IBM says its latest mainframe comes with the ability to run artificial intelligence models at high speeds and to provide quantum-grade encryption on transactions and data.
The IBM z17 features expanded on-chip AI capabilities and upgraded hardware and software designed to assist in fraud prevention and quantum security. The "Z" stands for zero downtime, according to IBM Z Chief Product Officer Tina Tarquinio.
The z17 model was unveiled during a launch event for IBM Z clients at IBM's new One Madison building on Tuesday.
The IBM z17 can now process up to 35 billion transactions per day (an increase from the previous model's 25 billion), according to the company. The IBM z17's main processing upgrade is through its Telum II chip, which features AI accelerators and 40% more on-chip capacity compared to the Telum chip in the previous z16 model, Tarquinio said.
"You can score every one of those 35 billion transactions with a one millisecond response time and with increased accuracy from running multiple models, expanding the scope of fraud detection," Tarquinio said. "We know fraud has so many forms and you need a model that's an expert in each one of those forms, and now you can run them all in a single transaction. The system really needs to be able to detect the difference between fraud and a valid transaction, so you spend more time on the things you should and less time on the things you shouldn't."
Elpida Tzortzatos, IBM fellow and chief technology officer of AI for IBM Z, told American Banker that many of IBM's clients face challenges in incorporating AI into their existing systems, due to AI processing times slowing down regular transaction and response times.
"When an enterprise really materializes the value of AI is when they deploy those models correctly," Tzortzatos said. "They can spend all the money in developing and building, but if they can't deploy them successfully in production environments and in mission-critical workloads, all that investment is for nothing at the end of the day. To answer this challenge, we had to build an on-chip AI accelerator that was optimized for very fast, ultra-low latency and high throughput, so our clients could embed AI in every transaction and have consistent response times."
IBM's z17 mainframe is also "quantum-safe" in anticipation of
"Our algorithms are engineered to be resistant to attack by quantum computers," Dames told American Banker. "We're using cryptography today to protect the system, and we've started making use of the algorithms that were selected for standardization by NIST to protect system infrastructure. For example, we leverage both classical and quantum-resistant algorithms to run whenever the mainframe is turned on to make sure that the firmware actually came from IBM, using quantum-safe algorithms for that authentication."
IBM is also finalizing an add-on product called the Spyre Accelerator, which is being built specifically for generative AI. The z17 mainframe allows for up to four additional AI accelerator drawers, and each processing drawer has room for eight peripheral component interconnect express, or PCIe, cards.
"Each of the eight chips on the processor drawer are all connected to each other," Rhonda Sundlof, product offering manager for IBM Z, told American Banker. "If you find that your utilization rates on your first chip are above where you want them to be, then it'll automatically flip over to take advantage of any of the other accelerators in the rest of the drawer. The system has four drawers, so now you effectively have a whole suite of AI accelerators that you wouldn't have had access to in a z16."
Speakers at the launch event emphasized IBM z17's ability to assist in fraud detection for their clients. Enhanced security and fraud mitigation are the
A
"If you look at the global economy, there's around a $500 billion annual cost of fraud," Tarquinio told American Banker in an interview. "If you could make even a small percentage dent in that, it's a material difference to our clients around the world."
Tarquinio also emphasized the scalability of the z17 mainframe in both directions, so both large banks and smaller community banks can use the same mainframe without cost being a barrier.
"One of the things that is really unique about our technology is that it can be dialed down all through code on the spot, so you can really start small and grow," Tarquinio told American Banker. "It's the same with our Spyre accelerator, we've designed it so that you can start with a starter set of eight cards and then grow. The starter set is aimed at clients who just need those lower capabilities, and then as you scale in production the capacity grows. We've really lowered the bar for everyone to get started on z17 with AI."
IBM partners with
"We like to call Z the backbone of the global economy," said Ric Lewis, senior vice president of infrastructure at IBM. "Seventy percent of transaction volume, by value, is done through Z systems. Last year, in 2024, one quadrillion dollars of transactions happened through IBM."
Those statistics were sourced from a
Mike Chuba, managing vice president of research at consulting firm Gartner, told American Banker that the savings from improved fraud detection can help banks offset the cost of purchasing or upgrading their mainframe.
"IBM customers need to do sophisticated, real-time analysis," Chuba said. "When you start talking about industries like banking and financial services, there's so much in the way of fraud detection that has to be done at the point of the transaction. If you don't capture it, then you lose that opportunity. For customers in those segments, the ability to do the real-time scoring and to determine at that point of transaction if it is fraudulent, over time, is going to save them more than what they spend on buying one of these machines, whether it's a $2 million, $5 million or $10 million configuration."
Chuba also expects that IBM will see stiff competition as it enters the AI development space.
"The challenges for IBM are that this will play very well within their established customer base. However, to date, their success has been fairly modest in terms of first-time customers," Chuba said. "Every vendor is talking about AI and generative AI, and the benefits that come from that. The competition that IBM has out there, who will try to attract workloads from the mainframe and get it moved to cloud providers, or whoever else, is going to continue to get stronger. As I try to modernize my applications, do I want to do it on the platform, taking advantage of the AI capabilities that IBM has, or are there reasons to move it off-platform, taking the increasingly improved AI capabilities that the hyper-scalers and others are starting to deliver? It's a question of whether IBM can step up to that challenge."
Bank of Montreal, an IBM client, is anticipating using IBM z17 for bank use cases such as fraud detection, loan decisioning and code explanation, according to Julie Trevisan, director of retail lending development at Bank of Montreal.
"What I'm really excited about with z17 is the generative AI," Trevisan said in a panel discussion at the launch event. "That's what we're pushing. Having the AI right on the Z mainframe, right next to the core data, is huge for security. The Spyre accelerator and the Telum II chip are going to move us forward, and that's going to allow us to leverage Z and our core systems and bring them into the next world."
The new IBM z17 could also help banking core providers such as the London-based core banking platform Hogan, according to Duncan Alexander, product director of Hogan at DXC Technology.
"We have specifically developed capabilities within Hogan and IBM to enable instant replication of data between two different data centers geographically dispersed," Alexander told American Banker. "If one has a major crisis, the other one is ready to go, and we've worked with IBM on that solution. We will equally adopt the development technologies and development capabilities of IBM Z in our own development shops, so that while we develop Hogan further we're using the same technologies as we recommend to the clients."
IBM z17 will be generally available June 18, 2025, according to a press release. The AI accelerator add-on is expected to be available starting in late 2025.
A company spokesperson told American Banker that the company "expects no impact on product availability" due to
"With supply chain costs continuing to fluctuate, the associated costs of manufacturing and delivering IBM's hardware products are reflected in our current pricing," the IBM company spokesperson said. "IBM continues to honor outstanding prices and terms in existing contracts."
