Mainframes are finding a new life in the AI era, as some banks, insurance providers, and airlines look to use them to run AI locally rather than in the cloud
Belle Lin / Wall Street Journal :
Context & Ripple Effects
The renewed interest builds on a long-lived installed base: IBM mainframes remain a business mainstay among the roughly 10,000 systems discussed in earlier coverage. That makes local AI an extension of existing enterprise infrastructure rather than a wholesale replacement project.
It also follows a broader reassessment of where AI workloads run, after companies shifted some data and ML work back toward on-premises and hybrid infrastructure amid cloud-cost concerns.
First-order effects
- Banks, insurers, and airlines considering local AI gain a path to keep selected AI workloads closer to the systems and data already housed on mainframes, rather than routing all of them to public clouds.
- Mainframe operators and their technology partners face new demand to make legacy environments usable for AI workloads; cloud providers may lose some workloads that would otherwise have moved off-premises.
Second-order effects
- Cloud providers and on-premises hardware vendors will have to compete more directly on workload placement, integration, cost, and operational control, not simply on access to AI capacity.
- Enterprise AI programs are likely to become more hybrid: organizations can allocate workloads between mainframes, other on-premises systems, and cloud services rather than treating cloud deployment as the default.
Third-order effects
- If adoption broadens, AI infrastructure could become more heterogeneous, with established systems retaining strategic roles alongside cloud platforms instead of being displaced by them.
- The pattern would weaken the assumption that advanced AI must be centralized in a small number of large cloud environments, though it depends on whether local deployments can meet organizations' technical and operating requirements.
The trend: This is one data point in the reallocation of AI workloads across cloud, on-premises, and legacy enterprise infrastructure as buyers prioritize fit for particular data and systems.