Satya Nadella on AI’s Business Revolution: What Happens to SaaS, OpenAI, and Microsoft?

All right, everybody. We're thrilled to have the one, the only Satya Nadella here, the third CEO of Microsoft, for an impromptu fireside chat with David Sacks, czar of AI and crypto. Satya, third CEO of Microsoft, born in India—what an incredible story. You came here right after college, and you had a little round trip to pick up your wife, as you describe in your book, to bring her here. Tell everybody briefly how that occurred.

Well, that's a great story of the labyrinth that is the immigration policy of the United States, I think. My wife and I went to college together in India. I came here for grad school, and we then got married. I got my green card, but she couldn't come join me because we got married.

So the story goes, basically, that I had to give up my green card. The funny thing is, I went to the American embassy in Delhi and said, "Where's the line to give up my green card?" They said, "There is no such line." That would be a crazy thing to do in the '90s.

It was a strange thing to give up your green card and get an H-1B so that she could join me, but it all worked out. It's a long-lost memory, but it was a way to work around it.

I wanted to ask you: having launched Copilot first with GitHub, then having Copilot on the desktop, you made a very bold move for Microsoft to put that in the Windows product, which I use every day, on the desktop. But you did that before it really could recognize the file system and interact with applications. It got a little bit of a lukewarm reception, but now you've been doubling down, doubling down.

There seem to be, in my estimation, 3 modalities for knowledge workers. Elon is building at xAI what they're calling a human emulator, if you saw that leak this week, where they're just building employees and putting them into their chat rooms and email. Then you have Claude, which came out with Cowork this week. It's incredibly powerful. People are kind of losing their minds over it. I've been playing with it for the last 40 hours, and it's truly impressive.

What's your vision for Microsoft and how knowledge workers will actually put this to use? There seems to be a gap between playing around with ChatGPT and getting some interesting results and getting business results.

I think one of the most illustrative examples of trying to understand these various form factors is looking at coding, which is obviously a form of knowledge work—or probably the best example of knowledge work. If you think about the journey coding has been on, it started with essentially the next-edit suggestion. That was the first time my own belief in this entire generation of tech really got formulated, when I started seeing—there was a Codex model back in the day, pre-GPT-3.5. That's when next edits started working with some real accuracy.

Then we went to chat, then we went to actions, and now to full autonomous agents. The autonomous agents can be foreground or background, in the cloud or local. Those are all the form factors that exist today when you're coding.

Interestingly, if you look at it, you use all of them. It's not like there's only one form factor. That's probably one of the other lessons. For example, when I'm in a CLI, I can use a foreground agent and a background agent, and then literally go edit in VS Code. They're all happening in parallel. That shows how these form factors even compose.

Then you bring that to knowledge work. We started with chat. Chat with reasoning goes beyond just request-response because you now have that chain of thought, where you can see it work. Now there are actions, either through computer use or through APIs—basically skills and agent calls—so you can do actions. That's the state of Copilot today.

There's a way to think about the theory-of-mind evolution. If you remember, Jobs had the best line for PCs or computers, I would say: "It's a bicycle for the mind." Bill had a line that I liked as well: "It's information at your fingertips." We now need a new concept or metaphor for how we use computers in the AI age.

And you have one?

The one I like actually came from the CEO of Notion, which is an incredible product: a manager of infinite minds. That's a nice way to think about it when you look at all the agents that you're working with.

You need to understand what—I think the other term I like is that we macro-delegate and micro-steer. In fact, you need that in coding. You do a macro-delegation, and then I can, in parallel, give it instructions while it's doing work. That's the state even today of Copilot, or what have you.

You bring up one of the form factors I'm very excited about, and you'll see us do things even in the next week. While I'm sitting in GitHub Copilot, it's not as if software developers sit in isolation. It's not like the only thing I work on is my repo. I attend meetings. I write specs, or others have written specs that I'm implementing. I need to have my repo be consistent with that.

That means using either a straightforward MCP server or a skill. I want to be able to call into my Work IQ, which is Copilot, and bring that in. That's the type of composition of knowledge work that'll happen.

The same thing applies to security. Say you're a security professional and you have lots of logs. How do you analyze them? You drop them into a file system, then write code on top of it and create a dashboard, or what have you. Those are the types of knowledge work that we can enable.

I think you bring up one more thing: Can you create, quote-unquote, digital employees or digital coworkers, or what have you? It's all about credentials. Today, you could literally assign—

Are you working on that as well?

Yeah. We introduced something called Agent 365 as a way to give identities—in fact, extending the identities we have for humans today, and the endpoint protection we have for their compute devices—to agents.

So you might clone me working in the HR department or working in the marketing department and have a virtual version of me inside of Office?

That's correct. There are 2 modalities there. One is that you give every knowledge worker infinite minds. That's one. Then you create even infinite minds independent of your identity, because identity is one of the key things you have to get right for it to work.

Permissions and decision-making.

Permissions, decision-making, and one of the key things: Who did what to whom? That's the most important query in an organization. At the end of the day, the organization needs to understand what work got done and what's the provenance of that work, and how do you trace it back?

Therefore, you either want a human with a lot of agents—that's really macro-delegation and micro-steering by the human whose identity was passed on—or a separate identity. It's delegation versus a separate identity.

That was done by a level of management—product management—that you've eliminated. Alphabet has eliminated it. Meta has started to eliminate it. Four years ago, you had the same number of employees you have at Microsoft now, but you put $90 billion onto the top line of revenue in that time, and you doubled your income during that time. How did that happen? Is that automation of those jobs? Were you a little bit overstaffed? Unpack.

I think you're pulling on a very interesting thread, which is: At some level, what's the big structural change that needs to happen? In fact, I would say this is probably the biggest change in knowledge work since PCs.

I always think about how work happened previously. Think about a multinational company like ours trying to do a forecast. Faxes went around, interoffice memos got sent, and then you created a forecast. Suddenly, PCs became standard issue. You put some numbers into an Excel spreadsheet, sent it in email, everybody entered their numbers, and you had a forecast.

The work artifact and the workflow all changed. That's what's happening now. For example, at LinkedIn, we used to have product managers, designers, front-end engineers, and CIS backend engineers. What we did was take those first 4 roles and combine them. In fact, we increased the scope and said, "Let's have them all be full-stack builders."

I like that because it's a structural change that allows us to increase the change in both the work and the workflow between these functions.

I would assume the velocity increases because you don't have 4 people communicating. The throughput of ideas is just one person and vibe coding.

Exactly. And there's a new workflow. At the same time, as you can imagine, to build an AI product today, there's a completely new workflow.

It starts with evals. Basically, there's evals, science, and infrastructure. Evals are done by these full-stack builders, and product managers in the new form, or what have you. The infrastructure is built by the systems engineers at the backend because they support the science that supports the product.

In some sense, there's a new loop, and you have to structurally change. A lot of what is happening inside of tech is that change, which I think is going to be pretty massive. At the same time, a company like ours, I have to do everything.

It's not like I can just go live in the future. I have to make sure we're doing a fantastic job of ensuring that hot patching on Windows is done with quality, while at the same time building the evals that are improving Copilot quality. Both of those have to be first-class.

I assume this is the most challenging moment of your career, because Microsoft was so dominant—a duopoly in some spaces. But you really weren't up against the level of competition you're up against now. I was talking to Elon, and he was sort of saying, “Building cars was pretty easy,” because he was up against the legacy carmakers. Now, look at the set of competitors you're up against.

Yeah, it's a pretty intense time. The way I always think is that it's helpful when you have a completely new set of competitors every decade, because that keeps you fit. I joined Microsoft in 1992, when we had Novell as the big existential competitor. Here we are in 2026, and you're absolutely right: it's a pretty intense time.

I'm glad there's competition. Honestly, at the end of the day, when I look at it, as a percentage of GDP, where will tech be 5 years from now? It will be higher. We're blessed to be in this industry. There's a lot of intense competition, but it's not as zero-sum as some people make it out.

It's getting much bigger.

Much bigger. The TAM and the impact of this technology are going to be so massive. The question, then, of course, is: What is the brand identity? Microsoft has brand permission. What do customers expect from us?

Sometimes we overthink this idea that every customer wants the same thing from all of the competitors. It's a different take on the Peter Thiel thing: You have to avoid competition by really understanding what customers want from you, versus thinking everybody's a competitor.

David.

Yeah. There are a lot of heads of state here, obviously, at Davos, as well as CEOs of Fortune 500 companies. I think you got asked a question last night at the dinner about how they should think about AI and how to be successful. I recall that they used the word “diffusion,” and I was wondering if you could expand on those remarks, because that really resonated with some of the policy work I've been doing.

No, absolutely. In fact, what you all have been doing to make sure that, in this context, the American tech stack is broadly used and trusted around the world is important. When I look back, David, to me, at the end of the day, you create the technology, but the benefits come only through intense use.

One of my favorite studies has always been this work that an economist out of Dartmouth did—his name is Diego Comin—where he studied what happened during the Industrial Revolution. How did countries get ahead? The simple takeaway from that was that any country that brought the latest technology into the country and then did value-added technology on top of it got ahead. Don't reinvent the wheel. Bring the latest technology and build on top of it. That's what happens when you have diffusion.

Especially with general-purpose technologies like AI, they need to spread. In our own country, the United States, we now have the technology. The question is: Is it being used in health care? Is it being used in financial services? Is it being used in every sector of the economy by large businesses, small businesses, and the public sector? Unless and until we see that diffusion and intense use, we're not going to have success.

That's the phase we're in. It's diffusing faster, and some of the policy work you've done—and, in general, all the good news here—is that the technology is there. The rails around cloud and mobile that were laid out make it possible for this thing to spread. It's not impossible to get the tokens. The question is: What are the use cases, and how do you manage the change in all of that?

One of the questions, at least in Davos, is one for the West and the developed nations: What about the Global South? I think the Global South has a huge opportunity too, quite frankly, because, let's say, 40% or 50% of the GDP of most Global South countries is the public sector. Just imagine this technology making a difference in how governments parlay taxpayer money into services for citizens. If there are efficiency gains, that's probably a couple of points of GDP growth right there.

I'm very optimistic that there's going to be a pull, and that we should, as the United States, given the technology stack we have—in Europe, in Asia, in South America, in Africa, and everywhere—get it broadly deployed.

One of the questions I get asked a lot about the AI race is how you know if you're winning, or how you know if the United States is ahead of its global competitors. The answer I give is market share. If we look around the world in 5 years and see that American companies and American technology have, say, 80% market share, it means we did a good job.

If we look around the world in 5 years and see that it's Chinese chips and Chinese models being used all over the world, it means we probably lost. Ultimately, the proof of the pudding is in the eating of it. In this case, the way you know that you're succeeding is through market share, through usage.

I would agree with that. But, David, since you even worked at Microsoft for a few years, you know one of the things I'm very grounded on is that Bill Gates line about a platform. One of the things I always think about is that it's market share, but it's also ecosystem effects.

What the United States has always done is not just focus on our market share or even the revenues to U.S. companies. One of the things I learned at Microsoft is that whenever I did a country visit, the data I would first study in, let's say, the U.K. or Switzerland was the total employment created in Switzerland through our channel. That used to be the number-one thing in our country reports: the total number of IT workers, the number of office workers, the channel partners, and the number of ISVs who were there.

We used to have a complete marker of how the ecosystem around the platform got built, one country at a time. That is what the United States has always done. In fact, the U.S. tech stack, including in China, got built because others built around our tech stack. The same thing is going to happen.

That's why I think the work you're doing around diffusion is about really increasing the size of the pie and the trust in the platform, so that there is true economic opportunity, quite frankly.

Well, you're right. I remember, actually—you brought back some memories—about a decade ago, when my company, Yammer, was acquired by Microsoft. We were part of the SharePoint group, and I remember that the product managers there were very proud of the fact that the revenue from the SharePoint ecosystem—meaning non-Microsoft revenue, the consulting community and the implementers who would go into companies and implement SharePoint—was something like 7 times greater than Microsoft's own software revenue in aggregate.

I think Bill had a line about how you're not an ecosystem or a platform until the revenue on top of your platform is some factor of your own revenue. What's really important about this is that when we talk about diffusion, and obviously want the United States to have this leading position, it doesn't mean it's bad for the rest of the world, because they're able to build on top of those platforms and create even more value.

100%. In fact, that's the most important point. This is not about American tech or American revenues to the United States. It's actually about creating opportunity using a new platform everywhere.

I remember working on our database products in the 1990s with SAP. The combination of SQL Server and SAP R/3 was successful on both sides. There's a lot talked about Intel and Microsoft, but one of the other things I grew up in, which has been foundational in how I look at the world, is what we did with a European software company that is still a giant.

Who knows what the next big AI app will be, where it will come from, and what will happen? I go in with the attitude that there will be tech companies—maybe even top-5 tech companies—that could emerge everywhere using the American tech stack.

You've done some amazing acquisitions, and you're quite a dealmaker on top of being a technologist. It's probably the least reported aspect of your spectacular tenure and the massive growth you've had. But you did a deal with OpenAI and probably one of the most savvy and controversial dealmakers of all time, Sam Altman.

That deal was looked at as though you were set up to get a windfall in cash, which you don't need as Microsoft. It's always nice, I'm guessing, if they IPO. But did you potentially create—and this was the criticism of it—the ultimate competitor to Microsoft? How do you think about that?

And how can Microsoft—which missed the mobile revolution, Steve Ballmer's biggest regret—how can you not have a Gemini, an xAI, or a Claude that is your own? Or, in your mind, do you have that because you have the source code of OpenAI?

Yeah, I think that's right. When people say, “Where is your foundation model?” at the end of the day, we do have the IP. That said, I think you bring up a couple of different things.

One is that, to us, the most important thing when I look at Microsoft's strategy today is that we want to build token factories. Our biggest business today is the Azure business, and the Azure business's TAM, given what's going to happen, is so huge that we now need to be fantastic at building these token factories.

That means a heterogeneous fleet of infrastructure, which is what every hyperscaler has always done: use software to make maximum use of it, for TCO and utilization.

So that's one side of it. Then there's the app-server business. If everyone's going to be building agents, having infinite minds, RL gyms, evals, and what have you, there's an entire app-server business. Just like every platform has had an app server, this one has an app server; that's what we're doing with Foundry and what have you.

In that app server, one of the things that's now structurally pretty clear is that anyone building any application, or any company, is going to use not 1 model but all the models. Why would I not? In fact, I will orchestrate even multiple models for any given task. There's this nice thing that we came out with in our healthcare practice called the Decision Orchestrator. What it proves is that assigning roles—investigator, data analyst, domain expert—and even just giving prompted roles to models and then orchestrating them gets better results than any single frontier model.

Am I right to read into that, then, that you're bullish on the open-source models and think large language models will largely be commoditized, and that's not where the value will occur?

In fact, the way I think about it is just like what happened—

And Apple thinks that too, by the way.

By the way, the way I think about what happened in the database market is that I used to think everything was just a SQL database until it was not. There were databases—think about it: there were document databases, there were NoSQL databases. The proliferation of databases—who would have thought that the database market would have such richness to it?

Or that it could ever be open source? That was—

That's true. I mean, talk about Postgres, or what has happened even with open-source databases. There are even companies that have backed them. To me, that's what's going to happen. To me, a model is like the database market. It's going to have differences, but I somehow think that there are definitely going to be frontier models that are closed source, and there are going to be open-source models that are going to be frontier class.

If anything, I think in this next year, what will probably be a big part of the discussion is the future of a firm. A firm should be able to take the tacit knowledge it has and embed it inside the weights of a model that it controls. When somebody asks me how many models should there be, I'll say, "As many models as firms in the world." That's sort of an extreme way of putting it, because to me, that's how this knowledge economy becomes an AI economy.

Are you secretly—and you can say it here since we're on All-In—working on an LLM that can exist on the Windows desktop? Because you have it today: there's a Phi Silica model that's completely resident, using NPUs and, of course, GPUs. In fact, one of the fascinating things is that the workstation is back.

Which is great for Microsoft because you have a nice desktop business.

Absolutely. We think that form factor is especially important. I always say this: I started my career on a command line. Who knows? I may just end it on a command line.

Well, you started at Sun, which was the original $5,000-$10,000 workstation. Do you see a time when you'll be meeting with your customers here and advocating for a $10,000 or $20,000 desktop machine that has an LLM and the hardware? You can put a DGX card in it and have a fantastic machine.

By the way, we are 1 architecture tweak away from even having some kind of distributed model architecture, even an architecture that knows how to really distribute itself. That's the type of breakthrough that can completely change what hybrid AI may look like. We're absolutely committed to and focused on making the PC a great place for local models, and local models that then do a lot of the prompt processing and call into the cloud. There's a whole lot of work that can happen, and that's definitely underway.

Yeah, I think that Claude Cowork has shown the power of tapping into the local file drive and being able to use that. That brings up another point. You got me thinking about Yammer, and for people who don't know, Yammer's claim to fame—this was about 15 years ago—was that it pioneered a lot of consumer-growth tactics to attack enterprise software.

I'm wondering, as you think about enterprise adoption of AI, how do you think it's going to spread over the next year? It feels like we're at a critical point. Do you think it's going to be top-down? Is it going to come from the CEO directing a team, giving them a strategic transformation project, and they're going to do an RFP? Or do you think it's going to spread bottom-up in the enterprise through AI-native employees who are adaptable, who are using the tools in their own lives, and who start to bring these things to work and accomplish amazing things?

Yeah, no, I think, like all things, David, it's both top-down and bottom-up. The reason I say top-down is that, if I look at the ROI of applying AI in customer service, supply chain, or HR self-service, those are the easy projects where IT and CXOs can make calls. That's where you'll see the first drop of real AI adoption.

But the bottom-up is what will ultimately happen. Even with PCs, lawyers brought Word in, finance brought Excel in, and then email came, and it became standard issue. That's what's happening right now. For example, with these agents, when I talk about everybody building agents, people are figuring out ways to create these things that are changing workflow and removing drudgery from their work. That's the beginning of what is a bottom-up transformation.

The thing I'm most excited about is this bottom-up change, even at Microsoft. For example, we manage something like 500-odd fiber operators around the world in Azure today. I didn't myself realize that a lot of it—you know, it's called DevOps, but it's physical assets. Things get cut. When you say DevOps, that means you're literally emailing people and saying, "Hey, what happened to that fiber cut? How do we repair it?" So there's a lot of back and forth.

The person who runs our global network has basically built, to your point about these personas, digital employees that are doing all of that DevOps. That's completely bottom-up. You see the tools and think, "Hey, I have a new way to build agents. I'm going to use it to create levels of automation that remove drudgery, improve efficiency, and improve quality." Ultimately, that's a skilling thing, which is the big issue.

Skilling is not mystical; it's just by doing. It's not like I go to a class per se. It's the diffusion of the tools and using the tools, and I think that's what is really going to be happening.

We're in a very interesting moment. Empowering an existing employee with these tools is so much easier than hiring, mentoring, and bringing up the next generation. It feels like we're in a little bit of an indigestion moment at Microsoft. Who's going to have my job in 30 or 40 years if the company stays the same size?

Given your technology-first approach, there's really no reason to ever add another Microsoft employee at the pace this is going, and you haven't for 4 years. You may have swapped some in and out and changed the texture of the company. How do you think about this next generation? What advice would you have for these college graduates who maybe don't have an offer from Microsoft right now? You used to spend a lot of time building that group, but maybe you don't have that luxury now. Do you think about it ever?

No, I mean, it's a great question. There's a little bit of a debate about what happens to early-career employees and how college recruiting works. I still am a big believer in college recruiting because, at the end of the day, this is going to change the curve by which anyone can pick up proficiency in a codebase.

What has changed is that someone who comes in new to a team can ramp up much faster, thanks to all of the models and skills and the fact that they can go ask the agent. Think about it: it's like having an unbelievable mentor who is getting you onboarded onto a codebase faster. In some sense, the productivity curve of a college hire is going to be much steeper than it has ever been before.

There might be a difference. In fact, one of the things we're experimenting with is a different type of apprenticeship. You take somebody who's an IC senior developer and have a cohort of college hires working with them because it's a new way of working.

I remember everybody who joined Microsoft would ask, "How did Cutler implement malloc, or whatever?" They would try to read his code to understand what great craftsmanship looks like. Nowadays, I think that great craftsmanship comes by looking at how the 10x and 100x engineers use AI to build high-quality products. That is what these new college graduates will learn, and they will learn it faster.

That's a beneficial thing for a company like us because, at the end of the day, until we see longevity or something, we need people to come into the workforce and be successful at Microsoft. We are very committed, but we're also making sure that the scope of the jobs makes sense for the aspirations of people who are currently in the workforce and people who are entering the workforce.

Okay, on that note, Satya Nadella, thank you so much.