Tech and AI
One processor core that executes two different instruction sets at the same time is the first result of a collaboration announced in April
By Staff Writer | 25 August 2026
IBM said on 24 August that it is building a dual architecture processor for its mainframes and Linux servers. There are no separate cores for the two platforms. Each core runs both, which is the part that had not been done before.
IBM announced at the Hot Chips conference on 24 August that it is designing a processor which will let its mainframe and Linux server systems run operating systems and applications built for two different architectures. It is the first result of a collaboration with Arm entered into in April.
The design is described as an 11 core part on a 2 nanometre process running above 5.7 gigahertz, with inference accelerators for fraud detection inside a transaction, a dedicated on chip unit for input and output, and a large cache. The company is explicit that it does not contain separate cores for each architecture: every core can execute both instruction sets concurrently, while keeping the encryption, error recovery and availability behaviour the platform is bought for.
This processor represents a significant architectural advancement for IBM Z and LinuxONE. By bringing Arm natively to our platform, we're combining access to one of the industry's fastest-growing software ecosystems with the qualities that have made IBM systems the foundation for how businesses run today.
Christian Jacobi, Chief Technology Officer and IBM Fellow, IBM Systems Development
What this answers is a migration argument, not a performance one
Mainframes still carry the transaction processing of banks, insurers, payment schemes and a good deal of government. The standing argument against them has never been that they are slow. It is that the modern software estate, the cloud native tooling and now the artificial intelligence libraries are all written for a different architecture, so anyone wanting to use those tools has to move the workload off the machine that already runs the ledger correctly.
A core that runs both instruction sets answers that by reversing it. The software comes to the machine. IBM puts the size of the addressable developer base at more than 22 million, and the practical claim is that Arm native Linux environments will run alongside the existing operating systems on the same silicon, with the platform's own reliability and encryption applied to both.
The correct reading of this announcement is a design, not a product. The release is written in the future throughout and carries an express statement that the company's stated direction may change or be withdrawn without notice.
The read across for a business sitting on old systems
Most large contractors and consultancies have at least one system nobody wants to touch: a costing engine, a payroll, a plant register, something written decades ago that works and that no one will sign off replacing. Every proposal to modernise runs into the same three numbers, the cost of the rewrite, the cost of the parallel run and the cost of being wrong.
The approach on display here is worth borrowing even by those who will never own this hardware. It treats the old system as the fixed point and the new tooling as the thing that must be made to fit, which is the opposite of how most replacement business cases are written. Where the old system is genuinely correct and genuinely understood, the cheaper project is usually the one that leaves it where it is and builds an interface around it.
Arm's own executive framed the same point from the other side, saying that as artificial intelligence scales more computing is converging on that architecture and that bringing it to these platforms extends that into regulated industries. Both parties are describing one movement: the workloads are not moving to the tools, the tools are moving to the workloads.