Diane Greene: Building the Elastic Sky at VMware
By: Terah Lyons
To call Diane Greene an “industry veteran” is an overwhelming understatement. Diane is a seasoned technology entrepreneur and executive, who has had a hand in developing and bringing to market some truly revolutionary companies and technologies. She was most recently the CEO of Google Cloud and a former director of Alphabet, Google’s parent company. Before that, she co-founded and helmed three successful technology companies: In 1998 she co-founded VMware, which went public at a $19 billion first-day closing valuation. She also co-founded and was CEO of VXtreme, a low-bandwidth streaming video company bought by Microsoft; and Bebop, a software-as-a-service (SaaS) startup acquired by Google. She has worked at Sybase, Tandem, and SGI as a software engineer, and she was a naval architect prior to transitioning into the tech industry. Today, she acts as an advisor to a host of organizations: she sits as Chair of the MIT Corporation and serves on several other boards, including Wix.com, Stripe, and AP Moller Maersk.
We were lucky to sit down with Diane to draw on her career-spanning insight for the Zetta AI Founder’s Playbook. This conversation summary is part one of two, focused on her experience leading VMware. The second installment, soon to be published, is about her time as the founding CEO of Google Cloud.
VMware was the first cloud computing and virtualization company — a category-definer for so many reasons. Diane walks us through the unique challenges facing a company presenting a new technology to a nonexistent market, the challenges they faced from incumbents when they got it all up and running, and customer acquisition strategies from the earliest days of what grew to be one of Silicon Valley’s most successful companies.
And, importantly, she explains how close we all came to calling the “cloud” the “elastic sky,” instead. (We sort of wish that had stuck — right along with the namesake company band started by early VMware employees. I won’t not encourage you to search for their old gig videos on YouTube…)
Without further ado: Diane Greene.
Terah Lyons: The process of commercializing an invention is extremely relevant to our AI founders. VMware is a great story of a new technology that turned into a really successful company. The server virtualization technology that you and Mendel [Rosenblum] introduced and you brought to the market revolutionized so many aspects of the technology industry as we know it today — even though it may not have been an obvious business proposition at the time. It really made what we know today as the Cloud possible.
There are so many challenge areas associated with navigating the process of first building a hammer, so to speak — and then finding the nails. How did you go about the process of identifying and building a market for the technology?
Diane Greene: The original technology was similar in concept to what IBM had done in the '50s, and early '60s, where they were using virtualization to do timesharing, because machines were so expensive back then. There was some precedent, but it had totally died out. In fact, Intel's architecture didn't support virtualization. They were so sure that virtualization was the best approach to timesharing. As we looked at it, we thought of virtualization as a Swiss army knife.
There were so many valuable things that it did that when we decided to do the company, the two graduate students that joined at first were just thinking they were getting offered lucrative consulting contracts around virtualization because the Disco paper had rekindled interest in researching virtual machines. Mendel and I kind of said, "Well, you want to go off and do that, that's fine, but we think that there's something much bigger to be done here." They decided to come along.
Ed Bugnion and Scott [Devine] joined. Then, we brought in another founder, Ellen Wang, who I'd been in grad school with at Berkeley, and who had written a lot of the original Unix.
TL: Once you assembled the team, what was the first order of business?
DG: The original mission statement that I wrote and everyone signed off on said that this technology was going to, one day, run on every computer and become ubiquitous.
That was before we even had it working very well. It was very difficult technology to build on the x86 architecture — you could look back at how IBM had used it, you could bridge isolated operating systems and let you run multiple different operating systems at the same time. The technology let you consolidate workloads on a single computer by running multiple OSes at the same time. IBM wanted to do this because machines were precious. On the x86 it was the desire to run multiple OS environments. We were confident that x86 virtualization would be useful, but we didn’t know what the dominant use case would be.
The first thing we did: We initially self-funded just to bootstrap it and get the software working but then we wanted to do it on the server and we thought, “What's the lowest friction way?” It became a sort of mantra at VMware: What's the lowest friction way to bring the most value to someone? People tend to like to use things that fit into how they're already working, that solves a particular problem they have.
It became a sort of mantra at VMware: What's the lowest friction way to bring the most value to someone? People tend to like to use things that fit into how they're already working, that solves a particular problem they have.
TL: What did you hear from customers – or potential customers – when you first started outreach? Were you targeting an archetypal customer? Did you have to first go to technical users, because it was such a technical product?
DG: Back then, Linux was a big deal among developers, but it wasn't widespread yet. Linus Torvalds had not yet been on the cover of Time Magazine. We thought, Okay, all these Linux users are having to work in companies that are completely Microsoft shops. They were having to either reboot their computers or have two computers in order to read their email, because they wanted to do development on Linux. There was a product out there that would let you reboot and bring up another operating system, and that was the only option.
We thought: That is the perfect first user for us. They have a need: to be able to just switch windows instead of having to reboot or go to another computer. They're technical so that they will understand this. We initially planned to give the VMware desktop product away to the Linux users for free but it soon became clear that they would pay money for it so we didn’t need to do that. We didn’t have to do a heavy explanation and sell. We decided to make it a desktop product, which originally assumed you had Linux on your machine, not Windows. It was hosted with Linux, used all the Linux device drivers and it would let you run Windows in a virtual machine.
That was a giant hit.
TL: What was the big unlock for go-to-market? Was there a specific moment/event? Who, specifically, ended up being your first customers and what was their reaction to the product?
DG: We launched the product at DEMO, Chris Shipley's conference. I remember: We brought up a Windows VM and ran an app that we knew would kill the operating system, producing the Windows blue screen of death. Then, we just switched to another window–and of course, the machine was still running because it was all virtual machines.
We got a standing ovation.
It was fun. But I have to tell you, before we got into that conference, we were having a lot of problems convincing non-technical people [of its value]. We tried to get a marketing person to help us in the demo, and they said, "Oh, you won't even get a table at the conference. You won't get a stage appearance." We said, "Okay." We fired that person and did the marketing and conference application ourselves, and we got in. I also remember trying to get a public relations firm, and they were not interested. This was during the .com boom and they all thought we were a dinosaur software company, one very well known firm told us that. The concept was too technical to sell.
TL: What about fundraising? Who did you go to first?
DG: For fundraising, interestingly enough, we decided to go to people that were super technical. We talked to a few VCs and they didn't understand. They wanted to change what we were doing. Even though they were better at building companies, they didn't understand what we were trying to do, and it wasn't going to work with them. We were disappointed but it just wasn't going to work.
Then, we called up Andy Bechtolsheim, John Hennessy — the former president of Stanford, who was a computer scientist and co-founded MIPS Computer Systems — David Cheriton, and another computer science professor. It was a half an hour phone conversation, like, "Oh, God, this is brilliant." I remember Andy said, "How much do you need?” He just gave us a check. All of the founders put in money as well, I was able to fund a significant amount out of my previous startup, which Microsoft had bought.
It was a half an hour phone conversation, like, "Oh, God, this is brilliant." I remember Andy said, "How much do you need?” He just gave us a check.
TL: Speaking of Microsoft: Were you thinking about competition at that early point? How were you interacting with the existing hardware and software providers?
DG: We knew we were going to do a server product from day one. We were trying to build up that market while we were building the desktop product. I spoke to the CTO of Hewlett Packard. I spoke to a lot of enterprise people. They all thought we were a solution in search of a problem.
In fact, before we launched, well, we knew we wanted to be able to package Windows in a virtual machine, which would make us an OEM–an Original Equipment Manufacturer–of Windows, but we weren't sending it out on hardware. We were sending it packaged in a virtual machine. Fortunately, at that time, Macs were running on power PCs. There was a company called Connectix that built Mac utilities, and they had an emulation product that went from x86 to PowerPC so you could run the x86 apps on the Mac. We were able to tell Microsoft, "You let Connectix package Windows in their virtual machines, you need to let us do the same thing."
As an aside, we always knew if someone was going to challenge us, it was probably going to be them because they had this sort of binary translation technology. It's not trivial technology.
We went up, and we met with the head of research at Microsoft and several engineers. They more or less made fun of us. They said, "This is incredible technology. You guys are so smart but you're not going to be able to build a market for this. Maybe someone would use it to debug device drivers…we could maybe partner with you on that."
But it served a purpose because I wanted the OEM relationship. I was glad they were not threatened by it. We became an OEM. As the CEO of the company, I got invited to Microsoft's OEM conferences and went to them because all the hardware vendors that I wanted to sell to were at those conferences.
We got going on desktop with the Linux developers. When we launched a demo, 75,000 people downloaded our beta overnight — which was a big number for that time.
TL: Can you say more about your initial go-to-market strategy? How did you develop traction after those initial launch successes? How did you keep your customers coming back?
DG: Again and again at VMware, we would search for the people that would understand the technology, often former IBM VM users, older technical people. We’d also look for system administrators and people that needed to run multiple operating systems for backwards compatibility or system administration, and ask, “Well, what makes sense for us?”
We knew from the beginning that we were building the server product. We wanted to test that product as much as possible to make it rock solid because no one was going to run their server on it if it had any failures, any bugs.
We wanted to get tons of people using the desktop product. We said, “Look, here it is. It's free. If you're using it for commercial reasons, please pay us but if you're using it as a hobbyist (and most Linux users were) it's free. Every 30 days, you have to renew your license and you can pay us if you want." That worked incredibly well and gave us viral adoption. The approach was being used at the time, it was called shareware and not viewed as a way of making money. Contrast us at that time with Connectix — you couldn't use printing, or other features — they limited the functionality in their free trial version.
That worked incredibly well and gave us viral adoption. The approach was being used at the time, it was called shareware and not viewed as a way of making money.
We wanted to get as many users as possible banging on it — and we did something that people thought was not the smart thing to do but it worked out incredibly well. That was the biggest lesson over and over again: learn the best practices out there, get the best thinking, think from first principles as well; put all that information in your own context and see if it makes sense. Where it doesn't make sense, adapt your strategy to what makes sense for your company.
We got widespread usage of the desktop product. It was hosted, so people were running either Windows or Linux. We launched that hosted product for server usage and charged a bit more but were working towards a native, raw hardware version; native on the hardware was needed for a server because it would facilitate better, and what we regarded as the required, performance. The server was our end game.
TL: How did you break into the hardware vendors? Were those different target customers, or the same ones? Did you use a different go-to-market approach?
DG: We studied Sun Microsystems and the way they got adoption of Solaris, and the SUN workstation. The system administrators loved that SUN workstation.
Then, when they brought out a server, all those system administrators told the engineering departments to buy Solaris servers because they wanted to administer them. We did the same thing. We made sure our virtualization product got in the hands of system administrators. We'd put in functionality that was really useful for the sys admin — for example, we had a checkpoint functionality. You could be installing a new operating system and get halfway through and have something go wrong, instead of having to go back to the beginning, you could have checkpointed it and then keep going from there.
Well, that functionality was a game changer for a system administrator. Then, when we thought they would all go for the server product. They did go for the hosted product but it was more difficult to get them to adopt the native hardware version, ESX. We ended up launching it twice since no one noticed the first time.
A funny story: We called it Elastic Sky. The word Cloud had not been formed yet. The “ES” in our ESX Server stood for “Elastic Sky.” The full story is that we hired a hipster SF marketing firm to help us name the desktop and server product. That exercise produced Elastic Sky and Ground Storm and the suggestion that our company colors be brown. This didn’t go over to well with the engineers, lots of grumbling and the suggestion that they would not check those names in. So in an act of desperation, we named the two new products ESX and GSX (elastic sky and ground storm with an X to make it seem appropriate for technology). That resulted in the company band naming themselves Elastic Sky. We had a lot of fun at VMware and that band played at every and the many company parties, you can listen to them on YouTube!
Since nobody was buying what we regarded as our superior product, ESX, we knew we had to go to market with the hardware vendors. We had originally guessed that Edge computing would be the first use, and the product was designed for that edge computing purpose, with Akami in mind. After the dot-com bust Akami lost interest in pushing edge computing with ESX and we turned elsewhere.
Since nobody was buying what we regarded as our superior product, ESX, we knew we had to go to market with the hardware vendors.
TL: Where did you start that process?
DG: About that time, an IBM fellow contacted us, Rich Oehler. He had noticed that IBM had these big Unix servers that they couldn't sell, because you could only run one operating system and most apps on Windows and Linux had interference problems if they all ran in the same operating system. There would be resource leakage and security problems and so forth.
He said, "I've been watching you guys. Could you bring this to IBM's Linux servers?" We're like, "Yeah. The product already works. Nobody seems to want it." He said, "Well, you know what you need at IBM? You need an internal evangelist because it's just too hard to get through from the outside." In fact, IBM's hardware division had never resold software before.
So Rich introduced us to Bob Stevens. He was super smart and creative and he immediately understood what we did and set out to bring it to IBM's Unix server division. We navigated the organization with Bob and finally got IBM to change their rules and allow for software to be resold with their servers.
But then we ran into a wall.
We were in the final throes of signing the reseller contract. We were so excited, and then the IBM lawyers said no. We used the GPL, open source license for the Linux boot of our system. It wasn't VMware but to boot VMware, we required an open source Linux kernel, and IBM lawyers said, "We can't resell that because the GPL license might contaminate every bit of software IBM sells. You could destroy the company." We were like: Oh my God, we're never going to get this thing to market.
Then, I had an idea: Maybe we didn’t have to sell it directly. So I asked, "Could you resell it through your partners?" They [said], "Perfect." In perhaps our luckiest break, they invited me to keynote at the IBM World Premiere Partners Event. They let us train their partners on how to use VMware and what it could do. In many ways, that launched the company.
TL: Selling through partners does not seem like the easiest path…
DG: It is hard to be resold through someone else. You have to offer a huge advantage. In our case, the server computers were big, expensive pieces of equipment and once you sold them, you had the sell, a big SAN (Storage Area Network) storage array and interconnects. It made for a big (dollarwise) deal for a sales guy and we were just a fraction of the total amount.
It is hard to be resold through someone else. You have to offer a huge advantage.
Adding us to the deal didn't make it harder for them to close since the amount didn’t go up much. It made it easier for them to sell a bigger and more expensive server with all the other needed hardware and storage. It was almost no cost to them — but it was a big sale for us. We were a sales accelerator, as opposed to added friction.
TL: Brilliant. How long roughly did it take you to cultivate that relationship with IBM? Did you use that partner program approach as a template for other customers?
DG: The IBM courtship was over a year. In fact, I was trying to do the same thing at Dell, Compaq, and HP. I remember, we were so frustrated that I said, "I know what we're going to do, we're going to have the VMware premier hardware partner program."
We created some urgency and then put in requirements like having a support contact, a sales contact, and a quarterly business review. We wrote all these very professional terms that we barely understood into this partner program. They all fell in line, and we got them all to sign up successfully.
That was huge because now we had a comprehensive relationship with every hardware vendor. Once the partner program happened, all of a sudden, IBM became big in Unix servers. Linux servers were outselling Dell and HP. They could win every time because they had VMware. All of a sudden, VMware became the long pole in the tent. The whole thing just took off.
After the dot-com bust companies became much more frugal, and server consolidation became very important. VMware’s server product allowed them to replace what used to be ten servers with one server. We gave them somewhere between 3 to 1 and 10 to 1 server consolidation.
TL: So you figured out how to advantage your partners with your product, and made your product the invaluable feature. A game changer. How else did you take advantage of that moment, and your momentum?
DG: We would sell it as if you could take three servers and bring them to one, but actually, it was more. Then, we still wanted to accelerate the sales process, so we built some tools. For example, we built a tool that would go out on a company's network, measure the utilization of all their servers, and generate a report. Most servers were utilized at under 20%, meaning they had a lot of opportunity to consolidate servers with VMware. We built another tool that would suck up the full software installation from a physical machine and reinstall it in one of our virtual machines automatically. “P-to-V,”- — we called it “physical to virtual” tool. Those two tools accelerated our ability to sell the products, and get them installed and working. You can't just sell it, you need the customer to use it and love it and want to buy more and that accelerates the whole scaling process. That was good.
You can't just sell it, you need the customer to use it and love it and want to buy more and that accelerates the whole scaling process.
Those kinds of things helped us even when we introduced new products. We invented and built an invaluable product that we called VMotion; it allowed you to take a virtual machine while it was running on one physical machine and move it to another physical machine with no interruption in service. What that meant was, if you had to upgrade the hardware on a machine, you could migrate everything off with no service interruption, so you could do it in the middle of the day. You didn't have to do it at two in the morning or something. This capability is now standard for all virtualization, actually, maybe AWS still claims that they don’t do migration.
At first we couldn't get anyone to use VMotion. We decided to try bundling it with the ESX server product and then get professional services to teach the customer how to use it and what it was for. Then, it became super high value and nobody would think of buying virtualization without that capability. Whenever you introduce something that is a completely different way of doing things, it's hard to get adoption in the beginning.
We kept learning how to present our products as if they were just like a physical machine. We didn't want to advertise how different it was because the antibodies would go up.
Installing, using, and administering the product needed to be easy. Slowly, but surely, the employees that pushed virtualization within a company became heroes, because they were saving time and money. Then, we would celebrate them. We would help them. We would even give them slideshows to use to show their management about what they had done. Then, all of a sudden, all of the engineers in the company wanted to be able to use virtualization. There's so many psychological elements to getting people to accept something new.
TL: What a journey. You identified your first customers: technical users. You got widespread, viral adoption by making the product free and easily accessible for hobbyists. When you decided to make a move into hardware, you targeted system administrators, and solved real pain points for these organizational buyers and decision makers. The partner program with the hardware providers really accelerated things. And you packaged all of this new technology — which no one had ever used before — in ways that presented it as analogous to physical machine technologies already in existence, until it was so invaluable that no one could imagine working without it.
So, you achieved go-to-market effectively, and commercial success. What happened when the massive incumbents woke up and tried to take the ball away from you — when they realized that this was actually something seriously useful? How did you protect your intellectual property, and the independence of the company?
DG: Well, there are a few stories about that, because when we were first starting out, there were a few changes that Intel could make that would have helped our performance. We said, "We cannot require any changes to Microsoft Windows, or to Intel architecture, because if we require those changes, they will control our fate." We don't have a company unless we can do it independently. That was our requirement. However, whatever we could get Intel to do, would make our product run better.
We started talking to Intel. First, they couldn't believe we had full X86 virtualization working. We proved we had it working and we said, “We want to work with you. We want to partner with you.” They said, "Well, just a minute," and the next thing we knew, we got a term sheet from them where they wanted to invest. But they wanted to have rights to our intellectual property, and they wanted to have the right of first refusal if we ever sold the company. We would not accept any of that from Intel, we wouldn’t have had a company had we accepted their terms.
Then, there was Connectix. We always knew they would be a challenger, though their product was not nearly as good as ours. Then, Microsoft bought them to compete with us. I actually believe that that was the best thing that could have happened because it turned out their implementation wasn't as strong as ours. Yet that was now the Microsoft team they had to compete with us and they had to adapt that implementation and they somehow couldn't get a viable product for years.
It was amazing how long it took them. We had patent protection but against Microsoft, that was a little dicey. Easier to compete against companies your own size. We survived all that because had built an amazing engineering team that was moving fast, maintained superb quality, and kept inventing.
In fact, Microsoft came in and tried to buy us at one point when we were about a hundred people. I remember them saying to us, "We're not very impressed if it takes a hundred people to build this product. We think we could do it with about twenty." They were totally underestimating us and what it took. We told them to give it a shot. As it turned out, they couldn't do it for a long time.
TL: What was the VMware advantage, beyond the sheer amount of engineering talent required to create the product?
DG: VMware really had unbelievably good products, well-engineered, very stable, they didn't have many or severe bugs, they didn’t crash. We were devoted to our customers. We had a user conference every year with our customers where we would set up a whole data center at Moscone Center [in San Francisco]. We would run these “raves” where our customers would be in this datacenter all night long, building virtual machines and doing interesting challenges with them.
We would bring all our engineers to the conference to interact with the customers. It was just culturally a group that was really enjoying having the kind of impact we were having. Our engineers would fight when we wanted to release a product if they didn't think it was ready.
TL: Datacenter raves. What a note to end on. That ruthless focus on engineering quality seems to have really paid off. Thanks so much for talking with us, Diane.