IBM recently achieved its highest market cap ever. How did it do this? It comes down to CEO Arvind Krishna's strategy and his team's execution. When Krishna took the reins as CEO, he wanted to have the company lead in "hybrid cloud and AI." We all understood the "hybrid cloud" part on day one (this is where Red Hat came into the picture), but what about AI? IBM followed up shortly with watsonx, the world's first-to-GA end-to-end enterprise generative AI platform. Yes, IBM did this before Azure, AWS or Google Cloud did -- and it runs in the cloud and on-prem. Krishna also spun off low-growth services businesses, turning IBM into a product and technology company again and investing where the puck was going.
Fit-for-purpose infrastructure is cool now, but IBM has never stopped working on it in areas including Z for mainframe, semiconductor research and quantum computing. In that context, we've recently covered the Z17 launch and weighed in on IBM's quantum roadmap. But what about IBM's Power servers, which we hadn't written about since Patrick Moorhead did in 2021?
This is an opportune moment to return to the subject, because IBM launched its new Power11 platforms last week to address growing modernization needs in the enterprise datacenter. While the Power platform has traditionally been considered mainframe-adjacent, Power11 is the company's attempt to expand the product line's target deployment scenarios and use cases.
How might IBM use Power11 to address an expanded market opportunity? And what will IBM need to do to be successful as it makes a play beyond what some would consider the mainframe-adjacent market?
(Note: IBM is an advisory client of our firm, Moor Insights & Strategy.)
As analysts who often confer with leaders from both enterprise IT and lines of business, we tend to notice a disconnect between the two groups. While business leaders understand that activating AI is a significant effort, their understanding tends to be abstract. Meanwhile, it's the IT leaders who must deal with the real and complex challenges of integrating technology stacks and, more importantly, data.
Further complicating this challenge, historical data that adds context to large language models often exists on platforms that have been powering the largest enterprises for decades. More specifically for our context here, this data often sits on "big iron" IBM Power servers that were originally designed with data in mind. IT leaders understand the importance of using this data to add texture and deliver more accurate intelligence with AI.
IBM Power has built a reputation over the years for the highest levels of performance, scale, reliability and security. That reputation is well deserved because Power is the compute engine for some of the most mission-critical workloads in the most regulated industries: large-scale database, transactional and ERP workloads. This is the bread and butter of the enterprise.
It's with this backdrop in mind that we come to Power11. While still emphasizing performance, reliability and security, Power11 has expanded in scope to address a larger segment of the enterprise datacenter market. By combining Power-based systems with IBM's Red Hat portfolio (OpenShift, RHEL) and watsonx, these big-iron platforms now deliver a highly performant hybrid cloud environment made for modern workloads such as AI and SAP HANA in addition to the traditional workloads.
IBM has delivered a portfolio of Power11 systems that spans many deployment scenarios in the enterprise:
For those unfamiliar with the Power portfolio, it's worth pointing out a few things. First, trying to compare an IBM Power processor and core with an x86 offering such as EPYC from AMD or Xeon from Intel is like comparing apples and oranges. IBM designs its RISC-based Power processors for absolute performance and sustained high utilization. Whereas an x86 CPU will operate most efficiently in the 40% utilization range, Power is designed for efficiency at upwards of 80% utilization.
The same kind of design approach extends across Power's security, reliability and power efficiency. While EPYC and Xeon also address these aspects, the Power architecture is seemingly almost maniacal about them. This is why IBM touts its "six nines" of availability and can point to differentiating features such as quantum-safe encryption built into the firmware of Power11 systems.
On top of this, Power systems are designed and built as fully integrated stacks. The silicon is fully leveraged in hardware. And the hardware is fully leveraged in firmware and the operating system. This is what enables the highest levels of reliability and performance.
For native AI support, IBM has embedded its Vector Scalar Matrix Engine v2. This is a similar approach to what Intel has been offering in its Xeon CPU for the last few generations -- dedicated silicon "engines" to offload CPU functions that are not performed as efficiently by the CPU. These accelerator engines have proven to deliver significant lift, specifically for inference. Finally, when IBM’s dedicated AI accelerator, Spyre, is available later this year, customers will be able to order it included in Power systems.
Power11 truly does combine the best of the traditional big-iron platforms that IBM has been designing for decades with the flexibility of the distributed x86 platforms that support today's hybrid cloud and virtualized environments.
With all of this in mind, it's worth exploring IBM's market play in more detail. As touched on earlier, under Arvind Krishna IBM looks at the evolution of the datacenter toward data-centric and cloud-native technologies. And in the case of the Power portfolio, the company sees an opportunity to increase its total addressable market significantly.
Whereas Power10 focused squarely on a big-iron space that IBM has dominated for quite some time, the Power11 architecture aligns nicely with these new IT market shifts -- and with the work IBM has done in recent years with its Red Hat portfolio and watsonx platform. The implications of this could be quite large; whereas the traditional TAM for Power servers running IBM's AIX or IBM i operating systems is estimated at $2 billion, adding new markets for hybrid as-a-service, modernization and SAP HANA use cases brings the TAM to $47 billion. To be clear, we don't think that IBM plans to grow Power's market share by such an exponential amount. But it does demonstrate that Power11 could compete in some very rich market spaces -- from which it was previously excluded.
So, can IBM effectively compete across the four pillars of as-a-service, modernization, SAP HANA and its core AIX/IBM i business? We think the answer is yes, though the details are important. For example, we don't expect IBM to start replacing traditional distributed computing solutions such as Nutanix and Dell for modernization projects. However, for IT shops already using IBM Z or LinuxONE, going with Power11 could be a very compelling option instead of using Nutanix, Dell or other vendors. Its performance-per-watt advantage (which IBM says is about twice that of x86-based servers), combined with its deep security and reliability, makes the discussion of upgrading to the latest x86 server versus deploying Power11 very interesting.
Overall, the integration story is good. Leveraging Red Hat Enterprise Linux and OpenShift to deliver a cloud environment presents a platform and tooling familiar to IT administrators who are not otherwise familiar with Power. Now, it's not so much about making a change to a Power server -- it's simply adopting a high-performing IBM cloud platform.
We believe Power11 offers a compelling value proposition for the enterprise, and that it does open up an expanded TAM for IBM. That said, the company will be challenged to cut through the noise even in current customer datacenters. In these environments, teams focused on distributed computing tend to be separate from those using Z or LinuxONE. Further, IBM has done a strong job of servicing its install base and building a strong but very tightly focused brand for Power. This brand will need to be deconstructed a bit through education and amplification to reach the broader audience that IBM is aiming for.
The road from $2 billion to $47 billion is long. However, in some ways, the transition from supporting large databases and big-iron workloads to hybrid cloud environments is even more daunting. As IBM looks to expand its footprint, it would be well served to play the long game. As previously mentioned, Power11 will not completely replace x86 across the enterprise today or tomorrow. However, there are cases where Power11 is a more logical choice based on where a workload is hosted, its integration with existing IBM systems or the desire to use Red Hat or watsonx to activate generative or agentic AI. From these adjacent IBM use cases, the company should be able to expand its coverage to environments that are perhaps less directly aligned with traditional IBM coverage areas.
Another area that IBM product marketing could focus on is adjacent workloads that have been running in VMware environments for the last decade or so. The disruption introduced by Broadcom's acquisition of VMware two years ago has led to many VMware customers looking for alternatives. We believe there's an OpenShift/L1124 or E1150 proof of concept waiting for these customers.
There are undoubtedly countless other campaigns IBM can initiate to drive interest that supports its TAM expansion. But the key is to execute this effort in smaller, manageable chunks, focusing on the lowest-hanging fruit first to build momentum.
IBM's launch of Power11 is potentially disruptive. By combining the extreme automation and reliability of the Power platform with the scale-out capabilities of distributed computing environments, it has the chance to significantly increase IBM's footprint in the enterprise datacenter by enabling modernization efforts anchored in a hybrid cloud architecture.
However, potential means nothing without execution. The path to expanding the enterprise datacenter footprint requires highly concentrated go-to-market efforts around awareness and education. Strangely enough, building Power11 was, in some ways, the easy part.
We'll check in over the next few quarters to see how Power11 adoption is shaping up.