All right. Thank you, Michal, for that nice introduction. Arvind, my welcome to you also. I think we are going to have a good conversation. Much to discuss. Michal mentioned a few things, whether it's quantum, AI, hybrid infrastructure. As I was thinking about this conversation today, I was reflecting on the fact that we're in a unique moment here as AI increases the capability of compute and quantum increases the power of compute. It's a unique moment in history. We have these two powerful forces coming together, and I was trying to reflect on when in history I've had similar moment in time. I thought, 40 years ago, you've got the advent and proliferation of the personal computer coupled with the beginnings of the internet. You have to rewind 125 years, maybe, to the turn of the 20th century in the early 1900s, with the advent of transportation, communication, aviation, motorcars, electrification. This is really a unique moment, and as I think Michal alluded to, who better positioned to talk about this than you? Again, welcome. Let's start with the core IBM strategy. IBM, and again, Michal alluded to this, has spent years repositioning itself around hybrid cloud and AI. As you look at the explosion of foundation models, particularly in the last two years, what did you understand earlier than the rest of the industry? Are there things maybe that you feel you were late on? Yeah. Andy, first, thank you for having me here. It's great to be back. I think I did this a couple of years ago, if I remember. Michal, thank you for those warm words. You took a very long arc, Andy, all the way from a century and a quarter, electrification to roads to transportation. I always like to think about, A, is there something in the technology? Cloud was pretty obvious. There is something in there. The question that you have to ask is the average for the moment, let me say, large client, government, going to put all their eggs in one basket, AKA put everything into a singular public cloud? We were pretty comfortable saying no. Homogeneity is a bad thing in nature, it's a bad thing in business. You'll give too much asymmetry maybe in the relationship if there's only one vendor. History has shown that's generally not a great idea. Regulators are going to get worried about in case innovation begins to come down at one place. Not that it ever has, but maybe it has. We were pretty sure people would pick a few. That led to the conviction on hybrid, and that led to the conviction on Red Hat, which Michal mentioned. I do that because there is an exact analogy going on in foundation models and large LLMs. It takes a huge amount of investment to make these. There is an incredible amount of productivity to be gained, which both of you alluded to, and we'll get to that. For a moment, let's not question it, just say there is. If you go and look a year and a half back, I would turn around and said, "Looks like Claude, Anthropic is ahead on a number of dimensions." Six months later, you might have said, "Well, maybe OpenAI is within 90, 95% of that." Maybe two months later, you'd have said, "Well, Gemini 4.5 is within 95% of that." This is going to be back and forth, and we haven't yet talked about what xAI might do, what Meta might do, what Microsoft might do, what Amazon might do, what the big three Chinese models might do, because it is a global stage. You kind of reach the conclusion that these are going to become commodities. Commodities don't have no value. By the way, platinum is a commodity. Gold is a commodity. They have a lot of value. Commodity to me means it's from the economic perspective, there is low switching cost. If there is low switching cost, the same thing that played out in cloud and before that, by the way, in servers, if you go back to pre-PC. The same exact thing is going to play out where people are going to say, "I need a way to go across these." The productivity, by the way, as you said, why did you see this? The amount of data is multiplying like crazy. The only tool we know which can digest and make some sense out of the volumes of data we got is AI. That led to sort of those convictions and that trying to sort of see around the corner that these are going to be structural tailwinds for a decade, maybe two. Right. Let's talk about how companies incorporate that. Think 2026 is IBM's flagship annual business and tech conference. It happened last month in Boston, and you argued at the time that the winners in AI will be companies that redesign their operating models around and incorporating AI as opposed to just deploying models. What does an AI native operating model actually look like inside of a Fortune 500 company today? Yeah. I'll give you an example from us, but which I think applies to a lot of the companies that this room follows. We are reasonably acquisitive, as Michal certainly knows. We typically buy 8-10 companies a year. Two, three make big headlines. The rest are much smaller, so they don't tend to make a lot of headlines. Historically, what did you do when you bought something? You looked at their expenses, you looked at their various silos, you looked at their various functions, you brought them in. By the way, typically, five years ago, most companies would say for the first couple of years, you actually make them less profitable than they were. You get revenue growth rates, and you go on. What do you think the economics become if I can look at you and say, "On day minus one-" The day before you close the acquisition, I can add 10-15 points of profitability instantly. How much does the aperture open up of how many more companies you can buy? You look at me, how can you do that? Well, typically you take in the entire what I call back office. How do you do contracts? How do you do sales management? How do you do accounting? How do you do revenue recognition? AKA the back office of G&A, which is 10 points, by the way, at most small companies, 15 if it's not perfectly run. You can now use AI to pretty much put all of that into your existing systems, and that means you're adding that, which immediately then means that you could actually invest more in sales, more in R&D, and increase the growth rate even more. That is really what I meant by thinking about an AI operating system. It's not like a tweaking of, "Oh, I'll take 10 points of G&A down to eight." A lot of people will do that by squeezing. Take it out, because you can now use all these tools to do it. You got to approach it from a risk lens, because people will say, "Well, you can make a mistake." Believe me, when you have humans doing these things, you also make mistakes. The AI actually makes fewer mistakes on rote work than most people does. You then sort of say, "Well, you're cutting out all those people." Well, you're actually adding a lot of people, too, because the functions I mentioned. Now you take that approach and you say it's not just back office. I think that code productivity is going to happen. I think that's a given. 40%-80% you can debate. I think the holistic number is probably 40%. If you just take a narrow thing like writing code, that's probably 80%. Holistically, you got to test, you got to talk to people, you got to design. If you take across the full gamut, you can get 40%. Customer service, I would turn around and tell anybody who runs a consumer customer service, "If you're not automating at least three-fourths of it with AI, you're going to fall behind to people who are." By the way, this has gone on for 40 years. For the first 30 years, this was offshored largely. People talked about putting large call centers elsewhere. Well, machines are even more productive and more effective, you're going to go ahead and do that. Enterprise back office, rethink how customer service works and how you get software development done. Those are pretty big, massive changes to how companies have to operate. To be clear, I wasn't talking about those who are building models, by the way, if I remember the statement that at least two of these six companies made is that 80% of the code today is written by the models themselves, and three years ago it was zero. That tells you the rate of improvement and the rate of progress, that is one of the proof points, and I gave you one from ourselves on the back office, and I think customer service is obvious. Yeah, I think that's consistent with other speeches you've given in terms of that the enterprise AI will be all about orchestration, governance, and workflow integration, not just bigger and bigger models, right? Here's the counter. Those that, OpenAI, Anthropic, you've alluded to, they're enjoying tremendous momentum, right? Tremendous absorption of capital, ability to receive capital. I'm sure everybody in this room is well aware that OpenAI confidentially filed yesterday for its public offering. Anthropic was a week or two ago. Between the two of them, that's $150 billion-$200 billion. Dwarfs all the IPOs in history, other than SpaceX, which will price sooner, most likely. What is the state of play in enterprise today? Are you seeing accelerating demand for orchestration and governance of multiple agents, or is it still about demand for vertically integrated systems? We have a book of business, which Michal referred to. It's about $12.5 billion as of December. Once it reached that size, we kind of stopped even talking about it because it was big enough to show it. I think we have $0 from direct users of models who don't need us. I think a customer will directly go ahead and use the model. If you want to have the model run around inside the enterprise, look deep inside your SAP system, look deep inside your Salesforce system, come and tell somebody what to do, worry about compliance, I'm not sure that any of them have built that directly. I think it's an and. I would not paint it as an or. The models are going to be incredibly useful, and they will do a lot of individual productivity. They will build some unique applications of their own. The moment you get to have to touch the enterprise as is, and it's not a greenfield, you are going to need people with the expertise to say, "How do I put the right guardrails?" I'll put that word rather than pure governance right now, because if you're living inside with your own data, it's more about putting the guardrails to make sure that only the correct people are getting access to the data. You're making the agents work on the right data. There is still a lot of dirty data inside the enterprise, so that has to be cleaned up. That is kind of where we will play a role. I don't paint it as a it's instead of or it's an opposition to what they might do. The same way as you mentioned the internet. In that era, there were a lot of companies who will do direct access to people. Think of all the social media companies. Think of streaming companies. That is there. Is there also an internet improvement of the business model for enterprises in the B2B world? Yes. It's an and the same exact thing is going to play out in the AI world. I think maybe your and comment ties into your position on whether this should be an open or closed proposition, and you've articulated that you think customers should avoid being locked into any single vendor. Want to expand upon that and sort of explore that philosophy? Sure. If you I just talked about how these different models kind of bounced around all over the place. One does video, then they refuse to do video. One does photographs, they refuse to do photographs. If you lock yourself into one, you're completely at the mercy of what they might choose to innovate on or not. If I look at the history of technology, maybe I have a little bit of age on me, as Michal said 30, I'll be precise, 36 years at IBM. Companies that are very innovative sometimes are not. The best people get persuaded to go elsewhere or maybe the capital structures are such that companies cease to exist. Not saying that will happen to any one of them, I'm simply pointing out what has happened if you look at the arc of history. I think it would be irresponsible for a company to get locked into only one. I believe that they will, over time, figure out what they're really good at, and they'll do those. It behooves a company to say, "I'll leverage innovation in the best possible way across a few." The second thing I would always say is pricing power. In order to get pricing, you got to use at least two things. If you don't use two things, what negotiation do you have? Yeah, you're at their mercy. I'll switch gears slightly. We'll talk about one of IBM's business lines, the consulting practice. IBM Consulting's been under pressure at times while both software and infrastructure have continued to accelerate. I'm wondering as to your thoughts on how AI changes the economics of consulting itself. Does it automate more implementation? Does it become more strategic? By virtue of AI, does it actually become structurally smaller? I think that the number of people in consulting, if the revenue was to stay constant, will likely decrease. Why do I say that? If you're going to say that AI agents and models can do some of the work that people did, and we can debate, is it 10%? Is it 20%? Is it 50%? Is it 70%? Then by definition, you need fewer people to get the same amount of work done. Now, I think that's a given. That's going to play out. There are two tailwinds then that help mitigate some of that reduction. One, I think the amount of work needed to be done is incredible, if the unit cost for doing it has come down because you're using AI tools and technology, not just people, then there will be more demand for the total amount of work, albeit at a lower unit cost. We can see that play out right now. The second thing that's going to happen is those who embrace it a lot more are going to be more effective and efficient at gaining market share. You will also gain market share from those who are reluctant to embrace it to that extent because they don't know what to do with their people, they have a sort of a labor based business model. I think that those things are going to play out for sure. I would be surprised if consulting is not at least half aided by digital workers or AI agents, take your terminology of choice, in the next 3 to 5 years. It's not going to be that long. That is already happening. We have over 200 large clients where a substantial amount of the work is being done by AI agents in addition to humans, so that speaks to the rate and pace of adoption. To make it very tangible, if I think about 5 years ago, if we walked into somebody and said, "We'll do an SAP implementation," we probably had to have 30 people spend 6 months figuring out what their existing process is and how would you map that in. Today, that's about 6 weeks with a half dozen people. That tells you how things are improving because what are those agents doing? They're reading the process documents. They understand how to map it onto SAP. They can go ahead and do all that, the people are left to go fill in the gaps where there is no documentation or where you have to conduct interviews in order to go get information as opposed to the raw gathering of data and assessments and reports. This is very real. Yeah. I see that play out. The strategy we have chosen is we're going to embrace it hard. We will give our clients that benefit. Will that let us gain market share, both in terms of volume of work as well as from others? I would say early signs are there. We grew 1% in the first quarter, down from negative numbers in the first half of last year. If I look at that vector, it's sort of turning around into growth, but this is going to play out over the next few quarters. Not next few years, next few quarters. Stepping outside of IBM, where I know you're deploying it as you just discussed, views on sort of general employment figures for the U.S. Are we augmenting knowledge workers? Are we complementing them? Are we replacing them? Just curious as to your macro effect. I'm going to take a very long arc since you gave me that permission by the opening comment. 125 year view. I'm actually going to go back to 1776. Okay. Only one point. 97% of the U.S. population worked in agriculture. Personally, meaning they worked on a farm. Year 1900 was 47%. Year 1950, just after World War II, was somewhere in the high single digits. I forget whether it was nine or 11, something like that. It's 3% today. I don't think 94% of the population is unemployed. It gave rise to things which you couldn't do until automation took those tasks away. As people began to get time because of automation with machines, harvesters, tractors, et cetera, you opened up, you said automobile. That combined innovation let people go to nearby towns. That gave rise to malls, gave rise to fast food, gave rise to restaurants, gave rise to hospitality, and that was 20% of the economy. I think as this is going to go on, I personally believe we will get about 10% job displacement if I take a decade. I think 20% is possible, but I'm squinting because I don't think it'll get adopted that fast. 10% displacement will happen, but more than 10% new jobs will get created. That is going to be the balance. By the way, 10% over five years is something that any advanced economy ought to be able to do because that's 2% to 3% a year that you got to do in terms of flexing. You got to give people. We give a lot of upskilling, reskilling, can you change your role opportunity inside our company, I think everybody ought to go do that because without that, then you're going to have a disaffected 10% of the population, which would not be good for all of us. All right. That's good to hear. There's obviously been a lot of speculation, different articles as to this doomsday effect from AI, which about a month ago there was one. It's good to hear that. I'll give you one simple stat. Lot have been talked about, "Oh, in the tech industry, you're losing jobs, has GenAI already causing a dearth of jobs?" All sorts of just look at facts, I'll call it, not speculation. The number of non-farm job openings exceeded the number of people looking for jobs. I think the jolt of the job openings to those looking is at 1.1. That tells you that, okay, the facts argue the opposite. Maybe some jobs are being lost, but the kind of new jobs being created exceeds the number of jobs being lost. Yeah, I think. Which is just a fact. It's a pivot, too. You think about two years ago, kids graduating from college with computer science engineering degrees, and the ability to code was the hottest thing, and two years later, it's virtually irrelevant. The focus is on how do you prompt AI models. The design, it's just tweaking, but you still have. Actually, all those who, if they have a good degree from a good place and they can't find a job, please send us an email. We tripled our entry level hiring, 2026 to 2025, for those kinds of jobs because I believe that people are getting scared and they're under-confident. We want to hire these kids who really know what to do because they are much more adept at AI than actually some of their older peers. We'll take them in, and we'll get them to produce even more products so we can go take market share. Our number of tripling entry level hires is a real number. It's actually out there. Fantastic. Michal mentioned the, you've mentioned the Red Hat acquisition again, 2019, $34 billion. You were a primary architect behind that. Was that primarily about cloud architecture at the time, or was it actually about preparing IBM for the AI era? Again, back to that, what was the hint back then that made that such an attractive acquisition for you? I was 100% sure that people are going to modernize their entire infrastructure. At that time, they meant that some of it is going to get lifted onto clouds. I was also sure that sovereignty would play a role, especially outside the U.S., people would not be willing to depend only on a public cloud. Anybody's public cloud. You can cut a fiber optic cable, trawlers can pull things out from the seabed, lots of things can happen. Unfortunately, it actually played out that way. if you look at the last five years. Because of that, I was sure people are going to need both sovereign infrastructure and multi-cloud infrastructure. Red Hat looked to be the most attractive set of capabilities to play into that. I will point out, the day we announced it, we took a 15% stock hit. The complaints from the community at large were, "You paid too much. You're going to destroy this company, and we don't understand the synergy with IBM." Since then, it was $3.4 billion in revenue when we bought it. It's probably a run rate today is $8 billion of revenue. The profitability has increased, the growth rate has increased. I don't know. You tell me whether for $8 billion of revenue, $34 billion sounds like too much or too little. Look, I think your stock price also tells a nice story, right? Pretty flattish in the 120, 150 range up until 2023, a couple of years after Red Hat, and crossing 300 last week. Hybrid infrastructure, you just talked about it, right? It's on-prem, it's public cloud, it's private cloud, it's edge devices, it's mainframe. Is there an ideal SKU? Is there an apportionment that is the best for a particular enterprise? If I go across enterprise globally first, I used to think that it's about 50/50 is probably ideal. Because I think certain workloads, it's fit for purpose. I think certain workloads ought to be only on public cloud. Why would you put your marketing website on anything but a public cloud? Why would you put streaming on anything but a public cloud? If you have a consumer application, why would you put it on anything but a public cloud? I actually think that that is a better answer. Now, if you need to have triple active payment systems like credit card authorizations, where you need it to be triple active to make sure that it can never go down, and even if one of them gets attacked, because attacks will happen, that it can recover, and you can pass the workload over. Not that it ever happens, but you have a network outage in a fourth of the country. For those kinds of systems, actually, the mainframe is a superior economic and technical answer. It is for those kinds of workloads that are of large volume and need the resilience. You put those in, and it's those. Then you have the middle, which is people are going to have their own data centers because of economics. If you have a vendor, you are going to have to give them 30-40 points of margin. There is no way around it. If your workload is big enough but it's not volatile, you should look at the margin and say, is it better to go run that in your own data center? Those are the three choices I think that have to get made. If you're running a factory, you are going to have to put local stuff because the factory, in order to keep working and to make sure the machine doesn't do something awry, cannot actually live with a 100-millisecond round trip because 100 milliseconds is a lot of time when you're at machine speeds. There you're going to need local infrastructure. Yeah What you just call the edge. Data for an operating theater, I don't think you're going to take the risk of it going all the way back to a data center across the country or across continents and back. It's going to need something local. You may have answered my next question. I do not want the question to be in and of itself the answer, the mainframe business has remained surprisingly resilient, and I think, 5 years ago, everybody thought, "Okay, mainframe is going away," exactly for what we just talked about. Everything is going to the cloud. Yet AI has suddenly made it strategically very relevant again. Agreed? I think, like I said, the question sort of is the answer. I will add a couple of pieces. Historically, when people wanted to do analytics or AI on the mainframe, they would take a copy of the data off, they would go train models or analytics someplace else, then they would do sampling. That would inform maybe some rules that you would put back on the mainframe. The ability to do AI now on the mainframe, I am not going to say the training, but to do inferencing on the mainframe opens up the aperture to say you can do inline fraud on actually every single transaction because the latency is now sub-millisecond. That is an innovation that we did to add the capability to do 450 billion inferences a day on a single mainframe. That opens up new workloads, and it is part of the growth that you talked about. Right. As you said, 5 years ago, people were predicting, "Look, this is just going to keep decreasing year-over-year." This last one, I think, is sitting at 120%, roughly, of its previous program. The prior one sat at 120% of the prior one. For 5 years now, we have had systemic growth in terms of both revenue, but also in terms of overall capacity. Mainframe capacity is roughly double today of what it was 3 years ago. You just talked about speed a number of times, the importance of speed. A perfect segue into a quantum compute discussion. IBM has the largest fleet of quantum computers, I think 90 systems across the globe. You have talked about delivering the world's first large-scale fault-tolerant quantum computer in 2029, clearly doubling down on quantum. The government is behind a quantum chip manufacturing initiative with you, as well as, I think, last week's announcement that IBM would invest $10 billion over the next 5 years in quantum. You have said it is moving from science toward engineering. What are the specific milestones you would like to see that are going to convince skeptical C-suite officers that we have crossed from experimental technology into actual commercial quality things? Look, I think that the skeptics should always look at what problems can you solve on quantum that are either economically or just infeasible to do otherwise. We have one of our partners, the Cleveland Clinic, and we're incredibly proud to be able to work with them, and they've been doing cutting-edge medical research for 100 years. They recently said, "We want to understand if we can model how proteins' properties are, because then we can figure out which drug, AKA small molecule, will attach and change the properties of that protein." Here is the rate of progress. Working with us, June of 2025, you could model maybe a five-atom protein. That's not a protein. That's a toy. And to be honest, five atoms, a really good chemist could probably do it by hand on a piece of paper. Last November, they did about 300 atoms. Okay, you can't really do that on paper, but you can easily do that on a classical supercomputer. This April, they did 12,000 atoms. Think of that progress, five to 300 to 12,000, and they're en route to about 30,000 as we speak. You can't do this on any other method. You can now begin to understand the 12,000 and 30,000 is a protein called trypsin, which is actually one of the building blocks that breaks down other proteins into different pieces. Suddenly, you can begin to do problems that are really interesting. I think that is what people ought to pay attention to. It's a mixture of how good is the hardware and how good are these new, I'll use the word, algorithms. That mixture has always been what drove new computing. It's not going to replace AI, it's not going to replace CPUs, just like people who are confused that, "Wait, do GPUs replace CPUs?" I think this last year, the semi boom has shown, no, you need both. Right. I think QPUs or quantum is going to be the and that solves certain problems that are economically infeasible to do on the other two technologies. This is what I'm going to play on. Well, they're complementary, right? That's my opening comments. Yeah. You have this amazing marriage going on between the two. I'm conscious of time. I want to hit a couple topics. The government's investment in this quantum effort, are we entering an era where AI infrastructure and quantum infrastructure could start to look like our energy infrastructure? Is there a risk that it sort of gets nationalized in all but name because it's that important on the global stage? Yeah. I want to be clear, the government did an investment. We were very proud to get that investment. However, it's a minority position in the entity we are standing up to build chips for quantum. It's not a major. I believe that it is largely what I will call that the government wants to help endorse. What is the purpose of government investment and government industrial policy? For all effective industrial policies, it has always been, A, can we nudge the industry to do something which they might not do? The second one, which I think is playing out here, is can we get much faster speed to get to the endpoint than we would otherwise? The fact that they invested with us, one, is an endorsement that they think we are in the right direction. Two, since it's about a foundry and it's not about doing the R&D, it's also their assessment that you're going to need this amount of scale because we can see this coming in the next two, three, four years. That is why that was a really interesting investment from our perspective. No, I don't think they are out to nationalize it and to say, "This is highly, highly regulated." If I look at what they did, even on the AI models in the last I/O, I thought it kind of hit a Goldilocks. Look, we want you to give us 30 days. We're not going to force you, but if you're a good citizen, you'll give us 30 days. My reaction is, that seems appropriate. If they had done six months, that slows people down. The voluntary would not have worked. 30 days is something that I think most responsible people can stomach. Yeah. Talk about navigating the relationship with the government and the interesting things, because I appreciate everything you've just said, but also in the back of my mind is the government investment in Intel, right? A different moment in time, but they're actually a stakeholder in Intel. Not the case with you. Well, yes, and in our case, it's in the new entity, not in us itself. I don't know. Intel stock was 20 when they made the investment. Yes. It's over 100. It's a good return. I don't know. Is there any employee, shareholder, or customer of Intel who's unhappy with that? To me, it's not just the pure investment. Just like in TARP, just like in GM, just like in Ford, and going back to the crisis, I think it's appropriate that the government gets a return if they're helping industry out. By the way, in all those examples, yeah, eight out of 10 gave them a great return. There were a couple of ones out there that went flush, just like happens in investments. I think an appropriate return is there. The biggest piece is what I think about is the endorsement of a technology that otherwise there would be skeptics on. If I go back all the way to post-World War II, I'm not sure that investors or individuals would have invested in rockets and jets until the government first proved that they can work. Once the government put their money in, that's one way. They could have equally put their money into companies that were doing that, which is what they're doing now. I think that that leads to more speed. Okay. I get a little philosophical with you for our last couple questions. Michal mentioned the 100-plus year history of IBM. You're simultaneously one of the oldest technology companies, and yet you are one of the ones that is betting hardest on the frontier technologies. How do you keep IBM culturally ambitious? How do you keep the workforce true to the identity? How do you maintain the reliability that enterprises expect from IBM as you traverse these new pathways? I have a deep conviction that the biggest risk is taking no risk. Let me put it in very simple economic terms. If you reach a place where you're kind of comfortable with your portfolio set, who you sell to, how you sell, you're kind of muddling along, I'll call it. Maybe you're growing at inflation, maybe sub-inflation. What happens? If you've got a profit pool and a revenue pool, people are going to come after you. If you're not innovating and increasing that gap, you're giving them time to catch up. They will figure out a better way to come at you, and they will not take everything, but they may take away your 20% most profitable clients. Or they might take away the 20% who are okay with a slightly inferior answer, to use the old Clayton Christensen dilemma. You're in some decline. If your model is to, "Oh my God, I'm conservative, I got less, I need to invest even less," you are going to eventually fall over a cliff. The only way I believe to sustain is to actually take risk and to innovate, to say, "I'm going to increase the gap because I'm offering our clients more." These are not dilemmas to me. I actually think that that is the worst behavior, to kind of get conservative, to say, "Look, I need to be ultra-reliable, so I won't do anything new." That is almost guaranteed going to get you out of business. That is why we do that. The art of it, maybe, is to figure out where do you have unique skills? Why do you have the right to go win in this? We talked about quantum. To win in quantum, you need to understand material science, if you're actually building a quantum computer, not just the algorithms. You need to understand semiconductors. You need to understand how do you build a complex system that you can keep up and running. That's why we have 90, and I don't think anybody else has more than three. You need all those skills. I said, "Okay, maybe we are structurally advantaged here. Not a guarantee, but we are advantaged. Can we leverage that advantage?" At some point, you've got to say, and you've got to pour money in to increase your speed, which is why the $10 billion. That is that sort of mix there. You got to pick. We could not pick five areas to do that in, we had to pick. We picked hybrid cloud, we picked AI, but we're not building the foundation models, we're using them, and then quantum. That's sort of three very big bets on which we do spend that kind of money. The returns are there for two of them, and they will come for the third as well. Well, the market is a believer so far. Congrats on that. You've given some insight into your leadership philosophy in your answer there. I thought I would key on that to say, what are the bets or the things that worry you the most right now, and what excites you the most right now as you look at the next handful of years? Look, what excites me is that I think we have really unlocked a lot of risk-taking inside the enterprise. What worries me is it's really easy, if you look at history, if people get insular or if people begin to get into decline as opposed to growth, it's very easy to fall into a risk-averse attitude and to sort of fight over the same pie. That becomes then internal as opposed to external. I think what you have to always be focused on is where does the client find value, make sure that that's where you're going, and to make sure that you do it in the areas where you have significant strength and innovation. Don't try to do it across the board. That's kind of what I worry a lot about. We made some right bets. Back in 2020, we made a decision to make sure that we didn't have a single constraint in our supply chain. Took us two, three years to get it done. I think when tariffs rolled around, we were ecstatic that we'd made that decision. Those are the kinds of things that you have to sort of think about and see, it's not a guarantee, but you think it's going someplace. What should you do to make sure that you're well positioned despite those headwinds? As you said, you were kind enough to grace this stage two years ago. As a final question, that we have some time for the audience to ask questions, I reviewed the conversation you had two years ago, and frankly, there's very little overlap to the things that we're talking about today, which is exciting in many respects. Is there anything that we haven't talked about that's been a big difference for you, for IBM over the last two years that might round out the conversation today? I think the geopolitics is one that we can't help but not mention. I think two, three years ago, we were seeing the beginning of it. Russia, Ukraine had broken out. If I remember, the war in Israel, I think, had started. I would've felt that those were maybe two, three-year things. The fact that they've gone on for five, and has definitely been much more extreme than I would have thought. The other thing is, I think a lot of people are convinced of doom and gloom. These things are going to cost two, three, four points of GDP headwind. I actually thought it'll cause only a half point of GDP headwind, and that seems to have played out. Yeah. I think that economies today are a lot more robust and policymakers have more tools that, at least in the short to medium term, they can actually keep the impact minimized. Of course, we'll play out the debt story over time. That's not going to be a short-term issue. Okay. Well, thank you for that. I've got a few minutes remaining, I'd love to take any questions from the audience. Should be regulated going forward? He'll probably mention Mythos. Yeah I assume so. Yeah. Look, it is really interesting questions, and let us just postulate. There are opinions here on every end of the spectrum, from there should be no regulation, no guardrails, to we need massive regulation and massive guardrails. These opinions also somewhat align with the political spectrum, sort of right to left, in what I mentioned. I believe that in the early days of a nascent technology, over-regulation is a terrible idea, and I think we see this exemplar, how many AI companies are there in Europe compared to the U.S. If people feel there is more of a headwind, I have to convince everybody to do things, it just puts a bigger burden. That 10, 20, 30% headwind is enough for capital to decide, "I'd rather do it someplace else." If it's going to be anywhere near as big as we think, that will be a terrible outcome. As time goes, I actually believe that there should be a risk-based assessment. If your use case is extremely risky, let's suppose it's doing surgery on you. Well, we've already got the FDA, they already regulate medical devices, so they're going to come in and want to regulate what's it doing and how is it doing it, and how safe is it? If it's making credit decisions, you already have the Treasury who's going to regulate a bank to go ahead and do it. I think there ought to be a risk-based assessment as we get down the road. Maybe early on, don't take those extremely risky use cases. I think it's the balance to do. If you look at the current one, as Andy mentioned, Mythos. AI can also do bad things. People talk about is it going to do bioweapons or cyber weapons? If it is going to do cyber, look, in the world of software, once an idea is out there, within three months somebody's going to copy it. Once the idea is known, it's actually not that hard to figure out how to do it. You have to give yourself a bit of time to say, "Well, can the same thing which is going to do the attack, can it also be used for defense?" In the case of Mythos, sure it's going to find vulnerabilities. By the way, it's not true for Mythos. Any LLM is going to go ahead and find those. Admittedly, Mythos is a bit easier to use. It's not like it's doing something that's off on an island. The same techniques that you can use to do the attacks can be used to do patches and to create the remedy also. You have to go ahead and do that. In that case, the 30-day waiting window looked like a reasonable answer, because if they had gone, as I said, to six months, people would've said, "Forget it." By the way, who's stopping the Chinese for six months? That would not be worth it. I think early days, be really careful. As it gets more mature, regulate the use case, not the tech per se. All right. I have 10 more questions, but I don't think we have time. Thank you very much for participating today. We really appreciate it. Thank you