A mainframe is a large, powerful computer that processes millions of transactions per second for organizations that cannot afford downtime
A mainframe is a room-sized or cabinet-sized computer built to run the critical operations of large institutions — banks, insurance companies, government agencies, airlines. It is not a personal computer or a server. It is a machine designed to handle thousands of users at the same time, process enormous amounts of data reliably, and keep running without interruption even when parts fail.
The mainframe sits at the center of an organization's computing infrastructure. A bank's mainframe processes your deposit, your withdrawal, and your balance inquiry simultaneously with thousands of other customers' transactions. An airline's mainframe books your seat, prints your boarding pass, and updates inventory across hundreds of airports in real time. These machines are built for that scale and that reliability. They cost millions of dollars and run for decades.
Mainframes are not obsolete. They still process the majority of the world's financial transactions. But they are invisible to most people because you never interact with them directly — you interact with the app or website that sits on top of them.
Key Takeaways
- A mainframe is a large computer designed to process millions of transactions per second for banks, insurance companies, and government agencies.
- Mainframes can serve thousands of users at the same time and continue operating even when individual components fail.
- The three major mainframe manufacturers are IBM, Unisys, and Fujitsu, with IBM holding the largest market share.
- Mainframes remain in use because replacing them would cost more money and create more risk than maintaining them.
- You use mainframe services every time you check your bank balance, book a flight, or file taxes, but you never see the machine itself.
How a mainframe differs from other computers
A personal computer or laptop runs one operating system and serves one user at a time. A mainframe runs a specialized operating system — usually z/OS, z/VM, or z/Linux — and serves thousands of users simultaneously. The mainframe divides its processing power among all those users so that each one feels like they have the machine to themselves.
A server in a data center is designed to be reliable and to handle many users, but it is still fundamentally a scaled-up version of a personal computer. A mainframe is a different architecture altogether. It has redundant processors, redundant memory, redundant storage, and redundant power supplies. If one processor fails, the others keep running. If one memory module fails, the system detects it and routes around it. The machine is designed so that a single failure almost never causes an outage.
Mainframes also use a different programming model. Code written for a mainframe often runs in a batch — a set of jobs scheduled to run at specific times, usually overnight. A bank might run a batch job at 2 a.m. to process all the day's transactions and settle accounts. Modern applications run in real time, but many mainframe applications still use batch processing because it is efficient for large volumes of data and because the code has worked for decades.
The three companies that build mainframes
IBM dominates the mainframe market. IBM's System z mainframes (the current line is called z15 and z16) run most of the world's financial transactions. IBM has held this position since the 1960s and continues to release new models every few years.
Unisys manufactures the ClearPath mainframe line, which is used primarily by government agencies and large enterprises in specific industries. Unisys mainframes are less common than IBM's but are still in active use and receive regular updates.
Fujitsu builds the SPARC M-series mainframe, which is used mainly in Japan and Asia. Fujitsu's mainframes are less visible in North America but represent a significant portion of the global installed base.
Why organizations keep mainframes instead of replacing them
A mainframe costs millions of dollars to purchase and millions more to operate and maintain. The obvious question is: why not replace it with a fleet of smaller, cheaper servers?
The answer is that the cost of replacement is higher than the cost of keeping the mainframe running. A bank with a 40-year-old mainframe has decades of code written specifically for that machine. That code handles edge cases, regulatory requirements, and business logic that nobody fully understands anymore because the original programmers retired. Rewriting that code for a different platform would take years and cost hundreds of millions of dollars. During the rewrite, the bank would have to run both systems in parallel, which doubles the cost. And the new system would have to be tested exhaustively because a single bug could lose customer data or violate financial regulations.
The mainframe, by contrast, is proven. It has processed trillions of transactions without data loss. The organization knows how to operate it. The cost of keeping it running is predictable. So organizations choose to keep the mainframe and add new systems around it — a web server that talks to the mainframe, a mobile app that talks to the mainframe, a cloud service that talks to the mainframe.
This is why mainframes are not going away. They are not fashionable, but they are reliable and the cost of replacing them is too high.
What mainframes actually do day to day
A mainframe spends most of its time processing transactions. A transaction is a request to read or write data — check a balance, transfer money, book a seat, update an inventory count. The mainframe receives the request, finds the data, modifies it if necessary, writes it back, and sends a response. It does this millions of times per day.
Mainframes also run batch jobs, which are large computational tasks scheduled to run at off-peak hours. A bank might run a batch job to calculate interest on all savings accounts. An insurance company might run a batch job to process all claims submitted that day. These jobs can take hours and process gigabytes of data, but they run when the system is less busy so they do not slow down the transaction processing.
Mainframes also store data. The mainframe's storage system is designed to be extremely reliable — data is written to multiple copies, and the system checks constantly for corruption. A mainframe might store decades of transaction history, customer records, and regulatory filings. That data is the organization's most valuable asset, and the mainframe is designed to protect it.
The skills gap and the future of mainframes
The average mainframe programmer is in their 50s or 60s. Fewer young people are learning mainframe programming because the work is not visible and the technology seems old. This creates a risk: as experienced mainframe programmers retire, organizations may struggle to find people who can maintain their systems.
Some organizations are addressing this by training younger programmers in mainframe technology. IBM and Unisys both offer training programs. Some universities teach mainframe programming. But the demand for new mainframe skills is not as high as the demand for cloud and web development skills, so the pipeline is smaller.
This does not mean mainframes will disappear. It means organizations will continue to run them but may struggle to find staff. Some organizations are also investing in tools that allow non-mainframe programmers to write code that runs on the mainframe, which could help bridge the gap.
Mainframes versus cloud computing
Cloud computing — renting computing power from Amazon, Microsoft, or Google — is cheaper than buying and operating your own mainframe. But cloud computing is not a direct replacement for a mainframe. A cloud service is designed for flexibility and scalability. A mainframe is designed for reliability and consistency.
Some organizations are moving workloads to the cloud. But most large financial institutions keep their mainframes because the cost of migration is too high and the risk is too great. Instead, they use the cloud for new applications and keep the mainframe for the systems that run the business.
The future is probably hybrid: a mainframe at the center, cloud services on the edges, and middleware connecting them. The mainframe will not disappear, but it will become less visible.
Frequently Asked Questions
Do mainframes use the same software as regular computers?
No. Mainframes run specialized operating systems like z/OS or z/VM that are designed for the mainframe architecture. Code written for Windows or Linux will not run on a mainframe without significant modification. However, some mainframes can run Linux, which allows organizations to run some standard software on the mainframe.
How much does a mainframe cost?
A new mainframe costs between $5 million and $15 million depending on the model and configuration. Operating costs — power, cooling, maintenance, software licenses — run into the millions per year. This is why organizations only buy mainframes if they process enough transactions to justify the expense.
Can a mainframe be hacked?
Mainframes are targets for hackers because they hold valuable data, but they are also heavily secured. They run in isolated networks, have multiple layers of authentication, and are monitored constantly. Breaches do happen, but they are rare compared to breaches of web servers or cloud services.
Why is the mainframe still called a mainframe?
The term comes from the 1960s, when the mainframe was the main computing resource for an organization and everything else — terminals, printers, storage — was peripheral to it. The name stuck even though the architecture has changed completely. Today, the mainframe is one part of a larger system, but the name remains.
What programming languages are used on mainframes?
COBOL is the most common language for mainframe business applications. It was designed in the 1960s specifically for business data processing and is still used because decades of code is written in it. Java, C, and other modern languages can also run on mainframes, but COBOL remains dominant.