Talking Power Systems And IBM i With Bargav Balakrishnan

January 20, 2025 Timothy Prickett Morgan

In last week’s issue of The Four Hundred, we told you about how the Power Systems division had a new vice president of product management – to be specific, Bargav Balakrishnan, who has spent decades in various technical and management roles within Big Blue. Balakrishnan takes over from Steve Sibley, who has been steering Power Systems hardware development since July 2007 and who was the longest serving executive in that role since the launch of the AS/400 back in 1988.

This week, we had a chat with Balakrishnan about the Power Systems business and the IBM i platform, and how AI and other factors are transforming the Power Systems business and putting it on a path of sustainable growth.

Timothy Prickett Morgan: As you take over this role, we are coming to the tail end of Power10 generation and we are looking forward to Power11 and Power12. I don’t need a roadmap dump – we got one from Bill Starke, the chief architect of the Power chips, back in November and then followed up with an analysis of the memory subsystem of the forthcoming machines in December. But as you take over this role from Steve Sibley, what I think we all really want to know is what are you thinking about with regard to systems and what are your goals for the business?

It could be lots of different things, obviously. You could want to expand the revenue share in a footprint inside of existing customers. That seems like an obvious thing. Your strategy looks like that to me. You could have a desire to expand into adjacent markets because you think there are other places beside the typical IBM i and AIX database serving where Power Systems can do well.

Bargav Balakrishnan: I know you have talked to Tom McPherson, our general manager, and we have about the same tenure at Power Systems for the past two years, and I was a peer of Steve Sibley, leading our AI and application modernization area for Power. Before that, I was on the mainframe side of systems for two decades. So I have been adjacent to Power, and then rolled up my sleeves two years ago.

To me, the objectives are very clear from a business perspective. What I have been able to see is that we have a huge, loyal, and growing incumbency of Power Systems within thousands upon thousands of clients who rely on the platform for all the reasons you know.

I think what we have opportunity to do – and your spidey sense is spot on – is to increase wallet share within our existing clients for growth, for new workloads, for AI, modernization, et cetera that we haven’t quite aggressively leaned into from an investment perspective to create the value our customers need. So all that said, there is, I think, phenomenal amount of growth and more value to be delivered to our incumbency, which will drive growth for years to come according to our models.

We will drive our incumbency, but then being able to deliver new value in dimensions like AI application modernization, which is what you see our roadmap geared towards, as well as providing things like cloud consumption models with PowerVS or our subscription with as a service. So there’s a lot of business model changes you’re seeing us make as well, all aligned with the ways clients want to consume Power and the ways clients want to gain value from Power – that is, infusing AI into their workloads.

We also see net new clients coming to the platform, aligned with key workloads where we have a very differential point of view, like SAP.

So in a nutshell, I would say that in the last two and a half years, as you have been following, the business has been growing pretty nicely, and in some cases with quarter on quarter increases and year to year increases. So we feel we have the right foundation and playbook, and it’s really not around not just continuing the business, but accelerating it.

TPM: With PowerVS, I struggle every quarter when IBM’s financial results come out to try to figure out how much money did PowerVS generate. There’s no data, but I have to not ignore the fact that there is a growing revenue stream from PowerVS.

We all want to have a sense of how much money is coming from the PowerVS cloud and how much is coming from transactional deals. I’ll give you an example that’s fresh in my mind, just because I just went to the Super Computing 2024 conference in Atlanta. If you look at the revenue stream from on premises HPC servers and you look at the revenue stream for HPC instances sold on the cloud in a given year – this was 2023 data – it looked to me like the revenue stream from the cloud was about 20 percent compared to 80 percent for on premises HPC system sales revenues per year. The cloud HPC is growing faster.

Do you think that in the long run that IBM Power will get something that looks like a 25-75 or 20-80 split in revenues for the year? Will it ever go 50-50? Will everything except the biggest Power systems sold to the bigger customers flip to the cloud? What’s the scenario, and what are the error bars on those scenarios? Because I can make a compelling case why nobody should be doing anything except cloud stuff. And I can also make a compelling case for data sovereignty and knowing your business and cutting costs by staying on premises.

Bargav Balakrishnan: I think that’s fair. So look, we have models. I can’t share everything, but I think your commentary, to me, is spot on. So here’s the first thing. The good news is we’re talking about tens of thousands of clients, right? You know, not 2,000 or 3,000 but in the tens of thousands of clients.

From what I’ve seen, there is ample opportunity and growth coming from to me, both dimensions. It is true today that the growth vector of our as a service and PowerVS business is in the triple digits, but it’s a very small percentage of our overall pie. You can almost say less than 10 percent or whatever, whatever that is, which is to be expected as you’re ramping up a business, especially when it when it’s subscription-based and you go to annuity and all those things. But over time, I could easily see “as a service” being equal, if not the dominant share of the overall estate. And the reason I can say that is because I said “as a service.” I didn’t say “cloud” per se.

While you know, significant numbers of clients do have a cloud first strategy, and that to them that really means get me out of my datacenter, I need to be in a hyperscaler, or in the IBM Cloud, or whatever. But beyond that, what starts to get interesting is the PowerVS Private because, you know, I was just with the client just the other day – it’s a large bank – and they are happy to go to cloud datacenters in a particular region because there’s an IBM Cloud presence. But in other subsidiary countries where there is no cloud datacenter presence, whether that’s IBM or any of our competitors, they need a cloud model, but they can absolutely not leave the country due to data sovereignty reasons. And they are very much happy with and looking at PowerVS Private. So you have a powers Private deployments in certain regions, and you have PowerVS in our public cloud in other regions, but you have a very consistent billing, metering, user experience, all of which would get counted as our “as a service” business, which, as I said, continues to grow.

So I do expect “as a service” to be the dominant, if not at least 50-50, part of the business in the near or in the midterm. And that at least the level of focused investment that we must have to go after that opportunity. Whether the demand, you know, quickly gets there, is a different story. But, I mean, we’re certainly seeing it and trying to race to it.

TPM: What’s the difference between AIX customers and IBM i customers in that statement? IBM has a larger number of AIX customers than IBM i customers. . . .

Bargav Balakrishnan: AIX customers do tend to be on larger machines. I would say they tend to be larger machines. But most of our large customers have pretty large AIX or IBM i instances, and in general, I’m seeing traction across both AIX and IBM i. Now, it is true that in the IBM i space, there are a whole bunch more SMBs that tend to be less sticky to the on premises world, and therefore they are happy with cloud-oriented solutions. So you could maybe argue that the IBM i side would be a little faster in terms of the percentage adoption. But I think we are seeing traction across both of them.

TPM: I don’t think “as a service” has taken the world by storm across the OEMs, including IBM, Hewlett Packard Enterprise, Dell, Cisco Systems, and Lenovo. Whether you count it by customer accounts, or percentage of aggregate computing being done, or whatever metric you want to use, we are still at the very beginning of this “as a service” thing for mission critical, back office systems. But I don’t have any problem believing that half of the computing capacity could be – let’s call it utility priced computing.

Bargav Balakrishnan: Which could be on prem, which could be at an MSP hosting center, or on a cloud.

TPM: Exactly.

I think the economic model of cloud, of utility pricing, is powerful, and that it is more powerful than buying a machine with a perpetual license that they try to use forever, and then they go off maintenance for the system and then for the apps. Yes, they pay a lot less for their system and apps, but this is a dead end and customers will find themselves in a much deeper technical debt and economic hole if they would just stay current with everything.

I don’t understand why there’s so much resistance to cloud pricing and the subscription model. Customers get a premium product, and while they are paying a premium price for it, whether it runs AIX or IBM i, it is highly reliable and available – and it runs the application estate and will continue to do so for at least the next decade.

Bargav Balakrishnan: Just out of curiosity, how much of that do you think is an economic hurdle or just a behavioral thing like people buying cars versus leasing them?

TPM: Well, it’s all money. In smaller companies, any incremental spending to keep current comes out of the owner’s pocket. And for bigger companies as well as smaller ones, there are so many fires, so many other things that need to be done, that keeping the Power System platform, which just works, current doesn’t feel like a priority. Power is a victim of its own success. An “as a service” model forces you to be current, and that’s a business value that’s hard to quantify – right up to the moment when you hit a brick wall where you can’t get security patches, and you just been hacked, and now your business is offline. And the funny bit is that scare tactics generally don’t work, so even saying this does not raise the pulse in many cases because so many other things are already doing it. I don’t know how to do anything other than tell people the truth.

New topic: What can you tell me about how 2024 turned out and what you think 2025 will look just generally like?

Bargav Balakrishnan: So in general, the business has been in a growth trajectory for the last few years, as we have doubled down in our investing in areas that are better aligned to our customers.

There are really three areas. One is continuing to build and deliver differentiated infrastructure; things like business resiliency matters more than ever. Operational risk matters more than ever. Security matters, and this infrastructure needs to be AI ready – to be able to deploy AI fit for purpose. So the infrastructure investments have been really important.

The second thing we’ve been investing in drives our application modernization and ability to leverage AI, which means thoughtful investments in the stack needed to deploy inferencing at the edge, or deploy inferencing co-located with your applications.

And then the third investment area has all been all about PowerVS as a service, the flexible consumption model.

So differentiated infrastructure for the workloads that their businesses run, providing them the ability to modernize their applications, surrounding them with cloud native services, and then having flexible workload placement. Those three pillars of investment are what have ultimately driven the growth over the last few years, and we are accelerating our investments in these area, which should give you a sense that we see more room to run and grow as we exited last year and head into this year.

Again, we have continued enhancements with PowerVS when it comes to things like the IBM i migrate while active, to help with migrating in a more seamless, non-disruptive way, from on prem IBM i to PowerVS, as well as introducing PowerVS Private.

Introducing our Power11 processor is all about bringing an integrated stack value. Every one of our competitors are doing the same thing, which is seeing improvements from silicon, but it’s more than silicon, right? It’s about silicon. It’s about system. It’s about networking, cooling, packaging, all those things together. Integration is what’s going to drive the innovation that we need, from a performance perspective, but also other dimensions, like AI acceleration, security, acceleration, business resiliency, innovation, and so on. So we are excited about the Power11 processor. It will have your usual higher clock speeds and more throughput, but beyond that, it’s setting the foundation for a lot of additional stack value, which we’ll be excited to talk about this year. Our IBM Spyre accelerator is a significant investment and area of focus for us, and it’s all about enabling more use cases with AI acceleration, whether it’s your traditional AI or more of the demanding generative AI, which we are seeing from our customers who want to deploy it on prem. But the most important thing being in a you know, being able to deploy it in an accelerator that comes with the security, resiliency, performance, efficiency that you expect from IBM infrastructure, highly integrated.

TPM: The genius of Spyre it that it is highly integrated, and in the mainframe, there are instructions that address it inside of the “Telum” chips. I don’t want to have to learn how to program in CUDA, I want integrated AI acceleration all within the security perimeter of the processor and under the metal skins of the server.

Bargav Balakrishnan: Our approach is really extensible. We introduced the MMA matrix math unit in Power10. Now you have on chip acceleration, which is getting better just because the system is getting faster, combined now with this IBM Spyre accelerator, which leverages the IP in the Telum processor. We got extensible on chip and off chip, and it’s seamless so that it just can support more use cases.

And then the third thing we previewed, which I’m super excited about, is our Code Assistant for IBM i, which is working hand in hand with the hardware and which is just a no brainer. We have so many customers who have not only contributed code, but are also signed up for sponsor users to help us bring this thing to the market this year. And it’s going to be around use cases like code explanation, the obvious one to code generation. Help me just create new free form RPG code to help me generate test validation scripts and help me modernize my code from old fixed format to free format RPG. So we are bringing that, and we’re excited.

TPM: I presume it is tuned for Spyre?

Bargav Balakrishnan: Eventually, but not out of the gate.

TPM: Will it run natively on the Power10 and Power11 processors and use the MMA? Or do I have to buy a GPU at first?

Bargav Balakrishnan: We are thinking initially, out of the gate, it will be a delivered as a service, but your comment is spot on about where I expect things to go.

TPM: PowerVS plus whatever, and then eventually put it on premises. . . .

Bargav Balakrishnan: As you can see, there’s a set of innovation pipelines, and you are going to see more and more datacenters and usability. You are going to see more AI and modernization coming from us. And then certainly you’ll see not just Power11 and the stack value we’re going to deliver then, but subsequent generations. To me, those things together are what’s going to help continue to drive growth this year and well into the future.

This Issue Sponsored By

Table of Contents

Content archive

Recent Posts

Subscribe

Pages

Search