IBM made a bet most of its rivals could not: instead of renting an AI engine from someone else, it built its own. It owns the models (Granite, released open under Apache 2.0), the control plane that runs them (watsonx), the hybrid platform they sit on (Red Hat), and the real-time data plumbing underneath (Confluent). Where a typical integrator wires a client up to OpenAI or Google and bills for the labor, IBM can point at the whole stack and say, this is ours top to bottom.

That means IBM shows up in the places where who-owns-what actually matters: banks, governments, regulated industries that cannot hand their data and their decisions to a black box they do not control. It pairs roughly 150,000 consultants with its own software, so the same firm that designs the system also sells the engine it runs on. That pull-through, software dragging consulting and consulting dragging software, is the structural edge the rented-model crowd does not have.

The real catch: most of that $12.5B GenAI book is consulting work that was signed, not high-margin software revenue that recurs, so near-term AI money is still services-led and exposed to the same fee-deflation pressure hitting every integrator. watsonx is also narrow next to what the hyperscalers offer. And investors noticed the gap: IBM is not selling a finished product so much as a promise that the full stack pays off later.

Read against the share price: IBM briefly became the exception, re-rating to an all-time high of $329 on June 2 on the quantum rally, then gave back about 24% to roughly $249 by late June as the AI halo deflated. The structural advantage, owning the engine others rent, is intact. The premium investors briefly paid for it is not.