Mainframes continue to lead the way, processing nearly 70% of all financial transactions globally and managing tens of thousands of insurance claims with unmatched levels of security, resilience, and availability. For mainframe customers, especially those in regulated industries, a critical question is how to leverage their mainframe infrastructure for AI use cases, especially the ones related to high-speed transaction processing workloads and real-time analytics, such as fraud detection, insurance claims analysis, and medical image analysis, running alongside their mission-critical, time-sensitive applications.

As part of IBM's strategy to leverage mainframes for AI workloads, IBM announced the next generation of enterprise compute for the AI era—the IBM Telum II Processor and IBM Spyre Accelerator. These technologies will significantly enhance the acceleration capacity available in next-generation mainframes, supporting a broader and more complex set of AI models. The new chipset will power the next version of mainframes, encompassing both IBM Z and LinuxONE models, available in the first half of 2025. These technologies are designed to scale processing capacity significantly, accelerating the use of traditional AI models and LLMs together.

Infusing AI into Enterprise Transactions

Integrating AI into enterprise transactions can be essential certain AI use cases. AI-driven fraud detection solutions, Advanced Anti-money Laundering, complex AI Assistants and Manufacturing Quality Control, for example, are designed to save clients millions annually while ensuring compliance with regulatory requirements.

These use cases may use ensemble AI approach, which leverages multiple AI models like machine learning, deep learning, and gen AI capabilities, in concert with each other, to improve the overall performance and accuracy of a prediction as compared to individual models. The enhanced AI accelerator on the new processor is tailored to support large language models, enabling comprehensive analysis of both structured and textual data.

When considering this approach for AI projects?

Implementing AI on IBM Z can be triggered by several factors, as example the need for enhanced data processing capabilities, energy efficiency, data gravity aspects, demand for real-time analytics and decision-making, among others, as described below:

• Energy efficiency and advanced security: the new processor improves energy efficiency and supports advanced encryption techniques, including quantum-safe encryption and enhancing data protection against current and future cyberthreats.

• Low latency and high processing power: it is optimized for low latency, crucial for high-speed transaction processing and real-time analytics, offering improved speed while managing large-scale transactions efficiently.

• Data protection and privacy: for AI applications that necessitate access to sensitive data, and failure to adequately secure this information can result in serious consequences.

• Accelerated AI integration: enable near real-time analytics and decision-making directly on the processor. This allows businesses to gain immediate insights.

• Data gravity: allowing data to be processed close to where it resides, can help organizations to derive insights from their large volumes and sensitive data, without migrating it other platforms, making it a suitable platform for addressing data gravity challenges.

Workloads enhanced with AI on IBM Z will provide organizations the opportunity to harness the full potential of their data, scalability, and security, perfectly suited to the demands of some key AI applications. Although it is specifically designed for certain AI use cases, it is a compelling option to consider for mission-critical AI enterprise transaction workloads and complex AI models that necessitate the combination of traditional, predictive AI, and even large language models (LLMs) to enhance prediction accuracy. It is built to deliver in-transaction inference in real-time and at scale, ensuring low-latency AI embedded directly in transactional workloads.