IBM and Arm have teamed up to operate AI software on mainframes, but a date has not been specified yet.

      In summary: On April 2, 2026, IBM and Arm announced a strategic partnership aiming to facilitate the execution of Arm-based software on IBM Z and LinuxONE mainframes, which are critical for processing a significant portion of the world’s regulated enterprise transactions. This collaboration focuses on three main areas: virtualization to enable Arm software environments on IBM hardware, ensuring security and compliance for regulated sectors, and promoting long-term interoperability within the ecosystem. The aim is to integrate the Arm-native AI software stack—comprising frameworks developed for major cloud services like AWS, Google, and Microsoft—closer to the enterprise data housed on IBM Z systems, which cannot be migrated to public cloud infrastructures. IBM did not provide a shipping timeline. Both companies describe this collaboration as a future-oriented initiative rather than a reflection of currently available products.

      IBM and Arm are collaborating to bridge the gap between the most widely utilized AI software stack and the most reliable enterprise hardware. On April 2, 2026, this partnership was revealed, aiming to allow Arm-based software to operate on IBM Z and LinuxONE mainframes, which support the transaction processing needs of banks, government entities, and regulated businesses that are unable to transfer their data to the public cloud. This announcement signifies mutual recognition that the enterprise computing landscape demands the coexistence of these architectures on a unified platform.

      The challenge that the alliance seeks to address involves IBM Z and LinuxONE mainframes, which utilize IBM’s s390x architecture. The AI and cloud-native software ecosystem—which includes tools like PyTorch, TensorFlow, llama.cpp, ONNX Runtime, and Kubernetes container workloads—has largely been developed for x86 and increasingly for Arm. According to Arm’s estimates, nearly 50% of the compute delivered to major hyperscalers in 2025 was Arm-based, with AWS Graviton, Google Axion, and Microsoft’s AI efforts centered around Arm silicon. Arm has embedded its Kleidi AI libraries into various prominent frameworks. Consequently, while there is a robust ecosystem of AI tools that operate efficiently on Arm, transitioning them to s390x requires a lengthy and costly porting process that lags behind ongoing development.

      For enterprises that use IBM Z as their primary data system, managing transactions and sensitive customer information, there exists a growing disconnect. AI inference needs to occur near the data that resides on the mainframe, but the AI frameworks are designed for different architectures. Thus, the partnership aims to bridge this gap without compelling businesses to choose between their existing systems and access to contemporary AI software.

      The collaboration is structured around three key areas. The first is virtualization, which involves creating tools that enable Arm-based software to run on IBM Z and LinuxONE platforms without needing to convert applications to s390x. The second focus is security and compliance, ensuring that Arm workloads on IBM hardware adhere to the necessary standards for data residency, encryption, and availability required by industries like banking, government, and healthcare. The third area is ensuring long-term interoperability within the ecosystem, allowing for a broader range of software options across both platforms as the partnership develops.

      A notable point, clearly stated in IBM's press release, is that none of these features are currently available. IBM indicated that while specifics are still to come, the intention is for the same qualities—security, performance, resilience, and cost-effectiveness—that characterize IBM Z and LinuxONE to be extended to Arm64 workloads. There was no shipping date or technical details provided for the anticipated dual-architecture systems. The declarations from both companies convey objectives and intended trajectories, rather than available products for purchase.

      The announcement comes amid a hardware landscape that IBM has been developing for several years. The IBM z17 mainframe, which became generally available in June 2025, is powered by the Telum II processor, featuring eight cores at 5.5GHz, 360MB of L2 cache, and a 50% increase in AI inference throughput compared to its predecessor, the z16. IBM claims that the z17 can handle over 450 billion AI inference operations daily. The IBM Spyre Accelerator, which became available for z17 and LinuxONE 5 systems on October 28, 2025, includes 32 AI-optimized cores per card with support for int8 and fp16 data types, up to 1TB of system memory, and a maximum power draw of 75W per card, designed to enable large language models to run on-premises without the delays and costs associated with cloud-based inference.

      The Arm collaboration effectively represents the software layer being developed atop this hardware foundation. IBM has invested significant time in creating a mainframe suitable for large-scale AI operations. The critical question this partnership seeks to answer is whether the AI software enterprises want to deploy will be compatible.

      For IBM, this partnership mitigates a strategic risk. As AI inference workloads expand and companies aim to run models closer to their transaction data, the inability of IBM Z to directly execute the native Arm software stack presents a challenge