Orlando Bravo and IBM CEO Arvind Krishna on AI, the Future of Software, and IBM’s $34B Red Hat Deal

AIR DATE:

January 29, 2026

LENGTH:

56:31 minutes

ORLANDO BRAVO (00:00):

Welcome to Thoma Bravos Behind The Deal. I'm Orlando Bravo, founder and managing partner at Thoma Bravo. Today we're bringing you a special fireside chat recorded live at Thoma Bravo's 2025 AI Summit. In this session, I sit down with one of the world's most influential technology leaders, a driving force behind IBM's massive transformation over the last decade, IBM CEO, Arvind Krishna. Arvind has led the company through a pivotal shift towards hybrid cloud, AI, quantum computing, and a new era of enterprise innovation. Under his leadership IBM sharpened its focus and helped some of the world's largest organizations rethink how they operate in a rapidly changing digital landscape. We talk about what it takes to build technology that truly scales how to navigate waves of disruption and why the next decade of enterprise innovation will look nothing like the last. And of course I get a taste of the competitive edge that sets Arvind apart. Let's dive in.

(01:25):

Arvind, how are you?

ARVIND KRISHNA (01:26):

I'm doing good. So these are all your investors.

ORLANDO BRAVO (01:29):

Can you believe that? Lots of money here. So tell me where'd you come from? How's your week been? Tell us a little bit about your life. You're spending your life on a plane with customers running the business. How's your week been?

ARVIND KRISHNA (01:44):

Week's been a good week. Let me think. This morning was different investors here in the city. Friday I was at Illinois at the Grainger College of Engineering. I was also in DC headed back there early in the morning.

ORLANDO BRAVO (02:00):

You have a big dinner there.

ARVIND KRISHNA (02:02):

Between DC, between clients. My ordering is very clear. Clients first, then employees, then all of the external forces, whether it's government or investors and then that gives us an outcome that we can serve our communities with. Very straightforward,

ORLANDO BRAVO (02:22):

Just amazing. I want to really, really thank you for being here. I think this is being recorded so I'm going to say it. I can think of nobody better, no better leader in the world to have this conversation with our investors. I've gotten to know Arvind over the years. You hosted me at IBM for an awesome dinner. We had many zooms during COVID and I've never met a leader that combines leadership, operations, sales, and a deep, deep knowledge of technology. So you want to know about AI and quantum and software? Just ask Arvind on any of your investments. So thank you. Thank you for being here.

ARVIND KRISHNA (03:06):

I'd say I always have an opinion may not always be right.

ORLANDO BRAVO (03:10):

We were just talking in the back. Every time I try to sell you a company, you pick the good ones to talk about. So they tend to trend, like you say, they're trending. I want to start by having people getting to know you a bit. How does one get to be CEO of IBM? What happened? What did you do?

ARVIND KRISHNA (03:30):

Random oscillatory path, but I call it look my path is not a common one. Very rarely do people start off in the bowels of the research organization by choice and decide that they want to get to be much more in the business than just there and then proceed down that path. So that's why it's not common. On the other hand, I would say I think 90% of the world is blind to opportunity. With respect, that probably includes some of you, you kind of have blinders on because you kind of don't think about everything that could be possible. And I've always believed keep your eyes open, there's opportunity. So I'll talk to anybody actually inside the company or outside. I accept I have my own filter so I may not follow 99% of what is there, but listen, because you can't predict what is that 1% that has incredible opportunity tied up in it. What is it 99% perspiration, 1% inspiration is a piece of it.

ORLANDO BRAVO (04:32):

How about growing up? Can you tell us a little bit about how you grew up and how you got into business in the first place and research?

ARVIND KRISHNA (04:38):

I grew up in India. I spent my first 21 years in India and I would call it almost idyllic. I was a good student, not a great student until I got to college. College was a little competitive, so if your competitive spot kicks in, then you go compete and that is when I learned also, I went to a really tiny high school. I went to a high school with 11 kids. My graduating class of which only four, graduated seven did not actually complete enough credits to graduate. So that tells you what the school was like. I was a good test taker. So I got into probably one of the most competitive colleges in India. So then you get in and you meet all these other kids, they're off from the big city. They kind of knew everything. They kind of knew what they were talking about and I had no idea what the hell they were talking about.

(05:29):

So you kind of say, okay, I got to work at this otherwise I'm going to sink. And then you discover after two or three months that if you are willing to put some work in. Not insane amounts, just a decent amount of work. All these people who are very smart, they might have been smart but they were not actually flourishing as much and that taught me that you don't always have to depend upon what you knew coming in. If you're willing to learn, read, listen, you can do pretty damn well and that probably helped me throughout including then my decision to come to the United States. Remember 1985 was way pre-internet. You were reading brochures and manuals that kind of decide where to go. So I said if I can get into a top five engineering school for a graduate program, I'll do it. That was really the criteria. And why Illinois? It looked a lot like where I went to my undergraduate. Okay, that was wheat, this was corn. But it was in the middle of an agricultural area, that was kind of a weird, but it was a top five schools, so that helped.

ORLANDO BRAVO (06:37):

That's amazing. Now how many years have you been at IBM?

ARVIND KRISHNA (06:42):

35 years this coming December.

ORLANDO BRAVO (06:45):

Just phenomenal. I mentioned at the beginning that combination of leadership, operations, technology, I mean all of it. Tell us about IBM. Did you have along the way any specific mentor that really, really influenced you or was it the whole system? How was that?

ARVIND KRISHNA (07:03):

It's a bit of both. When I went in IBM, at least in the r and d spaces was incredibly competitive. They didn't say it publicly, otherwise they might have scared off too many people, but when I went in, you were in your second year going to present what you had done over two years, not just to your upline manager but to all their peers. And it was a pretty public grading. After 30 minutes they would come out and tell you what the thought of your work, and I remember it was going to fall into three buckets. We're not going to fire you, but you're probably never going to get a raise again in your life, so you should decide what you want to do. And probably 20, 30% fell into that bucket, then 50, 60% fell into the bucket off. You're doing good work, we encourage you, but you're not kind of a star or you fall into the third bucket.

(07:59):

And so it was competitive but I felt that that was how it should be because you can sort of have a sense of where you are. And I went from there about 6, 7, 8 years and there was a mentor who really pushed me to think about what's possible, what can you do. That got me out from research to think about going into the business and he took a big risk. He took this guy who had only done research who had only done technology, never run a business, no P&L and said here, here is cybersecurity. We have no idea what we want to do with it. Why don't you figure out a strategy? And by the way, if you want to buy some things, I'm willing to spend some money. We bought Dascom, which did policy management. We bought Access 360. I divested our PKI business. I'm so glad nobody's made money in public key infrastructure yet. I think.

ORLANDO BRAVO (08:54):

I'm looking around, have we made money in that? Chip, Chip? You're like a little bit. Oh Intrust. That's true from Nortel

ARVIND KRISHNA (09:00):

And I think you make my point. I think you made a little bit. We sold ours to Verisign. I'm sorry, I'm very direct. I mean...

ORLANDO BRAVO

and very competitive, as we can see!

ARVIND KRISHNA

...and that was one big moment. Then another five years after that, somebody said, we'd like you to ride the business strategy for software. And I said, I'm a technologist. I've never run a business. I mean, well I don't know this stuff, but it was a great experience because you kind of learn of taking a long term view and putting some math behind some, I'll call it macro trends really. And in 2013 somebody looked at me and said, we want you to run all of hardware manufacturing. I'm a software guy. I don't know anything about semiconductors and those things, but the serendipity and keep your eyes open. I never took more than 10 minutes to say, okay, that's what you want me to do. I'll go do it. It was not like, okay, I need to think for a week. But I think all those collective experiences probably prepared me to be a good candidate for the role I'm in because that gave me the depth, you said operations. That gave me a really good sense of how the different parts of the company functioned, how things really get decision, not the decision at the top in a boardroom, but what just impact who does what. And that breadth I think has really served me really well in my current role.

ORLANDO BRAVO (10:28):

Really broad management, broad experience throughout. Let's talk, you mentioned decisions, one of the best deal decisions of all time is the Red Hat deal, the largest deal ever done at the time you mentioned that there was one bigger one, but it happened five years later, 36 billion deal.

ARVIND KRISHNA (10:49):

Yep.

ORLANDO BRAVO (10:50):

Take us through that. How did you come up with it? Why Red Hats? Just start from the beginning on that.

ARVIND KRISHNA (10:56):

So if I go back to 2016, 17, I was running our cloud business. I came to the conclusion and we were nowhere compared to what are now called the hyperscalers. They weren't quite called the hyperscalers. They were just called public cloud at the time. So you have two choices. It's too big a business to ignore. You can't just say, I'm going to put my head in the sand and just forget that it exists. I mean that doesn't work if it's that big a trend. So you get two choices, are you going to compete with them, which is what everyone wanted to do or do you find a way to be a partner and to ride their success? And I said because IBM had bought a company called SoftLayer for a couple of billion. The thought was we're going to go compete. I kind of came in my head to the conclusion you can't compete.

(11:48):

So I was very direct, you can spend 10 billion a year and I said five years from now we'll still be five years behind, so that doesn't seem like a great place. Second, I said, best case, I can see ourselves being fourth or fifth because I don't see any way in which you can catch up to Amazon, Microsoft, Google at the time. And I actually said fifth because I thought Oracle would actually go in even though this was well before they were that big in there. That was actually my conclusion. So I said, if you want to write them, what are the characteristics of what one has to do? It has to be on a piece of software that everybody wants to use and it'll write both in public and private. Now part of that bet was that 50% of the workloads could be 40, could be 60, doesn't matter, I'll call it 50 for argument's sake, are not actually going to migrate to the public cloud.

(12:39):

If that's the case, people need something that can straddle both. Then you say what is the most likely thing that can straddle both and it has to be open, otherwise people are going to be afraid of proprietary. So that is why one and not the other one, which was even bigger. So that is why Red Hat and so I said now the secret thing inside Red Hat was something called OpenShift, which is a container platform, tiny 110 million, not that I'm precise. When we bought the company and I said, I think this will be a multi-billion dollar software platform as big as the Linux business. So the Linux business would give you the credibility, it'll be the way to go back and forth and if we can get a platform going on which people can do development both private and public, then that'll be a great platform and we'll be number one on private if we succeed.

(13:33):

There were three competitors at the time, by the way, when we bought it Orlando, our stock fell 20% the day it got announced and I remember all the headlines, you guys have overpaid, you don't know what you're doing. What is this thing? I mean what does IBM understand about open source, et cetera, et cetera. It really told me that people are really shallow because people felt that the growth rate would stall and it would fail inside IBM. So that is why the stock fell. Well, we've doubled the revenue, we have increased the margins. OpenShift which is 110 million is over a billion five. Ansible, which was no revenue to talk about is probably half a billion plus. So not a bad software business to double the revenue of it and the growth rate increased after we bought it.

ORLANDO BRAVO (14:24):

We follow your numbers and the numbers from the business and OpenShift has been ripping just an amazing macro bet technology bet. Tell us about your financial model around the deal. How did you think about making money financially from it?

ARVIND KRISHNA (14:42):

So it helped that it was a good cashflow business. Look, most of you know this in the software business, if you have a good A&R business and your NRR is over a hundred percent, which it should be, if you have a good product and a set of products and good execution, you need both those things, then your cashflow is actually quite a bit ahead of profit because you're collecting cash ahead of delivering the revenue. So it was running at not quite a billion but close to a billion of cashflow on about 3.4 billion of revenue at the time. So we said we are going to preserve that, but we should be able to take some cost out because you don't need investor relations, you don't need public treasury, you don't need all those functions. So there is some cost that can come out and we should be able to improve this margin over time. It'll take a year or two to get in there. I was pretty convinced I had committed 12% revenue growth to the board when we bought it. We actually for the first five years did I think 14 if I remember. So when you combine those two, then actually the return is incredible because I would turn around and look at you and say at seven and a half billion of software growing in double digits with margins approaching 30%, my gut is this would be on itself worth about 70, which means you've doubled our investors' money.

ORLANDO BRAVO (16:07):

That's what I was going to say easily. What about the decision how decisions get made at IBM? How were you able to convince everybody?

ARVIND KRISHNA (16:18):

Yeah, we'll talk about that, but I want to make a joke first... a couple of times you and I had discussions about some potential deals.

ORLANDO BRAVO (16:26):

I know we were disappointed you didn't do that deal, but then we were happy.

ARVIND KRISHNA (16:30):

I know, but you made a lot of money on that deal.

ORLANDO BRAVO (16:32):

That was a good deal.

ARVIND KRISHNA (16:33):

But if you think about the decisions really at the end of the day, was it deeper than a conversation between the two of us? So at the end of the day you always have to come down to do we want rigor in the financial model that it does meet all our things? Yes, but on Red Hat, to be honest, because I was not in my current chair, I was running our software business at the time, it was a year long process to convince everybody to take the risk. Oddly one, people didn't see the vision of what OpenShift could be. So then the answer was, okay, you're buying a mature product called Linux, how fast can you grow it because if we step back it should grow at the server rate which is like 3, 4, 5, 6%. That doesn't seem that attractive. And I said, well yeah, but you're also taking three points of share each year from Windows.

(17:21):

So it's not just the server rate you're doing that and by the way, if you add capability, I think you can get that up to the high single digits. But the bet is really on the container platform. That was not an easy thing to convince because you needed to have two bold assumptions. One, the world is not all going to a singular public cloud. And in 2016 and 17 that was the conventional wisdom and I said, this is never going to happen. I just don't see enterprise clients and people outside the US being dumb enough to get locked into one public cloud. I just don't see it. But you have to convince people of that point of view. And then the second was if that's the case, then you need something that can straddle that environment for people and hence the container platform because without that you would just use whatever is on one cloud and that's enough of an answer.

(18:18):

So what was happening around sovereignty maybe I got lucky that all geopolitics pointed that to people. I always felt you can't depend upon a cloud for everything because there are these little things called cables that go into the ocean and it's not very hard to cut them if you get upset with somebody. And so you have to first convince that there was a strategy and that could go there. Next, then you have to have a financial model that you could lay out and that took, but the guy who's now our CFO very much got on board that this is possible and once he got convinced that we could grow it a bit faster and you look at that cash return and you could say, well then the one will become two getting two a year, then the deal is absolutely worth it right there and then. So all those things combined helped make the decision and fast forward here, that's where we are.

ORLANDO BRAVO (19:16):

I remember you explaining that to me. We had this great dinner in New York with I think the person that runs your software business now and Seth and myself working on the other opportunity, that deal would've been like 8 billion or so. This was about five times the size. And I remember also because we were working together on something else that when you did the Red Hat deal and you were looking at the thing with us, it was a tough time in the market. The market had crashed there in that quarter. Right. And how did you deal with that and how did your board deal with it?

ARVIND KRISHNA (19:51):

Well, there were two reactions. One, because the Red Hat deal was announced and I don't think we ever reneged on a deal after it was announced, but the reaction was clear. The board reaction was, is this a moment because the stock had fallen 20 bucks that an activist is going to walk in, try to decimate management and the board. So the reaction was, honestly in hindsight was, we can't do anything else that'll attract attention. Everything else is off the table for the next 24 months. And that was October of 2014. We had also done a second thing at the same time, which was a big cause for the reaction. I don't know. The other thing was that we also had done, we had divested all of our semiconductor manufacturing, manufacturing that's a few years before and those things were all people are beginning to understand what the implications of all those things were. And so those combination got everybody to, we are going on a pause until you digest this and can prove that you can run it well.

ORLANDO BRAVO (21:03):

How long did you look at Red Hat before kind of really approaching them and looking to buy it?

ARVIND KRISHNA (21:08):

I was pretty convinced in early 2017 and the question is how do you give yourself confidence? So I decided to first partner, so I said, we are going to partner deeply, we are going to work together, we are going to take common clients, we're going to put all our software on top of RHEL and OpenShift. And so that gives you a much better sense of the management team, their true capability, what clients thought about the company and all those things. That gave me even more confidence then because containers was new and people were skeptical as it containers or is it virtual machines? We decided to kick the tires hard and we got a lot of work going inside on containers and that convinced me that this is the answer you just have to execute also, meaning it's not just enough to say it's technically better, you got to have a product that's consumable by the enterprise customer. So those things in common gave us a sort of kicking the tires approach and then six months after that we started serious work internally and then six months after that we approached them.

ORLANDO BRAVO (22:11):

Phenomenal. I can say from experience that you're an incredible deal maker. We spent a lot of time in deals and on that opportunity that we were working together, I would call you and you're running a huge company and either you would pick up the phone or within five minutes you would call back. The conversations were clear, they were direct, they were to the point and I really trusted everywhere, every step of the way where you were. So I was inspired by that deal making capability of an executive in a big company. Let's talk a little bit about your stock price. IBM stock was doing nothing for a long time. You became CEO in 2020 and since you became CEO, we have the numbers here, the stock including dividends has tripled and you became CEO at the peak of the market. Now the market is not that good at 3X over this time period. We have this chart as well that of the software stocks, it's kind of IBM and Microsoft since 2020 that are right there in terms of a return. What'd you do? How'd you do that?

ARVIND KRISHNA (23:19):

Look, I was pretty sure that we have certain advantages. The question is are we making use of them? We did have incumbency, we did have trust, but we were kind of put into the legacy bucket. This is the stuff I'm going to be decreasing over time. So in hindsight it sounds pretty simple. So you say what is it that people are going to be focused on? They're going to be focused on running themselves in a multi-cloud infrastructure. And even then I called AI because it was clear that with the amount of data being thrown at us, the only known technique we have that can take advantage of data was AI. So I said also, let's take a look at a hard look at businesses that are there and if it's not aligned to why we have a name and trust, then we should get rid of it.

(24:13):

Example the weather company, it's a good business actually I think it's a good business, but we don't have an advertising business. The only way you make money there is advertising. So it should belong to somebody who wants to grow a big advertising business because we are not going to go do that. And I think actually if you're not a B2C company, I don't understand how you can make money in advertising. So a hard look at what should not belong and how it off a hard look at where we want to grow and pretty clear to me that software is where you have to grow. And then the third piece, there are things that were very close to the heart of IBM, I'll call it our outsourcing business. And you say, okay, is it going to struggle or is it not? And I said In a world of cloud and SaaS, fundamentally outsourcing is going to be less differentiated and so it's also going to have a low innovation.

(25:09):

You've got to run it really lean. If you're going to make money, that's the opposite of a software business where you need to invest on innovation, you are better off separating the two and letting both run in the way that is ideal for both. That's a lot of hard decisions and if you think about it and in terms of the customer impact, but my view was let's do it and let's do it quick. I've always believed if you have to do something hard, do it quicker because then you just get through it. I'm not the only one who's made that observation. I think General Patton made it. He used slightly different words, more colorful words and it goes back a long way. So we got all that done in the heart of the pandemic. If you think about the decisions, these are all decisions in the summer of 2020 and I said, because this is going to be a reset, you can see it, but if you have to do it all, put it in one my point about do hard things quick and package it up. And we got through all that decisions done by the end of 2020. 2021 was execution and that implied that you could then go into future growth at the end of 21. We were 19, 20% software in 2019. We have 45% software today, so that speaks to it. We were 55% services, some would say 60 back in 2019 with 30% services now and I doubt that will increase. Software grows at twice the rate of everything else. So speaks to the strategy,

ORLANDO BRAVO (26:43):

Love the operating comment. If you want to do something hard, do it fast and do it at once. We subscribe to that big time. Let's switch gears to AI, the point of the conference and to ask you straight out, is AI good or bad for software?

ARVIND KRISHNA (26:59):

I think AI is great.

ORLANDO BRAVO (27:01):

Please explain. Thank you. This is not scripted either. Thank you very much. I'll begin. Arvind said it, there you have it.

ARVIND KRISHNA (27:11):

I'll begin with our own team. So because always, I mean if you're not willing to do it to yourselves then you should always question is that really true? So we have about 40,000 developers who write software that includes some of the people who write software for hardware. So all in. A year and a half ago we said, okay, let's apply AI to ourselves because this is a thesis. Oh, you don't need people, you're going to like this thing. Everything becomes commoditized so let's go do it. So we built our own tools. We use a lot of models by the way. We use some of our own models but we also use open source models. We also use Anthropic. We use plenty of models. We want to be somewhat model agnostic is the wrong word, but we want to be an and we want to be using whatever makes sense for the 10,000 people who have deeply embraced it. So we are not yet trusted upon everybody. We are at 10,000 out of the 40 are using it. It's all pull. I always like to first see whether people are going to pull on their own. That 10,000 is at 45% improved productivity as a hard measurement in terms of how much work they can get done. So you one conclusion becomes, well if that's the case then you only need half the number of people and by the way, everybody else is going to get there also and they'll drop prices because it's cheaper. That's one thesis.

(28:25):

Every way of technology by the way you can count eight in total in the world has gone the exact opposite. If I can make something cheaper, I'm actually going to make more things because right now I'm thinking wait a moment, it's going to cost me a hundred million to do this. I want a 10 to one. It needs to be a billion dollar market. Can't do it otherwise, skills are limited. Now if easier skills can quickly get up to speed because of AI tools, that means I can approach much smaller markets that I might otherwise. That means I can make more products and I can still make a return and an appropriate profit. So I'm actually increasing my hiring in software development, not decreasing. This is a product centric view of the world and because we all know there is so much of a appetite for software, if you can give people more, they will be able to consume it and they will be able to use it because the western world has no more people.

(29:19):

Let's just accept it. This is the first time in human history in 2000 years that population is decreasing, not increasing. I think this is not possible to reverse this is because economics allows people to have a different life, a better life. As you get both genders equally educated, participating in the workforce, these are not easy trends to reverse. This is like 300 years have gotten us here and so with the exception of Africa, you have declining populations all over the world. What is going to give us our quality of life? Go back and look the last a hundred years you have 3% people productivity. If you have declining people, we still want 3% productivity. You've got to need technology to go replace it. Long story for we need more and more and more software, not less. And hence my point about and humans, you can look at it, we all want GDP growth because that's the only way you get a better life. And so the only way to get it is through technology to make up for the lack of productivity from labor force. So hence my belief that AI is essential for all that we are going to do. So AI is essential for software productivity and for software growth.

ORLANDO BRAVO (30:35):

I love your tie in to what you just said on your positive view that many people are saying, well, with AI there are going to be no entry level jobs, especially in development. People won't hire new developers, they won't hire young people. We will hire, you have the opposite view on -

ARVIND KRISHNA (30:52):

Will hire more entry level people. This is a formal statement. You can go look at our job postings. We will hire more entry level people in 26 than in any single one of the years going backwards for the last five. So we're kind of putting our money where our mouth is. Why I believe that we are going to be on a long-term, probably one if not two decades worth of a boom in software. So I want to bring the people in and if a entry level hire can with the help of these tools be like a five to 10 year experienced person within three months, why wouldn't you want them? I actually think the logic of the world and the media is wrong when they say there's no entry level hires because you don't need them. I think it's the inverse entry level hire can be much like a 10 year experience person really quick.

ORLANDO BRAVO (31:35):

And you're doing that. You're doing that in development, which is amazing. Now tie that into you have been one of the biggest role models in business transformation using AI and software along with Michael Dell. You did it early. Can you walk through the business processes that you took that you've transformed? Where are you in your journey and where you're putting that money that you're saving?

ARVIND KRISHNA (31:57):

So four and a half by the end of this year, three and a half billion by the end of last year, two thirds goes into software R&D, a little bit into sales and marketing and about a billion of it to the bottom line so far, and I think we are only some fraction of the way on the journey. I will not say how much. So it's really easy if you try and squeeze money in a large enterprise that it'll reappear somewhere else. So we took a very big bucket. We took all of back office, all of operations and all third party services that for us was about a billion dollars bucket. This is about a 50 billion total spend in the company. So we took that 15. The reason I take big broad buckets because then it can't hide. It can't really go from the 15 to the remaining 35.

(32:50):

That's a very hard thing to do, but if you take little buckets, people will say, I won't do this, but it'll show up as works aren't getting done somewhere else. So then you say, what do you want to do? In that 15, it took us six months to get there. We broke it down into there are 200 workflows that kind of define what is being spent in those buckets. These are pretty big buckets. I would look at what internal HR is a workflow that's multi hundred millions procurement. What we spend on procurement, which is again a few hundred million accounts payable, accounts receivable service calls from clients. This is a kind of workflows and then the question becomes, if you look at this workflow, forget how we have done it, you need an outcome. You know what you're doing today, so I'm not ignoring that. You leaders, we'll help you with a team to help you do discovery. If you had all the AI in the world, what could you do? Out of the 200 in the first year, 60 of them raised their hands enthusiastically and said, we think we can actually take quite a bit out. By the way, they all begin to say you can take 10% out, 20% out four years data. Those 60 have taken close to 50% out.

ORLANDO BRAVO (34:09):

Wow.

ARVIND KRISHNA (34:10):

The next 60 are well on their way now because they just needed to get unlocked because they observed a lot of success. 60 or 70 are somewhat recalcitrant. I don't really want to do this. I don't see this is helping me. I'm already the world's expert. I can't be any bit better. I'm okay with the first conversation like that because sometimes there are some really hard things and they should go figure it out the second time they can come in and have the conversation. I mean it's not so pleasant. If they do that a third time, they're not having the fourth one.

(34:41):

I quickly discovered you don't need to change the teams, you just need to change the leader usually. And that was sufficient to get unlock and I fundamentally believe if you just want to be really, really cynical about it, how many enterprise processes are there? But it really is people reading something, reading something else, comparing things by reading. I include simple numbers in it and then coming to a conclusion. Really the way AI models read, half of that can't be automated. Of course it can, but you have to be willing to step in and say what is the risk you're taking? Yep, I got it. It's not a human signing off, but I kind of say if the pain is only internal, come at it. I don't care how much pain there is. If the pain involves a customer, I'm going to be really, really careful.

(35:30):

Yep. Am I going to be very careful on anything I submit to the SEC and revenue recognition? Not all financials because people will hide under that. Oh, finance can't be disrupted because it's no revenue recognition is this much of finance. There's a lot of other finance on forecasting and planning. Why can't I do that using these tools? And so that's a journey we went on. I'll give two tangible examples. HR is a side of HR, which I think is very human. What kind of team do you need? Who compliments the leadership team? I think that's a very human task for a long time. There's the other side of HR. Your employees want to get a mortgage. It's perfectly natural. What happens today, typically employee calls their manager and says they need a letter. The manager has no intention of running that letter. They call their HR partner.

(36:18):

The HR partner has no intention of writing that letter. They call the HR back office. If you're of any scale, the HR back office, somebody now has no idea who these people are. They spent 20 minutes looking up various systems. Yeah, it all lines up. We'll write a letter how it goes. So our HR team came up with this. If you're on our intranet, we know exactly who you are. There is no way you're on our intranet unless we know who you are. Why do you even need to say anything? Why can't you just go to a bot and say, I need an EVL employee verification letter. Only question the bot should ask you is email or physical mail to whom. At which point, what is the total human work across the organization? Zero. 30 seconds later the letter is either emailed or printed and mailed as you think about this.

(37:08):

So what were all those checks doing before the bot can look up the systems, you give the bot access to your internal payroll and employee systems and directories and it can go do all that. At this point, 94% by volume of all HR transactions are done by our bot and our people have no idea of what the HR systems are and what they should be used for because they only interact with the bot. It's called ASK HR. All of them are in one bot. Think about now that unlocks your thinking about internal enterprise processes. Next one we began to look at was limits of liability and contracts. Oh my god, you need an expert lawyer, you can't do this. Well, you guys are fools. You need the finance team to think about all this. Throw them all into a big LLM and like you come back with really 90% fall into one of 12 buckets.

(38:03):

Got it. So I need something to understand this is the exception and it can have all of those people. Why can't the rest just not come back as, yep, it fits my template. Go. Now this does cross sales, contracting, legal and finance. This is a more complicated example. You get a sense if you're just willing to take a little bit of a risk and lean in. I go back to my reading. I mean that's what people do. Probably my son's a lawyer, so I have no hesitation in making fun of him, but he's probably, he might not even know him. Orlando may be his client. We are not his client. He only does M&A.

ORLANDO BRAVO (38:41):

Ah, okay.

ARVIND KRISHNA (38:44):

So these kinds of things I think as you unlock yourself, so I use this very simple example where you could easily see it and I use a slightly more complicated example and I think that there's a lot of these that you unlock across the enterprise. So in aggregate, we made progress on about a third of the workflows, about 200 and some progress in the remaining 70 ish. Also that is the four and a half we committed by the end of this year, three and a half was done already at the end of the second quarter. So let's cut it forward as a current round rate. It'll be four and a half by December out from the original 15, and I fully expect that this is now a muscle memory that's going to keep going. The enterprise has learned it and people don't shy away from it. Now they embrace it actually.

ORLANDO BRAVO (39:32):

I love hearing those examples and I think my partner's here as well. I love the fact that you started with HR. We've had this incredible mentor, one of the most important people that taught us operations, and he would sound exactly like you. He would start with HR and then I love what you said about the cost. Part of it goes to margin improvement. Part of it goes to tactical, which is sales, which pulls the train and then the bulk of it goes to innovation, which is long-term growth, especially now that you can get a 45% improvement in productivity. It all makes sense together. I really, really appreciate the way you spoke about that.

ARVIND KRISHNA (40:13):

The bottom margin one, because we're an acquisitive company, if you have more cashflow, does the market that you acquire more things?

ORLANDO BRAVO (40:20):

Exactly.

ARVIND KRISHNA (40:21):

So actually you get two levels on growth, the R&D, which is longterm, but also from improved cashflow. You can buy more things.

ORLANDO BRAVO (40:29):

Exactly. We got taught that operational bootstrapping through the financial bootstrapping from it. You don't need outside money to do it because you build your balance sheet and your income statement that way. It is very, very refreshing to hear that kind of level of operations and technology. We appreciate it. Let me switch to the revenue side of software and agents are now super popular, right? Everybody's selling agents and Agentic and all this. Do you see a world where agents kind of work on their own and do their own process and do their own thing? Or do the agents have to work with the existing business process and workflow? Where do you see it going?

ARVIND KRISHNA (41:14):

I probably have a very pointed point of view. At the end of the day, we do live in a world where there are capital markets that are regulators. Anything that is public has to report. You are going to need systems of record, not necessarily applications. So you can satisfy all those things. And I don't think those things are go away. You are going to need a way to say, yep, these are the people. This is how I'm paid revenue recognition. I used that example before. So now back to my HR example. If almost no employees ever interacting with the screens and the UIs and the GUIs, effectively it's becoming a system of record. Because if what's interacting with it is a machine, it doesn't really care whether it's a nice GUI or you are doing API, which the humans looks like, right? So I do think the systems of record are important because they satisfy many other purposes including recovery.

(42:16):

If a company, if something happens, but the agents are going to then operate on the systems of record, I think this is the big unlock yet to come. But the chances that I'm going to take people's salaries and feed it into a big public LLM is like zero or the chances that I'm going to take every transaction we did with every client in Nigeria and feed it into big public LLM is zero. So you are going to have agents that operate on those things, but then they might want to intersect with some summary of public information that could go outside. And so this knitting to put all this together is what is going to be the next five years of enterprise evolution. This is so reminiscent to me of the early days of the internet, 95 to 2000, all people thought of was here's a website, I'm just going to put data out there.

(43:05):

Anybody can scrape it. Yeah, okay. If all you're putting is information, who cares? The unlock came circa around 2000 people drive there moment. If I can put my inventory tied to what's on the website and I do the pricing tied to what's going on there and I knock out a tie to my supply chain, and that's a demand signal that's very different than a website with information. And suddenly you needed to make sure, hey, what comes out? The backside's got to be protected. I don't really want that to be mocked about. But yeah, if a front side something breaks, okay, who cares? And so much like that world, this is going to be, I think play out in very much the same way. And yes, to experiment, you might try something. If I want a mixer, an SMP transaction with my CRM data with something else and I want to understand what's happening and I want to pass it as a hint to my sales team to first prove it all works, I might use a public model because what I'm leaking is like this much. Once it works, then I'm going to get into what are my economics, what is my data protection, what is my leakage? Where do I want to run it? And that is I think the big unlock and the big opportunity.

ORLANDO BRAVO (44:12):

That's amazing. You once told me when I was super nervous about AI, your quote, if I can use it, there is zero chance you will not need an ERP system, right? You stand by that you see the agents working with all these systems, bringing everything together and potentially becoming the UI and then how software companies deal with that is up to them, especially the application companies.

ARVIND KRISHNA (44:37):

I do think, again, some are going to go to zero, there will be always, there's some you can debate 10, 20, 30% that is left as human, but the rest will be an agent or a machine leveraging these systems.

ORLANDO BRAVO (44:50):

How about the customers? Are the customers getting sick of AI right now or where are they? Where do you see them?

ARVIND KRISHNA (44:58):

I don't always love this, but I think the Gartner hype cycle is probably the best way to explain it. You first have your chasm of death that in this case was pre LLMs. People made it through, you reached a trough of illusionment first. I guess. Disillusion comes next. You're not yet quite there. I think you're on the way down right now. So you were at that extreme hype probably mid late 24, and now you're coming down towards the disillusionment because these are all those studies you see, but I think some of them are a little exaggerated. It's gives great hype to the people who do them. The one from MIT that said 95% see nothing and 5% only see something. We did a survey across three, 4,000 companies and we got 75, 25. I do think that people haven't thought through the scaling question and the economics question.

(45:55):

So if you do run and do a hundred experiments and now you suddenly say, I want to scale them all and each one of them is completely different, how much time I going to spend on your legal side, on your security side, on your compliance side? And I think that's where people are caught. So you've got to say, I give people the advice. Pick three things you're going to scale, do them really well. Now you have a methodology and the next 97 will go fast, not the other way around because now you have anarchy. So I think that to the point you're asking, I think this is going to play out now over the next few years.

ORLANDO BRAVO (46:35):

And pick three. I love that. It's like three priorities, three key areas. Another thing you've said is pick some that are low risk. Yes. Don't go for your riskiest process. You might not have the guts to make those decisions.

ARVIND KRISHNA (46:50):

Correct. If somebody tells me, take your internal revenue forecasting and move it into an AI system, I'm going to say, wait a moment, this is kind of important to me because I have to decide how much I'm going to spend or not. Am I willing to run it in parallel just to see how good it is? Yes. But replace No, but I go to my point about internal pain is only by your own employees. So if you're going to go do something like go take areas, like that's why I pick HR because in the end, outside of EEOC compliance, all the rest of it is just internal. I tell my employees, Hey, this is pain internally, you're welcome to complain. Don't expect I'll listen.

(47:36):

I am pretty upfront about it because my experiences that people are willing to complain about anything. I'll give you a funny example. I always hate redundancy. See, that's I, my people came to me and said, oh, we have this tool that lets people plan their vacation. So that in complicated things, when there's hundreds of people servicing a client, we can all make sure that it's staggered and nothing is missing. I don't know about you guys. That's called a calendar. So why did you guys just use your Outlook calendar to do that? No, we can't do it. It doesn't do enough. I'm very cynical. It doesn't do it because they don't want to expose their calendar to all their peers. Okay, so I said, get over it. We are canceling this. If you don't want to do it on the calendar, do it on paper. They all started doing it on Outlook, which is what I meant by internal pain is fine, no big deal. Then that's the low risk point because it's not impacting those things that can get you into regulatory trouble or jail. So you don't want to break laws. That's the highest risk. Anything that impacts a customer is in that boundary of highest to medium because you don't want to give a customer pain, absolutely not. But if the pain is to our team servicing the customer, maybe I'm willing to do that. And internal is low risk.

ORLANDO BRAVO (48:52):

There it is. You're helping us here because we preach a lot of the same. I have a general management question for all of us here that oversee companies and leadership. When you see revenue softness and you see that pipeline tightening up and you see a difficult time, what do you do about the cost? How do you do it?

ARVIND KRISHNA (49:21):

Well, so before you talk about cost, you got to ask yourself the following question. Is it the market? Meaning you can do whatever you like, but the market is softened for those things completely. Or is it your capability, meaning you are less competitive than others, or is there a business model change that is happening and you haven't woken up to it? If it is one of those three, that's a very different answer than cost. Now, if it is the market truly, and you convince yourself there's no point in just spending a lot of money, you can try, but maybe you'll take some share from others, but then very quickly it'll go back down. Then you have to think about, I would first rip out if it's softening that much and why am I spending that much on sales and marketing? That's the first cost to go start to rip out because if it's just going to become a fixed set of customers, you Fdon't need that much sales and marketing. And the second thing to say is why and how much innovation are we delivering? If it's too much, customers can't swallow it, if that's the case, but that's second. And then third, what is the cost in what you're doing, whether you're delivering it as a SaaS service, et cetera. So depending on the evolution and the stage, you have to track it. It's unemotional, it's a business. It sounds too simple.

ORLANDO BRAVO (50:39):

That's exactly right. Depending on the core issue, you have to find the core problem and first and then you address it. How about the other way? How about when you see revenue acceleration? How much cost do you release?

ARVIND KRISHNA (50:52):

So we are lucky that we already have a large sales team. I tend not to release a lot of sales costs initially because salespeople are wonderful. If they see something is hot, they'll gravitate to it on their own. So the first bit of acceleration you can get by just people realizing that the friends are all making money on doing something. But that I agree comes because we are already at scale. If you're not, then you have to think about, do I need money for a country or a continent or a approach or a partner, but we have at least some coverage in most of those things. The question you have to always ask yourself is how much innovation or R&D cost do you have to begin to release? Because that is where then will that help growth? Will that hurt growth? And is it A lane? If it's A lane, then actually a narrow team is better. You don't need to release a lot. But if there are opportunities that could be coming in a year or two, if you are going to see that same success, then you have to release some for those. Or you say, Hey, this is going really fast. Maybe I should do M&A because I don't want to wait two to three years to get something built. So there's different kinds of answers, Orlando, depending on where you're

ORLANDO BRAVO (52:03):

At, the root cause where you're at.

ARVIND KRISHNA (52:05):

Yeah.

ORLANDO BRAVO (52:06):

Wonderful. Let's go back to talking about AI for a minute on LLMs. My partner Holden, wrote this amazing article, Holden, I think it was a couple years ago now, on small models on their virtue. You've been a big proponent, you've been pretty vocal on how models will evolve over time. Can you talk about that in terms of the cost, the cost of goods sold, and what's going to prevail in the enterprise?

ARVIND KRISHNA (52:35):

Look, I think it's add, and I would not say that it's one or the other. Large models are very, very useful, but they're also very, very expensive both to train and to run. Why do I'm separating the training and the running who is training them eventually has to make the money back. So it's not just the cost of running them, they also have to make back the cost of training them. So effectively the cost of a very large compared to a small model is about a hundred to one. If you look at the very large, they're all talking about spending 50, a hundred billion a year or more for training. And then if you're running a 500 billion power model, you're running it on a large cluster of the latest and hottest GPUs. A 30 billion parata model takes you a few million to train, not billions, and you can run it on one GPU.

(53:30):

So you can look at that difference and say that's a hundred to one. But if you don't know what you're wanting and if you're in a consumer business, a very large model is very useful because if you can have a model and that's got a few hundred million customers and you now add finish capability, you can add 5 million more. You add French to finish, you probably add 5 million more. You add Japanese haiku, you probably add 10 million more. So I get it, that's the usefulness. But if all I want to do is compare one legal document to another, why do I need that model? You don't, it turns out that a 10 billion parameter model is as effective if not more than the largest on that problem. So for the enterprise use case, that's not the consumer use case. I would say if it is 10% cheaper, who cares if it's 99% cheaper?

(54:22):

I think you care. And so that's why I believe that small models have a huge place. The second isn't domains. We look at things like climate change. We look at things like time series analysis. We look at things like chemistry and discovery. Small models are far better because they don't come with all kinds of crap that they've ingested from the internet. And so that's a hallucination because it's got something in there that is contradictory, and once in a while it'll serve up the contradictory fact as opposed to a strained on a tighter known corpus. I'm not saying there's no contradictions even in the tight tone corpus, but far less. And so that's why I'm in the and world. You use very large models both in the consumer space and to maybe prove to yourself something's possible. Then you got to say, where should it run? And then you hone in on both the economics and the feasibility. And in some cases people are just worried about what's proprietary. If you put thing way out into the public, there's always chance. Nobody's going to steal the data. I know people worry about that's possible, but unlikely. But is the model implicitly learning from what you're asking it? Of course. That's how they work. So you are implicitly teaching them something. So if you're water deeply about dual code and certain proprietary, I would never put that on a public model.

ORLANDO BRAVO (55:45):

Look, I want to end with you sharing with all these incredible investors here, what your plan is for IBM. What's the strategy? What's your plan?

ARVIND KRISHNA (55:57):

Look, we started off, we said we'll do low single digit revenue growth, then we upped it to five. You should imagine that's not the end of the road. We began when I began, actually 2020 was a really rough year, so I got a huge advantage. We did 6 billion of cashflow. This year we'll do 14, and we were a one fifth software company. We are almost half now, and Quantum is going to be an accelerator that's not inside any of our numbers or projections in terms of what we put out to the street. That's the plan.

ORLANDO BRAVO (56:31):

Unbelievable. I trust that there's more software in that plan for sure. We have some nice companies here. So when you produce all that cash flow, maybe you could make some great investments like Red Hat with us. I really, really appreciate your time. I know how valuable it is. The fact you agreed to do this means a lot to us. So everybody, please give a round of applause to Arvind and thank you.

Listen to Thoma Bravo's Behind the Deal, Season 4 on Spotify, Apple Podcasts, YouTube, or wherever you get your podcasts.