.
Red Hat OpenShift gives it a hybrid cloud story. HashiCorp gives it an infrastructure automation story. Confluent gives it a real-time data story. watsonx gives it an AI governance and model management story. IBM Sovereign Core gives it a sovereignty story. IBM Concert gives it an operations story. Even IBM Z can be recast as AI-era transaction infrastructure for companies still running core business systems on mainframes.
None of that makes IBM’s argument wrong. It does make the argument commercially convenient.
If the AI race moves below the model layer, IBM suddenly has a cleaner way to connect its strongest businesses, protect slower-growing ones and explain to Wall Street why old enterprise plumbing still matters.
The sharpest part of the keynote may have been Krishna’s repeated attack on AI pilot culture.
Too many companies, he argued, are still confining AI to “little pilots and little projects” and “POCs and experiments.”
“The real question,” Krishna said, “is how deeply is AI embedded in your business processes?”
IBM is unlikely to dominate the frontier model race against OpenAI, Google, Anthropic or Microsoft. But Think 2026 made clear the company wants to be the Switzerland of enterprise AI infrastructure. It’s goal is to be enterprise’s open, hybrid, sovereign and deeply focused leader when it comes to managing the messy operational layer most AI hype tends to skip over.
Image of IBM Think 2026, courtesy of IBM
