Microsoft CEO Satya Nadella at WEF Davos 2026 | FULL SPEECH | India Today

Opening and AI as a platform shift

00:00I want to talk about AI and I would really want to because I think this is on everybody's mind more than almost any other subject today um related to intersection [clears throat] of business technology society um so Satia um you know we're we're moving AI from something that was experimental something that we always talked about in the future and now it's it's today um and it's now more foundational and it's not just foundational for companies but it it really is becoming now foundational for countries and throughout society and I think you know you have an advantage over so many other people you know being at the forefront of this technology transition. Um so um with that I wanted to ask a few questions related to that

00:57you have described that AI is a plat plat uh platform shift and what does what does that mean question one where do you see that shift going in the next few years and importantly the third part of my question would be fast forwarding a few years five years what's going to seem obvious in hindsight that feels less clear today. You >> know, first of all, um it's great uh to be back here, uh Larry, and it's um I had a chance, in fact, yesterday when you put out the letter [clears throat] uh to kick off the forum and read it and um and in there you sort of had this one line of uh really I think when it comes to AI, the real question in front of all of us is how do you ensure that the diffusion of AI happens and happens

01:48fast? I mean I think you had that line of how do the models, the data and the infrastructure spread more evenly to create surplus everywhere. If you sort of think about it, the the way I come at this is not that um this has always been the arc of computation, right? You can sort of take it in the last 30 years or the last 70 years. It's always been about can you digitize artifacts on about people, places, and things and then build analytical and predictive [clears throat] power. Right? That's what the mainframes did. That's what the mini computer did. That's what the client server error did. That's what the web error did. The mobile cloud error did. So it like it depend irrespective of which paradigm or

02:37platform. It has been one continuous arc of saying let's make better sense of this world um by reasoning about it in digital form because in some sense once you have these artifacts in digital form you can use a more malleable resource like software >> right >> um which doesn't have the same type of you know I'll call it >> [clears throat] >> um marginal cost economics associated ated with it that allows us to then build more insight and more uh more capability and in that context AI I would say is of the same class at least like the web or the internet um or mobile or PC or the cloud or and maybe even greater and so to me right now where we are is you know let's take just what's happened with software

From code completions to autonomous agents

03:34engineering right which is one you know is knowledge work um you know you could say it's elite knowledge work >> it started off uh you know in fact my own belief in this generation of AI and its capability uh really got built up when I first saw GitHub copilot do code completions right so for the longest time we had the dream that if you're a software developer can you predict the next sort of word or the next uh line of code uh and suddenly it started working with these models u then You said okay if I can do that then can I actually go and bring back you know the flow for a software developer by going to a chat session and asking any question and it comes back with answers that then uh you can use in your coding

04:21flow right so that was the next thing then you said well if that's working can I assign it small tasks that was the agent mode uh now you have complete autonomous agents where you can give it your entire project right it can work uh you know >> 24/7 >> it can work for 24/7. I mean it's still we've got some ways to go for these things to remain coherent long time but nevertheless it's getting better and better and interestingly enough you look at it uh the software developer still is got a lot of agency in it right so that's why I kind of still think that you know going and thinking of these as somehow living outside of the realm of human agency is probably not the right way to think about it in fact the way to

05:03perhaps conceive it like it's let's say in early8s If somebody had come to us and said whatever four billion people are going to wake up every morning and start typing you would have said why right you know we have a like we have a typist pool that's good enough we don't need 4 billion people but we that's what happened right we invented this entire class of thing called knowledge work where people started really using computers uh to go amplify what we were trying to achieve uh using software I think in the context of AI that same thing is going to happen. >> Um it's not like you know what is hardcore coding is going to remain hardcore coding forever. It's just that the levels of abstraction are going to

05:47change. Uh but we also are going to have code as output just like documents. In fact, one of Bill's things at Microsoft from the day I joined in '92 always was what's the real difference between a document, a website and an application, right? It's the lack of sort of software that can transform itself. Interestingly enough, AI finally gives us that, right? Which is I can write a document. I can just say no, I don't want it as a document. I want it as a website. It'll just transform that document using code into a website. I say I don't like the website. I want an app. It'll write more code to transform it. So that reasoning and cap reasoning capabilities that prediction capabilities that ability to

06:30take action remain long-term coherent is all improving. Um and our job though is to parlay this like take even what you at Black Rockck are doing right when you are bring taking something like say co-pilot plus Aladdin and bringing those things together [clears throat] >> to improve the productivity in the firm for the decisions you want to make right with your data >> I could just tell you from in our firm things that would take 12 hours to compute now takes minutes for us processing 14 trillion dollars of other people's money with hundreds of thousands of different um mandates um we could do that instantaneous and we you know that to me if it wasn't for the technology and AI today we would not be

Diffusion, surplus and social permission

07:15able to function to the scale that we're operating >> that's right and so to me that one firm at a time one country at a time if we can really take these tokens and bend the curve of productivity then there is surplus everywhere and that's really the goal >> well surplus could be scary too does does surplus mean fewer workers? What do we mean by surplus? And so, you know, the I want to tie that into my second question about AI diffusion. >> Yeah. >> To me, the the whole realization of AI for any society and also for a more balanced world is making sure that it's diffused and accessible and available across the world. So what you know can you describe how this process of diffusion uh across economies across

08:04companies across people and countries how does that play out? >> Yeah I think that this is the real question right because one of the uh things right now the zeitgeist is a little bit about the admiration for AI in its abstract form or as as technology. uh [clears throat] but I think we as even a global community um have to get to a point where we're using this to do something useful uh that changes uh the outcomes of people and communities and countries and industries right otherwise I don't think uh this makes much sense right in fact I would say we will quickly lose even the social permission uh to actually take something like uh energy which is a scarce resource and use it to generate these tokens if these

08:54tokens are not improving health outcomes, education outcomes, public sector efficiency, private sector competitiveness across all sectors, small and large, right? And that to me is ultimately the goal. So therefore I think really diffusion is everything. And so the way it happens is let's sort of unpack this uh on the supply side what needs to happen in each country is the tokens per dollar per watt have to sort of monotonically get more efficient and better right so to some degree even what we're trying to do with the investments the two firms are doing uh around the world is to just say that like let's make sure that the supply is there which is uh everything from the chips on down ultimately ely to these

09:42token factories that get deployed everywhere. By the way, there's not one token factory. This token factory is the first thing that's going to be diffused all around the world. It's just like electricity, right? You just need a ubiquitous grid of uh energy and tokens that then will power the rest of the economy, right? So, that's I think one side of it. Then the demand side of this is a little bit like every firm has to start by using it. If I look back even, you know, when the PCs first came out or the personal computing era started, I I loved, you know, I think Jobs had a nice metaphor. He called it the bicycle for the mind. Uh Bill had a metaphor which I remember was like information at your

10:27fingertips, right? These two metaphors were great like which allowed us to say that's what it is. It's a tool that I will use to get information at my fingertips. I'll use it as a cognitive amplifier. Now I think that's what we have. But you know 10x 100x right? So in some sense you as a as every knowledge worker you now have access to infinite minds. That's the way I think about it. Right. So there's a Turing award winner uh called Raj Ready who had this nice metaphor of AI and he had this long before uh even generative AI. He said either either it's a cognitive amplifier or it's a guardian angel right so if you think of AI as that um then that in the global workforce right when a doctor can

11:13get to a patient spend more time with the patient because the AI is doing the transcription and entering the records in the EMR system entering the right billing code so that the health care uh you know industry is better served across the payer the provider and uh and the patient ultimately right that's an an outcome that I think all of us can benefit from so I feel ultimately it's going to require real leadership on the private sector and the public sector to ensure that diffusion happens and the one thing other thing I'll mention Larry is skilling right so in some sense the thing that diffusion is very strongly correlated to one thing alone which is How broadly are people skilled in using this? Um, and interestingly enough, I

12:06think if mobile has taught us one thing is it it's actually distinct from what happened in the PC, right? Uh, I remember even growing up in the global south, uh, there used to be a real relationship between learning Excel skills or word skills and getting a job. um you know right now um what's the model in in mobile it's kind of created the same opportunity but it's been a lot more consumption-led it's these creator economy and what have you but it has not been about sort of oh wow here is how you get a healthare job or here is how you get a finance job or here is how you get you know you get ahead professionally um and that needs to come back right people need to say I pick up this AI skill and now I'm a better provider of

12:54some product or service in the real economy. >> So it's it's very [clears throat] easy to see how mobile and the diffusion of mobile how it transformed economies especially in the global south. How does how does this you know to me I I I just read a research report that said the applications for AI so far are heavily weighted towards those who are educated or educated economies. And so does that create that, you know, more of a bifurcation, a more polarization? How do we ensure that that that diffusion is spread evenly? How do we make sure that we're [clears throat] not leaving major portions of society or the world behind because I think that's that's [clears throat] going to be the big issue for us going forward.

Global south, capital and the grid

13:39>> Yeah. So it's it's it's interesting, right? This is one of those times when uh by definition and because of the rails that have been established you know as you said right which is uh what's happened with mobile as well as what's happened with um >> uh you know essentially connectivity >> right >> you have the ability [clears throat] to sort of deliver the tokens pretty evenly around the world >> right >> a lot more so than let's say uh the PC era or even the beginning of the mobile era, right? Because it took a long time for even the mo smartphone in particular to penetrate um all of the world. Whereas now it's not the case, right? These models and their outputs are pretty much available everywhere. And so

14:26the question to me is what's the use cases that make sense, right? One, in fact, one of the demos I always go back to, I think this was even in the beginning of 23 was a rural Indian farmer was able to use a bot built on I think a very early GPT3 or 25 even uh essentially to reason over some farm subsidies that he had heard about in a local language and had it even in that very early days uh have it even show some agentic behavior right like go complete a form for me. So in some sense it took uh you know it brought back agency to someone uh who perhaps didn't have that because the technology was so much more accessible. So I I do think it's in our hands even in the global south to use it uh to create I would say

15:19more of that opportunity where there isn't one. Um but I think what the [clears throat] necessary conditions still are do you have uh the capital investment being put in uh do you even have an environment for capital because in an interesting way we are for example as hyperscalers investing all over right including the global south uh so as long as there's an environment which attracts the capital investment >> and you see the demand >> and you see and then yeah the demand is there >> um and so the question is how do you have a set of policies that allow for both the capital to come in for it to find nexus where there are certain things by the way private capital can do certain things that public capital only

16:00can do for example the grid right it's not I mean grid in most countries is sort of fundamentally driven by governments >> public >> and public and so if [clears throat] you don't have a sophisticated sort or rather if you don't have a a real approach to modernizing the grid uh that will hold things back I mean there's a lot of talk about behind the meter and so on and Yes, there's some amount of that we can do ourselves. >> We can do that in the US. Many countries can't. >> Exactly. And it's not long-term scalable, right? I mean, like to me, a long-term scalable solution is to have uh you know, all of these token factories, part of the real economy, connected to the grid, connected to the telco network, delivering uh just like

16:42we delivered bits, you have to deliver tokens plus bits. Um and that's kind of what's going to drive an at scale whether it's in the global south or in uh on the developed world. >> So so many people talk about there may be an AI bubble. I mean the most important thing that we see as an investor is the the democratization of technology is and the diffusion of that technology really does then transform the demand and the the companies or the countries that diffuse it fastest are going to be the ultimate winners. not the technology creator. >> That's that's you know it's it's it's for this not to be a bubble >> y >> by definition it requires that the benefits of this are much more evenly spread. I mean I think a tellt tell sign

17:33of if it's a bubble would be if all we're talking about are the tech firms. Uh right? If uh all we talk about is what's happening to the technology side that then that's by you know it's just purely supply side >> right >> uh ultimately if we are not talking about wow here is a drug comp you know drug that was sort of uh brought into the market that's super successful because it was uh AI accelerated the clinical trial it's not even the magical molecule right it's kind of even the rest of what is uh needed in order to make something much more relevant right um and So the more [clears throat] we have uh and by the way it's happening right. So I'm not sort of saying that that's why I'm much more confident that

Is there an AI bubble?

18:19this is a technology that will in fact build on the rails of cloud and mobile diffuse faster >> and bend the productivity curve and bring local surplus and economic growth all around the world. Not just economic growth driven by capital expenses. uh right because that's it's a narrow point in time uh calculation. right now that's what we're seeing more >> that's what we're seeing you know you know in in the developed world in particular uh but remember my capital like that the one thing that you know is definitely we're spending a lot of it in the United States but 50% of it is also all over the world >> right >> um and so interesting enough it depends on uh demand all over the world and the demand all over the world will only be

19:07there if there is local surplus all over the world and so that's sort of the way I see the equation >> so let's drill down a little more. As AI diffuses, obviously organizations, companies, governments are going to have to evolve. I'm now getting to the demand side. So, how do you think the structure of organizations is changing in an AI world across roles, across teams, management? I'm I'm sure um Microsoft has evolved itself. So it probably be good to tell the audience how do you see this diffusion occur in the utilization at the corporate level or at a government level that which ultimately then creates that demand which eliminates any fears and bubbles. >> Yeah. No, I think it's probably one of

How organizations must change

19:53the the the big challenges with all of these new technologies is when uh work work artifact and work flow changes uh the means we as firms have to change how we work. In [clears throat] fact, I remember meeting um uh the CEO of General Ali, you know, a few years back and he was describing he had joined um uh the firm, you know, prec [clears throat] era and uh uh and he was describing how for example they worked with their agents in the field uh with faxes inter office memos and um and and suddenly the PC showed up and people would then put a spreadsheet in an email and send it around and the entire workflow and the work process has changed right so similarly I think with AI uh you are going to start seeing uh

20:46actual change in how workflow happens right I mean even [clears throat] in fact for me coming to Davos you know whatever 50 bilateral meetings I have preparing for those uh had a particular workflow right which is uh there is to be my field team would prepare notes and that would come to my HQ and that would get further refined and nothing had really changed right since I joined in '92 to essentially even a few years back. Whereas now I just go to co-pilot and say hey I'm meeting Larry. Please give me a brief and it comes back and gives me by the way the one nice thing is it gives me a 360 right? It knows what we're doing with you as a client what we're doing as a client of yours and everything in [clears throat]

21:28between as an investment. It's so it captures even information unlike anything else. In fact, what I do is I take that and immediately share that back with all my colleagues across all the functions. Right? Think about it. It's a complete inversion of how information is flowing in the organization. It's not like this classic we have an organization, we have departments, we have these specializations and the information trickles up. No, no, no. It actually it flattens the entire information flow. So once you start having that you have to redesign structurally uh so the current structure may not make sense um because you want people to be able to work in a way that allows them to have this information flow freely. So what all

22:19this leads me to if I had to sort of say what's the formula the formula I think it starts with the mindset. So the mindset we as leaders should have is we need to think about changing the work the workflow with the technology then that needs skill set. So you can't sort of talk about this in the abstract. You got to use it like so if I'm not using the >> you have to trust it. >> You have to trust it. You have to use it. You have to learn even how to put the guard rails to trust it. Right? So you can't again you can't just be afraid of it. uh it's going to it's going to be diffused. So the question is as a firm you have to use it to learn how to even uh put the guardrails that allow you to be able to trust it. So skills uh so

23:06mindsets skills the other big consideration u really is how do you make sure you have the data set that you're feeding like context like it's kind of like you have a new intelligence layer but the intelligence layer is only as good as the context you give it. So people describe it even as context engineering but that is what firms do right if you think about what do firms do it's all about the tacit knowledge we have by working as people in various departments and moving paper and information so the question is how do you really have this AI also have that context so these are sort of some of the new things that have to percolate throughout an organization uh to take advantage in fact that's why I think you

23:54you're going to see that challenge of why am I not seeing immediate results in productivity? Because you have to do the hard work. In fact, that's why it's not going to be some, you know, it's going to there going to be firmwide differences. They're going to there could be sectorwide differences, but it's going to fundamentally be because of the leadership will in an organization. >> Do you see the applications being used across large companies and medium and small companies or is it still the domain of mostly the large companies at this moment? I I think that what you're seeing is it's easier the because if you have a green sort of um uh you know if you start fresh it's easier to adopt these tools and you construct your

24:39organization knowing that these tools exist. So >> is it a barbell then? >> It is a barbell. So small companies that are just starting use that platform >> 100% and I think in fact I would say even for large organization there's [clears throat] a fundamental challenge right because unless and until your rate of change keeps up with >> right >> uh with with what is possible uh you're going to get schooled by someone small being able to achieve scale because of these tools. So but I think scale large organizations have an inherent strength. You have the relationships, you have the data, you have um you have knowhow. But the bottom line is if you don't translate that with a new production function, uh then you really will be

25:22stuck. And so therefore the change management challenge for large organizations is going to be bigger. The structural challenge for small organizations of how to overcome scale issues is going to be harder. So it's sort of the two sides. in an interesting way. It's going to be a very competitively intense world uh where neither side like whether you're a new entrant or an incumbent can't take it as like I I can just coast. >> What about country to country? Are you seeing big differences in how the applications are being used? Is it is AI still the domain of developed countries or is it becoming rapidly a domain of all countries? I I I'm seeing there [clears throat] are two things I would say Larry as I travel around the world

Country differences and energy costs

26:04the quality of um whether it's the knowhow the software developers the startups um or even large or large organizations it's not that different it's fascinating you can show up in Jakarta you can show up in Istanbul you can show up in Mexico City it's not that different than showing up even in say Seattle or San Francisco right it's uh for the first time just because access to what's happening uh is there that said at [clears throat] scale the commitment to using this the risk capital being there the large companies pushing it hard I mean I you know again the US you know is in fact if I compare it uh take financial sector financial sector's adoption of the cloud versus AI night and day, right? Because

27:02in an interesting way, it's much faster uh when it comes to AI versus it was with the cloud and cloud because for a variety of reasons regulatory issues too until the regulators allowed banks to bring their data off out of campus that was a big issue. >> Yeah. So I would say I think [clears throat] wherever you know so in the west in particular in the US uh there is clearly a real I would say more of an energy around it in terms of going and using it uh but it's sort of a lot more uniformly spreading around the world than any technology at least I've seen but are you [clears throat] are you mentioned about the power the grid is that going to be one of the determinants of of the accessibility ility if you do

27:50not have cheap power it the demand is costly >> 100%. So if you sort of look at the tokens per dollar per watt right which I think in [clears throat] some sense I would claim that GDP growth in any place will be directly correlated like you if you sort of buy my entire argument that look we've got a new commodity its tokens >> right >> and the job of every economy uh and every firm in the economy is to translate these tokens into economic growth then if you have a cheaper commodity it's better uh And so that's sort of what why there's tokens per dollar per watt. And by the way, there are many many elements to this, right? Which is uh it's not just um the production side. That's why I think even

28:35having the grid is important. Um uh construction costs, right? So if you like if you think about the total TCO uh everything it's like the how are you a cheap producer of energy? Can you build the data centers? Uh then what's the cost curve uh of the silicon and the systems? uh the and by the way look at the token pricing right token pricing basically drops by you know a half uh every three months uh I mean this is a so how so that that's why I think you can sort of really plot how you use the tokens to create surplus knowing that you have a commodity that's whose prices are just going to monotonically come down in a pretty fast curve >> we're sitting in Europe uh and there is a real fear because of the co Europe does not have

Europe's competitiveness and sovereignty

29:27its own power and has to import mostly of its power. Um do you have any messages for Europe related to this? >> Yeah, I mean I think so there are two sets of things right. is you know here we are in in Switzerland and I look at uh the the pharma or the financial sector you know obviously Switz they're uh they they do do a big job in in in this country as in in Europe but they're also international brands and international operations. So one thing that uh whenever I think about Europe is the Europeans are producing products and services that actually are going everywhere in the world and so therefore uh European competitiveness is about the competitiveness of their output globally not just in Europe. I think sometimes

30:12when you come to Europe there's a lot of conversation about just Europe. Uh but European economy is thrives and has you know thrived in the last whatever 200 years 300 years. the miracle of the west is fundamentally because of what has happened in Europe uh is because they were able to produce things that the world needed and so I would say that's number one and in order to do that you again I go back to the human capital here is just fantastic and world class uh you have to absolutely invest in uh producing uh you know having the energy and the tokens here which again you're attracting like as I said we're investing and others are investing uh the the data centers here so the question is what's that next generation

30:54of output that comes from here. Right? I always think about the German middle. Whenever I go to a jeweler or a dentist in the United States, I'm surrounded by German middle. Right. Totally. >> Um it's just unbelievable engineering provice of that country. uh and now the question and by the way that's the point that they are producing industrial products which today are built in into it all the intelligence as well that data right so I know whenever we come to Europe everyone's like talking about sovereignty and data this data that guess what Europe actually should be much more concerned about access to their industrial companies their financial services companies of data from you, US and the rest of the world.

31:44Uh as opposed to just thinking that somehow by protecting Europe, you're going to be competitive. You are only going to be competitive if the products coming out of Europe are globally competitive, right? >> Um and so that's I think what needs to change. Uh you know, Europe has led in privacy, that's fantastic, has led in many aspects of even safety around AI and what have you. And that's a feature uh that's great but you also have to complement it with by building locally and then also thinking globally what's the contribution this continent will make uh to the rest of the world which it has historically been a leader >> a leader so do you think the whole idea around sovereignty of data is that being

32:27misunderstood >> I I think that that when people talk about sovereignty first of all it's very important clearly And who owns and in a week like this it's more important. >> Uh but that said uh it is you have to kind of think about how what is sovereignty mean like for example in the AI era the topic that's least talked about but I feel will be most talked about in in this uh this calendar year will be the sovereignty of a firm. Just imagine if your firm you're not able to embed the tacit knowledge of the firm in a set of weights in a model that you control. By definition, you have no sovereignty. That means you're leaking enterprise value through some model company somewhere. In fact, that it's

33:22sort of fascinating that nobody's talking about that, right? It's like everybody's talking about everything else that is sort of you know outside of that whereas the most important thing is it really doesn't matter if you in fact the data center where it runs is the least important thing quite honestly but like even there first of all the data centers all are all over just because speed of light is a real constraint uh and so therefore the data centers will be spread yes uh you will have digital you'll be able to encrypt everything you'll be able to have the keys with you all of these are much more techn technically solve problems but the one problem that will only be solved is by you having much more sovereignty over

34:02the you know tacid knowledge uh and control over the models and it's not a one-way enterprise value transfer um and so to me I think sovereignty requires real thought on what is it um you know control of destiny means that your your ability to produce something that is unique is preserved uh David Ricardo was not wrong. There's comparative advantage in countries. Uh there is comparative advantage in firms that needs to be preserved even in the AI era. That's what'll give you real sovereignty. >> One last question. I know we're running out of time. Um in 5 years or 10 years, is there going to be one dominant model that we're all going to be using or are and how is Microsoft preparing for this? Are

A multimodel future

34:48you going to be are we going to be using one model for for enterprise, one model for other other traits? >> You know, even in the last whatever 3 years, four years that we've been at it, um the the reality at this point is it's a multimodel world, right? I mean the in fact if you think about it the in both there are going to be multiple models and the trick is really how do you take advantage of these multiple models and in fact build your own model by distilling these right uh so think of these models uh that you orchestrate to build your own model and more importantly you do what is described as orch orchestration or harness engineering. So the IP of any application or any firm is how do you

35:43use all these models with context engineering or your data? Yes. >> Right. So it's that three part. So can I bring in all the models by the way uh which is close source, open source, build my own model, orchestrate them and feed it my data to change the trajectory of some outcome that I care about. That's it. That's the entire picture. So you can do it in like oh I produce a particular product or service. Uh first I got to do better better job in sales or better job in R&D or better job in finance or what have you. And you take that outcome and then you say can I use all the models orchestrate them and feed it my context and then in as a result of it the reasoning traces are really leading to some capability and models

36:32that I control as my IP. As long as firms can answer that question, they're going to be getting ahead. >> Ladies and gentlemen, let's uh thank Satia, my friend. Thank you [applause] for and hopefully this is the beginning of many great dialogue and conversations here at the World Academic Forum. Thank you everyone. >> Thank you.