On August 24, 2026, during the annual Hot Chips conference, IBM unveiled a processor designed for the next generation of IBM Z and LinuxONE systems. Each of its cores can execute IBM Z instructions and Arm instructions and switch between the two modes within the nanosecond range. The company describes the design as the first main processor of its kind with dual architecture: rather than adding separate Arm cores alongside Z cores, it makes the same cores capable of running both architectures.
A Unified Design Instead of Side-by-Side Arm Cores
The processor contains 11 high-performance cores operating at a base frequency exceeding 5.7 gigahertz, and is built using 2-nanometer process technology. It also includes artificial-intelligence inference accelerators integrated into the processor, a unit dedicated to accelerating input and output operations, and a large cache-memory structure. IBM says complete systems will scale to hundreds of cores and tens of terabytes of memory.
Operation of the two environments relies on the open-source KVM hypervisor. This allows Arm64 Linux virtual machines to run alongside virtual machines running Linux on Z, while the hypervisor switches the core’s mode when assigning each virtual machine to the hardware. According to Christian Jacobi, an IBM Fellow and the technical leader for IBM systems development, the switch takes about a nanosecond, making its practical impact close to zero during the longer operating periods of a virtual machine.
Traditional z/OS workloads run in a separate partition outside KVM. This makes it possible to place critical transaction software, such as core banking and fraud-detection systems, alongside monitoring applications or services originally built for Arm on the same chip, while sharing the memory system and the platform’s reliability features.
What Changes in Practice?
The main goal of the move is not merely to add another instruction set architecture, but to reduce the need to port every application to s390x separately. Many monitoring and security tools, cloud software, and artificial-intelligence frameworks, such as PyTorch and ONNX Runtime, are strongly available in the Arm ecosystem. Arm says that more than 22 million developers use its ecosystem worldwide, and that approximately half of the computing capacity shipped to major cloud-service providers in 2025 was based on Arm, according to its estimates.
IBM says Arm Linux binaries are designed to be 100% compatible without modification, including applications running on Red Hat Linux for Arm. However, these are design promises associated with the announcement, not independent evidence of every application’s performance in production environments. In addition, the announced Arm support is currently limited to Linux and does not include running z/OS on Arm.
Artificial Intelligence Moves Closer to the Data
Alongside the processor, IBM is showcasing the next generation of its Spyre artificial-intelligence accelerator, with high-bandwidth memory and capabilities aimed at large language models and agent workflows. This complements the inference accelerator present in the processor since the Telum chip in 2022, which is used for low-latency tasks such as calculating a fraud score during transaction execution.
IBM believes that combining Arm and Spyre could allow artificial-intelligence tools to run close to the data residing on mainframe systems, rather than separating these workloads from systems of record. This is particularly relevant to organizations that rely on IBM Z for regulated transactions, although the source provides no independent performance results or final details about Spyre models and availability.
Launch Timing and Open Questions
The processor will appear in the successor to the z17, which shipped in the second quarter of 2025, suggesting an expected launch around 2028 based on IBM’s cycle of approximately three years. IBM emphasizes that the path does not mean abandoning the traditional Z architecture, but rather adding a new capability to a hardware roadmap that, according to Tina Tarquinio, extends from 10 to 15 years.
However, the crucial part has not yet been tested against actual customer workloads: running Arm instructions with production-level performance, along with fault detection and recovery at the level expected from mainframes. IBM also states that forward-looking statements represent goals and objectives only. Therefore, the significance of the announcement lies in integrating the two architectures at the core level, while production testing and final launch details will determine the extent to which the design can turn this promise into a practical option for organizations.