AI on IBM Z

AI-powered innovation to fuel business growth

What's next for mainframes and AI? Learn about Telum and Spyre
A high-tech microchip featuring a layered architecture is displayed against a sleek black background. The design highlights individual components, including connectors and circuit boards, with a glowing purple light emphasizing the layers

Unlock AI insights and securely run gen AI

AI on IBM Z® brings real-time insights by applying machine learning directly to transactional data, eliminating the need for data movement.

By using the advanced hardware and software stack of IBM z17™, businesses can scale multiple AI models to power predictive use cases such as fraud detection and retail automation. This process also includes supporting generative AI (gen AI) capabilities securely on premises. With high throughput, low latency and industry-leading cyber-resilience, IBM Z is built for mission-critical AI.
Get real-time insights when needed

Infuse AI into every transaction without data movement while meeting stringent service level agreements (SLAs) and response times.
Keep data secure and compliant

Run AI where your data resides to protect sensitive information and meet regulatory requirements.
Scale seamlessly with transaction volume

With IBM z17, process up to 450 billion inference operations per day with 1 ms response time for real-time use cases.1
Boost inference throughput

Route inference requests to any idle Integrated Accelerator for AI to boost throughput up to 7.5x over IBM z16.2

Use cases

Generative AI with Spyre Accelerator

Increase productivity with agentic AI
Employee autonomy is enhanced by simplifying access to information and automating routine tasks with agentic AI.
Minimize the learning curve for IBM Z professionals
Faster onboarding and knowledge transfer for IBM Z users is enabled, reducing reliance on experts and enhancing workflows for experienced users.
Secure gen AI behind your firewall
Generative AI can run securely on-premises with Spyre™cards, ensuring privacy, compliance and complete control over operations.

Predictive and multi-model AI on Telum

Anti-money laundering
A multiple AI model approach on IBM z17 accelerates AML efforts, improves accuracy and streamlines compliance.
Insurance claims processing
Combining encoder large language models (LLMs) with predictive AI creates a faster, more efficient claims process and improves service quality.
Real-time fraud detection
Fraudulent transactions can be detected instantly by deploying AI models in transactions on IBM z17 with Machine Learning for IBM z/OS, reducing risk and saving costs.

Featured products

Big data technology Data science analysing artificial intelligence generative AI
Generative AI and agentic AI

IBM watsonx Assistant for Z delivers secure, AI-powered virtual agents at scale on IBM Z for smarter customer interactions and agentic workflows.

 IBM watsonx Assistant for Z
Transparent glass cubes with reflected light effects on a dark blue background
Deploy models on IBM Z

Machine learning for IBM z/OS allows users to deploy machine learning models within transactional applications while maintaining SLAs. 

 Machine learning for IBM z/OS
Isometric illustration showing data monitoring and compliance auditing
Enterprise open-source frameworks

AI Toolkit for IBM Z is a family of supported open source AI frameworks optimized for the Telum processor and use on-chip AI acceleration in IBM z16® and z17 systems.

 AI Toolkit for IBM Z
Badge of Best Software Top 50 2025
Artificial training data

IBM Synthetic Data Sets is a family of artificially generated datasets designed to enhance predictive AI model training and LLMs.  

 IBM Synthetic Data Sets

Related software

IBM Concert for Z enables anomaly detection, smart event correlation, and expert advice through a unified user interface.
IBM Threat Detection for z/OS® identifies anomalies in data access that might indicate a potential cyberattack.
IBM watsonx Code Assistant for Z accelerates mainframe application development and modernization with generative AI and automation.
IBM Db2 for z/OS delivers secure, agile data serving for hybrid cloud, transactions, and analytics.
Python AI Toolkit for IBM z/OS provides open-source tools to run AI and ML workloads.
IBM Deep Neural Network Library for TensorFlow deploys AI models using the Integrated Accelerator for AI.
IBM Z Platform for Apache Spark enables high-performance in-memory analytics with Java, Scala, Python, and R.
IBM Z Deep Learning Compiler runs ONNX AI models as optimized libraries using the Integrated Accelerator for AI.

Resources

How mainframes are rewriting the AI playbook
Learn about the key elements of current and next-generation mainframes and how they are transforming the AI landscape.
Mainframes as mainstays of digital transformation
Discover how mainframes are more relevant than ever for hybrid cloud optimization, AI innovation and digital transformation.
Optimized inferencing and integration with AI on IBM Z
Learn how to enable AI solutions in business-critical use cases, such as fraud detection and credit risk scoring, on the platform.
What if IBM Z might help stop fraud?
Discover low-latency AI on a highly trustworthy and secure enterprise system: the modernized IBM Mainframe.
Take the next step

Discover how to use AI and machine learning to convert data from every transaction into real-time insights. 

 Get started
More ways to explore Documentation Support Lifecycle services and support Community
Footnotes

¹ DISCLAIMER: Performance result is extrapolated from IBM® internal tests running on IBM Systems Hardware of machine type 9175. The benchmark was executed with 1 thread performing local inference operations using a LSTM based synthetic Credit Card Fraud Detection model to exploit the Integrated Accelerator for AI. A batch size of 160 was used. IBM Systems Hardware configuration: 1 LPAR running Red Hat® Enterprise Linux® 9.4 with 6 IFLs (SMT), 128 GB memory. 1 LPAR with 2 CPs, 4 zIIPs and 256 GB memory running IBM z/OS® 3.1 with IBM z/OS Container Extensions (zCX) feature. Results may vary.

2 DISCLAIMER: Performance results are based on internal tests exploiting the IBM Integrated Accelerator for AI for inference operations on IBM z16 and z17. On IBM z17, each IBM Integrated Accelerator for AI allows any CPU within a drawer to direct AI inference request to any of the 8 idle AI accelerators on the same drawer. The tests involved running inference operations on 8 parallel threads with batch size of 1. Both IBM z16 and z17 were configured with 2 GCPs, 4 zIIPs with SMT and 256 GB memory on IBM z/OS V3R1 with IBM Z Deep Learning Compiler 4.3.0, using a synthetic credit card fraud detection model. Results may vary.