You’d be forgiven for thinking mainframes were a relic of a bygone era. In reality, they’re responsible for processing nearly every financial transaction today. But that could soon change if public cloud providers like Amazon Web Services (AWS) have their way.

“Over the last 15 years, as the cloud has become mainstream, your options for where you run applications have proliferated,” AWS CEO Adam Selipsky said during his re:Invent keynote last winter. “We continue to believe the vast majority of applications and workloads will run in the cloud.”

And, no surprise, this includes mainframes, which he argues are expensive, complicated, and nearly impossible to find people that can program them. “Maintaining a mainframe is kind of like trying to shoot a basketball with your two feet planted on the ground,” Selipsky said. “You know you can do it, but you know there’s got to be a better way.”

AWS’s answer? Make it easier for customers to migrate their applications written in common business oriented language (COBOL) — an esoteric programming language that dates back to the earliest days of computing — to the cloud.

The service, dubbed Mainframe Migration, takes code written in COBOL and rebuilds it to run on AWS’s infrastructure. Meanwhile, integrated testing tools ensure the rebuilt application still functions as intended.

Amazon claims migrated workloads not only benefit from improved scalability, but substantially lower operating costs. “We’ve seen customers reduce their costs by up to 70% or more after migrating,” Selipsky boasted.

An Architectural Advantage

However, this only addresses the COBOL problem and doesn’t necessarily address the architectural advantages inherent to mainframes, Dell’Oro analyst Baron Fung told SDxCentral.

“IBM mainframes seem to be holding their own, given that they haven’t lost much market share in the last several years, despite increased cloud-based offerings,” he wrote in response to questions. “These mainframes are still hard to replicate in the cloud given the large parallel computing that can be enabled with these mainframes that is not very common in the cloud yet.”

Fung also highlighted the high degree of redundancy and security designed into modern mainframes. For example, large financial institutions may be bound to data sovereignty regulations that make porting applications to the cloud impractical if not impossible.

Custom silicon tailored to these workloads as a means to further improve efficiency and performance isn’t out of the question. While Amazon has yet to announce mainframe-specific silicon, the company hasn’t been shy about developing chips that compete directly with leading chipmakers anywhere it sees an opportunity to deliver higher performance at lower costs.

Amazon’s Graviton CPUs, Trainium and Inferentia artificial intelligence (AI) accelerators, and Nitro smartNICs represent just a handful of custom silicon the cloud provider has developed to date.

Amazon declined to comment on this story.

Is AWS Mainframe Migration a Threat?

Amazon’s efforts to eat away at the mainframe stalwarts hasn’t gone unnoticed at IBM, the company responsible for popularizing the systems in the early '60s.

“We definitely see the efforts by AWS as a threat, even if they are starting out and appear to be primarily a marketing message,” Barry Baker, VP of product management for IBM Z and LinuxOne, told SDxCentral.

However, Baker doesn’t believe many of its core mainframe customers will find the prospect of AWS’s offering all that appealing.

“If you tell them the first step is you have to migrate — you have to migrate that to a non-optimized, generic-cloud infrastructure before you can even get going on that — it's really not palatable," Baker said.

Instead, Baker said many customers are looking for ways to deploy applications in the cloud — including on AWS — and on IBM mainframes. “In the context of AWS or other hyperscalers, it is much more of an ‘and us’ versus ‘an or,’” he added.

In fact, Baker argues Amazon’s pivot to custom silicon only goes to show how valuable highly-specialized compute architectures are.

“I don’t view it necessarily as a threat so much as I view it as an acknowledgement of what we’ve been saying, which is there are different architectures that exist and they exist for good reason, and IBM Z and mainframe is one of them,” he said.

IBM Fights to Maintain Mainframe Relevance

IBM hasn’t been resting on its laurels either. The company has continued to innovate in the mainframe space, despite the platform’s prehistoric perception.

The company’s z16 mainframes, announced earlier this month, are capable of real-time artificial intelligence (AI) inferencing, which promises to improve fraud detection and combat financial crimes.

By embedding the AI accelerator directly into the z16’s Telum processors, IBM is able to achieve significant gains over GPU-based AI acceleration. This is particularly true when it comes to latency, something that’s critical when processing financial transactions.

“The accelerator can scale up to 300 billion inference requests per day at one millisecond of latency for each inference request, enabling state of the art AI in workloads that were not possible before,” Patrick Moorhead, principal analyst at Moor Insights & Strategy, explained in a recent article contributed to Forbes.

These capabilities, he said, enable financial institutions to run fraud detection algorithms on a broader range of transactions and proactively identify patterns associated with financial crimes like money laundering.

The z16 launch comes amid a resurgence of mainframe adoption, according to a Forrester Research report, which found 74% of those surveyed still saw mainframes as a long-term strategic platform.