IBM is pushing generative AI, the kind that produces text and code from a prompt, into the core of how large companies run their operations.
Chief executive Arvind Krishna casts the effort as a redesign of the operating model rather than a scatter of standalone trials.
The distinction matters, because it commits IBM to rebuilding the plumbing of enterprise software rather than bolting AI onto the side.
The bets behind the pitch
The strategy rests on product and infrastructure moves.
IBM bought Confluent in March 2026 and wired it on day one into watsonx.data and its other integration software.
It launched two enterprise platforms, Sovereign Core and Enterprise Advantage, in February.
In August it released Granite 4.2, a set of open-weight models, the kind whose parameters are published so customers can run them on their own machines, built to power AI agents that act on a user's behalf.
Traction, and a repricing
The commercial signals point up.
IBM's watsonx and agent products have reached multibillion-dollar scale in recent quarters, and generative AI makes up a growing share of new contracts and order backlog.
Investors have been less patient with the cost of getting there.
The heavy spending cycle drew a sharp repricing of the shares, wiping out a chunk of market value even as the sales numbers climbed.
The test ahead
The strategy now turns on execution.
Customers have to adopt these joined-up agent and data workflows at scale, not just trial them.
And IBM has to show the Confluent and Granite integrations deliver a measurable return, the step from proof of concept to something that actually runs the business.
Until that evidence lands, the redesign is a credible plan rather than a proven one.
