Plenty of companies have AI pilots. Far fewer have AI in production. IBM is wagering that its older strengths, including mainframes, hybrid infrastructure and a long record on governance, give it an edge as buyers shift their attention from experiments to results. That wager sits at the core of IBM enterprise AI strategy.
As first reported by SiliconANGLE, the pitch draws on a Deloitte finding that 54% of organizations plan to move at least 40% of their AI initiatives into production within six months, while only 25% have done so today. The gap between plan and practice is where IBM enterprise AI offerings are aimed.
Key Takeaways
- IBM enterprise AI focuses on bridging the gap between AI plans and production, with a strategy built on four pillars: agent orchestration, trusted data, intelligent operations, and governance.
- The company leverages its historical strengths, such as hybrid infrastructure, to ensure AI applications can access and utilize on-premises data efficiently.
- IBM acts as its own first customer, with more than 80,000 employees using AI tools, reporting significant productivity gains.
- Clients prioritize data sovereignty, requiring control over their AI workloads while reducing reliance on third-party infrastructure.
- Analysts see potential in IBM’s differentiated approach, but they question whether it will become the system of record for enterprise AI or just another layer in the tech stack.
Table of contents
The four pillars of IBM enterprise AI
The IBM enterprise AI operating model is built on four pillars: agent orchestration, trusted data, intelligent operations and governance. Each one maps to a product or acquisition announced over the past several months.
On orchestration, IBM partnered with Deepgram in February to add voice capabilities to watsonx Orchestrate, its platform for building, deploying and managing agents. In April, it made IBM Bob, an AI-assisted coding tool for planning, writing, testing, modernizing and deploying code, globally available.
For data, IBM completed its acquisition of Confluent in May. The aim is to feed streaming, real-time data into watsonx.data so that agents work from continuously updated context rather than stale snapshots.
For operations, IBM repositioned Concert at Think 2026. It was a point product. It’s now pitched as an agentic IT operations platform covering observability, optimization, security and automation across hybrid environments. Governance falls to IBM Sovereign Core, which builds compliance and operational control into the infrastructure itself.
The business is already sizable. IBM reported AI-related software sales and consulting engagements of more than $12.5 billion as of Q4 2025, up from $9.5 billion in the prior quarter.
Why old infrastructure matters to IBM enterprise AI
IBM chief executive Arvind Krishna told media ahead of Think that more than 70% of enterprise data still sits in on-premises systems. If that’s right, any AI plan that assumes everything moves to a public cloud is missing most of the data. Hybrid cloud lets companies point AI at that data without tearing out what they run today, which is why it’s central to IBM enterprise AI.
The recent numbers lend some support. Red Hat, whose OpenShift Kubernetes platform underpins IBM’s hybrid software, grew revenue 13% year over year. Revenue from IBM Z, the mainframe line, rose 51% year over year.
Ric Lewis, who leads IBM’s Infrastructure Division, said IBM is embedding AI directly into Z systems for demanding workloads. He also made a bold claim about the installed base: IBM handles 70% of the world’s financial transactions, and he called sovereignty its home court. Banks, insurers and other financial institutions run these machines on-premises, which is exactly where regulators and risk teams want sensitive data to stay.
Sovereignty moves up the agenda
In IBM enterprise AI, sovereignty comes down to customer control. Critical data and AI workloads stay in environments the customer owns and governs, instead of depending wholly on third-party infrastructure.
Javier Olaizola, a global managing partner at IBM Consulting, says clients now raise the topic at the start of an engagement. It used to show up around slide 40 of a pitch deck. Now the first question is how much exposure a company carries.
Sripriya Srinivasan, a general manager at IBM, pointed to a second driver beyond regulation: agent sprawl. As teams across a company build their own agents, chief information officers need standards, consistency, orchestration and governance. Sovereign Core is IBM’s answer to both pressures, and it’s meant to keep operations steady as AI rules and geopolitics shift.
IBM as its own first customer
IBM calls itself Client Zero for IBM enterprise AI, meaning it uses the tools internally before asking customers to trust them. Chief financial officer James Kavanaugh said AI now touches service delivery, the software portfolio, infrastructure and internal productivity.
The company offered some figures to back that up:
- More than 80,000 IBM employees use IBM Bob, and users report productivity gains of more than 45%.
- IBM Consulting has deployed more than 4,000 digital workers across 450 active projects.
- Those digital workers include agents from watsonx, Anthropic and OpenAI.
These are IBM’s own reported numbers, so treat them as claims rather than independent audits. Still, the management approach behind them is worth noting. Mohamad Ali, who heads IBM Consulting, said Krishna pushed him to treat digital labor the way a company treats people, with HR-style management applied to both human and software workers. For leaders wrestling with who is accountable when an agent makes a mistake, that framing is a practical starting point.
Customers and open questions
BNP Paribas offers an outside example of IBM enterprise AI ideas in practice. The bank has nearly 1,000 AI use cases underway, run under a group-level framework with governance spread across business units and functions. Its group chief technology officer, Jean-Michel Garcia, described the goal as using AI to serve customers better and help employees focus on the right tasks, not becoming an AI company.
IBM has also been adding partners. In June it joined OpenAI’s Daybreak Cyber Partner Program, which targets AI-driven cybersecurity threats, and it teamed with Google Cloud so IBM consultants can design, build and govern enterprise agents there. Those moves suggest IBM enterprise AI is meant to work across ecosystems, not only inside IBM’s own stack.
Analysts quoted in the report were positive but not uncritical. Dave Vellante of theCUBE Research said IBM has found a differentiated path by productizing workflow-level outcomes, though he acknowledged the stack can be confusing even if customers trust it. His colleague John Furrier framed the open question more bluntly: does IBM become the system of record for enterprise AI, or just another layer in the stack?
Read Next
A few related pieces worth your time:
- Jensen Huang Spy Report Points to a Megaspeed Executive
- Government Share of SpaceX Revenue Falls Below 10%
- OpenAI Acquires Pierre Computer to Strengthen Codex
What to watch next
Three things will tell us whether IBM enterprise AI holds together. First, whether the AI revenue run rate keeps climbing now that the Confluent deal has closed. Second, whether Concert and Sovereign Core find real buyers beyond IBM’s existing mainframe and Red Hat customers. Third, whether the product line gets simpler, since a confusing portfolio is the one weakness analysts flagged.
For business leaders, the takeaway is less about IBM than about the buying criteria it’s betting on. Governance, data control and measurable outcomes are becoming the questions that decide which AI projects reach production. IBM enterprise AI is a test of whether trust in existing infrastructure can count as a competitive advantage, and the next few quarters of results should show how well it works.
Source: SiliconANGLE
