Satya, many thanks for being here. Great to talk to you.
Let's start with AI. Let's talk AI. So, analysts estimate that the AI hyperscalers could invest over $500 billion in AI technology till 2026. AI is revolutionizing work and life. There are also some critical voices. For example, just yesterday, Google CEO Sundar Pichai called out the irrationality of the AI boom. He said that every company would be affected if the AI bubble were to burst. How would you describe the current situation? Are we in a bubble?
The way I think about these shifts, Felix, is from time to time in our own industry we've had these every 10 years massive platform shifts, whether it was the PC, client server, web, internet, mobile, cloud, and now we have, at least in my own career, this is the fourth one that we are seeing. So the way I think about this is through the arc of that shift. This shift, just like the cloud one, comes with both capital intensity and knowledge intensity. After all, the cloud wave also required us to build data centers and also do software innovation, and the AI one in some sense is more of a hyperdrive version of that. So I think the reality is for us all to keep that long-term arc and keep going, because if you think about what's going to be true 5 years from now, tech as a percentage of GDP in any country, Germany or in the United States or any place, is it going to be higher or lower? Absolutely higher. So once you say it's going to be higher, then you really should not worry too much about what happens quarter over quarter. You have to be mindful that you want to make the investments in such a way that there is an ultimate return. But these are shifts that are secular in nature. They are economy-wide impact in nature. So therefore I think of this as I'm very, very optimistic about this trend. Ultimately it has to deliver real value in the world. That is where the focus should be, and we see it already.
Your new Fairwater data center in Atlanta has 10 times the performance of the world's largest supercomputer. It holds hundreds of thousands of Nvidia chips. This year alone, Microsoft is investing more than $80 billion in AI. Talk about that a bit. Why is that necessary? When will this investment pay off? And excuse the question, what happens if the bubble bursts to Microsoft and the economy?
First of all, you also have to not only look at our capital spend but also at our free cash flow and our revenue. The only reason why we're able to spend that is because it's not future revenues, it's not about future free cash flow; it is Microsoft's ability to have existing businesses continue to scale. Think about it: our Azure business by itself is larger than all of Microsoft was 10 years ago. So we already have permission to invest. And even AI revenues, in fact, Microsoft is pretty unique in that sense even amongst the hyperscalers: we are not talking about future AI revenues; we have at-scale AI revenues today, and that's what shows up in Azure growth rates. So the way I think about the ROIC or return from an investor perspective is pretty straightforward: we are going to build out a very fungible fleet. Even with Fairwater, we have Fairwater in Atlanta, Fairwater in Wisconsin. These are massive AI factories. We've even connected them with an AI van. They're also connected to all of our other regions in the United States. So it's a fungible AI fleet which will support training, inference, and everything in between like data generation. So we are really making sure that the asset here is something that will truly be helpful in the long run for the scale out of our revenue. That's why I think we are in good shape on the investment. The other thing I'd also mention is our portfolio. Not only is the fleet fungible, but even our portfolio: we have Copilot in consumer, we have Microsoft 365 Copilot, we have GitHub Copilot, we have DAX Copilot in healthcare, and Security Copilot. So we have a family of Copilot products, a fungible fleet, and a third-party business. If you take those three, I feel like we have a very good mix of things that allow us to make sure there's a real asset long term.
In 1987, Nobel Prize winner Robert Solow said, speaking of real returns here, 'You can see the computer age everywhere but in the productivity statistics.' Is that true for the AI age so far, and what has to happen?
I remember that quote very well because that was around the time I joined the computing industry. The other critique of technology and productivity has been Robert Gordon out of Northwestern as well. In fact, if you look at his book, productivity started showing up, or tech-driven productivity started showing up in the GDP in the early 2000s, and guess what? It was mostly because of the PC and the workplace. So PC in the workplace started the broad diffusion in the early 90s; it developed in the 80s, late 80s, but really by the mid-90s it hit mainstream, and by 2000 it had, at least in the West, diffused and showed up in the statistics. The lesson I picked up then is these things first require to be broadly diffused. People have to start using it. You can't suddenly say 'I used Excel once, where's my productivity?' You have to use Excel and email and change the business process around forecasting in order for productivity to show up. So it's not just the tool and its presence; it's your intense use of the tool. That's where the reality is: we care about not just the technology. Right now, we're still in the hype phase of just talking about technology. We need to get into the phase where everybody is using, let's call it, agentic models and tools and scaffolding to actually do something that they are proud of.
Let's talk about this AI future that's probably some years away. How will AI, how will this technology change our lives, our work, our world, let's say in the next 10 years?
If you take one of the frontier use cases, which is software development, look at the arc of software development. I remember distinctly when I started developing software: I learned, thank God I was in the era after punch cards, but I started on a PC on a Z80, and it was all about writing assembly or Basic. But the bottom line is I first learned assembly language, then I learned there were compilers, and we got a level of abstraction and I was able to write in a higher-level language. I remember using Lex and Yacc in school and building my own compiler for a fourth-generation language. So with every generation we have had more and more abstractions, and now we have one more with GitHub Copilot. I'm still doing all the cognition and the metacognition with an AI agent. For example, if I want to bootstrap into a new framework, it does the bootstrap and I then do the steering. One of the descriptions I give is macro delegation and micro steering; that's the new interface. The way we'll work with AI as humans is using that metaphor. So it'll impact everybody. I'll give you another use case: when you go to a doctor today who is using DAX Copilot, what are they doing? They're spending more time with the patient, having DAX Copilot diarize their interaction, enter that into their EMR system, do the billing codes for them, so that they can spend more time thinking about health outcomes for the patient. A security analyst sitting there getting all these false alerts or alerts on phishing campaigns: AI does the triage, allows me to focus on the things that matter. So AI is making its way into everyday workflow, giving a doctor, a security analyst, a software developer, or a knowledge worker levels of abstraction where we can start focusing on things that matter and not be distracted by the drudgery of knowledge work. That's the first phase. Based on that, humankind has always been able to figure out new tasks. In the 80s, if somebody had come to us and said there's going to be a grand future where three billion people will get up every morning and start typing, we would have said, 'What are you talking about?' That was in the pre-PC era. Yet knowledge work is what everyone does around the world. I think that's the type of thing that will happen in the future.
Do you know what happened almost exactly 2 years ago on November 17, 2023? November 17, 2023. Is that quite busy?
Oh, this is the blip side of it. Yeah, I got it. I think November was when we did even ChatGPT the year before, so we're talking about the ousting of Sam Altman by the OpenAI board.
Satya, back then you stepped in and you saved the shop, basically. That's at least what many observers are telling us. First you did the turnaround of the century with Microsoft, you bet the company on the cloud, tremendous success. Then you bet the company on OpenAI, a small startup. And there were some people that said, 'Wow, Satya is betting the future of a company with 200,000 employees on a small startup in San Francisco.' Were you afraid that weekend two years ago that you bet on the wrong horse?
I must say that weekend I mostly remember the outcome of a cricket match between India and Australia, which was not that great. But other than that, the best way to characterize it is what I said to all parties then and what I say even today: at the end of the day, I've grown up in a company that understands how to partner and partner well. I take that seriously. We had backed at that time OpenAI, a research lab, a nonprofit research lab, and we were very comfortable working with them. Even during the November episode, I was consistent with that, saying, 'Look, I am going to partner with OpenAI, and if something happens there, I'll partner with Sam even inside or outside OpenAI.' I'm glad it ended well for all parties, and here we are, and OpenAI continues to thrive. When I look at today, if I think back at it, Microsoft's partnership with OpenAI has created perhaps one of the most valuable private companies, one of the most valuable nonprofits next to the Gates Foundation. I'm glad that Microsoft is associated with two of the largest nonprofits, and we continue to have a fantastic partnership between the two companies. I'm also glad that they have other partners and we have other partners.
Your partners, you're also competitors to some extent. Frenemies. OpenAI opened its first office in Munich. It's also approaching German customers of Microsoft. There are some CTOs that told me in the past months, for example, that their employees prefer ChatGPT to Microsoft's Copilot, that it's just working better. Some of them even have an agreement with Microsoft and their employees privately use ChatGPT, not Copilot. What's your response?
