'The world needs only five computers'

But Sun Microsystems Chief Technology Officer Greg Papadopoulos thinks the idea will pan out eventually.

"The world needs only five computers," Papadopoulos said on his blog. He then listed seven--Google, eBay, Amazon.com, Microsoft, Yahoo, Salesforce.com, and what he called the Great Computer of China--but let's not split hairs. He was trying to make the point that "there will be, more or less, five hyperscale, pan-global broadband computing services giants."

Sun caters to pan-global hyperscale customers, so it's not surprising that its top technologist sees the world through data center-colored glasses. But the Santa Clara, Calif.-based company's vision has been right on target more than once--notably in forecasting the rise of the Internet.

Papadopoulos detailed his vision in an interview recently with CNET News.com's Stephen Shankland.

Q: For starters, describe your thesis. You think there are going to be five or six computers at some point in the future, and that's all the world needs?
Papadopoulos: I think what we're going to be seeing is a huge consolidation around scale. Scale matters, and efficiency of scale matters. You've got to get beyond some critical size, and once you're there you'll start absorbing other, less-efficient people into your system. It's a very large-scale distributed system.

What's your definition of a computer? Most people think of a computer as a PC or maybe a server, but I think few people think of a computer as what eBay runs on--8,000 computers or whatever.
Papadopoulos: What I'm saying is a computer--Google runs a computer. It happens to have hundreds of thousands of processors in it, and millions of disk drives, but it's a computer. The important distinction is there is a point of control that determines what software is going to run, and then the systems work collectively to provide some service.

So under this definition, you think that there are only going to be four or five or six of these?
Papadopoulos: I think it's going to look more like the five or six, like there are five or six multinational energy companies. There are hundreds of multinational energy companies, but there are five or six that really get to the scale of being able to efficiently do the whole thing. If you have someone who is 10 times the scale of someone else, they're able to spend more money on R&D and architecture. They're able to invest more in engineering to get it to be more efficient. That inevitably tips towards larger systems. So whether it's five or six or 12 is not the issue; it's the consolidation around the very large scale.

What happens to everybody else? (Sun CEO) Jonathan Schwartz used the corner dentist office as an example--these people, they end up being a tenant or a client who taps into one of these large systems. So if you're one of the survivors, then you end up hosting software for some huge constellation of customers?
Papadopoulos: Exactly. It's called software as a service. It really is the running of what we think of as IT through the network. You don't buy software, you buy the consequence of the software. That starts with the small and medium enterprises. eBay, in my mind, is the leading example of small businesses being absorbed by services. Anybody who clicks their store on eBay is in fact consuming a service. They are contributing to a larger-scale eBay rather than them buying some server and sticking it on their desk.

If there's a tremendous diversity of small clients running on these very large infrastructures, then these very large infrastructures are going to have to accommodate that variety. When I think of eBay today, I think of auctions and direct sales. When I think of Amazon, you think of e-commerce. But when I think of all the businesses out there in the world that use computers today, they use computers for everything under the sun. Does that mean that Amazon is going to become some generic computing infrastructure that gets tailored for all kinds of different things? That seems like a very different incarnation of eBay or Amazon than what they have today.
Papadopoulos: Amazon, I think, sees a bit of this direction. Their Elastic Computing Cloud is at least an experiment in the area. You'll get people specializing in particular kinds of businesses and caring for those businesses. If you are a small business today, or you're a start-up, mostly you'll get services from service providers. You'll get your e-mail somewhere, you'll get your sales and customer relationship service from somewhere else, you'll do a storefront somewhere else. There will be people who will consolidate that for you in the future.

So you envision the big guys like Amazon becoming generic infrastructure, and then you have a whole bunch of middlemen packaging that up, and tailoring it to the needs of the actual clients?
Papadopoulos: Yeah, initially. I don't know whether the intermediaries will survive or they ultimately get absorbed. For example, you see what American Express is doing with small businesses, and you could certainly imagine them filling out their repertoire of things that they offer. I just wonder whether the intermediaries get in front of them, and they broker those folks as wholesale services, or if ultimately (the intermediaries) get absorbed into them. I would doubt that the intermediaries will ultimately get absorbed simply because knowing the customer and controlling the customer is so important.

When will this drastic consolidation happen?
Papadopoulos: It's happening now. Perhaps a different question to ask is, "When does it become most of computing?" The total worldwide spending on information and communications technology is somewhere in the $2.5 trillion to $3 trillion range, and most of that is not spent on computing stuff. It's spent on people and software and a lot of services. At these very large computers, a (much larger fraction) of their spending is on equipment because they're operating in a much more efficient way...This new world is brutally efficient, and most of the dollars that get invested are going into actually getting work done instead of paying for organizations or people. In terms of (spending on) computing stuff, we'll see the crossover before the end of this decade.

A lot of people think that Sun is long on vision and short on execution. How is it that you guys are going to be supplying all this infrastructure when you've been struggling for the last few years?
Papadopoulos: At the core of what we do, we believe in investing in R&D and the return you get off of innovation. A lot of what we use the vision for is to guide that portfolio of investment. Our track record on reality meeting vision is really, really good. Our track record in being able to extract a lot of profits from that, or go in and monopolize a piece of the market because of that insight, ain't as good.

If you are inside Sun, you are getting a big earful that the R&D has to reshape to meet the reality--what I've been calling this brutal efficiency of that landscape. You can't sell soft products into a world that looks like that. It's: What's the performance per watt, per dollar, per rack unit? What's the productivity? What is its service level under load? It's much more about how one would think about approaching designing large-scale power plants that have to service the city vs. designing portable generators. It's that kind of holistic engineering that we're really trying to drive toward.

When you look at Sun and its competitors, are these brutal economies of scale also going to apply to the companies that supply these data centers in the sky with equipment? Do you think that also is going to be equally consolidated?
Papadopoulos: I think if you don't shape your engineering to meet the needs of this new class of end users, then basically what you will be doing is irrelevant and too expensive, and the end users will do it themselves. You have to be responsive to it.

But if the end users are the center of gravity of this new computing universe, why should they not be acquiring the computing equipment manufacturers and designers themselves?
Papadopoulos: If we look at other markets, every time that kind of extreme vertical integration has been tried, it fails. Basically once you own something from the engineering side, your choice gets really limited. If you look at analogies like in the, I think, the oil business, there are lots of people who go and provide tooling and specialization, who supply all of the majors--the Schlumbergers of the world. They do really well. People like GE or Rolls Royce who are providing aircraft engines, or Boeing, who are providing airframes into an equally consolidated-looking airline industry. If you want to make money in air transportation now, you would much rather be Boeing than American Airlines. I don't think that the model that Google pursues now is in the long term a sustainable one, particularly when you have technology underneath that moves so rapidly.

You're talking about Google rolling their own operating systems, running their own networks--a whole lot of customized computing technology.
Papadopoulos: Yeah, they get boards manufactured. I think they do that now simply because the traditional suppliers in the IT market are way off in terms of efficiency.

If you assume these six remaining companies have slightly different specialties, that raises the prospect that (customers) are going to be buying that from a monopoly. Monopolies are famous for their ability to raise prices at will. Is that a problem? Do you think this situation fosters monopolies?
Papadopoulos: I don't know whether we're going to go repeat the same mistakes that we made in IT, which is allowing people to come in and establish control points inside our IT architectures. It is in the power of all of us not to do that again. (The reason) why are we so vocal and progressive around open-source and communities is that it's really all about getting the switching cost down, so we don't end up back in that same situation where the whole economies of IT get distorted because of these very high switching costs.

I note with some interest that Salesforce.com recently announced that generic programming infrastructure?
Papadopoulos: AppExchange.

...where you not only can run their own prebuilt services, you can write your own services. But you have to write in their special proprietary, custom language. As soon as I heard that, I thought of switching costs (the expense of moving to another technology foundation) and proprietary lock-in.
Papadopoulos: I think the service provider would want to provide barriers to people switching. The flip side is competition erodes that. The competitors to Salesforce.com will ensure that they will be able to extract whatever data they need to offer people to help them move from Salesforce.com to their service.