The Mainframe Is the Story at IBM Think 2026

May 15, 2026

Research By: Mark Tauschek, Helena Lang, Info-Tech Research Group

Think 2026 positions IBM’s AI products as middleware between modern tools and the mainframe customers who can’t leave. The agentic and integration announcements ride on top of a moat IBM has been deepening for forty years.

The GM of IBM Z Software, Skyla Loomis, could not have been more direct at the hybrid cloud keynote: “Activating AI on the mainframe is a strategic business decision. It is about removing friction, unlocking greater value from your applications and data, while protecting the core of trust and cost efficiency.”

“Mainframe” is not a term you’d expect to hear at a big software vendor’s annual flagship conference, at least not in our age of AI innovation and of hyperscalers racing to dominate AI compute. Although IBM’s Deep Blue was the first computer to beat a reigning world champion, Garry Kasparov, in a chess match under tournament rules in 1997 and IBM’s Watson beat Ken Jennings on Jeopardy! in 2011, when it comes to the big AI players today, IBM stands in their shadow.

Think 2026 was not about a new model, or even about the agentic innovations from orchestration capabilities to canned agents that were announced, but about how IBM is bolstering its strength in mainframes (specifically the Z series), which act as its moat.

To understand this, you had to understand and also somewhat look past the glitz of more than twenty product announcements during Think 2026. New products ranged from IBM Bob SaaS (set to compete with Copilot and Cursor) to IBM Concert (an agentic platform for hybrid operations), which, according to IBM Senior VP of Software and CCO Rob Thomas, is IBM’s reply to organizations that are looking to get control of their unknown vulnerabilities, a topic that should be fresh in everyone’s mind in light of frontier models such as Claude Mythos as well as projects such as Glasswing by Anthropic and Daybreak by OpenAI.

Thus, considering Loomis’s statement and where IBM sits in the AI race, the real focus was on how AI can work together with what IBM still owns. Among all the announcements, there was a red thread in the form of new AI products for the mainframe and for the regulated-industries franchise built around it.

Z Secretly Dominated the Agentic AI Announcements

Of more than twenty products announced at Think 2026, four were explicitly Z-targeted. IBM zSecure Secret Manager offers automated certificate lifecycle management for IBM z/OS environments and integrates with IBM Vault Self-Managed for Z. IBM Z Database Assistant does agentic operations for Db2 and IMS on Z. IBM SQL Data Insights Pro is an intelligence layer for Db2 for z/OS. And last but not least, Bob Premium Package for Z, an evolution of watsonx Code Assistant for Z, was announced in private preview, “with early users reporting 10x productivity gains” according to IBM. Add to that the Confluent product announcement, AI Editions of Core Software, and Vault Enterprise 2.0, which all mentioned Z but are framed to apply across multiple portfolios and clouds, and a Z-centric picture of the announcements emerges.

Z-oriented products, however, did not feature prominently in what IBM called its “biggest announcements” at Think 2026. Instead, the headline releases covered IBM Bob SaaS and IBM Concert, both mentioned above, as well as IBM Sovereign Core, which was made generally available with 160 compliance frameworks and partners such as AMD, Cegeka, Dell, Intel, Mistral, MongoDB, and Palo Alto Networks, clearly positioned ahead of the EU AI Act compliance deadline for high-risk AI systems on August 2, 2026. And Confluent, now part of IBM, offers native integration with watsonx and IBM Z in its data streaming platform.

Z Products? AI Interpreters!

Essentially, all the newly announced Z products act as interpreters between the modern AI tooling that enterprises want to use and the mainframe estate that they currently run. One handles languages relevant for IBM Z that mainstream AI coding assistants have not been trained much on and brings an integrated development environment to coding for z/OS applications (Bob Premium Package for Z). zSecure Secret Manager repurposes a tool originally designed for the cloud-native world for z/OS certificate lifecycle management. Db2 and IMS, two mainframe data stores that had not previously been well-integrated into the rest of the AI ecosystem, get an agentic interface through Z Database Assistant. And SQL Data Insights Pro makes it possible to conduct AI-enabled data analysis directly on Db2 for z/OS while the data stays there.

Modern AI tools were built for x86 and for the data formats that live in them. Without a translation layer, a customer running their core business on a mainframe faces two choices: They can either keep AI confined to whatever runs alongside the mainframe, which keeps the mainframe isolated from the AI stack, or they can invest in a multiyear migration off the mainframe entirely. But this latter option moves the AI investment to someone else’s platform and keeps the business in rebuild mode for years before the payoff arrives. In letting the AI work directly on the mainframe, the four Z products let you have your cake and eat it too.

Why the Lock-In Holds: Forty Years of Accumulated Code

What makes this strategy possible is that banks, insurers, healthcare payers, and large public-sector institutions sit on millions of lines of legacy code in old languages that they have accumulated over forty-plus years of business. If they were to migrate those stacks to x86 or pure cloud-native environments overnight, the cost would be astronomical, not to mention the execution risk of rewriting a mission-critical settlement system from scratch. In short, this would be a mammoth project only further compounded by the regulatory friction in banking, insurance, and healthcare. And the Db2-on-Z customer base in particular is concentrated in regulated industries where the lock-in is most durable.

What Think 2026 announces, then, is that IBM intends to reinforce that lock-in by making the mainframe a first-class destination for agentic AI so that the next decade of AI investment from these customers gets routed through, rather than around, IBM.

The Coda: Seven Years of Acquisitions on Stage Together

An open question remains, however. Can the rest of the stack fit together in the same way? After all, the agentic and platform announcements tie together the numerous acquisitions IBM has been digesting for seven years. Concert, for example, consolidates four products that were never one platform: Instana (acquired in 2020), Turbonomic (acquired in 2021), SevOne (acquired in 2021), and IBM’s own Cloud Pak for AIOps. Then there is Confluent, of course, IBM’s latest acquisition and now prominently featured as a recent announcement, but also Red Hat, Apptio, HashiCorp, and webMethods. Think 2026 is the first event where IBM is showing how those pieces can work together and not just sit there by themselves.

At the very least, the agentic announcements are an attempt to extend IBM-style integration to a multivendor stack. In a Fox Business interview during Think 2026 week, IBM CEO Arvind Krishna pushed an agnostic-orchestration framing, with the idea being that agents can come from anywhere and that IBM will provide the layers of governance and policy for them.

Krishna complicated this picture by deliberately not showing an architecture slide in the Think 2026 keynote, which is an unusual omission for a vendor making a stack-level claim of this kind. And in the past, IBM has bought excellent technology and then spent years, sometimes decades, before weaving it into the rest of the company, a fate that Watson Health and Cognos went through. Whether the acquisitions that are part of this newest agentic stack break that pattern is a question for after Think 2026.

Strategic Implications for IT leaders

• The “legacy” reframe. Mainframes are not legacy in any technical sense, even if they might have that reputation. They are modern transaction-processing platforms that are now getting agentic AI overlay.

• Vendor concentration in regulated workflows. Bob Premium for Z and Sovereign Core both push IBM deeper into customer infrastructure that’s operationally and legally hard to exit.

• Sovereign AI is the regulated industry lever. The EU AI Act will be in full effect on August 2, 2026.

• The contestable claims. Bob’s competitive position against Cursor and GitHub Copilot is as yet unproven. Watsonx Orchestrate as “agentic control plane” depends on Microsoft, Salesforce, and ServiceNow letting their agents be orchestrated by an IBM layer, which is unlikely.

Our Take

• Treat the mainframe announcements as the actual differentiator and the agentic and integration announcements as something IBM must ship to stay in the conversation.

• Revisit any “off mainframe by 20XX” roadmap your organization has been carrying. The cost-benefit math has changed.

• Reframe IBM in vendor evaluations as a regulated industry AI vendor, not a generalist agentic AI competitor.