Bridging the gap between cutting-edge research and enterprise-grade infrastructure, our collaboration between research and IBM Z is accelerating AI innovation in bringing a new era of optimized business solutions. These innovations span across hardware AI inferencing acceleration via IBM Spyre Accelerator, enhanced AI platform capabilities and tooling for ecosystem and clients.

Building out Agentic AI futures with Accelerated Compute on Spyre

The IBM Spyre Accelerator and Telum II processor with key developments from IBM Research were first announced at HOT CHIPS semiconductor conference in August 2024, marking a new wave of optimizations across the platform. Telum II has been enabling predictive AI acceleration since its release to the public in 2Q 2025, marking billions of inferences per day with <1ms response time, targeted for 100% in-transaction analysis. Now the continuous delivery waves continue, with the availability of Spyre Accelerator for IBM Z and LinuxONE and a wealth of platform capabilities optimizing LLMs and generative AI for business ROI.

Client interests around generative AI often center around streamlining business operations, considering chatbot-enabled assistants and agents to free up time for more strategic work, or speed new employee training.  Other client interests are clearly rooted in optimizing business decisions that improve the bottom line, considering enhanced insurance approval automation, advanced risk classification analysis with multiple AI model inputs, or expanding end-client offers and revenue including document AI techniques.

IBM introduced agentic AI capabilities in the latest release of IBM watsonx Assistant for Z, purpose-built to simplify IBM Z IT operations. These new features help IT operators resolve issues more efficiently by understanding conversational context, reasoning through multi-step interactions, making goal-driven decisions and automating complex workflows. This represents a meaningful shift from reactive troubleshooting to proactive, intelligent system management. IBM watsonx Assistant for Z powers the AI chat and agent runtime experience behind products like IBM Concert® for Z, enabling context-aware responses that support effective incident remediation. Read the full blog here to see the full suite of agentic AI on Z announcements.

AI Acceleration and Hybrid Flexibility with IBM Z Qualities of Service

The Spyre Accelerator for IBM Z enables generative AI such as watsonx Assistant for Z and the assistants and agentic AI capabilities to run with the IBM Z qualities of service. Generative AI on premise offers strategic and operational benefits, especially for enterprises with strict requirements around data security, compliance, and performance. Sensitive data or prompt inputs do not need to leave your environment – thus, reducing exposure to external threats.  This can be ideal for industries like finance, healthcare, and government with highly regulated environments.

This enables full control over access, encryption, and audit trails over the IBM Z and LinuxONE environments. It also allows the ability to integrate proprietary data and domain-specific knowledge without external dependencies. Hosting the data and models on premises makes it easier to meet industry-specific regulations. Finally, on-prem deployments can reduce the latency for real-time applications when the AI is hosted where the data and applications reside, on the performance, availability, and resiliency of the mainframe.

Continuous deliveries and enhancements in AI platform and tooling capabilities

For AI software and tools across IBM Z and LinuxONE, you can expect waves of continuous deliveries (see statement of direction)* with further AI capabilities and enhancements.

·      AI Toolkit for IBM Z and LinuxONE 

       IBM AI Toolkit for IBM Z and LinuxONE components Including PyTorch, TensorFlow, TensorFlow Serving, Triton Inference Server, and SnapML will transition to Universal Base Images (UBI) from Ubuntu in 4Q 2025 enhancing portability, security compliance, and consistency across IBM Z and LinuxONE environments. Building on this foundation, the AI Toolkit for IBM Z and LinuxONE will deliver capabilities to fully exploit the Spyre Accelerator in 2026.

·      Machine Learning for IBM z/OS

Machine Learning for IBM Z (MLz) continues to actively collaborate with clients to drive enhancements that align with evolving AI workloads and enterprise needs. IBM intends to deliver capabilities to exploit the Spyre Accelerator with Machine Learning for IBM z/OS.

·      Red Hat OpenShift AI

IBM and Red Hat intend to support deployment of Red Hat Openshift AI across IBM LinuxONE and IBM Z (via Linux on IBM Z, and z/OS Container Extensions) for a broad spectrum of options to deploy and manage the full lifecycle of predictive and generative AI models at scale.

·      Red Hat AI Inference Server

IBM and Red Hat intend to support Red Hat AI Inference Server on IBM LinuxONE and IBM Z (via Linux on IBM Z and z/OS Container Extensions) to deliver vLLM benefits for optimized generative AI. As part of z17 and IBM LinuxONE 5, the Spyre Accelerator is designed for vLLM and LLM usage, enabling on-premises integration of AI solutions on a platform designed for mission-critical performance.

·      IBM Z Platform for Apache Spark (zSpark)

For large-scale data processing of batch use cases with z/OS data, IBM Z Platform for Apache Spark (zSpark) is a high-performance execution engine to perform in-memory computing. In this quarter’s release of zSpark 1.4, the Z supported version now aligns with Apache Spark 4.0.

·      IBM Synthetic Data Sets Core Banking and Money Laundering

In this quarter’s update of IBM Synthetic Data Sets, the Core Banking and Money Laundering dataset now includes artificially generated data designed to be similar to data used by Peer-to-Peer (P2P) institutions: Venmo, Zelle, PayPal, Cash App, ApplePay, GooglePay, and MetaPay. Clients can use this artificial data to train fraud detection models to better identify patterns of P2P and instant payment fraud.

Ecosystem and client highlights in design thinking

Our partnership and collaborations drive innovation every day, especially across our software ecosystem, such as a recent collaboration with Quantexa.  Quantexa Decision Intelligence Platform on IBM Z allows organizations to “turn fragmented data into connected, contextual insight. Using AI-driven entity resolution and network analytics, it unifies data across silos of data sources to reveal hidden relationships, patterns, and risks… right where their most critical data already resides and is being processed.”

This has been an incredibly exciting year with all the enhancements enabling more AI capabilities on the platform, and the momentum will only continue with more innovation into 2026.