The Pioneer of the Cloud

What does it take to ask one question — *how do you free software from hardware?* — and end up reshaping the entire computing industry? In this episode of Decoded, Dr. Jurgi Camblong speaks with Edouard Bugnion, co-founder of VMware and Vice President for Innovation and Impact at EPFL, the Swiss Federal Institute of Technology in Lausanne. VMware didn't just build a product. It built the foundation on which the modern cloud was built — and was acquired by Broadcom in 2023 for $69 billion. From a childhood in Neuchâtel to a PhD at Stanford, from research lab to billion-dollar company, and back to Switzerland as a professor and public advisor, Edouard has spent his career at the precise intersection of deep science and real-world impact. You'll hear how a single idea about separating software from hardware gave birth to an industry, what the dot-com crash taught him about building through uncertainty, the three factors he believes determine whether a company succeeds or fails, why the energy cost of AI is a problem that can't be wished away, and what it actually means to be responsible when innovation moves faster than society can keep up. This episode is about what happens when a scientist becomes a builder, and why the most important question in technology right now may not be *what can AI do* — but *what should it do.*

Transcript

SPEAKER_01 0:01

Welcome to Decoded, the podcast where we explore how data technology and human insights are reshaping patient care worldwide. I'm Jorgie Kamlong, biologist, dreamer, and founder at Sophia Genetics. In each episode, we decode ideas with pioneers who have helped shape the way industries think, build, and move forward. Today I'm excited to welcome one of the architects of modern cloud and virtualization infrastructure. Before the cloud, before virtualization went mainstream, Edouard Bugnon was working on one idea. How do you free software from hardware? He co-funded VMware, acquired by Broadcom in November 2023, for about $69 billion, as the company was generating over $13 billion in revenues. Since then, in 2012, he is back in Switzerland as a professor of the EPFL, the Swiss Federal Institute of Technology in Lausanne, where he also serves as vice president for innovation and impact. In this episode, we will explore how AD evolved from technologist to entrepreneur, from professor to public advisor, bridging deep science with real-world impact. This is a conversation about making machines work as one, about turning deep tech into global impact, and about what AI means when innovation moves faster than society can keep up. Ed, welcome. Thank you, Jorgi. Good to be with you. And great pleasure being in the beautiful uh EPFL Scientific Park. So, Edward, you are being uh, as I mentioned, a very successful entrepreneur, but actually you grew up in a little town called Neuchatel in Switzerland, surrounded by mountains and a beautiful lake, and end up doing a PhD at the very prestigious University of Stanford. How we moved from Neuchatel to Stanford?

SPEAKER_00 2:03

That's a good question. So uh relatively easily I grew up in Neuchatel, then Geneva. Then I actually studied at ETH Zurich, computer science at the time. And I think I was fortunate to be exposed to research doing my undergraduate degree at ETH. And because I had this research exposure, I had relationships with professors and then got interested in research and very simply applied for a PhD program at various places in the United States. And uh Stanford was kind enough to take me, and uh that was it.

SPEAKER_01 2:34

And uh internet was not used then. So basically, was it easy for you to create these relationships? Had you ever been in the US before?

SPEAKER_00 2:44

Yeah, well, thank you for mentioning that I grew up in the Middle Ages. But yes, of course, at the time applications were paper-based. So you submit your resume, you had the full application file on paper. Uh the letter writers would send the letters directly on separate envelopes, and then you got an answer back on paper. It was uh the early days, some people had email, but and I had an email, but email was not used for actual formal exchange of attachments. That came just a handful of years afterwards.

SPEAKER_01 3:11

And uh, well, you mentioned your research at ETH, but why Stanford? Was there something that was basically triggering your interest? Is it because you would see that the tech industry would be bubbling there?

SPEAKER_00 3:23

So the tech industry was already bubbling. Stanford was already, I mean, Silicon Valley existed. Stanford, of course, is in the heart of Silicon Valley. So that was part of the appeal. And I moved to Stanford in 1994, and 1994 was uh the beginning of the web. Netscape had just been created, it would go IPO the following year. So I was actually very fortunate to arrive at Stanford at the time when the commercial web happened. And my one of my fun memories of that time is I remember in the lab seeing the first ad. It was a poster with a URL on the poster. So the first commercial ad for a website with a URL, it was Gatorade. It was a big thing. And then this is perhaps when we recognized collectively that the internet was going to be something significant.

SPEAKER_01 4:11

Amazing. So at the time when uh you went to Stanford, you went for your PhD, correct? Yes. You were working on operating systems and virtualization. Is this how the cloud was born?

SPEAKER_00 4:23

Well, so let me explain. So if you think about a traditional computer, you have hardware, an operating system, and applications. And there is this intuitive understanding that the operating system is actually bolted onto the hardware. And virtualization, it's another approach. It's another approach in which you add another layer of software between the hardware and the operating system so that you free the operating system and their applications from the underlying hardware. That's what people call virtual machines. Virtual machines were a really big thing in the 1960s, 1970s, and then completely went out of fashion. And uh our research program at Stanford ended up identifying the use of virtual machines for a solution to a new problem that we were faced, which was to build very large computers. And we had tried other solutions, and virtual machines turned out to be the better solution to solve that one problem. And with that, we had some experience with the return of virtualization. And it turns out that virtualization had a much bigger use case than we had identified originally academically.

SPEAKER_01 5:32

And if I understand correctly, before virtualization, one computer, one disk would be only used for one individual, right? And with virtualization, what you're doing is demultiplying the access to these disks for multiple users, is that correct?

SPEAKER_00 5:52

Yeah, it's not just disks, it's the entire machine. So the the notion that users, whether they're individuals or IT operators, they think about a computer as a unit of management. This is how you configure applications, access control, resources, file systems. But that unit of operation does not need to be a physical computer. It could equally be a virtual machine if as long as the virtual machine is operates exactly as if it were to be a physical computer. And so virtualization allows server consolidation, it allows hardware independence, it allows mobility of applications. One of the most fascinating examples of the use of virtualization is that you can actually move a computer as it is running from one machine to another machine. You can't do that with a physical computer. But if that computing and that workload is running inside a virtual machine, you can actually run it. And as it is moving from one physical box to another physical box, it can keep running without any interruption.

SPEAKER_01 6:52

Yeah, well, definitively uh no cloud without virtualization. Now that uh uh you explained me that uh it's something super fundamental and super powerful. So at the time, as I said, you were doing your PhD in that topic, and at some point you decided to stop everything and start from scratch to be involved in a company that was called VMware. What were the problem or the opportunity you were trying to solve?

SPEAKER_00 7:17

So the work we had done at Stanford generated a signal from the industry. We got industry interest. This is when we realized that virtualization could be useful for more than that one academic application. And we quickly turned to identifying that the real opportunity was on x86 machines, desktop, servers. At the time, there were multiple incompatible versions of Windows that were coexisting very painfully in environments. Windows 95, 98 would come later, Windows NT, and then of course Linux was also becoming a more visible operating system in the workplace. And so this was an opportunity to have different types of operating system coexist. And we started VMware in 1998 with the premise of decoupling operating systems from the underlying hardware.

SPEAKER_01 8:06

And who are your main uh partners and your main uh targets? When I refer to targets, I refer to customers.

SPEAKER_00 8:14

Well, the first product was aimed at IT administrators, people that basically needed to install computers for a living, and they needed a development environment and a test environment. But very quickly we realized that these customers were using VMware not just for test purposes, but for actually production and consolidation of applications. And the cloud is actually nothing more than the deployment of very large farms of virtual machines inside data centers. And VMware was and is still today the major provider of private cloud solutions for enterprise data centers.

SPEAKER_01 8:49

So given you mentioned that the Silicon Valley started before you would uh go uh to the to Sanford for your PhD, if I'm not mistaken. HP was funded as well in the Silicon Valley. So was HP one of your partners, or you you had uh, I mean, where those, I guess those type of companies were probably important, right? To basically um get traction with your technology.

SPEAKER_00 9:11

Yeah, so HP, Dell, IBM, Compaq. Compaq had been purchased by HP. Those were all key partners. There were OEM partners uh reselling VMware to their customers because that would provide us a scalable route to market. The earliest and the most interesting partner of them all, of course, was IBM, because IBM had 40 years of experience with virtualization and virtual machines. And working with them was fascinating because we not only worked with the people in charge of the x86 business who actually knew nothing about virtualization, but at times we actually got to interact with the people from the mainframe divisions who had been working on the same problem for the last 40 years.

SPEAKER_01 9:49

So big names, very hot topic. Uh, we are at the end of the 90s, just at the time that there is as well the dot-com bubble. How did that crisis impact uh you personally, uh VMware, and uh, how have you been able to navigate through this crisis as an executive team?

SPEAKER_00 10:08

So it's probably difficult for people who did not live it to recognize how intense the crash was. I mean, driving on the highway went from being painful because of the traffic jams to effectively being a perfectly easy experience in a matter of weeks. This is how brutal the crash was. For VMware, it was an opportunity because a lot of really, really good people were looking for jobs. And we were able to grow and our head count with very high quality people at the time. And our customers were largely enterprise customers unaffected by the crash. And they realized that they actually need to reduce their costs. And VMware provided an effective way to reduce operating costs, and so we were able to actually grow during the crash period when the rest of the industry was actually imploding and retreating.

SPEAKER_01 10:52

So basically, it was a kind of a catalyzer to your business.

SPEAKER_00 10:56

Yeah, this is how a chemist would probably describe it. Yeah.

SPEAKER_01 10:59

I'm a biochemist, so maybe yeah, that's my uh, you know, I don't know, my nature is coming back. So very successful company, as we said, then uh uh sold for $69 billion, generating over 13 billion revenue. I don't think you guys had raised so much money neither, isn't it?

SPEAKER_00 11:17

Well, it's a the the story of VMware, and we're not gonna go here in all the details, but it was a complicated one because it was sold first and then went was public again. The parent company went private and then it got sold again. I was there for only the first eight years of VMware. So uh by the time the company had grown, uh when I left the company had about 700 people. Uh by the time it was sold to Broadcom, it's 25,000 people. So the second phase of the growth uh had nothing to do with it. Uh, it has been a fascinating story to watch.

SPEAKER_01 11:47

And you remember uh how much you had raised in uh these uh early years? How much money?

SPEAKER_00 11:53

22 million total.

SPEAKER_01 11:54

Yeah, so I would say you you had been very efficient, right? Uh in uh basically using those proceeds to grow a significant business.

SPEAKER_00 12:02

Aaron Powell Yes. So I think we were very fortunate of having found immediately a very good product market fit, and we were also able to take advantage of sort of the new ways of distributing software over the web and over the internet.

SPEAKER_01 12:18

In these years, what has been the hardest? So you were uh initially the CTO, you came from an engineering background. Was the engineering the hardest or was it something else?

SPEAKER_00 12:33

No, I think the the learning curve as a co-founder, as I was always you know, a co-founder and a CTO. Uh that was sort of my primary role. But underneath that primary role of being a member of the team, being the technology spokesperson, I played very different roles. I think one of the most fascinating aspects of anybody involved in a startup is learning new jobs, new tasks, and also learning to let go of the things that made you special before. So my key role in the first nine to 12 months of VMware was actually to write software. I was expert in a very narrow piece of software, which is called dynamic binary translators. And so I wrote the dynamic binary translator for VMware. And that was very special. And it was equally important to actually let go of that piece of software as quickly as possible. Fortunate that you were able to hire somebody who could take over that piece of software because I had to actually take other tasks. It's impossible to grow as a founder if you don't learn how to let go, delegate, and effectively transfer responsibilities to others, even if those were tasks you were doing where you were effectively making a big difference.

SPEAKER_01 13:46

Yes. I remember you visiting us in our company in the first years of SOFIA and telling us that basically everyone had three jobs, right? The one of yesterday, the one of today, and the one of tomorrow. And that indeed you had to work with the one that were taking your job and you had to work with the one eventually that was providing you some new capacity to be into a new job. Yes, absolutely. You told me as well another thing that struck me, and I thought it was really nice and genuine, uh, which helped me at the time that actually product and technology was a must, but that in the end the only thing that really mattered was sales. And I found it very powerful from someone coming from a technology angle.

SPEAKER_00 14:28

Yeah, so the technology, it's of course a necessary condition. Without technology and without innovation, it's very difficult to create at least a technology-driven company. But that is a necessary, it's not a sufficient condition. You need to have the right product market fit, the right business model. What today everybody understands and recognizes as the business model generation, the canvas, the seven boxes, the lean startup, the pivoting, all those concepts are now relatively well understood globally. At the time, we were iterating and navigating through them without really understanding the formalism. We were lucky to have the right product market fit. We were, I think, very lucky to have hired uh very, very good product managers early on in the team. And then we build a Salesforce that was effective because if you sell to enterprise customers, large accounts, you actually cannot sell it directly. The product will not sell itself. You actually need a team that will actually work with the accounts to generate the demand and then uh make sure that we have the confidence of these enterprise customers.

SPEAKER_01 15:31

So we covered you being uh one of the co-founders and uh CTO. You mentioned already some of your uh basically findings and experiences. But in your public talks, you often refer to three success factors. The fact that as a company grows, you know, the roles are evolving. We we discussed a bit that. But as well, there might be other learnings you've made there, and that might be nice from you know startupers to hear from you so that uh they know how they could face when they are navigating in difficult times, which I would say right now it's easy maybe for some companies which are in AI, but some other companies maybe we may consider that we are a time of war time, a crisis time.

SPEAKER_00 16:15

Yeah, so I think the the composition of the team is extremely important. It's important to recognize who is and only to have within the founding team people that you view as peers. They may have complementary experiences and profiles, and that is often a plus, but they need to be viewed as as a team and as uh as peers because as the company grows, it's very important that you have a level of cohesiveness uh among the founding team, even if some of the members of the team are destined to actually remain individual contributors and engineers and not necessarily be active uh executives for the long run. And then the the other thing that is extremely important is timing. Sometimes technology is ahead of the market. Sometimes the product is ahead of the technology trends. And so there's a the product innovation that relies on the other technology components that are not quite there yet, and you need one or two generations in order to reduce cost, reduce weight, reduce, increase efficiency. So I've seen many entrepreneurial projects fail because their products were either not appropriately a good fit to the market or because the technology wasn't there. But the common element is that they were too early. It would have been a great idea three years afterwards, but you don't have three years to wait when you start something.

SPEAKER_01 17:32

Well, actually, uh, you know, uh in the prior episode, I was interviewing uh Bernard Liotto, founder of Business Object, and he said very similar things to the one you said regarding the team members as well as the timing of your product market fit. I know that sustainability is important for you. Eventually, uh, as a vice president of innovation and impact, maybe an impact, it might be something that is part of your responsibilities at uh DPFL. The vast deployment of AI requires massive natural resources. I think uh we know that, not only uh for silicon chips, some minerals, but as well water to cool down data centers. Are these viable solutions or what can or we should we do so that AI can be a viable solution?

SPEAKER_00 18:22

This is a real concern. At some point, we are already hitting the natural planetary limits in terms of the amount of electricity that we can naturally dedicate to AI. And it's not the AI that is running in the phone, it's the AI that is running in these mega data centers that are each and every one hundreds of megawatts in size. So we're now seeing a situation where these hyperscalar companies are building massive data centers. The location is chosen explicitly so that there is expandable energy resources and cooling capabilities, sometimes in locations very far remote from where the population lives, which may actually be a reasonable trade-off. It actually is a perfectly sensible decision. The concern I have is that exponential growth cannot go forever, and exponential growth of electricity consumption cannot go forever. And I think the computer industry has a very good track record of systematically improving the efficiency of computer systems generations after generation. That is done through optimization in hardware, optimization in software. But of course, these optimizations are also happening right now, in fact, at a very high pace for AI and AI workloads, but they're not sufficient to keep up with the exponential increase in the demand. And so the question is how this, whether we reach some kind of an equilibrium, whether we have peak electricity of AI at any point in the next few years, it's an interesting question. It's certainly one that everyone collectively is facing.

SPEAKER_01 19:57

And beyond the ecological considerations and as well as the uh natural extraction considerations at the society level, how do you see society will be impacted by AI if you look back and you see how society had been impacted in the past by internet and the cloud?

SPEAKER_00 20:19

Well, I'm not a social scientist, I'm not an economist. But it's very clear that AI is transforming not only industrial sectors, but the workforce itself. So if we look at the internet, the internet had a profound impact in creating new distribution channels, and of course, that disrupted other existing distribution channels, and they're well-known examples of the impact of the internet on bookstores and DVD rentals and other types of stores and business models that don't exist anymore. AI is completely different because AI is about transforming the nature of jobs, and it is doing so at an extremely fast pace. So the consequences are everybody needs to become proficient and comfortable with these new tools. The learning curve is actually not that easy because you need to have in a position where you can actually supervise and interpret the insights provided by AI and of course not trust them blindly. The agentic model is even more difficult to reason through, and the impact on the workforce are likely to be quite significant. What worries me the most, perhaps as an educator, uh, being a professor of engineering, is that it also is changing the way students learn. And students need to learn how to embrace AI, but also to learn themselves and to really learn. And learning how to learn is one of the key things that one actually learns in higher education. Having the tools for lifelong learning and developing these tools for lifelong learning. And that requires going deep into topics without AI at times. And so students are challenged with having. This bimodal life, where at the same time we tell them to use AI and we tell them not to use AI, depending on the use case. Of course, the students that are in university right now are perhaps the greatest generation alive because they will be able to have all of the AI knowledge at their fingertips as they enter the workforce, or they may be actually faced with the worst of economic situations. And that I don't know. The future will tell.

SPEAKER_01 22:23

Yeah, actually, this made me think about some statistics, at least in France. I don't know whether this applies to Sweden as well, where it looks like companies now tend to not recruit junior software developers, but rather people who have experience because they believe that with AI, maybe some of the job of the junior software developers is not anymore required that AI assisted, or I would say senior software developers assisted by AI or with AI can make it happen more rapidly and more efficiently. But in the meantime, back to your point on the learning, if that's the case, how the new generation of software level developers are going to learn if they are not given the opportunity to go deep and produce software by themselves?

SPEAKER_00 23:09

So the answer is actually there's only one answer is that junior software developers will become senior software developers with the expected master of the nuance and architectural understanding of computer systems much faster than the previous generation. Now, will there be as many software developers in the future as there are today? That is an open question. But I think the career arcs will be faster. But that is not only limited to software development. The career arcs will be much faster for all branches of science and engineering because we now have with AI the ability to sort of develop things and develop insights at a much, much faster pace. Which is why we're seeing this acceleration across all disciplines, scientific disciplines and industrial sectors, at an unprecedented rate with AI as the common substrate.

SPEAKER_01 24:02

So to end um this fascinating conversation, Edouard, it's always a great pleasure uh meeting you. Uh, I learn always new things, so I um I very much uh like this moment. Looking back at your journey, so you started as a researcher, then founded one of the giants in the world of technology, came back to an academic world where I would say your passion is very clear. How do you see the role of science and research for the world of tomorrow?

SPEAKER_00 24:36

So academia has, of course, a multifaceted mission of education, research, innovation. It also has to be a trusted partner in the key conversations, not being in charge of policy decision making, but being a trusted source that informs policy decision making. If we look at the situation we have right now in various countries around the world, there is a growing distrust in science, in science itself, the scientific method, scientific results, what is considered established scientific consensus can be challenged and rechallenged. And that, of course, is something that should concern all of us. One of the goals that we have as an academic community globally is to maintain this role of being a trusted advisor. Otherwise, lobbies will play a role, uh, special interests will play a role, and they should play a role. They have a role in the overall political and societal debate. But academia has a role as well, which is to effectively inform with the appropriate constraint of the certainty of the information that we convey to be a trusted partner in those conversations. That is probably one of the biggest concerns that I have in a world that is changing faster than ever, both in terms of technological change. We talked about AI, but of course in terms of climate change.

SPEAKER_01 26:09

Fantastic. Maybe next time we meet, uh we spend more time to discuss about philosophy and ethics. Edouard, thank you very much for joining me on Decodi today. What I love the most from this conversation is the idea that great technologies can positively contribute to the world. From co-founding VMworld to leading innovation and impact at EPFL, your journey has been about scaling computing beyond physical machines and understanding the impact that comes with scale, including in society. I really enjoyed this conversation and I hope you did as well. If you enjoyed this episode, make sure to subscribe to Decode It. We continue meeting leaders and builders shaping the future of healthcare, technology, and I and society. I'm Diori Camlong, thank you for listening and let's keep moving forward.