关于 AI 的一切：@altcap、@sama 与 @satyanadella。万圣节特别篇。🎃🔥BG2 与 Brad Gerstner 对谈

Yeah, I think this has really been an amazing partnership through every phase. We had no idea where it was all going to go when we started, as Satya said. I think this is one of the great tech partnerships ever, and without Microsoft, and particularly without Satya's early conviction, we would not have been able to do this.

What a week. What a week. Great to see you both. Sam, how's the baby?

The baby is great. That's the best thing ever, man. Every cliché is true, and it is the best thing ever.

Hey, Satya, with all your time together, the smile on Sam's face whenever he talks about his baby is just so different. It's Dad, that, and compute, I guess, when he talks about compute and his baby.

Well, Satya, have you given him any dad tips with all this time you guys have spent together?

I said, just enjoy it. We had our babies, or our children, so young, and I wish I could redo it. In some sense, it's just the most precious time, and as they grow, it's just so wonderful.

I'm happy to be doing it older, but I do think sometimes, man, I wish I had the energy when I was 25. That part's harder.

No doubt about it. What's the average age at OpenAI, Sam? Any idea? It's young.

It's not crazy young. Not like most Silicon Valley startups. I don't know, maybe low 30s on average.

Are babies trending positively or negatively?

Babies are trending positively.

Oh, that's good. That's good.

Well, you guys, such a big week. I was thinking about how I started at NVIDIA's GTC, which just hit $5 trillion. Google, Meta, Microsoft—Satya, you had your earnings yesterday—and we heard consistently: not enough compute, not enough compute, not enough compute. We got rate cuts on Wednesday. GDP is tracking near 4%.

The president cut these massive deals in Malaysia, South Korea, Japan, and it sounds like with China—deals that really provide the financial firepower to reindustrialize America. $80 billion for new nuclear fission, all the things that you guys need to build more compute. But what certainly wasn't lost in all of this was that you guys had a big announcement on Tuesday that clarified your partnership. Congrats on that.

1. Microsoft’s Investment in OpenAI

I thought we'd just start there. I really want to break down the deal in simple, plain language to make sure I understand it, and others do as well. We'll start with your investment. Satya, Microsoft started investing in 2019 and has invested in the ballpark of $13.4 billion into OpenAI. For that, you get 27% ownership in the business on a fully diluted basis. I think it was about a third, and you took some dilution over the course of last year with all the investment. Does that sound about right in terms of ownership?

2. The Nonprofit Structure and Its Impact

Yeah, it does. But before even our stake in it, Brad, I think what's pretty unique about OpenAI is the fact that, as part of OpenAI's restructuring process, one of the largest nonprofits gets created. Let's not forget that. At Microsoft, we are very proud of the fact that we are associated with 2 of the largest nonprofits: the Gates Foundation and now the OpenAI Foundation. So that's, I think, the big news.

Obviously, we are thrilled. It's not what we thought. As I said to somebody, it's not like when we first invested our $1 billion that we thought, “Oh, this is going to be the 100-bagger that I'm going to be talking to VCs about.” But here we are.

We are very thrilled to be an investor and an early backer. It's really a testament to what Sam and team have done, quite frankly. They obviously had the vision early about what this technology could do, and they ran with it and just executed in a masterful way.

Yeah, I think this has really been an amazing partnership through every phase. We had no idea where it was all going to go when we started, as Satya said. I think this is one of the great tech partnerships ever, and without Microsoft, and particularly Satya's early conviction, we would not have been able to do this.

I don't think there were a lot of other people who would have been willing to take that kind of a bet, given what the world looked like at the time. We didn't know exactly how the tech was going to go. Well, not exactly—we didn't know at all how the tech was going to go. We just had a lot of conviction in this one idea of pushing on deep learning and trusting that if we could do that, we'd figure out ways to make wonderful products and create a lot of value and also, as Satya said, create what we believe will be the largest nonprofit ever. I think it's going to do amazingly great things.

I really like the structure because it lets the nonprofit grow in value while the PBC is able to get the capital that it needs to keep scaling. I don't think the nonprofit would be able to be this valuable if we didn't come up with the structure and if we didn't have partners around the table who were excited for it to work this way.

I think it's been more than 6 years since we first started this partnership, and it's a pretty crazy amount of achievement for 6 years. I think much, much more is to come. I hope that Satya makes a trillion dollars on the investment, not $100 billion, whatever it is.

Well, as part of the restructuring, you guys talked about it: You have this nonprofit on top and a public benefit corporation below. It's pretty insane. The nonprofit is already capitalized with $130 billion—$130 billion of OpenAI stock. It's one of the largest in the world out of the gates. It could end up being much, much larger.

3. Health, AI Security, and Resilience

The California attorney general said they're not going to object to it. You already have this $130 billion dedicated to making sure that AGI benefits all of humanity. You announced that you're going to direct the first $25 billion to health and AI security and resilience.

Sam, first, let me just say, as somebody who participates in the ecosystem, kudos to you both. It's incredible, this contribution to the future of AI. Talk to us a bit about the importance of the choice around health and resilience. Then help us understand how we make sure that you get maximal benefit without it getting weighed down, as we've seen with so many nonprofits, with their own political biases.

Yeah. First of all, the best way to create a bunch of value for the world is hopefully what we've already been doing, which is to make these amazing tools and just let people use them. I think capitalism is great. I think companies are great. I think people are doing amazing work getting advanced AI into the hands of a lot of people and companies. They're doing incredible things.

There are some areas where market forces don't quite work for what's in the best interest of people, and you do need to do things in a different way. There are also some new things with this technology that just haven't existed before, like the potential to use AI to do science at a rapid clip—really, truly automated discovery.

When we thought about the areas we wanted to first focus on, clearly, if we can cure a lot of disease and make the data and information for that broadly available, that would be a wonderful thing to do for the world. On this point of AI resilience, I do think some things may get a little strange, and they won't all be addressed by companies doing their thing.

4. Models, Exclusivity, and Distribution

As the world has to navigate through this transition, if we can fund some work to help with that—and that could be cyber defense, AI safety research, or economic studies—all of these things can help society get through this transition smoothly. We're very confident about how great it can be on the other side, but I'm sure there will be some choppiness along the way.

Let's keep busting through the deal. Models and exclusivity: Sam, OpenAI can distribute its leading models on Azure, but I don't think you can distribute them on any other big cloud for 7 years, until 2032. That would end earlier if AGI is verified. We can come back to that, but you can distribute your open-source models, Sora, agents, Codex, wearables—everything else—on other platforms. So, Sam, I assume this means no ChatGPT or GPT-6 on Amazon or Google.

No. First of all, we want to do lots of things together to help create value for Microsoft. We want them to do lots of things to create value for us, and there are many, many things that'll happen in that category.

We are keeping what Satya termed “stateless APIs” on Azure exclusively through 2030. Everything else we're going to distribute elsewhere, and that's obviously in Microsoft's interest, too. So we'll put lots of products in lots of places, and then this thing we'll do on Azure, and people can get it there or via us. I think that's great.

5. Revenue Sharing and AGI Milestones

And then the rev share: There's still a rev share that's paid by OpenAI to Microsoft on all your revenues. That also runs until 2032 or until AGI is verified. Let's just assume, for the sake of argument—I know this is pedestrian, but it's important—that the rev share is 15%.

That would mean if you had $20 billion in revenue, you're paying $3 billion to Microsoft, and that counts as revenue to Azure. Satya, does that sound about right?

Yeah, we have a rev share, and I think, as you characterized it, it's either going to AGI or till the end of the term.

I actually don't know exactly where we count it, quite honestly—whether it goes into Azure or somewhere else. That's a good question. It's a good question for Amy.

Given that both exclusivity and the rev share end early if AGI is verified, it seems to make AGI a pretty big deal. As I understand it, if OpenAI claimed AGI, it sounds like it goes to an expert panel, and you guys basically select a jury that's got to make a relatively quick decision on whether or not AGI has been reached.

Satya, you said on yesterday's earnings call that nobody's even close to getting to AGI and you don't expect it to happen anytime soon. You talked about this spiky and jagged intelligence. Sam, I've heard you perhaps sound a little bit more bullish on when we might get to AGI. So, I guess the question is to you both: Do you worry that over the next 2 or 3 years we're going to end up having to call in the jury to effectively make a call on whether or not we've hit AGI?

I realize you've got to try to make some drama between us here.

I think putting a process in place for this is a good thing to do. I expect that the technology will take several surprising twists and turns, and we will continue to be good partners to each other and figure out what makes sense.

That's well said. I think that's one of the reasons why this process we put in place is a good one. At the end of the day, I'm a big believer in the fact that intelligence, capability-wise, is going to continue to improve. Our real goal, quite frankly, is how you put that in the hands of people and organizations so that they can get the maximum benefits. That was the original mission of OpenAI that attracted me to OpenAI and Sam and the team, and that's what we plan to continue on.

Brad, to say the obvious, if we had superintelligence tomorrow, we would still want Microsoft's help getting this product out into people's hands, and we want them—

Of course. Of course.

Yeah, no, again, I'm asking the questions I know are on people's minds, and that makes a ton of sense to me. Obviously, Microsoft is one of the largest distribution platforms in the world. You guys have been great partners for a long time, but I think it dispels some of the myths that are out there.

6. OpenAI’s Growth and Compute Commitments

Let's shift gears a little bit. Obviously, OpenAI is one of the fastest-growing companies in history. Satya, you said on this pod a year ago that every new phase shift creates a new Google, and the Google of this phase shift is already known: It's OpenAI. None of this would have been possible had you guys not made these huge bets.

With all that said, OpenAI's revenues are still a reported $13 billion in 2025. Sam, on your livestream this week, you talked about this massive commitment to compute: $1.4 trillion over the next 4 or 5 years, with big commitments—$500 million to NVIDIA, $300 million to AMD and Oracle, and $250 billion to Azure.

I think the single biggest question I've heard all week, hanging over the market, is how can a company with $13 billion in revenues make $1.4 trillion of spending commitments? You've heard the criticism, Sam.

First of all, we're doing well more revenue than that. Second of all, Brad, if you want to sell your shares, I'll find you a buyer.

I think there are a lot of people who would love to buy OpenAI shares. I don't think people who talk with a lot of breathless concern about our compute stuff or whatever would be thrilled to buy shares. So I think we could sell your shares, or anybody else's, to some of the people who are making the most noise on Twitter about this very quickly.

We do plan for revenue to grow steeply. Revenue is growing steeply. We are taking a forward bet that it's going to continue to grow, and that not only will ChatGPT keep growing, but we will be able to become one of the important AI clouds; that our consumer device business will be a significant and important thing; and that AI that can automate science will create huge value.

There are not many times that I want to be a public company, but one of the rare times it's appealing is when those people are writing these ridiculous “OpenAI is about to go out of business” stories or whatever. I would love to tell them they could just short the stock, and I would love to see them get burned on that.

We carefully plan. We understand where the technology and the capability are going to go, how we can build products around that, and the revenue we can generate. We might screw it up—this is the bet that we're making, and we're taking a risk along with that. A certain risk is that if we don't have the compute, we will not be able to generate the revenue or make the models at this kind of scale.

Exactly. And let me just say one thing, Brad: As both a partner and an investor, there has not been a single business plan that I've seen from OpenAI that they have put in and not beaten. In some sense, this is the one place where, in terms of their growth and even the business, it's been unbelievable execution, quite frankly.

Obviously, everyone talks about all the success in usage and what have you, but even all up, I would say the business execution has been pretty unbelievable. I heard Greg Brockman say on CNBC a couple of weeks ago, “If we could 10x our compute, we might not have 10x more revenue, but we'd certainly have a lot more revenue.”

Simply because of the lack of compute power. It's really wild when I look at how much we are held back. In many ways, we've scaled our compute probably 10x over the past year, but if we had 10x more compute, I don't know if we'd have 10x more revenue. I don't think it'd be that far.

7. Compute Constraints and Scaling

We heard this from you as well last night, Satya—that you were compute-constrained and growth would have been higher even if you had more compute. So help us contextualize this, Sam. How compute-constrained do you feel today? When you look at the buildout over the course of the next 2 to 3 years, do you think you'll ever get to the point where you're not compute-constrained?

We talk about this question of whether there's ever enough compute a lot. I think the only—or the best—way to think about this is like energy or something. You can talk about demand for energy at a certain price point, but you can't talk about demand for energy without talking about different demand at different price levels.

If the price of compute per unit of intelligence, or however you want to think about it, fell by a factor of 100 tomorrow, you would see usage go up by much more than 100. There'd be a lot of things that people would love to do with that compute that just make no economic sense at the current cost, but there would be new kinds of demand.

On the other hand, as the models get even smarter and you can use these models to cure cancer, discover novel physics, or drive a bunch of humanoid robots to construct a space station—or whatever crazy thing you want—then maybe there's huge willingness to pay a much higher cost per unit of intelligence for a much higher level of intelligence that we don't know yet. But I would bet there will be.

When you talk about capacity, it's like cost per unit and capability per unit. Without those curves, it's sort of a made-up, not-super-well-specified problem.

Yeah. I think the one thing that Sam has talked about, which I think is the right way to think about it, is that if intelligence is a log of compute, then you try to really make sure you keep getting efficient. That means the tokens per dollar per watt and the economic value that society gets out of it is what we should maximize, while reducing the costs.

That's where the Jevons paradox point comes in, right? You keep reducing it, commoditizing intelligence in some sense, so that it becomes the real driver of GDP growth all around.

Unfortunately, it's something closer to log of intelligence equals log of compute. But we may figure out better scaling laws, and we may figure out how to beat this.

We heard from both Microsoft and Google yesterday. Both said their cloud businesses would have been growing faster if they had more GPUs. I asked Jensen on this pod if there was any chance that over the course of the next 5 years we would have a compute glut, and he said there was virtually no chance in the next 2 to 3 years.

I assume you guys would both agree with Jensen that, while we can't see out 5, 6, or 7 years, certainly over the course of the next 2 to 3 years, for the reasons we just discussed, there's almost no chance that you have excess compute.

I think the cycles of demand and supply in this particular case can't really be predicted. The point is, what's the secular trend? The secular trend is what Sam said.

At the end of the day, quite frankly, the biggest issue we're now having is not a compute glut, but power—and the ability to get the builds done fast enough close to power. If you can't do that, you may actually have a bunch of chips sitting in inventory that I can't plug in. In fact, that is my problem today. It's not a supply issue with chips; it's actually the fact that I don't have warm shells to plug into.

[Speaker?]

How some supply-chain constraints emerge is tough to predict because demand is tough to predict. It's not like Sam and I would want to be sitting here saying, “Oh my God, we're less short on compute,” because we just weren't that good at being able to project what the demand would really look like.

And, by the way, on the worldwide side, it's one thing to talk about one segment in one country, but it's about really getting it out to everywhere in the world. There will be constraints, and how we work through them is going to be the most important thing. It won't be a linear path, for sure. There will come a glut, for sure, and whether that's in 2 to 3 years or 5 to 6, I can't tell you, but it's going to happen at some point—probably several points along the way.

There's something deep about human psychology here, and bubbles. Also, as Satya said, it's such a complex supply chain; weird stuff gets built, and the technological landscape shifts in big ways. If a very cheap form of energy comes online soon at mass scale, then a lot of people are going to be extremely burned by existing contracts they've signed.

If we can continue this unbelievable reduction in cost per unit of intelligence—let's say it's been averaging 40x for a given level per year—that's a very scary exponent from an infrastructure-buildout standpoint. Now, again, we're taking the bet that there will be a lot more demand as that gets cheaper, but I have some fear that it's just, man, we keep going with these breakthroughs, and everybody can run a personal AGI on their laptop, and we just did an insane thing here. Some people are going to get really burned, like has happened in every other tech infrastructure cycle at some points along the way.

I think that's really well said, and you have to hold those 2 simultaneous truths. We had that happen in 2001, and yet the internet became much bigger and produced much greater outcomes for society than anybody estimated in that period of time.

Yeah, but I think the one thing that Sam said is not talked about enough, which is the optimizations that OpenAI has done on the inference stack for a given GPU. It's kind of like we talk about the Moore's law improvement on one end, but the software improvements are much more exponential than that.

8. The Future of AI Devices and Consumer Use

Someday, we'll make an incredible consumer device that can run a GPT-5- or GPT-6-capable model completely locally at a low power draw. This is so hard to wrap my head around.

That will be incredible.

And you know, that's the type of thing I think scares some of the people who are building these large, centralized compute stacks. Satya, you've talked a lot about the distribution both to the edge, as well as having inference capability distributed around the world.

Yeah, I mean, the way at least I've thought about it is more about really building a fungible fleet. When I look at the cloud infrastructure business, one of the key things you have to do is have 2 things. One is an efficient—in this context, a very efficient—token factory, and then high utilization. That's it.

There are 2 simple things that you need to achieve. In order to have high utilization, you have to have multiple workloads that can be scheduled, even on the training side. If you look at the AI pipelines, there's pre-training, mid-training, post-training, and RL. You want to be able to do all of those things. So, thinking about fungibility of the fleet is everything for a cloud provider.

Okay. So, Sam, you referenced—and Reuters was reporting yesterday—that OpenAI may be planning to go public late 2026 or in 2027.

No, no, no. We don't have anything that specific. I'm a realist. I assume it will happen someday, but I don't know why people write these reports. We don't have a date in mind, a decision to do this, or anything like that. I just assume it's where things will eventually go.

But it does seem to me, if you guys were doing in excess of $100 billion of revenue in 2028 or 2029, that you at least would be in a position to do an IPO at the rumored $1 trillion valuation.

What?

How about 2027?

Yeah, 2027— even better.

You are in a position to do an IPO. Again, just to contextualize for listeners, if you guys went public at 10 times $100 billion in revenue, which would be, I think, a lower multiple than Facebook went public at, a lower multiple than a lot of other big consumer companies went public at, that would put you at $1 trillion. If you floated 10% to 20% of the company, that raises $100 billion to $200 billion, which seems like it would be a good path to fund a lot of the growth and a lot of the stuff that we just talked about.

So, you're not opposed to it. You're not. But you guys are going to fund the company with revenue growth, which is what I would like us to do. I've also said I think that this is such an important company, and there are so many people, including my kids, who like to trade their little accounts and use ChatGPT. I think having retail investors have an opportunity to buy one of the most important and largest companies would be great.

Honestly, that is probably the single most appealing thing about it to me.

One of the things I've talked to you both about, shifting gears again, is part of the Big Beautiful Bill. Senator Cruz had included federal preemption so that we wouldn't have this state patchwork—50 different laws—that mires the industry down in needless compliance and regulation.

9. Regulation and the Patchwork Problem

Unfortunately, it got killed at the last second by Senator Blackburn because, frankly, I think AI is pretty poorly understood in Washington, and there's a lot of doomerism that has gained traction there. So now we have state laws like the Colorado AI Act, which goes into full effect in February, I believe, and creates this whole new class of litigants: anybody who claims any unfair impact from algorithmic discrimination in a chatbot. Somebody could claim harm for countless reasons.

Sam, how worried are you that having this state patchwork of AI laws poses real challenges to our ability to continue to accelerate and compete around the world?

I don't know how we're supposed to comply with that Colorado law. I would love them to tell us, and we'd like to be able to do it, but from what I've read of that, I literally don't know what we're supposed to do. I'm very worried about a 50-state patchwork. I think it's a big mistake. I think there's a reason we don't usually do that for these sorts of things. I think it'd be bad.

Yeah. I mean, I think the fundamental problem of this patchwork approach is, quite frankly, between OpenAI and Microsoft, we'll figure out a way to navigate this. We can figure this out. The problem is anyone starting a startup and trying to kind of do this—it goes to the exact opposite of what I think the intent here is.

Obviously, safety is very important, and making sure that the fundamental concerns people have are addressed is important. But there's a way to do that at the federal level. So, if we don't do this, again, the EU will do it, and then that'll cause its own issues. I think if the U.S. leads, it's better, as one regulatory framework.

For sure. And to be clear, it's not that one is advocating for no regulation. It's simply saying let's have agreed-upon regulation at the federal level, as opposed to 50 competing state laws, which certainly firebomb the AI startup industry. I think it makes it super challenging even for companies like yours, who can afford to defend all these cases.

Yeah. And I would just say, quite frankly, my hope is that this time around, even across the EU and the United States, that'll be the dream, quite frankly, for any European startup.

I don't think that's going to happen.

What is that?

That would be great, but I wouldn't hold your breath for that one.

That would be great. No, but I really think that if you think about it, if anyone in Europe is thinking about how they can participate in this AI economy with their companies, this should be the main concern there as well. Therefore, I hope there is some enlightened approach to it, but I agree with you that today I wouldn't bet on that.

I do think that with Sacks as the AI czar, you at least have a president that I think might fight for that in terms of coordination of AI policy, using trade as a lever to make sure that we don't end up with overly restricted European policy. But we shall see. I think, first things first, federal preemption in the United States is pretty critical.

10. Looking Ahead to 2026 and Beyond

You know, we've been down in the weeds a little bit here, Sam, so I want to telescope out a little bit. I've heard people on your team talk about all the great things coming up. As you start thinking about much more abundant compute, ChatGPT-6 and beyond, robotics, physical devices, and scientific research, what do you think surprises us the most? What are you most excited about in terms of what's on the drawing board?

I mean, you just hit on a lot of the key points there. I think Codex has been a very cool thing to watch this year, and as these go from multi-hour tasks to multi-day tasks—which I expect to happen next year—people will be able to create software at an unprecedented rate and really in fundamentally new ways.

I'm very excited for that. I think we'll see that in other industries too. I have a bias toward coding. I understand that one better. I think we'll see that really start to transform what people are capable of.

I hope for very small scientific discoveries in 2026, but if we can get those very small ones, we'll get bigger ones in future years. That's a really crazy thing to say: AI is going to make a novel scientific discovery in 2026. Even a very small one. This is a wildly important thing to be talking about. So, I'm excited for that.

Certainly, robotics, computing, and new kinds of computers in future years. That'll be very important. But my personal bias is, if we can really get AI to do science here, that is superintelligence in some sense. If this is expanding the total sum of human knowledge, that is a crazy big deal.

Yeah. I think one of the things—to use your Codex example—is the combination of the model capability. If you think about the magical moment that happened with ChatGPT, it was the UI that met intelligence and just took off right there. It was unbelievable, right? The form factor was part of it, and some of it was also the instruction-following piece. The model capability was ready for ChatGPT.

I think that's what Codex and these coding agents are about to help us with. A coding agent goes off for a long period of time, comes back, and then I'm dropped into what I should steer. One of the metaphors I think we're all working toward is: I do this macro-delegation and micro-steering. What is that? It's a UI that meets this new intelligence capability.

You can see the beginnings of that with Codex, right? The way, at least, I use it inside of GitHub Copilot is just a different way than the chat interface now. I think that would be a new way for the human-computer interface.

That might be the departure. That's one reason I'm very excited that we're doing new form factors of computing devices, because computers were not built for that kind of workflow very well. Certainly, a UI like ChatGPT is wrong for it.

But this idea that you can have a device that's always with you, but able to go off and do things, get micro-steering from you when it needs it, and have really good contextual awareness of your whole life and flow—I think that'll be cool.

And what neither of you have talked about is the consumer use case. I think a lot about—again, we go on this device and we have to hunt and peck through 100 different applications and fill out little web forms, things that really haven't changed in 20 years.

But just to have a personal assistant—we take it for granted, perhaps, that we actually have a personal assistant—and to give a personal assistant for virtually free to billions of people around the world to improve their lives, whether it's ordering diapers for their kid, booking their hotel, or making changes in their calendar. I think sometimes it's the pedestrian that's the most impactful.

As we move from answers to memory and actions, and then the ability to interface with that through an earbud or some other device that doesn't require me to constantly be staring at this rectangular piece of glass, I think it's pretty extraordinary.

I think that's what Sam was teasing.

Yeah. Yeah. Hope we get it right. I've got to drop off, unfortunately.

Sam, it was great to see you. Thanks for joining us. Congrats again on this big step forward, and we'll talk soon.

Thanks for letting me crash.

See you, Sam. Take care. See you.

As Satya well knows, we're certainly a buyer, not a seller. But sometimes, I think it's important because the world—you know, we're a pretty small group, and we spend all day long thinking about this stuff, right? Conviction comes from the 10,000 hours we've spent thinking about it. But the reality is, we have to bring along the rest of the world.

The rest of the world doesn't spend 10,000 hours thinking about this. Frankly, they look at some things that appear overly ambitious, right, and get worried about whether or not we can pull those things off. You took this idea to the board in 2019 to invest $1 billion into OpenAI. Was it a no-brainer in the boardroom? Did you have to expend any political capital to get it done?

Dish for me a little bit what that moment was like, because I think it was such a pivotal moment—not just for Microsoft, not just for the country, but I really do think for the world.

Yeah, it's interesting when you look back at the journey. We were involved even in 2016, when OpenAI initially started. In fact, Azure was even the first sponsor, I think, and then they were doing a lot more reinforcement learning at that time. I remember the Dota 2 competition—I think it happened on Azure—and then they moved on to other things.

I was interested in reinforcement learning, but quite frankly, it speaks a little bit to your 10,000 hours, or the prepared mind. Microsoft, since 1995, was obsessed—I mean, Bill's obsession for the company was natural language. Natural language. After all, we're a coding company. We're an information-work company.

So, it was when Sam, in 2019, started talking about text, natural language, transformers, and scaling laws.

That's when I said, "Wow, this is interesting." This was a team that was going in a direction—the direction of travel was now clear—and it had a lot more overlap with our interests. So, in that sense, it was a no-brainer.

Obviously, you go to the board and say, "Hey, I have an idea of taking $1 billion and giving it to this crazy structure, which we don't even understand. What is it? It's a nonprofit, blah blah blah," and saying, "Go for it." There was a debate.

Bill was, kind of rightfully so, skeptical. But once he saw the GPT-4 demo, that was the thing Bill's talked about publicly: when he saw it, he said it was the best demo he saw after what Charles Simonyi showed him at Xerox PARC.

The moment for me was: let's go give it a shot. Then, seeing the early Codex inside of Copilot, inside of GitHub Copilot, and seeing just the code completions and seeing it work, that's when I would say I felt like we could go from 1 to 10, because that was the big call, quite frankly. 1 was controversial.

But the 1 to 10 was what really made this entire era possible, and then obviously the great execution by the team and the productization on their part and our part. If I think about it, the collective monetization reach of GitHub Copilot, ChatGPT, Microsoft 365 Copilot, and Copilot—you add those 4 things, that is it, right? That's the biggest sort of AI set of products out there on the planet, and that's what obviously has let us sustain all of this.

I think not many people know that your CTO, Kevin Scott, an ex-Googler, lives down here in Silicon Valley. To contextualize it, Microsoft had missed out on search, had missed out on mobile. You became CEO and almost had missed out on the cloud, right? You've described it as catching the last train out of town to capture the cloud.

I think you were pretty determined to have eyes and ears down here so you didn't miss the next big thing. So I assume that Kevin played a good role for you as well.

Absolutely.

DeepSeek and OpenAI.

Yeah. In fact, I would say Kevin's conviction—and Kevin was also skeptical—was one of the big things. I always watch for people who are skeptical, who change their opinion, because to me that's a signal. I'm always looking for someone who's a nonbeliever in something and then suddenly changes, and they get excited about it. I have all the time for that, because I'm then curious: Why? What?

So, Kevin started with all of us kind of skeptical. In some sense, it defies the idea—we've all gone to school and said, "God, there must be an algorithm to crack this," versus just scaling laws and throwing compute at it. But quite frankly, Kevin's conviction that this is worth going after is one of the big things that drove this.

11. Microsoft’s Strategic Value from OpenAI

Well, we talk about that investment that's now worth $130 billion, I suppose, and could be worth $1 trillion someday, as Sam says. But it really, in many ways, understates the value of the partnership, right?

You have the value in the rev share, billions per year going to Microsoft. You have the profit you make off the $250 billion Azure compute commitment from OpenAI. And, of course, you get huge sales from the exclusive distribution of the API.

So talk to us about how you think about the value across those domains, especially how this exclusivity has brought a lot of customers who may have been on AWS to Azure.

Yeah, absolutely. To us, aside from all the equity parts, the real strategic thing that comes together and remains going forward is that stateless API exclusivity on Azure. That helps, quite frankly, both OpenAI and us and our customers.

When somebody in the enterprise is trying to build an application, they want an API that's stateless. They want to mix it up with compute and storage, put a database underneath it to capture state, and build a full workload. That's where Azure coming together with this API is important.

What we're doing even with Azure AI Foundry—because, in some sense, let's say you want to build an AI application—but the key thing is: How do you make sure that the evals are great? That's where you need even a full app server in Foundry. That's what we've done.

Therefore, I feel that that is the way we will go to market in our infrastructure business.

The other side of the value capture for us is going to be incorporating all this IP. Not only do we have the exclusivity of the model in Azure, but we have access to the IP. Even forgetting all the know-how and the knowledge side of it, having royalty-free access for 7 more years gives us a lot of flexibility business-model-wise. It's kind of like having a frontier model for free, in some sense, if you're an MSFT shareholder.

That's kind of where you should start from: We have a frontier model that we can then deploy, whether it's in GitHub, whether it's in Microsoft 365, whether it's in our consumer Copilot. Then we can add to it our own data and post-train it. That means we can have it embedded in the weights there. Therefore, we're excited about the value creation on both the Azure and infrastructure sides, as well as in our high-value domains—whether it is in health, knowledge work, coding, or security.

You've been consolidating the losses from OpenAI. I think you just reported earnings yesterday, and I think you consolidated $4 billion of losses in the quarter. Do you think investors may even be attributing negative value because of the losses, as they apply their multiple of earnings?

Satya, when I hear this, I think about all of those benefits we just described, not to mention the look-through equity value that you own in a company that could be worth $1 trillion unto itself. Do you think the market is misunderstanding the value of OpenAI as a component of Microsoft?

Yeah, that's a good one. I think the approach that Amy is going to take is full transparency because, at some level, I'm no accounting expert. Therefore, the best thing to do is to give all of the transparency, I think, this time around as well. I think that's why there's a non-GAAP gap, so that at least people can see the EPS numbers.

The common-sense way I look at it, Brad, is simple: If you've invested, let's call it, $13.5 billion, you can of course lose $13.5 billion, but you can't lose more than $13.5 billion. At least the last time I checked, that's what you have at risk. You could also say, hey, the $135 billion that is today our equity stake is sort of illiquid, what have you. We don't plan to sell it, so therefore it's got risk associated with it.

But the real story, I think, you were pulling is all the other things that are happening. What's happening with Azure growth? Would Azure be growing if we had not had the OpenAI partnership? To your point, the number of customers who came from other clouds for the first time—this is the thing that we really benefited from.

What's happening with Microsoft 365? In fact, one of the things about Microsoft 365 was, what was the next big thing after E5? Guess what? We found it in Copilot. It's bigger than any suite. We talk about penetration and usage and the pace. It's bigger than anything we've done in information work, which we've been at for decades.

We feel very, very good about the opportunity to create value for our shareholders and, at the same time, be fully transparent so that people can look through the losses. Who knows what the accounting rules are, but we will do whatever is needed, and people will then be able to see what's happening.

But a year ago, there were a bunch of headlines that Microsoft was pulling back on AI infrastructure. Fair or unfair, they were out there, and perhaps you guys were a little more conservative, a little more skeptical of what was going on. Amy said on the call last night, though, that you've been short power and infrastructure for many quarters, and she thought that you would catch up, but you haven't caught up because demand keeps increasing.

I guess the question is: Were you too conservative, knowing what you know now, and what's the road map from here?

Yeah, it's a great question. The thing that we realized, and I'm glad we did, is the concept of building a fleet that truly was fungible for all the parts of the lifecycle of AI—fungible across geographies and fungible across generations.

Because one of the key things is, when you have—let's take even what Jensen and team are doing, right? They're at a pace. In fact, one of the things I like is the speed of light. We now have GB300s that we're bringing up, so you don't want to have ordered a bunch of GB200s that are getting plugged in, only to find that GB300s are in full production.

You kind of have to make sure you're continuously modernizing, you're spreading the fleet all over, you're truly fungible by workload, and you're adding to that the software optimizations we talked about. To me, that is the decision we made, and we said, look, sometimes you may have to say no to some of the demand, including some of the OpenAI demand.

Sometimes Sam may say, “Hey, build me a dedicated, big, multi-gigawatt data center in one location for training.” That makes sense from an OpenAI perspective; it doesn't make sense from a long-term infrastructure build-out for Azure. That's where I thought they did the right thing to give them flexibility to go procure that from others, while maintaining, again, a significant book of business from OpenAI, but more importantly, giving us the flexibility with other customers and our own 1P.

Remember, one of the things that we don't want to do is be short on—you know, we talk about Azure. In fact, sometimes our investors are overly fixated on the Azure number. But remember, for me, the high-margin business is Copilot. It is Security Copilot, GitHub Copilot, and the healthcare Copilot.

We want to make sure we have a balanced way to approach the returns that the investors have. That's kind of one of the other things that are perhaps misunderstood in our investor base in particular, which I find pretty strange and funny because I think they want to hold Microsoft because of the portfolio we have. But, man, are they fixated on the growth number of one little thing called Azure.

On that point, Satya, Azure grew 39% in the quarter on a staggering $93 billion run rate. I think that compares to GCP, which grew 32%, and AWS, closer to 20%. But could Azure, because you did give compute to 1P and because you did give compute to research, have grown 41% or 42% had you had more compute to offer?

Absolutely. Absolutely. There's no question. There's no question. That's why I think the internal thing is to balance out what we think, again, is in the long-term interests of our shareholders, and also to serve our customers well.

One of the other things was, people talk about concentration risk. We obviously want a lot of OpenAI, but we also want other customers. We're supply-constrained, not demand-constrained, so we are shaping the demand such that it matches the supply in the optimal way with the long-term view.

To that point, Satya, you talked about $400 billion. It's an incredible number of remaining performance obligations. Last night, you said that that's your booked business today. It'll surely go up tomorrow as sales continue to come in, and you said your need to build out capacity just to serve that backlog is very high.

How diversified is that backlog, to your point? And how confident are you that that $400 billion does turn into revenue over the course of the next couple of years?

Yeah, that $400 billion has a very short duration, as Amy explained. It's a 2-year duration on average. That's definitely our intent. That's one of the reasons why we're spending the capital outlay with high certainty that we just need to clear this backlog.

To your point, it's pretty diversified, both on the 1P and the 3P. Our own demand is, quite frankly, pretty high for our own first-party. Even among third-party, one of the things we now are seeing is the rise of all the other companies building real workloads that are scaling.

Given that, I think we feel very good. Obviously, that's one of the best things about RPO: You can be planful, quite frankly. Therefore, we feel very, very good about building. This doesn't include, obviously, the additional demand that we're already going to start seeing, including the $250 billion, which will have a longer duration, and we'll build accordingly.

Right, so there are a lot of new entrants in this race to build out compute: Oracle, CoreWeave, Crusoe, et cetera. Normally, we think that will compete away margins, but you've somehow managed to build all this out while maintaining healthy operating margins at Azure.

I guess the question is, for Microsoft, how do you compete in this world where people are levering up and taking lower margins, while balancing that profit and risk? And do you see any of those competitors doing deals that cause you to scratch your head and say, “Oh, we're just setting ourselves up for another boom-and-bust cycle”?

I'd say, at some level, the good news for us has been competing even as a hyperscaler every day. There's a lot of competition between us and Amazon and Google on all of these. It's one of those interesting things where everything is a commodity—compute, storage. I remember everybody saying, “Wow, how can there be a margin?” Except at scale, nothing is a commodity.

Therefore, yes, we have to have our cost structure, our supply-chain efficiency, and our software efficiencies all continue to compound in order to make sure that there's margins. But scale—and to your point, one of the things that I really love about the OpenAI partnership is it's gotten us to scale. This is a scale game.

When you have the biggest workload there is running on your cloud, that means not only are we going to learn faster about what it means to operate at scale, it means your cost structure is going to come down faster than anything else. And guess what? That'll make us price-competitive. So I feel pretty confident about our ability to have margins. This is where the portfolio helps.

I've always said—you know, I've been forced into giving the Azure numbers, right? Because at some level, I never thought of allocating—I mean, my capital allocation for the cloud, whether it is Xbox cloud gaming or Microsoft 365 or Azure, is one capital outlay. Everything is a meter, as far as I'm concerned. From a Microsoft perspective, it's a question of, hey, the blended average of that should match the operating margins we need as a company. Because after all, otherwise, why are we not a conglomerate? We're one company with one platform logic. It's not running 5 or 6 different businesses; we're in these 5 or 6 different businesses only to compound the returns on the cloud and AI investment.

Yeah, I love that line: nothing is a commodity at scale. There's been a lot of ink and time spent, even on this podcast with my partner, Bill Gurley, talking about circular revenues, including Microsoft's Azure credits to OpenAI that were booked as revenue.

Do you see anything going on like the AMD deal, where they traded 10% of their equity for a deal, or the NVIDIA deal? Again, I don't want to be overly fixated on concern, but I do want to address head-on what is being talked about every day on CNBC and Bloomberg. There are a lot of these overlapping deals going on out there.

When you think about that in the context of Microsoft, does any of that worry you, again, as to the sustainability or durability of the AI revenues that we see in the world?

Yeah. First of all, our investment of, let's say, $13.5 billion—which was all the training investment—was not booked as revenue. That is the reason why we have the equity percentage. That's the reason why we have 27% of $135 billion. So that was not something that somehow made it into Azure revenue. In fact, if anything, the Azure revenue was purely the consumption revenue of ChatGPT and anything else, and the APIs they put out that they monetized and we monetized.

To your point on others, to some degree, it's always been there in terms of vendor financing, right? So it's not like a new concept: when someone's building something and they have a customer who is also building something, they need financing for it. They're taking some exotic forms, which obviously need to be scrutinized by the investment community. But that said, vendor financing is not a new concept.

Interestingly enough, we have not had to do any of that. We may have either invested in OpenAI and essentially got an equity stake in return for compute, or essentially sold them great pricing on compute in order to be able to bootstrap them. But others choose to do so differently, and I think circularity ultimately will be tested by demand, because all this will work as long as there is demand for the final output of it. Up to now, that has been the case.

Certainly, certainly. Well, I want to shift. As you said, over half your business is software applications. I want to think about software and agents. Last year on this pod, you made a bit of a stir by saying that much of application software was this thin layer that sat on top of a CRUD database. The notion that business applications exist—that's probably where they'll all collapse, right, in the agent era? Because if you think about it, they are essentially CRUD databases with a bunch of business logic. The business logic is all going to these agents.

Public software companies are now trading at about 5.2 times forward revenue. So that's below their 10-year average of 7 times, despite the markets being at all-time highs. There's lots of concern that SaaS subscriptions and margins may be put at risk by AI.

So how is AI affecting the growth rates of your software products today, those core products, and specifically as you think about databases, Microsoft Fabric, security, and Office 365? And the second question, I guess, is what are you doing to make sure that software is not disrupted, but is instead superpowered by AI?

Yeah, I think that's right. The last time we talked about this, my point really there was that the architecture of SaaS applications is changing because this agent tier is replacing the old business-logic tier. If you think about it, the way we built SaaS applications in the past was you had the data, the logic tier, and the UI all tightly coupled. AI, quite frankly, doesn't respect that coupling because it requires you to be able to decouple. Yet context engineering is going to be very important.

I mean, take something like Microsoft 365. One of the things I love about our Microsoft 365 offering is that it's low ARPU, high usage. If you think about it, Outlook or Teams or SharePoint—you pick Word or Excel—people are using it all the time, creating lots and lots of data, which is going into the graph, and our ARPU is low. So that's what gives me real confidence that this AI tier, I can meet it by exposing all my data.

In fact, one of the fascinating things that's happened, Brad, with both GitHub and Microsoft 365 is, thanks to AI, we are seeing all-time highs in terms of data that's going into the graph or the repo.

I mean, think about it. The more code that gets generated, whether it is Codex or cloud or wherever, where is it going? GitHub. More PowerPoints that get created, Excel models that get created, all these artifacts and chat conversations. Chat conversations are new docs. They're all going into the graph, and all that is needed, again, for grounding.

So that's what you turn into a forward index and an embedding, and basically that semantics is what you really use to ground any agent request. I think the next generation of SaaS applications will have to—if you are high ARPU, low usage, then you have a little bit of a problem. But we are the exact opposite. We are low ARPU, high usage, and I think that anyone who can structure that and then use this AI as, in fact, an accelerant—

If you look at the Microsoft 365 Copilot price, it's higher than anything else that we sell, and yet it's getting deployed faster and with more usage. So I feel very good. Or coding, right? Who would have thought? In fact, take GitHub. What GitHub did in the first—I don't know—15 years of its existence, or 10 years of its existence, it basically did in the last year, just because coding is no longer a tool. It's more a substitute for wages, and so it's a very different type of business model, even thinking about the stack and where value gets distributed.

So until very recently, clouds largely ran precompiled software. You didn't need a lot of GPUs, and most of the value accrued to the software layer—to the database, to the applications like CRM and Excel. But it does seem in the future that these interfaces will only be valuable if they're intelligent. If they're precompiled, they're kind of dumb. The software's got to be able to think and to act and to advise. And that requires the production of these tokens, dealing with the ever-changing context.

12. The Economics of AI and SaaS

In that world, it does seem like much more of the value will accrue to the AI factory, if you will—to Jensen producing, helping to produce these tokens at the lowest cost—and to the models. Maybe the agents or the software will accrue a little bit less of the value in the future than they've accrued in the past. Well, steelman for me why that's wrong?

Yeah. So I think there are 2 things that are necessary to try and drive the value of AI. One is what you described first, which is the token factory. And even if you unpack the token factory, it's the hardware, silicon, and systems, but then it is about running it most efficiently with the system software, with all the fungibility and maximum utilization. That's where the hyperscaler's role is, right?

What is a hyperscaler? If you said, hey, I want to run a hyperscaler, you could say, oh, it's simple. I'll buy a bunch of servers, wire them up, and run it. It's not that. I mean, if it were that simple, then there would have been more than 3 hyperscalers by now. So the hyperscaler is the know-how of running that maximum utilization and the token factories.

By the way, it's going to be heterogeneous. Obviously, Jensen's super competitive. Lisa is going to come, likely Hock is going to produce things from Broadcom. We will all do our own. So there's going to be a combination. You want to run, ultimately, a heterogeneous fleet that is maximized for token throughput and efficiency and so on. So that's one job.

The next thing is what I call the agent factory. Remember that a SaaS application in the modern world is driving a business outcome. It knows how to most efficiently use the tokens to create some business value. In fact, GitHub Copilot is a great example of it, right?

If you think about it, the auto mode of GitHub Copilot is the smartest thing we've done, right? It chooses, based on the prompt, which model to use for a code completion or a task handoff. You do that not just by choosing in some round-robin fashion.

You do it because of the feedback cycle. You have the evals, the data loops, and so on. So the new SaaS applications, as you rightfully said, are intelligent applications that are optimized for a set of evals and a set of outcomes, and then know how to use the token factory’s output most efficiently. Sometimes latency matters; sometimes performance matters. Knowing how to make that trade-off in a smart way is where the SaaS application value is.

But overall, it is going to be true that there is a real marginal cost to software this time around. It was there in the cloud era, too. When we were doing CD-ROMs, there wasn’t much of a marginal cost; with the cloud, there was, and this time around, it’s a lot more. Therefore, the business models have to adjust, and you have to do these optimizations for the agent factory and the token factory separately.

You have a big search business that most people don’t know about. It turns out that that’s probably one of the most profitable businesses in the history of the world because people are running lots of searches—billions of searches—and the cost of completing a search, if you’re Microsoft, is many fractions of a penny, right? It doesn’t cost very much to complete a search, but the comparable query or prompt stack today, when you use a chatbot, looks different, right? So, I guess the question is: assuming similar levels of revenue in the future for those 2 businesses, do you ever get to a point where that chat interaction has unit economics that are as profitable as search?

I think that’s a great point because search was pretty magical in terms of its ad unit and its cost economics, because there was the index, which was a fixed cost that you could then amortize in a much more efficient way. Whereas this one, to your point, each chat requires you to burn a lot more GPU cycles, both with the intent and the retrieval, so the economics are different. So, I think you do. That’s why I think a lot of the early economics of chat have been the premium model and subscription, even on the consumer side.

We are yet to discover whether it’s agentic commerce or whatever the ad unit is, how it’s going to be monetized. But at the same time, the fact is that, at this point, I kind of know what I use search for. In fact, I use search for very, very specific navigational queries. I used to say I use it a lot for commerce, but that’s also shifting to my Copilot. I look at the Copilot mode in Edge and Bing, or Copilot now—they’re blending in.

So, yes, I think there is going to be a relitigation, just like we talked about the SaaS disruption. We’re in the beginning of the cheese being moved a little in the consumer economics of that category.

Right. I mean, given that it’s the multitrillion-dollar—this is the thing that’s driven all the economics of the internet, right? When you move the economics of search for both you and Google, and it converges on something that looks more like a personal agent, a personal assistant chat, that could end up being much, much bigger in terms of the total value delivered to humanity. But the unit economics—you’re not just advertising against this one-time fixed index.

That’s right.

And so, that’s right. I think that the consumer could be worse. Yeah, the consumer category, because you are pulling a thread on something that I think a lot about, right? During these disruptions, you kind of have to have a real sense of what the category economics are. Is it winner-take-all? Both matter. The problem in the consumer space always is that there’s a finite amount of time, and so if I’m not doing one thing, I’m doing something else. If your monetization is predicated on some human interaction—in particular, if there was truly agentic stuff even on consumer—that could be different. Whereas in the enterprise, one, it’s not winner-take-all, and two, it is going to be a lot more friendly for agentic interaction. So, it’s not like, for example, the per-seat versus consumption. The reality is agents are the new seats.

And so, you can think of it as the enterprise monetization is much clearer. The consumer monetization, I think, is a little more murky.

We’ve seen a spate of layoffs recently, with Amazon announcing some big layoffs this week. The Magnificent 7 has had little job growth over the last 3 years despite really robust top lines. You didn’t grow your headcount really from 2024 to 2025. It’s around 225,000. Many attribute this to normalization and getting fit—just getting more efficient coming out of COVID—and I think there’s a lot of truth to that. But do you think part of this is due to AI? Do you think that AI is going to be a net job creator? And do you see this being a long-term positive for Microsoft productivity?

13. Productivity, Jobs, and the Age of AI

It feels to me like the pie grows, but you can do all these things much more efficiently, which either means your margins expand or it means you reinvest those margin dollars and you grow faster for longer. I call it the golden age of margin expansion.

I’m a firm believer that the productivity curve does and will bend, in the sense that we will start seeing some of the work, and the workflow in particular, change, right? There’s going to be more agency for you at a task level to get to job complete because of the power of these tools in your hand, and I think that is going to be the case.

That’s why I think we are, even internally, for example, when you talked about our allocation of tokens, wanting to make sure that everybody at Microsoft has standard issue, right? All of them have Microsoft 365 Copilot to the hilt in the most unlimited way and have GitHub Copilot, so that they can really be more productive.

But here is the other interesting thing, Brad: we’re learning that there is a new way to even learn, which is how to work with agents. That’s kind of like when Word, Excel, and PowerPoint all showed up in Office. We learned how to rethink, let’s say, how we did a forecast.

Think about it, right? In the 1980s, forecasts were interoffice memos and faxes and what have you. Then suddenly somebody said, “Oh, here’s an Excel spreadsheet. Let’s put it in an email. Send it around. People enter numbers, and there was a forecast.”

Similarly, right now, any planning, any execution starts with AI. You research with AI. You think with AI. You share with your colleagues and what have you. So there’s a new artifact being created and a new workflow being created. That is the pace of change of the business process that matches the capability of AI. That’s where the productivity efficiencies come. Organizations that can master that are going to be the biggest beneficiaries, whether it’s in our industry or, quite frankly, in the real world.

And so, is Microsoft benefiting from that? Let’s think about a couple of years from now. 5 years from now, at the current growth rate—it will be sooner, but let’s call it 5 years from now—your top line is twice as big as what it is today. Satya, how many more employees will you have?

One of the best things right now is these examples that I’m hit with every day from the employees of Microsoft. There was this person who leads our network operations, right? If you think about the amount of fiber we have had to put in for this 2-gigawatt data center we just built out in Fairwater, and the amount of fiber there, the AI and what have you, it’s just crazy, right?

It turns out this is a real-world asset. There are, I think, 400 different fiber operators we’re dealing with worldwide. Every time something happens, we’re literally going and dealing with all these DevOps pipelines. The person who leads it basically said to me, “You know what? There’s no way I’ll ever get the headcount to go do all this. Not to forget—even if I approve the budget, I can’t hire all these folks.”

So she did the next best thing. She just built herself a whole bunch of agents to automate the DevOps pipeline of how to deal with the maintenance. That is an example of, to your point, a team with AI tools being able to get more productivity.

So, in answer to your question, I will say we will grow headcount. But the way I look at it is that the headcount we grow will grow with a lot more leverage than the headcount we had pre-AI.

And that’s the adjustment I think, structurally, you’re seeing first, right? One, you called it getting fit; I think of it as more getting to a place where everybody is really now learning how to rethink how they work. And it’s the how, not even the what. Even if the what remains constant, how you go about it has to be relearned. It’s the unlearning and learning process that I think will take the next year or so; then the headcount growth will come with max leverage.

Yeah, no, I think we’re on the verge of incredible economic productivity growth. It does feel like, when I talk to you or Michael Dell, that most companies aren’t even really in the first inning—maybe the first batter in the first inning—in reworking those workflows to get maximum leverage from these agents. But it sure feels like over the course of the next 2 to 3 years, that’s where a lot of gains are going to start coming from.

And again, I certainly am an optimist. I think we’re going to have net job gains from all of this. But I think for those companies, they’ll just be able to grow their bottom line, their number of employees slower than their top line. That is the productivity gain to the company. Aggregate all that up. That’s the productivity gain to the economy. And then we’ll just take that consumer surplus and invest it in creating a lot of things that didn’t exist before.

100%. Even in software development, one of the things I look at is that no one would say we're going to have a challenge in having more software engineers contribute to our society, because the reality is, you look at the IT backlog in any organization. The question is: Are all these software agents hopefully going to help us take a whack at all the IT backlog we have and realize that dream of evergreen software?

That's going to be true. Then think about the demand for software. So, to your point, I think the levels of abstraction at which knowledge work happens will change. We will adjust to that, and the work and the workflow will then adjust themselves, even in terms of the demand for the products of this industry.

I'm going to end on this, which is really around the reindustrialization of America. I've said that if you add up the $4 trillion of capex that you and so many of the large U.S. tech companies are investing over the course of the next 4 or 5 years, it's about 10 times the size of the Manhattan Project on an inflation-adjusted or GDP-adjusted basis.

14. Reindustrialization of America

So it's a massive undertaking for America. The president has made it a real priority of his administration to recut the trade deals, and it looks like we now have trillions of dollars. South Korea committed $350 billion of investment just today into the United States.

When you think about what you see going on in power in the United States—both production, the grid, et cetera—and what you see going on in terms of this reindustrialization, how do you think this is all going? Maybe just reflect on where we're landing the plane here and your level of optimism for the few years ahead.

Yeah, no, I feel very, very optimistic because, in some sense, Brad Smith was telling me about the economy around a Wisconsin data center. It's fascinating. Most people think, “Oh, a data center—that's going to be one big warehouse, and it's fully automated.” A lot of that is true, but first of all, think about what went into the construction of that data center and the local supply chain of the data center. That is, in some sense, the reindustrialization of the United States as well.

Even before you get to what is happening in Arizona with the TSMC plants, or what is happening with Micron and its investments in memory, or Intel and its fabs, there's a lot of stuff that we will want to start building. It doesn't mean we won't have trade deals that make sense for the United States with other countries, but to your point, the reindustrialization for the new economy—and making sure that all the skills and all that capacity, from power on down, are in place—I think is very important for us.

The other thing that I also say, Brad, is important—and this is something that I've had a chance to talk to President Trump, as well as Secretary Lutnick and others, about—is that it's important to recognize that we, as hyperscalers of the United States, are also investing around the world.

In other words, the United States is the biggest investor in compute factories or token factories around the world. Not only are we attracting foreign capital to invest in our country so that we can reindustrialize, we are helping, whether it's in Europe, Asia, elsewhere in Latin America, or Africa, with our capital investments—bringing the best American tech to the world that they can then innovate on and trust.

Both of those, I think, bode really well for the United States long term.

I'm grateful for your leadership, Satya. Sam is really helping lead the charge at OpenAI for America. I think this is a moment where I look ahead—you can see 4% GDP growth on the horizon. We'll have our challenges. We'll have our ups and downs. These tend to be stairs, stairs up, rather than a line straight up and to the right.

But I, for one, see a level of coordination going on between Washington and Silicon Valley, between big tech and the reindustrialization of America, that gives me cause for incredible hope. Watching what happened this week in Asia, led by the president and his team, and then watching what's happening here is super exciting.

So thanks for making the time. We're big fans. Thanks, Satya.

Thanks so much, Brad. Thank you.