Satya Nadella – Microsoft’s AGI Plan & Quantum Breakthrough

Summary

Highlights

• Satya Nadella discusses Microsoft's dual breakthroughs in AGI and quantum computing with Dwarkesh Patel. 🎉
• Reflects on the evolution of tech trends and Microsoft's strategic decisions over the decades. 💡
• Explains how scalability in AI infrastructure is a key to meeting growing global demands. 🏗️
• Details on how Microsoft's Majorana zero modes are a potential game-changer in quantum computing. ⚛️
• Importance of aligning tech developments with societal needs and legal frameworks. ⚖️

Key Takeaways

• Microsoft's AGI advancements and quantum breakthroughs are set to transform the tech landscape. 🚀
• Satya Nadella reflects on past tech trends, highlighting valuable lessons for future innovations. 📚
• AI infrastructure's scalability is crucial for meeting future demands. 📈
• Microsoft's Majorana zero modes could usher in the era of utility-scale quantum computing. 🖥️
• Technological advancements must align with societal trust and economic growth for maximum impact. 🌍

Overview

Chapters

• 00:00 - 00:30: Introduction The introduction chapter sets the foundation of the book by providing an overview of the main themes and objectives. It outlines the relevance and importance of the topics to be discussed, highlighting key questions and problems the book aims to address. Additionally, it offers a brief glimpse into the structure of the book, guiding the reader on what to expect in subsequent chapters. This chapter establishes the context and background necessary for understanding the detailed discussions that are to follow.
• 00:30 - 01:00: Microsoft's Recent Breakthroughs In this chapter, Satya Nadella discusses Microsoft's recent two major breakthroughs, which were significant enough to be featured in Nature on the same day. The breakthroughs include the development of the Majorana zero chip and advancements in world human action models.
• 01:00 - 02:00: Early Years in Tech and Lessons Learned The chapter titled 'Early Years in Tech and Lessons Learned' reflects on the technological developments and recurring patterns observed from the 1980s to the present. The speaker expresses excitement about noticing similar dynamics occurring again today, which takes them back to their initial years in the technology field. The initial part of the conversation also contains an appreciation for the host's podcast, noting the broad range of topics covered.
• 02:00 - 03:00: The Evolution of Client-Server Architecture The chapter discusses the historical context of client-server architecture evolution, focusing on debates from the 1990s. It highlights the technological debates between RISC (Reduced Instruction Set Computer) and CISC (Complex Instruction Set Computer), as well as the feasibility of building servers using x86 architecture. It also references the early days of Microsoft Windows NT and the development of a comprehensive technology stack from core silicon platforms to operating systems and application tiers.
• 03:00 - 04:00: Web's Impact on Technology and Business Models This chapter discusses the significant impact the web has had on technology and business models. It begins with a reference to ongoing litigations which have been part of the technological evolution. The transition to cloud computing and distributed systems is highlighted as a major turning point that changed traditional client-server models. The chapter suggests that the current changes in technology, likely driven by the web, are more comprehensive, affecting the full stack of technology and business models more thoroughly than any previous shifts witnessed by the narrator or others in the field. The chapter also encourages readers to consider various decision-making processes influenced by these technological advancements.
• 04:00 - 05:00: Long-term Trends and Secular Changes This chapter explores the differences between lasting technological trends and temporary ones, with a focus on the winners and losers of the 80s and 90s. It includes an analysis of Sun Microsystems' experience during the 90s dotcom bubble, discussing the concept of the data center build-out bubble, and its long-term impact demonstrated by the current prevalence of the Internet. The chapter aims to distinguish between secular (long-term, lasting) trends and ephemeral (short-lived) ones, setting the foundation for understanding what developments may stand the test of time.
• 05:00 - 06:00: The Cloud and AI: Market Predictions The chapter discusses the evolution of computing technologies, focusing on the transition from client-server architectures to modern cloud computing. It highlights the introduction of the graphical user interface and x86 architecture, which enabled the development of servers. Additionally, it recounts an experience from a developer conference in 1991 where Microsoft introduced the Win32 interface, signaling significant advancements in software development.
• 06:00 - 07:00: Hyperscalers and AI Workloads The chapter discusses the transition to x86 servers and the parallels with client-side computing. It highlights the secular trend towards centralizing computing power, predicting that what happened with client-side applications would occur on the server side as well. This led to the development of client-server applications, providing a clear model for application development.
• 07:00 - 08:00: Market Dynamics and Winner-Take-All Models The chapter discusses the impact of the web on market dynamics and the emergence of winner-take-all models. It mentions the launch of the Netscape or Mosaic browser in 1993 by Andreessen and his team, highlighting it as a significant game-changer. This development occurred just as Microsoft was gearing up for success in the client-server wave, signaling their impending victory in that sector as well.
• 08:00 - 09:00: Open Source and Model Diversity The chapter discusses the shift to the browser as a new application model. It describes the adaptations made to embrace this change, including integrating HTML into Word, developing a proprietary browser, and creating a web server to compete in the new landscape.
• 09:00 - 10:00: Consumer Markets and Network Effects The chapter discusses the often overlooked potential of business models on the web, particularly highlighting the rise of the search engine model as pioneered by Google. It emphasizes the importance of not just identifying technological trends, but also understanding where value creation will occur in the market. This is exemplified by Google's success in organizing the web through search, a path that others did not foresee as prominently.
• 10:00 - 11:00: Cloud Commoditization and Future of AI The chapter discusses the challenges and opportunities presented by shifts in business models and technology trends, specifically focusing on the commoditization of cloud services and the future of AI. It highlights the critical areas where value will be created in AI, emphasizing that there are two significant opportunities, particularly for successful hyperscalers.
• 11:00 - 12:00: Predicting Future Revenue Growth In the chapter titled 'Predicting Future Revenue Growth,' the discussion centers around the critical importance of computing power in determining the success of AI systems. It emphasizes that greater computing capabilities translate to a significant advantage. It highlights how technological advancements, particularly in GPUs and AI accelerators, are essential in scaling and enhancing AI workloads, such as those in ChatGPT. This inevitably impacts revenue growth prospects, as entities capable of handling extensive compute tasks position themselves as leaders in the AI field.
• 12:00 - 13:00: The Role of AI in Economic Growth The chapter discusses the increasing role of AI in bolstering economic growth. It highlights the expanding need for infrastructure that can support the voracious demand for computational power due to AI workloads. The conversation points out that AI systems require significant compute resources not only for their training phase but also during their testing phase. As AI agents become more prevalent, the economic systems must adapt to these increasing computational demands, thereby driving further growth in the underlying technological infrastructure.
• 13:00 - 14:00: The Balance Between Supply and Demand In this chapter titled 'The Balance Between Supply and Demand,' the discussion revolves around the impact of AI agents on compute usage. It highlights how AI isn't limited by a single user running a single program. Instead, one user can trigger numerous programs, massively increasing demand for computational resources. This shift is expected to significantly expand hyperscale businesses, such as Azure, and alter the competitive landscape, potentially leading to a winner-takes-all scenario.
• 14:00 - 15:00: Investment in AI and Return Expectations The chapter delves into the dynamics of investing in AI, emphasizing the importance of accurately identifying winner-take-all markets versus those that aren't. It highlights a personal anecdote regarding Azure and Amazon, illustrating early skepticism faced when entering markets dominated by major players like Amazon, with outsiders believing it was impossible to catch up. This reflects the broader theme of discerning market potential and investment returns in AI.
• 15:00 - 16:00: Compute Infrastructure and Scalability The chapter discusses the nature of hyperscale infrastructure in the context of competition and market dynamics. It contrasts consumer markets, which can be winner-take-all, with corporate and enterprise markets that prefer multiple suppliers for their compute infrastructure. This preference is attributed to the informed nature of buyers like IT departments, who strategically choose to avoid dependence on a single supplier.
• 16:00 - 17:00: Global Expansion of AI Capabilities The chapter discusses the future of artificial intelligence with a focus on the dual presence of open-source and closed-source models. It draws parallels to the software industry, particularly mentioning how closed-source operating systems like Windows have inspired the growth of open-source complements. This balance between open and closed sources acts as a regulatory mechanism, ensuring that open-source alternatives are always available alongside closed-source ones. The chapter implies that such dynamics will shape the global expansion of AI capabilities.
• 17:00 - 18:00: Role of AI in Knowledge Work The chapter discusses the potential impacts of artificial intelligence on knowledge work, with a focus on mitigating the closed-source and winner-take-all market dynamics. The conversation suggests that if AI technology proves to be as powerful as anticipated, governmental intervention may occur to regulate its effects. It contemplates a non-monopolistic future for AI development that fosters a competitive and fair marketplace.
• 18:00 - 19:00: AI and Change Management in Organizations This chapter discusses the impact of AI, using ChatGPT as a key example, on change management within organizations. It highlights the concept of 'winner-take-all' effects due to network advantages, where successful AI applications like ChatGPT can achieve significant market dominance and consumer recognition early on. The mention of ChatGPT's consistent high ranking in the App Store exemplifies its rapid adoption and integration into everyday consumer use, showcasing the competitive advantage gained by early market leaders in AI technology.
• 19:00 - 20:00: New Workflows Enabled by AI The chapter titled 'New Workflows Enabled by AI' explores the evolving landscape of AI in various sectors. It discusses how AI can transform workflows by providing competitive advantages, particularly in consumer and enterprise markets. Different categories may see different winners, as AI models might become commoditized over time. The analogy is made with the cloud's evolution over the past decades, transitioning from merely being a chip in a box to a now expansive service platform.
• 20:00 - 21:00: The Future of AI and Managing AI Agents The chapter titled 'The Future of AI and Managing AI Agents' discusses the evolution and potential of artificial intelligence, particularly the concept of Artificial General Intelligence (AGI). It highlights how achieving economies of scale and adding value in cloud computing have led to high-profit margins. The transcript suggests that the development of AGI, which could assist in creating more advanced AI systems, including synthetic data and reinforcement learning (RL), provides a strategic advantage. This advancement could eventually lead to the emergence of automated AI researchers, further entrenching the competitive edge of those leading in this field. The speaker underscores the importance of staying ahead in the AI race, noting that at a large scale, nothing remains a commodity.
• 21:00 - 22:00: Microsoft's Quantum Breakthrough The chapter discusses the misconception that cloud computing is a mere commodity and emphasizes the significance of the expertise required to run a large-scale cloud service, or hyperscaler. Initially, many believed that cloud hosting businesses were not profitable, questioning the viability of hyperscalers like Azure. However, the chapter highlights how mastering the operation of a hyperscale system, such as managing the world's computing needs across more than 60 regions, proves the existence of a substantial business opportunity. Azure's success in effectively managing this scale demonstrates the critical nature of cloud computing knowledge and the business potential in hyperscale operations.
• 22:00 - 23:00: The Road to Utility-Scale Quantum Computing The chapter discusses the concept of investing in fields with large Total Addressable Markets (TAMs). It questions whether certain markets are 'winner-take-all' and suggests that it's advantageous to enter categories where various competitors can thrive. This reduces the risk associated with a market being dominated by a single player, making it more appealing for investors and innovators.
• 23:00 - 24:00: Comparisons with Other Quantum Efforts This chapter discusses the intersection of hyperscale computing with quantum efforts, emphasizing the need for models to operate on hyperscale compute. It highlights the continuous necessity of storage and traditional computing resources for maintaining states and running agent-based environments.
• 24:00 - 25:00: AI, Quantum, and the Future of Computing The chapter discusses the limitations of a single entity dominating with one AI model and highlights the advantages of hyperscalers, especially in terms of leveraging data centers and GPUs for both AI model training and inference.
• 25:00 - 26:00: Quantum Computing Timeline and Applications The chapter discusses what kind of hyperscaler Microsoft Azure represents in the context of quantum computing and cloud services. It explores whether Azure focuses on pre-training, O3-type inference, or hosting and deploying models. The discussion also touches on Azure's strategy concerning Moore's Law.
• 26:00 - 27:00: The Role of AI in Simulating Nature The chapter discusses how artificial intelligence has been increasingly used to simulate nature, highlighting the importance of refreshing the technology fleet regularly. It emphasizes the strategy of asset depreciation and the effective placement of fleet resources to handle various tasks efficiently. The chapter points out the occasional necessity for highly concentrated peak periods to address extensive training jobs.
• 27:00 - 28:00: Long-Term Research Decisions at Microsoft The chapter discusses the challenges Microsoft faces in managing long-term research due to the growing scale of their data processing needs. As data processing requirements grow, particularly in pre-training, Microsoft needs to expand its data center capabilities. Challenges emerge as data processing scales to a point where it surpasses the capacity of individual data centers, requiring coordination across multiple centers.
• 28:00 - 29:00: Microsoft's Research Strategy and Goals The chapter discusses Microsoft's research strategy and goals, particularly focusing on the importance of building a robust distributed computing infrastructure. This infrastructure is essential for supporting large training jobs, test-time computations, and potential reinforcement learning (RL) applications. The strategy emphasizes the need for increased training capabilities to create specialized models.
• 29:00 - 30:00: Cultural and Managerial Challenges The chapter titled 'Cultural and Managerial Challenges' discusses the implementation of distilled models tailored for various tasks, focusing on the need for a globally distributed inference fleet. It highlights the importance of minimizing latency by ensuring that data centers are strategically located around the world, rather than relying on a single, centralized data center. Aside from inference fleet distribution, the chapter also underscores the importance of proximity between storage, compute resources, and their associated tasks to maintain efficiency in a hyperscale environment.
• 30:00 - 31:00: Economic Implications of AI and AGI The chapter discusses the infrastructure requirements for AI and AGI technologies, emphasizing the importance of stateless AI accelerators, adequate storage for training data, and the need for systems that support multiplexing of multiple training jobs. It highlights the necessity of memory and execution environments for agents to perform programs effectively.
• 31:00 - 32:00: The AGI Debate and Human Labor The chapter discusses the financial growth potential of Artificial General Intelligence (AGI) and its impact on human labor. It highlights a projected increase in revenue from AI, estimating it could reach $130 billion if current trends persist. The chapter raises questions about the deployment of such intelligence on an industrial scale, pondering whether it will be through software like Office or hosted deployments by others. The discussion emphasizes the need for AGI to achieve the projected revenue goals and questions what the future utilization of this intelligence will look like.
• 32:00 - 33:00: Legal and Ethical Considerations in AI The chapter focuses on legal and ethical considerations in artificial intelligence, particularly in the context of rapid growth and commoditization of intelligence. The speaker acknowledges the excitement around artificial general intelligence (AGI) but emphasizes the importance of observing GDP growth as a primary indicator amidst such advancements. The discussion hints at the necessity of cautious progression, especially when considering the implications for corporations like Microsoft. The speaker acknowledges that while there's significant potential, it's equally essential to consider the constraints imposed by existing economic structures, especially in the developed world, which comprises approximately 2% of the global context.
• 33:00 - 34:00: AI Alignment and Deployment Challenges The chapter discusses growth challenges, drawing parallels to the Industrial Revolution. It touches on the necessity of achieving high growth rates, such as 10% or 7%, particularly in the developed world, to mirror the growth experienced during the Industrial Revolution.
• 34:00 - 35:00: The Future of SaaS and AI Integration This chapter discusses the integration of SaaS (Software as a Service) and AI (Artificial Intelligence) and its implications on different industries. It highlights that while the tech companies are prominent players in this integration, the real beneficiaries will be the broader industries that use these technologies. As SaaS and AI become more available and abundant, they are expected to drive significant productivity growth across various sectors.
• 35:00 - 36:00: Microsoft's Legacy and Future Vision The chapter titled "Microsoft's Legacy and Future Vision" discusses the criteria for measuring progress and success in the tech industry. It critiques the self-proclaimed achievements of companies claiming advancements in Artificial General Intelligence (AGI) as mere 'benchmark hacking'. The real measure of success, according to the text, is a substantial growth rate of the global economy, cited as 10%. Such growth would translate to an increase of $10 trillion in value annually, marking true advancement and industry wellbeing.
• 36:00 - 37:00: Reflection on Leadership and Company Growth This chapter discusses the significant role of hyperscalers in the growth of the world economy. It questions whether current financial investments, like $80 billion, might be too meager compared to the actual potential and demand, suggesting perhaps a goal as high as $800 billion. The chapter highlights that one of the primary limitations to tapping into this growth is having the adequate computational power necessary for deploying AI technologies effectively. Additionally, the classic supply-side approach assumes a model of building the technological infrastructure expecting subsequent demand.
• 37:00 - 38:40: Final Thoughts and Future Directions The chapter emphasizes the importance of aligning supply and demand in business strategies. It warns against getting caught up in self-hype without recognizing real customer value. Successful strategies require balancing risk-taking with practical application and customer focus to ensure both supply and demand sides are effectively managed.

Satya Nadella – Microsoft’s AGI Plan & Quantum Breakthrough Transcription

• 00:00 - 00:30
• 00:30 - 01:00 Satya, thank you so much  for coming on the podcast.  In a second, we're going to get to the two  breakthroughs that Microsoft has just made,   and congratulations, same day in Nature: the  Majorana zero chip, which we have in front   of us right here, and also the world human action  models. But can we just continue the conversation
• 01:00 - 01:30 we were having a second ago? You're describing  the ways in which the things you were seeing in   the 80s and 90s, you're seeing them happen again. The thing that is exciting for me... Dwarkesh,   first of all, it's fantastic to be  on your podcast. I'm a big listener,   and I love the way that you do these interviews  and the broad topics that you explore.  The thing that is exciting for me… It reminds me  a little bit of my, I'd say, first few years even
• 01:30 - 02:00 in the tech industry, starting in the 90s, where  there was real debate about whether it's going   to be RISC or CISC, or, "Hey, are we really  going to be able to build servers using x86?"  When I joined Microsoft, that was the  beginning of what was Windows NT. So,   everything from the core silicon platform to the  operating system to the app tier- that full stack
• 02:00 - 02:30 approach- the entire thing is being litigated. You could say cloud did a bunch of that,   and obviously distributed computing and cloud  did change client-server. The web changed   massively. But this does feel a little more  like maybe more full-stack than even the past   that at least I've been involved in. When you think about which decisions
• 02:30 - 03:00 ended up being the long-term winners in  the 80s and 90s, and which ones didn't, and   especially when you think about- you were at Sun  Microsystems, they had an interesting experience   with the 90s dotcom bubble. People talk about  this data center build-out as being a bubble,   but at the same time, we have the Internet  today as a result of what was built out then.  What are the lessons about what will stand  the test of time? What is an inherent   secular trend? What is just ephemeral? If I go back, the four big transformations
• 03:00 - 03:30 that I've been part of, the client and the  client-server. So that's the birth of the   graphical user interface and the x86 architecture,  basically allowing us to build servers.  It was very clear to me. I remember going to  what is PDC in '91, in fact I was at Sun at   that time. In '91, I went to Moscone. That's when  Microsoft first described the Win32 interface and
• 03:30 - 04:00 it was pretty clear to me what was going to  happen, where the server was also going to be   an x86 thing. When you have the scale advantages  accruing to something, that's the secular bet you   have to place. What happened in the client was  going to happen on the server side, and then   you were able to actually build client-server  applications. So, the app model became clear.
• 04:00 - 04:30 Then the web was the big thing for us,  which we had to deal with in starting,   in fact as soon as I joined Microsoft, the  Netscape browser or the Mosaic browser came out   what, I think, December or November of '93, right?  I think is when Andreessen and crew had that.  So that was a big game-changer, in an interesting  way, just as we were getting going on what was   the client-server wave, and it was clear that  we were going to win it as well. We had the
• 04:30 - 05:00 browser moment, and so we had to adjust.  And we did a pretty good job of adjusting   to it because the browser was a new app model. We were able to embrace it with everything we did,   whether it was HTML in Word or building a new  thing called the browser ourselves and competing   for it, and then building a web server on  our server stack and go after it. Except,
• 05:00 - 05:30 of course, we missed what turned out to  be the biggest business model on the web,   because we all assumed the web is all about being  distributed, who would have thought that search   would be the biggest winner in organizing the web?  And so that's where we obviously didn't see it,   and Google saw it and executed super well. So that's one lesson learned for me: you have   to not only get the tech trend right, you also  have to get where the value is going to be created
• 05:30 - 06:00 with that trend. These business model shifts are  probably tougher than even the tech trend changes.  Where is the value going to be created in AI? That's a great one. So I think there are two   places where I can say with some confidence.  One is the hyperscalers that do well,
• 06:00 - 06:30 because the fundamental thing is if you sort of  go back to even how Sam and others describe it,   if intelligence is log of compute, whoever  can do lots of compute is a big winner.  The other interesting thing is, if you  look at underneath even any AI workload,   like take ChatGPT, it's not like everybody's  excited about what's happening on the GPU side,   it's great. In fact, I think of my fleet even  as a ratio of the AI accelerator to storage,
• 06:30 - 07:00 to compute. And at scale, you've got to grow it. Yeah.  And so, that infrastructure need for the world  is just going to be exponentially growing.  Right. So in fact it's manna   from heaven to have these AI workloads because  guess what? They're more hungry for more compute,   not just for training, but we now know, for  test time. When you think of an AI agent, it
• 07:00 - 07:30 turns out the AI agent is going to exponentially  increase compute usage because you're not even   bound by just one human invoking a program. It's  one human invoking programs that invoke lots more   programs. That's going to create massive, massive  demand and scale for compute infrastructure. So   our hyperscale business, Azure business, and  other hyperscalers, I think that’s a big thing.  Then after that, it becomes a little fuzzy. You  could say, hey, there is a winner-take-all model-
• 07:30 - 08:00 I just don't see it. This, by the way, is the  other thing I’ve learned: being very good at   understanding what are winner-take-all markets  and what are not winner-take-all markets is,   in some sense, everything. I remember even in  the early days when I was getting into Azure,   Amazon had a very significant lead and people  would come to me, and investors would come to me,   and say, "Oh, it's game over. You'll never  make it. Amazon, it's winner-take-all."
• 08:00 - 08:30 Having competed against Oracle and IBM  in client-server, I knew that the buyers   will not tolerate winner-take-all.  Structurally, hyperscale will never   be a winner-take-all because buyers are smart. Consumer markets sometimes can be winner-take-all,   but anything where the buyer is a  corporation, an enterprise, an IT department,   they will want multiple suppliers. And so  you got to be one of the multiple suppliers.
• 08:30 - 09:00 That, I think, is what will happen even on the  model side. There will be open-source. There   will be a governor. Just like on Windows,  one of the big lessons learned for me was,   if you have a closed-source operating  system, there will be a complement to it,   which will be open source. And so to some degree that's   a real check on what happens. I think in models  there is one dimension of, maybe there will be   a few closed source, but there will definitely be  an open source alternative, and the open-source
• 09:00 - 09:30 alternative will actually make sure that the  closed-source, winner-take-all is mitigated.  That's my feeling on the model side. And by the  way, let's not discount if this thing is really   as powerful as people make it out to be, the  state is not going to sit around and wait for   private companies to go around… and all over the  world. So, I don't see it as a winner-take-all.  Then above that, I think it's going to be  the same old stuff, which is in consumer,
• 09:30 - 10:00 in some categories, there may be  some winner-take-all network effect.   After all, ChatGPT is a great example. It's an at-scale consumer property that   has already got real escape velocity. I go to the  App Store, and I see it's always there in the top   five, and I say “wow, that's pretty unbelievable”. So they were able to use that early advantage and
• 10:00 - 10:30 parlay that into an app advantage. In consumer,  that could happen. In the enterprise again,   I think there will be, by category, different  winners. That's sort of at least how I analyze it.  I have so many follow-up questions. We have to  get to quantum in just a second, but on the idea   that maybe the models get commoditized: maybe  somebody could have made a similar argument a   couple of decades ago about the cloud – that  fundamentally, it's just a chip and a box.
• 10:30 - 11:00 But in the end, of course, you and many others  figured out how to get amazing profit margins in   the cloud. You figured out ways to get economies  of scale and add other value. Fundamentally,   even forgetting the jargon, if you've got AGI  and it's helping you make better AIs – right now,   it's synthetic data and RL; maybe in the future,  it's an automated AI researcher – that seems like   a good way to entrench your advantage there. I'm  curious what you make of that, just the idea that   it really matters to be ahead there. At scale, nothing is commodity.
• 11:00 - 11:30 To your point about cloud, everybody would  say, "Oh, cloud's a commodity." Except,   when you scale... That's why the know-how of  running a hyperscaler... You could say, "Oh,   what the heck? I can just rack and stack servers." Right.  In fact, in the early days of hyperscale, most  people thought “there are all these hosters,   and those are not great businesses. Will there be  anything? Is there a business even in hyperscale?”   And it turns out there is a real business,  just because of the know-how of running, in   the case of Azure, the world's computing of  60-plus regions with all the compute. It's
• 11:30 - 12:00 just a tough thing to duplicate. So I was more making the point,   is it one winner? Is it a winner-take-all or  not? Because that you've got to get right. I   like to enter categories which are big TAMs,  where you don't have to have the risk of it   all being winner-take-all. The best news to  be in is a big market that can accommodate
• 12:00 - 12:30 a couple of winners, and you're one of them. That's what I meant by the hyperscale layer.   In the model layer, one is models need ultimately  to run on some hyperscale compute. So that nexus,   I feel, is going to be there forever. It's not  just the model; the model needs state, that means   it needs storage, and it needs regular compute for  running these agents and the agent environments.
• 12:30 - 13:00 And so that's how I think about why the  limit of one person running away with   one model and building it all may not happen. On the hyperscaler side, and by the way, it's also   interesting the advantage you as a hyperscaler  would have in the sense that, especially with   inference time scaling and if that's involved  in training future models, you can amortize   your data centers and GPUs, not only for the  training, but then use them again for inference.
• 13:00 - 13:30 I'm curious what kind of hyperscaler you  consider Microsoft and Azure to be. Is it on   the pre-training side? Is it on providing  the O3-type inference? Or are you just,   we’re going to host and deploy any single  model that's out there in the market,   and we are sort of agnostic about that? It’s a good point. The way we want to   build out the fleet is [to], in some sense  ride Moore's law. I think this will be like
• 13:30 - 14:00 what we've done with everything else in the  past: every year keep refreshing the fleet,   you depreciate it over whatever the lifetime value  of these things are, and then get very very good   at the placement of the fleet such that you can  run different jobs at it with high utilization.   Sometimes there are very big training jobs  that need to have highly concentrated peak
• 14:00 - 14:30 flops that are provisioned to it that also need  to cohere. That's great. We should have enough   data center footprint to be able to give that. But at the end of the day, these are all becoming   so big, even in terms of if you take pre-training  scale, if it needs to keep going, even   pre-training scale at some point has to cross data  center boundaries. It's all more or less there.  So, great, once you start crossing pre-training  data center boundaries, is it that different than
• 14:30 - 15:00 anything else? The way I think about it is hey,  distributed computing will remain distributed,   so go build out your fleet such that  it's ready for large training jobs,   it's ready for test-time compute, it’s ready-  in fact, if this RL thing that might happens,   you build one large model, and then after  that, there’s tons of RL going on. To me,   it's kind of like more training flops, because  you want to create these highly specialized,
• 15:00 - 15:30 distilled models for different tasks. So you want that fleet, and then the   serving needs. At the end of the day,  speed of light is speed of light,   so you can't have one data center in Texas and  say, "I'm going to serve the world from there."  You've got to serve the world based on  having an inference fleet everywhere in   the world. That's how I think of our  build-out of a true hyperscale fleet.  Oh, and by the way, I want my storage and  compute also close to all of these things,
• 15:30 - 16:00 because it's not just AI accelerators that are  stateless. My training data itself needs storage,   and then I want to be able to multiplex  multiple training jobs, I want to be able to   then have memory, I want to be able to have these  environments in which these agents can go execute   programs. That's kind of how I think about it. You recently reported that your yearly revenue
• 16:00 - 16:30 from AI is $13 billion. But if you look at your  year-on-year growth on that, in like four years,   it'll be 10x that. You'll have $130 billion in  revenue from AI, if the trend continues. If it   does, what do you anticipate doing with all  that intelligence, this industrial scale use?  Is it going to be through Office? Is it going  to be you deploying it for others to host?   You've got to have the AGIs to have $130  billion in revenue? What does it look like?
• 16:30 - 17:00 The way I come at it, Dwarkesh, it's a  great question because at some level,   if you're going to have this explosion, abundance,  whatever, commodity of intelligence available,   the first thing we have to observe is GDP growth. Before I get to what Microsoft's revenue will look   like, there's only one governor in all  of this. This is where we get a little   bit ahead of ourselves with all this AGI hype.  Remember the developed world, which is what? 2%
• 17:00 - 17:30 growth and if you adjust for inflation it’s zero? So in 2025, as we sit here, I'm not an economist,   at least I look at it and say we have a real  growth challenge. So, the first thing that   we all have to do is, when we say this is  like the Industrial Revolution, let's have   that Industrial Revolution type of growth. That means to me, 10%, 7%, developed world,
• 17:30 - 18:00 inflation-adjusted, growing at 5%. That's the  real marker. It can't just be supply-side.  In fact that’s the thing, a lot of people  are writing about it, and I'm glad they are,   which is the big winners here are not going to be  tech companies. The winners are going to be the   broader industry that uses this commodity that, by  the way, is abundant. Suddenly productivity goes
• 18:00 - 18:30 up and the economy is growing at a faster rate.  When that happens, we'll be fine as an industry.  But that's to me the moment. Us  self-claiming some AGI milestone,   that's just nonsensical benchmark hacking to me.  The real benchmark is: the world growing at 10%.  Okay, so if the world grew at 10%, the world  economy is $100 trillion or something, if the   world grew at 10%, that's like an extra $10  trillion in value produced every single year.
• 18:30 - 19:00 If that is the case, you as a hyperscaler...  It seems like $80 billion is a lot of money.   Shouldn't you be doing like $800 billion? If you really think in a couple of years,   we could be really growing the world economy  at this rate, and the key bottleneck would be:   do you have the compute necessary to  deploy these AIs to do all this work?  That is correct. But by the way, the classic  supply side is, "Hey, let me build it and they’ll
• 19:00 - 19:30 come." That's an argument, and after all we've  done that, we've taken enough risk to go do it.  But at some point, the supply and demand have  to map. That's why I'm tracking both sides of   it. You can go off the rails completely when  you are hyping yourself with the supply-side,   versus really understanding how to  translate that into real value to customers.
• 19:30 - 20:00 That's why I look at my inference revenue. That's  one of the reasons why even the disclosure on the   inference revenue... It's interesting that not  many people are talking about their real revenue,   but to me, that is important as a  governor for how you think about it.  You're not going to say they have to symmetrically  meet at any given point in time, but you need to   have existence proof that you are able to parlay  yesterday's, let’s call it capital, into today's   demand, so that then you can again invest,  maybe exponentially even, knowing that you're
• 20:00 - 20:30 not going to be completely rate mismatched. I wonder if there's a contradiction in these   two different viewpoints, because one of the  things you've done wonderfully is make these   early bets. You invested in OpenAI in 2019, even  before there was Copilot and any applications.  If you look at the Industrial Revolution,  these 6%, 10% build-outs of railways and   whatever things, many of those were not  like, "We've got revenue from the tickets,   and now we're going to..." There was a lot of money lost.
• 20:30 - 21:00 That's true. So, if you really think there's  some potential here to 10x or 5x the growth   rate of the world, and then you're like,  "Well, what is the revenue from GPT-4?"  If you really think that's the possibility  from the next level up, shouldn't you just,   "Let's go crazy, let's do the hundreds of  billions of dollars of compute?" I mean,   there's some chance, right? Here’s the interesting thing, right?
• 21:00 - 21:30 That's why even that balanced approach to the  fleet, at least, is very important to me. It's   not about building compute. It's about building  compute that can actually help me not only train   the next big model but also serve the next big  model. Until you do those two things, you're not   going to be able to really be in a position  to take advantage of even your investment.  So, that's kind of where it's not  a race to just building a model,   it's a race to creating a commodity that  is getting used in the world to drive...
• 21:30 - 22:00 You have to have a complete thought, not  just one thing that you’re thinking about.  And by the way, one of the things is  that there will be overbuild. To your   point about what happened in the dotcom era,  the memo has gone out that, hey, you know,   you need more energy, and you need more compute.  Thank God for it. So, everybody's going to race.  In fact, it's not just companies deploying,  countries are going to deploy capital,
• 22:00 - 22:30 and there will be clearly... I'm so excited to  be a leaser, because, by the way; I build a lot,   I lease a lot. I am thrilled that I'm going  to be leasing a lot of capacity in '27,   '28 because I look at the builds, and I'm  saying, "This is fantastic." The only thing   that's going to happen with all the compute  builds is the prices are going to come down.  Speaking of prices coming down, you recently  tweeted after the DeepSeek model came out about   Jevons’ Paradox. I'm curious if you can flesh that  out. Jevons’ Paradox occurs when the demand for
• 22:30 - 23:00 something is highly elastic. Is intelligence  that bottlenecked on prices going down?  Because when I think about, at least my use cases  as a consumer, intelligence is already so cheap.   It's like two cents per million tokens. Do I  really need it to go down to 0.02 cents? I'm just   really bottlenecked on it becoming smarter. If you  need to charge me 100x, do a 100x bigger training   run. I'm happy for companies to take that. But maybe you're seeing something different
• 23:00 - 23:30 on the enterprise side or something. What is the  key use case of intelligence that really requires   it to get to 0.002 cents per million tokens? I think the real thing is the utility of the   tokens. Both need to happen: One is intelligence  needs to get better and cheaper. And anytime   there's a breakthrough, like even what DeepSeek  did, with the efficient frontier of performance   per token changes, the curve gets bent,  and the frontier moves. That just brings
• 23:30 - 24:00 more demand. That's what happened with cloud. Here’s an interesting thing: We used to think   “oh my God, we've sold all the servers in the  client-server era”. Except once we started   putting servers in the cloud, suddenly people  started consuming more because they could buy   it cheaper, and it was elastic, and they  could buy it as a meter versus a license,
• 24:00 - 24:30 and it completely expanded. I remember going, let’s say,   to a country like India and talking about  “here is SQL Server”. We sold a little,   but man, the cloud in India is so much bigger  than anything that we were able to do in the   server era. I think that's going to be true. If you think about, if you want to really have,   in the Global South, in a developing country,  if you had these tokens that were available   for healthcare that were really cheap,  that would be the biggest change ever.
• 24:30 - 25:00 I think it's quite reasonable for somebody to  hear people like me in San Francisco and think
• 25:00 - 25:30 “they're kind of silly; they don't know what it's  actually like to deploy things in the real world”.
• 25:30 - 26:00 As somebody who works with these Fortune  500s and is working with them to deploy   things for hundreds of millions, billions  of people, what's your sense on how fast   deployment of these capabilities will be? Even when you have working agents, even when   you have things that can do remote work for you,  with all the compliance and with all the inherent
• 26:00 - 26:30 bottlenecks, is that going to be a big bottleneck,  or is that going to move past pretty fast?  It is going to be a real challenge because  the real issue is change management or process   change. Here's an interesting thing: one of  the analogies I use is, just imagine how a   multinational corporation like us did forecasts  pre-PC, and email, and spreadsheets. Faxes went
• 26:30 - 27:00 around. Somebody then got those faxes and did an  interoffice memo that then went around, and people   entered numbers, and then ultimately a forecast  came, maybe just in time for the next quarter.  Then somebody said, "Hey, I'm just going to  take an Excel spreadsheet, put it in email,   send it around. People will go edit it,  and I'll have a forecast." So, the entire   forecasting business process changed because  the work artifact and the workflow changed.
• 27:00 - 27:30 That is what needs to happen with AI being  introduced into knowledge work. In fact, when we   think about all these agents, the fundamental  thing is there's a new work and workflow.  For example, even prepping for our podcast, I go  to my copilot and I say, "Hey, I'm going to talk   to Dwarkesh about our quantum announcement and  this new model that we built for game generation.
• 27:30 - 28:00 Give me a summary of all the stuff that I should  read up on before going." It knew the two Nature   papers, it took that. I even said, "Hey, go give  it to me in a podcast format." And so, it even   did a nice job of two of us chatting about it. So that became—and in fact, then I shared it   with my team. I took it and put it  into Pages, which is our artifact,
• 28:00 - 28:30 and then shared. So the new workflow for me is  I think with AI and work with my colleagues.  That's a fundamental change management of everyone  who's doing knowledge work, suddenly figuring out   these new patterns of "How am I going to get  my knowledge work done in new ways?" That is   going to take time. It's going to be something  like in sales, and in finance, and supply chain.  For an incumbent, I think that this is going  to be one of those things where—you know,
• 28:30 - 29:00 let's take one of the analogies I like to use  is what manufacturers did with Lean. I love   that because, in some sense, if you look at it,  Lean became a methodology of how one could take   an end-to-end process in manufacturing and become  more efficient. It's that continuous improvement,   which is reduce waste and increase value. That's what's going to come to knowledge.   This is like Lean for knowledge work, in  particular. And that's going to be the hard work   of management teams and individuals who are doing  knowledge work, and that's going to take its time.
• 29:00 - 29:30 Can I ask you just briefly about that  analogy? One of the things Lean did   is physically transform what a factory floor  looks like. It revealed bottlenecks that people   didn't realize until you're really paying  attention to the processes and workflows.  You mentioned briefly what your own  workflow—how your own workflow has   changed as a result of AIs. I'm curious if we  can add more color to what will it be like to   run a big company when you have these AI agents  that are getting smarter and smarter over time?  It's interesting you ask that. I was thinking,  for example, today if I look at it, we are very
• 29:30 - 30:00 email heavy. I get in in the morning, and I’m  like, man my inbox is full, and I'm responding,   and so I can't wait for some of these  Copilot agents to automatically populate my   drafts so that I can start reviewing and sending. But I already have in Copilot at least ten agents,   which I query them different things for  different tasks. I feel like there's a new
• 30:00 - 30:30 inbox that's going to get created, which is  my millions of agents that I'm working with   will have to invoke some exceptions to me,  notifications to me, ask for instructions.  So at least what I'm thinking is that there's  a new scaffolding, which is the agent manager.   It's not just a chat interface. I need  a smarter thing than a chat interface to
• 30:30 - 31:00 manage all the agents and their dialogue. That's why I think of this Copilot,   as the UI for AI, is a big, big deal. Each of us  is going to have it. So basically, think of it as:   there is knowledge work, and there's a knowledge  worker. The knowledge work may be done by many,   many agents, but you still have a knowledge worker  who is dealing with all the knowledge workers.   And that, I think, is the  interface that one has to build.
• 31:00 - 31:30 You're one of the few people in the world who  can say that you have access to 200,000… you   have this swarm of intelligence around you in  the form of Microsoft the company and all its   employees. And you have to manage that, and  you have to interface with that, how to make   best use of that. Hopefully, more of the world  will get to have that experience in the future.  I'd be curious about how your inbox,  if that means everybody's inbox,   will look like yours in the morning. Okay, before we get to that,
• 31:30 - 32:00 I want to keep asking you more about AI,  but I really want to ask you about the big   breakthrough in quantum that Microsoft Research  has announced. So can you explain what's going on?  This has been another 30-year journey for  us. It's unbelievable. I'm the third CEO   of Microsoft who's been excited about quantum. The fundamental breakthrough here, or the vision   that we've always had is, you need a physics  breakthrough in order to build a utility-scale
• 32:00 - 32:30 quantum computer that works. We took the path  of saying, the one way for having a less noisy   or more reliable qubit is to bet on a physical  property that by definition is more reliable and
• 32:30 - 33:00 that's what led us to the Majorana zero modes,  which was theorized in the 1930s. The question   was, can we actually physically fabricate  these things? Can we actually build them?  So the big breakthrough effectively, and I know  you talked to Chetan, was that we now finally   have existence proof and a physics breakthrough  of Majorana zero modes in a new phase of matter
• 33:00 - 33:30 effectively. This is why we like the analogy  of thinking of this as the transistor moment   of quantum computing, where we effectively have a  new phase, which is the topological phase, which   means we can even now reliably hide the quantum  information, measure it, and we can fabricate it.
• 33:30 - 34:00 And so now that we have it, we feel like with that  core foundational fabrication technique out of the   way, we can start building a Majorana chip. That Majorana One which I think is going to   basically be the first chip that will be capable  of a million qubits, physical. And then on that,   thousands of logical qubits, error-corrected.  And then it's game on. You suddenly have the
• 34:00 - 34:30 ability to build a real utility-scale quantum  computer, and that to me is now so much more   feasible. Without something like this, you  will still be able to achieve milestones,   but you'll never be able to build a utility-scale  computer. That's why we're excited about it.  Amazing. And by the way, I  believe this is it right here.  That is it. Yes.  I forget now, are we calling it Majorana? Yes,   that's right. Majorana One. I'm  glad we named it after that.
• 34:30 - 35:00 To think that we are able to build something  like a million-qubit quantum computer in a thing   of this size is just unbelievable. That's the  crux of it: unless and until we could do that,   you can't dream of building a  utility-scale quantum computer.  And you're saying the eventual million qubits  will go on a chip this size? Okay, amazing.
• 35:00 - 35:30 Other companies have announced 100 physical  qubits, Google's, IBM's, others. When you   say you've announced one, but you're saying  that yours is way more scalable in the limit.  Yeah. The one thing we’ve also done is we've  taken an approach where we've separated our   software and our hardware. We're building  out our software stack, and we now have,   with the neutral atom folks, the ion trap folks,  we're also working with others who even have
• 35:30 - 36:00 pretty good approaches with photonics and what  have you, that means there'll be different types   of quantum computers. In fact, we have what, I  think that the last thing that we announced was 24   logical qubits. So we have also got some fantastic  breakthroughs on error correction and that's what   is allowing us, even on neutral atom and ion  trap quantum computers, to build these 20 plus,
• 36:00 - 36:30 and I think that'll keep going even throughout  the year; you'll see us improve that yardstick.  But we also then said, "Let's go to the first  principles and build our own quantum computer   that is betting on the topological qubit."  And that's what this breakthrough is about.  Amazing. The million topological  qubits, thousands of logical qubits,   what is the estimated timeline to scale up to  that level? What does the Moore's law here,
• 36:30 - 37:00 if you've got the first transistor, look like? We've obviously been working on this for   30 years. I'm glad we now have the physics  breakthrough and the fabrication breakthrough.  I wish we had a quantum computer because  by the way, the first thing the quantum   computer will allow us to do is build  quantum computers, because it's going to   be so much easier to simulate atom-by-atom  construction of these new quantum gates.  But in any case, the next real thing is,  now that we have the fabrication technique,
• 37:00 - 37:30 let us go build that first fault-tolerant quantum  computer. And that will be the logical thing.  So, I would say now I can say, "Oh, maybe  '27, '28, '29, we will be able to actually   build this." Now that we have this one gate, can I  put the thing into an integrated circuit and then   actually put these integrated circuits into a real  computer? That is where the next logical step is.
• 37:30 - 38:00 And what do you see as, in '27, '28, you've got  it working? Is it a thing you access through   the API? Is it something you're using internally  for your own research in materials and chemistry? It’s a great question. One thing that I've  been excited about is, even in today's world…   we had this quantum program, and we added  some APIs to it. The breakthrough we had   maybe two years ago was to think of this HPC  stack, and AI stack, and quantum together.
• 38:00 - 38:30 In fact, if you think about it, AI is like an  emulator of the simulator. Quantum is like a   simulator of nature. What is quantum going  to do? By the way, quantum is not going to   replace classical. Quantum is great at what  quantum can do, and classical will also...  Quantum is going to be fantastic for anything  that is not data-heavy but is exploration-heavy
• 38:30 - 39:00 in terms of the state space. It should be  data-light but exponential states that you   want to explore. Simulation is a great one:  chemical physics, what have you, biology.  One of the things that we've started doing  is really using AI as the emulation engine.   But you can then train. So the way I think of it  is, if you have AI plus quantum, maybe you'll use
• 39:00 - 39:30 quantum to generate synthetic data that then gets  used by AI to train better models that know how to   model something like chemistry or physics or what  have you. These two things will get used together.  So even today, that's effectively what  we're doing with the combination of   HPC and AI. I hope to replace some of  the HPC pieces with quantum computers.
• 39:30 - 40:00 Can you tell me a little bit about how  you make these research decisions which,
• 40:00 - 40:30 in 20 years time, 30 years time, will  actually pay dividends, especially at
• 40:30 - 41:00 a company of Microsoft's scale? Obviously, you're  in great touch with the technical details in this   project. Is it feasible for you to do that  with all the things Microsoft Research does?  How do you know the current bet you're making  will pay out in 20 years? Does it just have to   emerge organically through the org, or  how are you keeping track of all this?  The thing that I feel was fantastic is when Bill,  when he started MSR back in '95 I guess. I think
• 41:00 - 41:30 in the long history of these curiosity-driven  research organizations, to just do a research   org that is about fundamental research and MSR,  over the years, has built up that institutional   strength so when I think about capital allocation  or budgets, we first put the chips in and say,
• 41:30 - 42:00 "Here is MSR's budget." We gotta go at it each  year knowing that most of these bets are not   going to pay off in any finite time frame. Maybe  the sixth CEO of Microsoft will benefit from it.   And in tech that is I think a given. The real thing that I think about is,   when the time has come for something like  quantum or a new model or what have you, can
• 42:00 - 42:30 you capitalize? So as an incumbent, if you look at  the history of tech, it's not that people didn't   invest. It's that you need to have a culture that  knows how to take an innovation and scale it.  That's the hard part, quite frankly, for CEOs and  management teams. Which is kind of fascinating.
• 42:30 - 43:00 It's as much about good judgment as it is about  good culture. Sometimes we've gotten it right;   sometimes we've gotten it wrong; I can tell you  the thousand projects from MSR that we should have   probably led with, but we didn't. And I always  ask myself why. It's because we were not able to   get enough conviction and that complete thought  of how to not only take the innovation but make
• 43:00 - 43:30 it into a useful product with a business  model that we can then go to market with.  That's the job of CEOs and management teams:  not to just be excited about any one thing,   but to be able to actually execute on a complete  thing. And that's easier said than done.  When you mentioned the possibility of three  subsequent CEOs of Microsoft, if each of them   increases the market cap by an order of magnitude,  by the time you've got the next breakthrough,   you'll be like the world economy or something. Or remember, the world is going to be growing
• 43:30 - 44:00 at 10%, so we'll be fine. Let's dig into the other big   breakthrough you've just made. It's amazing that  you have both of them coming out the same day,   in your gaming world models. I'd love if  you can tell me a little bit about that.  We're going to call it Muse. It's going to be the  model of this world action, or human action model.  This is very cool. One of the things is  that obviously, Dall-E and Sora have been   unbelievable in what they've been able  to do in terms of generative models. One
• 44:00 - 44:30 thing that we wanted to go after was  using gameplay data. Can you actually   generate games that are both consistent  and then have the ability to generate the   diversity of what that game represents,  and then are persistent to user mods?  That's what this is. They were able  to work with one of our game studios,   and this is the other publication in Nature. The cool thing is what I'm excited about is
• 44:30 - 45:00 bringing--we're going to have a catalog of games  soon that we will start using these models,   or we're going to train these models to  generate, and then start playing them.  In fact, when Phil Spencer first showed it  to me, he had an Xbox controller and this   model basically took the input and generated the  output based on the input. And it was consistent   with the game. That to me is a massive moment  of “wow”. It's kind of like the first time we
• 45:00 - 45:30 saw ChatGPT complete sentences, or Dall-E  draw, or Sora. This is one such moment.  I got a chance to see some of the videos  in the real-time demo this morning with   your lead researcher Katja on this. Only  once I talked to her did it really hit me   how incredible this is, in the sense that  we've used AI in the past to model agents,
• 45:30 - 46:00 and just using that same technique to model  the world around the agent gives consistent   real-time – we'll superimpose videos of what this  looks like atop this podcast so people can get a   chance to see it for themselves. I guess it'll  be out by then, so they can also watch it there.  This in itself is incredible. You, through your  span as CEO, have invested tens of hundreds of   billions of dollars in building up  Microsoft Gaming and acquiring IP.  In retrospect, if you can just merge all of  this data into one big model that can give
• 46:00 - 46:30 you this experience of visiting and going  through multiple worlds at the same time,   and if this is the direction gaming is headed,  it seems like a pretty good investment to have   made. Did you have any premonition about this? I wouldn't say that we invested in gaming to   build models. We invested, quite frankly, because-  here's an interesting thing about our history:   We built our first game before we built  Windows. Flight Simulator was a Microsoft
• 46:30 - 47:00 product long before we even built Windows. So, gaming has got a long history at the company,   and we want to be in gaming for gaming's  sake. I always start by saying I hate to   be in businesses where they're means to some  other end. They have to be ends unto themselves.  And then, yes, we're not a conglomerate.  We are a company where we have to bring   all these assets together and be better owners  by adding value. For example, cloud gaming is
• 47:00 - 47:30 a natural thing for us to invest in because that  will just expand the TAM and expand the ability   for people to play games everywhere. The same thing with AI and gaming:   we definitely think that it can be helpful in  maybe changing- it's kind of like the CGI moment,   even for gaming long-term. And it's great.  As the biggest, world's largest publisher,   this will be helpful. But at the same time,  we've got to produce great quality games. I mean,   you can't be a gaming publisher without, sort  of, first and foremost being focused on that.
• 47:30 - 48:00 But the fact that this data asset is going to  be interesting, not just in a gaming context,   but it's going to be a general action model  and a world model, it's fantastic. I mean like,   you know, I think about gaming data as perhaps,  you know, what YouTube is perhaps to Google,   gaming data is to Microsoft. And so  therefore I'm excited about that.  Yeah, and that's what I meant, just in the sense  of like, you can have one unified experience   across many different kinds of games. How does  this fit into the other, separate from AI,
• 48:00 - 48:30 the other things that Microsoft has worked on  in the past, like mixed reality? Maybe giving   smaller game studios a chance to build these AAA  action games? Just like five, ten years from now,   what kinds of ways could you imagine? I've thought about these three things   as the cornerstones of, in an interesting way,  even five, six, seven years ago is when I said   the three big bets that we want to place [are]  AI, quantum, and mixed reality. And I still   believe in them, because in some sense,  what are the big problems to be solved?
• 48:30 - 49:00 Presence. That's the dream of mixed  reality. Can you create real presence?   Like you and I doing a podcast like this. I think it’s still proving to be the harder   one of those challenges, quite honestly. I thought  it was going to be more solvable. It's tougher,   perhaps, just because of the social  side of it: wearing things and so on.
• 49:00 - 49:30 We're excited about, in fact, what we're  going to do with Anduril and Palmer, now,   with even how they'll take forward the  IVAS program, because that's a fantastic   use case. And so we'll continue on that front. But also, the 2D surfaces. It turns out things   like Teams, right, thanks to the pandemic,  we've really gotten the ability to create   essentially presence through even 2D. And that  I think will continue. That's one secular piece.  Quantum we talked about, and AI is the other one.  So these are the three things that I look at and
• 49:30 - 50:00 say, how do you bring these things together?  Ultimately, not as tech for tech's sake,   but solving some of the fundamental things that  we, as humans, want in our life, and more, we want   them in our economy, driving our productivity.  And so if we can somehow get that right,   then I think we will have really made progress. When you write your next book, you've got to have   some explanation of why those three pieces  all came together around the same time,   right? Like, there's no intrinsic reason you  would think quantum and AI should happen in
• 50:00 - 50:30 2028 and 2025 and so forth. That's right. At some level,   I look at it and say: the simple model I have  is, hey is there a systems breakthrough? To me,   the systems breakthrough is the quantum thing. Is there a business logic breakthrough? That's   AI to me, which is: can the logic tier be  fundamentally reasoned differently? Instead of   imperatively writing code, can you have  a learning system? That's the AI one.
• 50:30 - 51:00 And then the UI side of it is presence. Going back to AI for a second, in your   2017 book… 2019 you invest in OpenAI, very early,  2017 is even earlier, you say in your book, "One   might also say that we're birthing a new species,  one whose intelligence may have no upper limits."  Now, super-early, of course, to be talking  about this in 2017. We've been talking in   a granular fashion about agents, Office  Copilot, capex, and so forth. But if you
• 51:00 - 51:30 zoom out and consider this statement you've  made, and you think about you as a hyperscaler,   as the person doing research in these models as  well, providing training, inference, and research   for building a new species, how do you think  about this in the grand scheme of things?  Do you think we're headed towards  superhuman intelligence in your time as CEO?  I think even Mustafa uses that term. In fact he’s  used that term more recently, this “new species”.
• 51:30 - 52:00 The way I come at it is, you definitely need  trust. Before we claim it is something as big   as a species, the fundamental thing that we've  got to get right is that there is real trust,   whether it's personal or societal level trust,  that's baked in. That's the hard problem.  I think the one biggest rate limiter to the  power here will be how does our legal… call
• 52:00 - 52:30 it infrastructure, we’re talking about all  the compute infrastructure, well how does   the legal infrastructure evolve to  deal with this? This entire world is   constructed with things like humans owning  property, having rights, and being liable.   That’s the fundamental thing that one has to first  say, okay what does that mean for anything that
• 52:30 - 53:00 now humans are using as tools? And if humans are  going to delegate more authority to these things,   then how does that structure evolve? Until that  really gets resolved, I don't think just talking   about the tech capability is going to happen. As in, we won't be able to deploy these kinds   of intelligences until we figure out how to…? Absolutely. Because at the end of the day,   there is no way. Today, you cannot  deploy these intelligences unless and
• 53:00 - 53:30 until there's someone indemnifying it as a human. To your point, I think that's one of the reasons   why I think about even the most powerful AI  is essentially working with some delegated   authority from some human. You can say, oh, that's  all alignment and this, that, and the other.   That's why I think you have to really get these  alignments to work and be verifiable in some way,   but I just don't think that you can deploy  intelligences that are out of control. For   example, this AI takeoff problem may be a real  problem, but before it is a real problem, the real
• 53:30 - 54:00 problem will be in the courts. No society is going  to allow for some human to say, "AI did that."  Yes. Well, there's a lot of societies in the  world, and I wonder if any one of them might   not have a legal system that might be more  amenable. And if you can't have a takeoff,   then you might worry. It doesn't  have to happen in America, right?
• 54:00 - 54:30 We think that no society cares about it, right?  There can be rogue actors, I'm not saying there   won't be rogue actors; there are cyber criminals  and rogue states; they're going to be there.  But to think that human society at large  doesn't care about it is also not going   to be true. I think we all will care. We know  how to deal with rogue states and rogue actors   today. The world doesn't sit around and say  “we’ll tolerate that”. That's why I'm glad
• 54:30 - 55:00 that we have a world order in which anyone who is  a rogue actor in a rogue state has consequences.  Right. But if you have this picture where you  can have 10% economic growth, I think it really   depends on getting something like AGI working,  because tens of trillions of dollars of value,   that sounds closer to the total of human wages,  around $60 trillion of the economy. Getting that   magnitude, you kind of have to automate labor  or supplement labor in a very significant way.
• 55:00 - 55:30 If that is possible, and once we figure  out the legal ramifications for it,   it seems quite plausible, even within your  tenure that we figure that out. Are you   thinking about superhuman intelligence? Like,  the biggest thing you do in your career is this? You bring up another point. I know David Autor  and others have talked a lot about this which is,   60% of labor- I think the other question  that needs to happen, let’s at least talk   about our democratic societies. I think that  in order to have a stable social structure,
• 55:30 - 56:00 and democracies function, you can’t just have a  return on capital and no return on labor. We can   talk about it, but that 60% has to be revalued. In my own simple way, maybe you can call it naive,   we'll start valuing different types of  human labor. What is today considered
• 56:00 - 56:30 high-value human labor may be a commodity.  There may be new things that we will value.  Including that person who comes to me and  helps me with my physical therapy or whatever,   whatever is going to be the case that we value,  but ultimately, if we don't have return on labor,   and there's meaning in work and dignity in work  and all of that, that's another rate limiter   to any of these things being deployed. On the alignment side, two years ago,
• 56:30 - 57:00 you guys released Sydney Bing. Just to be clear, I  think given the level of capabilities at the time,   it was a charming, endearing, kind  of funny example of misalignment.  But that was because, at the time, it was like  chatbots. They can go think for 30 seconds and   give you some funny or inappropriate response.  But if you think about that kind of system--that,   I think to a New York Times reporter, tried  to get him to leave his wife or something--if
• 57:00 - 57:30 you think about that going forward, and you  have these agents that are for hours, weeks,   months going forward, just like autonomous swarms  of AGIs, who could be in similar ways misaligned   and screwing stuff up, maybe coordinating with  each other, what's your plan going forward so   that when you get the big one, you get it right? That is correct. That's one of the reasons why
• 57:30 - 58:00 when we usually allocate compute, let's allocate  compute for what is that alignment challenge?  And then more importantly, what is the runtime  environment in which you are really going   to be able to monitor these things? The  observability around it? We do deal with   a lot of these things today in the classical  side of things as well, like cyber. We don't   just write software and then just let it go.  You have software and then you monitor it.   You monitor it for cyber attacks, you monitor  it for fault injections, and what have you.
• 58:00 - 58:30 Therefore, I think we will have to build  enough software engineering around the   deployment side of these, and then inside the  model itself, what's the alignment? These are all,   some of them are real science problems.  Some of them are real engineering problems,   and then we will have to tackle it. That also means taking our own   liability in all of this. So that's why I'm  more interested in deploying these things in
• 58:30 - 59:00 where you can actually govern what the scope of  these things is, and the scale of these things   is. You just can't unleash something out there in  the world that creates harm, because the social   permission for that is not going to be there. When you get the agents that can really just do   weeks worth of tasks for you, what  is the minimum assurance you want   before you can let it run a random Fortune 500? I think when I use something like Deep Research,
• 59:00 - 59:30 even, the minimum assurance I think  we want is before we especially have   physical embodiment of anything, that I  think is kind of one of those thresholds,   when you cross. That might be one place. Then the other one is, for example,   the permissions of the runtime environment in  which this is operating. You may want guarantees   that it's sandboxed, it is  not going out of that sandbox.
• 59:30 - 60:00 I mean, we already have web search and  we already have it out of the sandbox.  But even what it does with web search and  what it writes -- for example to your point,   if it's just going to write a bunch of  code in order to do some computation,   where is that code deployed? And is that  code ephemeral for just creating that output,   versus just going and springing  that code out into the world?
• 60:00 - 60:30 Those are things that you could, in  the action space, actually go control.  And separate from the safety issues, as you think  about your own product suite, and you think about,   if you do have AIs this powerful, at some  point, it's not just like Copilot- an example   you mentioned about how you were prepping  for this podcast- it's more similar to how   you actually delegate work to your colleagues. What does it look like, given your current suite,   to add that in? I mean, there's one question about  whether LLMs get commodified by other things.
• 60:30 - 61:00 I wonder if these databases or canvases or  Excel sheets or whatever -- if the LLM is your   main gate point into accessing all these things,  is it possible that the LLMs commodify Office?  It's an interesting one. The way I think  about the first phase, at least, would be:   Can the LLM help me do my knowledge work using  all of these tools or canvases more effectively?
• 61:00 - 61:30 One of the best demos that I've seen is a doctor  getting ready for a tumor board workflow. She's   going into a tumor board meeting, and the first  thing she uses Copilot for is to create an agenda   for the meeting because the LLM helps reason about  all the cases, which are in some SharePoint site.   It says, "Hey, these cases -- obviously, a tumor  board meeting is a high-stakes meeting where you
• 61:30 - 62:00 want to be mindful of the differences in cases  so that you can then allocate the right time."  Even that reasoning task of creating an agenda  that knows how to split time- super. So, I use   the LLM to do that. Then I go into the meeting,  I'm in a Teams call with all my colleagues. I'm   focused on the actual case versus taking notes,  because you now have this AI copilot doing a full   transcription of all of this. It's not just a  transcript, but a database entry of what is in
• 62:00 - 62:30 the meeting that is recallable for all time. Then she comes out of the meeting, having   discussed the case and not been distracted  by note-taking. She's a teaching doctor;   she wants to go and prep for her class. And so she  goes into Copilot and says, "Take my tumor board   meeting and create a PowerPoint slide deck out of  it so that I can talk to my students about it."
• 62:30 - 63:00 So that’s the type. The UI and the scaffolding  that I have are canvases that are now getting   populated using LLMs. And the workflow itself is  being reshaped; knowledge work is getting done.  Here's an interesting thing: If someone came to me  in the late '80s and said, "You're going to have   a million documents on your desk," I would say,  "What the heck is that?" I would have literally
• 63:00 - 63:30 thought there was going to be a million physical  copies of things on my desk. Except, we do have a   million spreadsheets and a million documents. I don’t, you do.  They're all there. And so, that's what's  going to happen with even agents. There   will be a UI layer. To me, Office is not  just about the office of today; it's the   UI layer for knowledge work. It'll evolve as the  workflows evolve. That's what we want to build.
• 63:30 - 64:00 I do think the SaaS applications that  exist today, these CRUD applications,   are going to fundamentally be changed because  the business logic will go more into this agentic   tier. In fact, one of the other cool things  today in my Copilot experience is when I say,   "Hey, I'm getting ready for a meeting  with a customer," I just go and say,   "Give me all the notes for it that I should  know." It pulls from my CRM database, it pulls   from my Microsoft Graph, creates a composite,  essentially artifact, and then it applies even
• 64:00 - 64:30 logic on it. That, to me, is going to transform  the SaaS applications as we know of it today.  SaaS as an industry might be worth hundreds  of billions to trillions of dollars a year,   depending on how you count. If really  that can just get collapsed by AI,   is the next step up in your next decade 10X-ing  the market cap of Microsoft again? Because you're   talking about trillions of dollars... It would also create a lot of value in
• 64:30 - 65:00 the SaaS. One thing we don't pay as much  attention to perhaps is the amount of IT   backlog there is in the world. These code gen things, plus the   fact that I can interrogate all of your SaaS  applications using agents and get more utility   will be the greatest explosion of apps, they'll  be called agents, so that for every vertical,
• 65:00 - 65:30 in every industry, in every category, we're  suddenly going to have the ability to be serviced.  So there's going to be a lot of value. You can't  stay still. You can't just say the old thing of,   "Oh, I schematized some narrow business  process, and I have a UI in the browser,   and that's my thing." That's ain’t going to  be the case. You have to go up-stack and say,   "What's the task that I have to participate in?" You will want to be able to take your SaaS
• 65:30 - 66:00 application and make it a fantastic agent  that participates in a multi-agent world.   As long as you can do that, then I  think you can even increase the value.  Can I ask you some questions  about your time at Microsoft?  Yeah. Is being a company   man underrated? So you've spent most of your  career at Microsoft, and you could say that one   of the reasons you've been able to add so much  value is you've seen the culture, the history,   and the technology. You have all this context by  rising up through the ranks. Should more companies
• 66:00 - 66:30 be run by people who have this level of context? That's a great question. I've   not thought about it that way. Through my 34 years now of Microsoft,   each year I felt more excited about being at  Microsoft versus thinking that, oh, I'm a company   person or what have you. I take that seriously,  even for anybody joining Microsoft. It's not like
• 66:30 - 67:00 they're joining Microsoft as long as they feel  that they can use this as a platform for their   both economic return, but also a sense of purpose  and a sense of mission that they can accomplish by   using us as a platform. That's the contract. So I think yes, companies have to create a   culture that allows people to come in and  become company people like me. Microsoft
• 67:00 - 67:30 got it more right than wrong, at least in  my case, and I hope that remains the case.  The sixth CEO that you’re talking about, who’ll  get to use the research you’re starting now,   what are you doing to retain the  future Satya Nadellas so that they're   in a position to become future leaders? It's fascinating. This is our 50th year,   and I think a lot about it. The way to think about  it is, longevity is not a goal; relevance is.
• 67:30 - 68:00 The thing that I have to do and all 200,000 of  us have to do every day is: Are we doing things   that are useful and relevant for the world as we  see it evolving, not just today, but tomorrow?  We live in an industry where there's no franchise  value, so that’s the other hard part. If you take   the R&D budget that we will spend this year,  it’s all speculation on what's going to happen
• 68:00 - 68:30 five years from now. You have to basically go  in with that attitude, saying, "We are doing   things that we think are going to be relevant." So that's what you have to focus on. Then know   that there's a batting average, and you're not  going to get- you have to have a high tolerance   for failure. You have to take enough  shots on goal to be able to say, "Okay,   we will make it to the other side as a company."  That's what makes it tricky in this industry.
• 68:30 - 69:00 Speaking of- you just mentioned that  you're two months away from your 50th   anniversary of Microsoft’s founding. If you  look at the top 10 companies by market cap,   or top 5, basically, everybody else but  Microsoft is younger than Microsoft. It's   an interesting observation about why  the most successful companies often   are quite young. The average Fortune  500 company will last 10 to 15 years.
• 69:00 - 69:30 What has Microsoft done to remain relevant for  this many years? How do you keep refounding?  I love that, Reed Hoffman uses that term,  "refounding." That's the mindset. People   talk about founder mode, but for us mere  mortal CEOs, it's more like refounder mode.  To be able to see things again in a fresh way  is the key. To your question: can we culturally
• 69:30 - 70:00 create an environment where refounding becomes  a habit thing? Every day we come in and say,   "We feel we have a stake in this place to be  able to change the core assumptions of what   we do and how we relate to the world around us. Do  we give ourselves permission?” I think many times,
• 70:00 - 70:30 companies feel over-constrained by either  business model or whatever. You just have   to unconstrain yourself. If you did leave Microsoft,   what company would you start? Company I would start? Man.   That’s where the company man and me sort  of says, “I'll never leave Microsoft.”  If I were thinking of doing something, I think  picking a domain that has... When I look at the
• 70:30 - 71:00 dream of tech, we've always said technology is  about the biggest, greatest democratizing force.  I feel like finally, we have that ability.  If you say those tokens per dollar per watt   is what we can generate, I would love  to find some domain in which that can   be applied, where it is so underserved. That's where healthcare, education...
• 71:00 - 71:30 Public sector would be another place. If you take  those domains, which are the underserved places,   where my life as a citizen of this country or  a member of this society or anywhere, would I   be better off if somehow all this abundance  translated into better healthcare, better   education, and better public sector institutions  serving me as a citizen? That would be a place.
• 71:30 - 72:00 One thing I'm not sure about, hearing  your answers on different questions,   is whether you think AGI is a thing. Will there  be a thing which automates all cognitive labor,   like anything anybody can do on a computer? This is where I have a problem with the   definitions of how people talk about it. Cognitive  labor is not a static thing. There is cognitive   labor today. If I have an inbox that is managing  all my agents, is that new cognitive labor?
• 72:00 - 72:30 Today's cognitive labor may be automated.  What about the new cognitive labor that   gets created? Both of those things have  to be thought of, which is the shifting…  That's why I make this distinction, at least  in my head: Don't conflate knowledge worker   with knowledge work. The knowledge work  of today could probably be automated.
• 72:30 - 73:00 Who said my life's goal is to triage my  email? Let an AI agent triage my email.  But after having triaged my email, give me a  higher-level cognitive labor task of, "Hey,   these are the three drafts I really want you  to review." That's a different abstraction.  But will AI ever get to the second thing? It may, but as soon as it gets to that   second thing, there will be a third thing.  Why are we thinking that somehow, when we
• 73:00 - 73:30 have dealt with tools that have changed what  cognitive labor is in history, why are we   worried that all cognitive labor will go away? I'm sure you've heard these examples before, but   the idea that horses can still be good for certain  things, there are certain terrains you can't take   a car on. But the idea that you're going to see  horses around the street, they’re going to employ   millions of horses, it’s just not happening. And then the idea is, could a similar   thing happen with humans? But in one very narrow dimension?
• 73:30 - 74:00 It's only 200 years of history of humans where  we have valued some narrow sort of things called   "cognitive labor" as we understand it. Let's take something like chemistry. If   this thing, quantum plus AI really helped us  do a lot of novel material science and so on,   that's fantastic to have novel material science  being done by it. Does that take away from   all the other things that humans can do? Why can't we exist in a world where there
• 74:00 - 74:30 are powerful cognitive machines, knowing that  our cognitive agency has not been taken away?  I'll ask this question, not about you, but in a  different scenario, so maybe you can answer it   without embarrassment. Suppose on the Microsoft  board, could you ever see adding an AI to the   board? Could it ever have the judgment, context,  and holistic understanding to be a useful advisor?
• 74:30 - 75:00 It's a great example. One of the things we added  was a facilitator agent in Teams. The goal there,   it's in the early stages, is can that facilitator  agent use long-term memory, not just on the   context of the meeting, but with the context  of projects I'm working on, and the team,   and what have you, be a great facilitator? I would love it even in a board meeting,   where it's easy to get distracted. After all,  board members come once a quarter, and they're
• 75:00 - 75:30 trying to digest what is happening with a complex  company like Microsoft. A facilitator agent that   actually helped human beings all stay on topic and  focus on the issues that matter, that's fantastic.  That's kind of literally having, to your point  about even going back to your previous question,   having something that has infinite memory  that can even help us. You know, after all,
• 75:30 - 76:00 what is that Herbert Simon thing? We are all  bounded rationality. So if the bounded rationality   of humans can actually be dealt with because there  is a cognitive amplifier outside, that's great.  Speaking of materials and chemistry stuff,  I think you said recently that you want the   next 250 years of progress in those fields  to happen in the next 25 years. Now, when I   think about what's going to be possible in the  next 250 years, I'm thinking like space travel,
• 76:00 - 76:30 and space elevators, and immortality, and  curing all diseases. Next 25 years, you think?  One of the reasons why I brought that up was, I  love that thing of, the industrial revolution was   the 250 years. We have to take this entire change  from a carbon-based system to something different.  That means you have to fundamentally reinvent  all of what has happened with chemistry over   the last 250 years. That's where I hope we have  this quantum computer, this quantum computer
• 76:30 - 77:00 helps us get to new materials, and then we can  fabricate those new materials that help us with   all of the challenges we have on this planet.  And then I'm all for interplanetary travel.  Amazing. Satya, thank you so much for your time. Thank you so much. It's wonderful. Thanks.  Great, thank you.