IBM Unveils Dual-Architecture Mainframe Processor With Native Arm Support

IBM has unveiled a groundbreaking mainframe processor at the annual Hot Chips conference. The new chip combines IBM’s proprietary instruction set with the Arm architecture, allowing enterprises to switch between both environments in nanoseconds.

Designed for the next generation of IBM Z and LinuxONE systems, the processor is the first dual-architecture mainframe chip announced by IBM. It is engineered to run the rapidly expanding ecosystem of Arm-native Linux software—including modern artificial intelligence frameworks—alongside the z/OS transaction-processing workloads used by banks, insurers, governments, and other highly regulated organizations.

“As enthusiasts of both technologies, we’re really excited about what will be one of the most powerful dual-architecture processors on the market,” Tina Tarquinio, chief product officer for IBM Z and LinuxONE, told VentureBeat in an exclusive interview ahead of the announcement.

The announcement represents the first major hardware milestone from IBM and Arm’s strategic partnership, announced in April. It also addresses a long-standing question surrounding mainframe technology: Can the systems responsible for processing many of the world’s regulated financial transactions remain central to enterprise computing in the age of AI, where much of the software ecosystem is built for alternative processor architectures?

IBM Builds Mainframe Processor With Two Native Instruction Sets

One of IBM’s most important engineering decisions was what it chose not to do. Rather than adding a group of separate Arm cores beside the mainframe processor, the company designed every core to support both architectures. This approach is more complex than conventional heterogeneous chip designs, but it provides deeper integration between Arm and IBM Z workloads.

“This chip has 11 cores, each with an Arm software mode and a traditional Z software mode,” explained Rob Jacobi in an interview with VentureBeat. “This allows us to run mission-critical enterprise software on the same chip, right next to the broader software ecosystem of Arm applications.”

The dual-architecture design works with the open-source KVM hypervisor. Enterprises can run Arm64 Linux virtual machines and Linux on Z virtual machines in parallel. When KVM assigns a virtual machine to a physical core, that core switches to the required instruction-set mode.

According to Jacobi, the performance impact is negligible. The architecture switch occurs on a nanosecond timescale, while virtual machines typically run for many milliseconds. As a result, the switching overhead is effectively amortized and has little influence on application performance.

Traditional z/OS workloads run in separate partitions outside KVM. This architecture could allow a bank’s core ledger, fraud-detection systems, and Arm-native monitoring or AI services to operate on the same processor, memory fabric, and security infrastructure.

IBM considered a simpler approach but rejected it because separate Arm cores would not provide the same consistency or quality of service as the company’s existing mainframe technology.

“If you simply put a few standalone Arm cores in the corner of a chip, you’re not really addressing customer requirements,” Jacobi said. “To provide the same quality of service that our clients are accustomed to, we needed to integrate Arm deeply into the overall system design.”

The processor is designed as a high-performance system rather than a limited compatibility feature. Built using a 2-nanometer process, it includes 11 high-performance cores operating at a base frequency above 5.7 GHz. The chip also features an integrated AI inference accelerator for applications such as real-time fraud detection, a dedicated data-processing unit for I/O acceleration, and a large cache architecture.

A complete IBM Z or LinuxONE system based on the technology could scale to hundreds of cores and tens of terabytes of memory. Jacobi described the processor as evidence that modern mainframes are not outdated systems but highly advanced platforms designed for demanding enterprise workloads.

Why Arm’s Developer Ecosystem Matters to IBM Z

The strategic value of IBM’s dual-architecture processor is primarily about software. IBM’s s390x architecture supports a significant share of the world’s mission-critical transactions. However, much of the latest enterprise software—including cloud-native middleware, security tools, monitoring platforms, container technologies, and AI frameworks—is developed first for x86 and increasingly for Arm.

Arm estimates that nearly half of the computing capacity shipped to major hyperscalers in 2025 was based on Arm architecture. Cloud providers such as AWS, Google, and Microsoft have all introduced Arm-based processors, while Arm’s global ecosystem includes more than 22 million developers.

Porting every application to s390x is difficult and time-consuming. IBM would need to work with thousands of independent software vendors to adapt their tools and platforms individually. Tina Tarquinio said IBM wanted to address this challenge through a fundamental hardware change rather than relying solely on ecosystem-by-ecosystem software porting.

“No matter how strong our ecosystem team is, we will never be able to work with everyone and port everything,” Tarquinio said. “There are a lot of software vendors, so we wanted to take a bold, fundamental step forward from a technology perspective.”

Customers are not necessarily asking for dual-architecture processors as an end in themselves. Instead, they want to bring new applications and supporting workloads into production more quickly.

“I don’t think our clients were saying, ‘Can you give us a dual-architecture environment?’” Tarquinio explained. “What they were saying was, ‘Please help us bring surrounding workloads and different types of workloads to market faster.’”

The compatibility goal is ambitious. Arm Linux binaries are expected to run without modification, provided they are built for the supported Arm environment.

“New Arm features are designed to be 100% binary compatible,” Jacobi said. “For example, if you deploy Red Hat Linux for Arm and build applications that run on Red Hat Linux for Arm, those applications will run on the system without modification.”

Arm defines the instruction set architecture and supplies verification tools to ensure IBM’s implementation behaves consistently with other Arm-based processors. IBM is responsible for designing and manufacturing the silicon.

Jacobi described the relationship between the two companies as a strong engineering partnership that combines Arm’s software ecosystem with IBM’s expertise in secure, reliable enterprise systems.

IBM Previews the Next Generation of Spyre AI Accelerators

IBM is also previewing a next-generation Spyre AI Accelerator at Hot Chips. The timing is significant because the new processor and the accelerator are designed to address different layers of enterprise AI workloads.

IBM’s current architecture already provides two AI processing options. The on-processor accelerator in the Telum II processor is designed for ultra-low-latency inference, including fraud scoring during payment transactions. Spyre accelerator cards, installed through the system’s I/O subsystem, are intended for larger and more demanding models.

The next-generation Spyre accelerator is designed to support significantly larger AI workloads, including language models used in agentic workflows and business processes.

“We are also introducing much more powerful chips that can run large-scale language models for agent workflows,” Jacobi said. These applications could include AI operations that manage IT systems, as well as business applications that analyze documents, support insurance decisions, and automate other complex processes.

The new accelerators will include high-bandwidth memory to help support large AI models and more intensive inference workloads.

This is where IBM’s Arm strategy and AI strategy intersect. Enterprises want to run AI inference close to their operational data, and much of that data already resides on mainframes. At the same time, many AI tools and frameworks are optimized for Arm-based environments.

Mohamed Awad, executive vice president of Cloud and AI at Arm, summarized the partnership’s potential: “As AI scales, more computing environments are converging on Arm. Bringing Arm compute and its software ecosystem to these platforms will extend that momentum into mission-critical enterprise infrastructure and give organizations greater choice in how they deploy AI.”

The move comes as enterprises transition from experimental AI projects to production deployments. McKinsey’s State of AI research found that 88% of organizations use AI in at least one business function, while nearly two-thirds have not yet scaled AI across the enterprise.

Organizations that achieve the greatest value from AI are increasingly redesigning core business processes instead of running isolated pilot programs. For regulated industries that rely on IBM Z as a system of record, running AI where transactions and critical data already exist could provide a more direct path to enterprise-wide integration.

When Will IBM’s Dual-Architecture Mainframes Ship?

Customers will need to wait before the dual-architecture processor reaches commercial IBM Z and LinuxONE systems. The chip is expected to follow the IBM z17, which shipped in the second quarter of 2025. If IBM maintains its traditional product cycle of approximately three years, the new platform could arrive around 2028.

Tarquinio emphasized that the project has progressed well beyond the concept stage.

“This is not simply something on the drawing board. We are fully committed to the entire system,” she said, adding that IBM plans to share additional details before the platform launches.

For existing IBM Z customers, the announcement raises questions about whether Arm support signals a reduced commitment to IBM’s traditional architecture. IBM executives strongly reject that interpretation.

“This is an ‘and,’ not an ‘or,’” Jacobi said. “We have a 10- or 15-year roadmap for our hardware systems. Many people are working on the next system, and many others are working on the systems that follow it.”

IBM views Arm support as another stage in the long-term evolution of the mainframe rather than a replacement for z/OS or the company’s proprietary architecture.

“The traditional mainframes we use today, such as the z17, are not simply faster versions of what we built 25 years ago,” Jacobi said. “We did not have extensive encryption capabilities or on-processor AI capabilities back then. Adding Arm support is the next major step in that continuing evolution.”

Why Enterprises Could Run Arm Workloads on IBM Z

The dual-architecture platform will compete not only with conventional servers but also with public cloud providers. When asked why an organization might run Arm workloads on a mainframe instead of deploying them on a hyperscaler, Tarquinio pointed to IBM Z’s reliability and availability.

She cited IBM Z availability of 99.999999%, commonly known as “eight nines,” which represents approximately 0.3 seconds of downtime per year. For applications such as financial ledgers, fraud detection, payment processing, and government systems, that level of availability can be more important than following the latest infrastructure trend.

The argument is ultimately about matching infrastructure to service-level agreements. Organizations with mission-critical workloads may value the mainframe’s security, resilience, centralized data access, and operational consistency when deploying Arm-native applications.

There are still important uncertainties. IBM’s press release notes that statements about future products and capabilities represent goals and objectives rather than guaranteed outcomes. Arm support is currently focused on Linux, and IBM must still demonstrate that an external instruction set can operate at production-level performance while meeting the mainframe’s demanding requirements for fault detection, recovery, security, and reliability.

Those capabilities will be tested under real customer workloads over the next several years. Software compatibility, performance consistency, workload management, and the economics of running Arm applications on IBM Z will all influence adoption.

Nevertheless, IBM’s ambition is clear. For decades, the mainframe has survived major technology shifts—from minicomputers and client-server computing to virtualization and the cloud—by incorporating the capabilities enterprises need.

IBM’s new dual-architecture processor represents its most significant attempt yet to connect the mainframe with the AI software ecosystem. By allowing IBM Z workloads and Arm-native applications to run on the same silicon, IBM aims to help organizations bring AI and modern cloud-native services closer to the systems that manage their most valuable data.

“Bringing this first-of-its-kind technology into production is another demonstration of what IBM can achieve from a technology perspective,” Tarquinio said.

The mainframe is not standing outside the future of enterprise computing. With native Arm support, IBM is attempting to make it an active part of that future.