In the age of cloud computing, it's easy to forget that IBM (IBM +2.59%) still sells its hulking mainframe systems. Following IBM's fourth-quarter earnings report, it's clear that the mainframe business is not only surviving but also thriving.
Here's an incredible fact IBM CFO Jim Kavanaugh disclosed during the earnings call: IBM's mainframe business recorded its best fourth quarter revenue in more than 20 years. Revenue soared 61% year over year, adjusted for currency, driving a 17% increase in the infrastructure segment.
Image source: Getty Images.
IBM's mainframe systems are still widely used, particularly in industries that need extreme reliability and security. 71% of Fortune 500 companies use mainframes, according to Market Reports World. About 92% of large financial institutions and 63% of government agencies use mainframe systems in their operations, with IBM accounting for more than 90% of installed mainframe systems. Amazingly, over 87% of global credit card transactions are processed on mainframe systems.
How have mainframes remained relevant for so long? IBM has repeatedly iterated, upgraded, and improved its systems to meet client needs. The latest z17 mainframe systems, which are driving record results, are tailor-made for the AI era.
AI inference on the mainframe
IBM's z17 mainframe system is designed to handle more than 250 AI use cases, including loan risk mitigation, chatbot services, medical image analysis, and retail crime prevention. The z17 can handle up to 450 billion AI inferencing operations per day, up 50% from its predecessor. Importantly, those AI operations have average response times of just one millisecond, making the z17 ideal for real-time operations where latency isn't acceptable.
While the z17 has plenty of on-board AI processing power, IBM also sells the Spyre AI accelerator. The Spyre accelerators can be installed in z17 mainframe systems, enabling customers to run more powerful AI models, including IBM's Granite models.
IBM has a clear idea of where the AI inferencing market is heading. While most AI workloads are running in the cloud today, IBM CEO Arvind Krishna expects that to change drastically. "...if I look out three to five years, 50% of the enterprise usage of AI going to be in either a private cloud or is going to be in their own data centers. And the other 50% is gonna be usage of public models," Krishna said during the fourth-quarter earnings call.
This prediction makes a lot of sense. AI inference in the cloud using frontier models is expensive. For tasks that can work with smaller, fine-tuned models and have stable, predictable loads, moving away from the cloud could unlock major cost savings for enterprises.
IBM tends to launch a new mainframe system every 2.5 to 3 years, so the z17, which shipped in mid-2025, will headline the mainframe business until 2028. With AI a focal point of IBM's strategy, its next-gen mainframe will almost certainly further boost AI inference performance.
NYSE: IBM
Key Data Points
An unlikely AI powerhouse
The mainframe is only part of IBM's AI strategy. The company also offers AI consulting services and its watsonx AI software platform. With an enormous enterprise client base and a large mainframe install base, IBM has an advantage in selling AI products and services to its customers.
IBM expects its total revenue to grow by at least 5% in 2026, with software revenue expanding by 10%. Free cash flow is expected to improve by about $1 billion from the $14.7 billion reported in 2025. The mainframe, along with the rest of IBM's AI businesses, is a key component in the company's strong outlook.
While the mainframe isn't as important to IBM's business today compared with 20 years ago, it remains "a key contributor to IBM's profit and free cash flow," according to the company. IBM has adapted the mainframe for the AI era, and the result has been record sales. The AI industry is evolving rapidly, but IBM is emerging as a clear winner.
