The Last Snowflake - 2011 Gartner IOM Summit Presentation

George Slessman, CEO for IO, gives a presentation entitled, 'The Last Snowflake' at the 2011 Gartner IT Infrastructure, Operations & Management Summit.

Transcript

George Slessman: I'm George Slessman. I'm the CEO and founder of IO Data Centers. What do I mean by the last snowflake? A lot of people look at the title of the presentation and they are curious of what I'm talking about. I put it into pictures. We all have data centers. We've all seen data centers. We've built a lot of data centers. In my short life of data center infrastructure, the physical infrastructure side of data centers have built 100 megawatts of IT load capable data centers. That's net of redundancy and actual IT capacity. To give you a sense of size, currently at IO, we manage approximately a million and a half square feet of data center capacity in two data centers, one in central New Jersey and one in Arizona.

The point of it is that data centers classically that have been designed have been snowflakes, if you think about it. Every single one of them is different. Every single one of them looks different. They all have different attributes, features, physicality, systems, subsystems, technology managing it, and ways of working with it. Everything about them is custom. There has literally been very, very little standardization that we see across the rest of the IT stack delivered in the data center. The conjecture here is that we believe, I believe, that in the next 24 months, the transformation we're seeing is that the last facilities-based data center will be constructed in the world.

Essentially, we're in this cycle where we're going from seeing data centers being constructed as buildings, and we're going to see them turn into a modular delivery footprint that fits into the IT stack. Essentially, the economics of this, the operational sustainability, the improvement in overall management of capital, the IT stack, your ability to deliver on a service level, and your ability to tier the solution so that you can actually match your data center requirements with the IT requirements are going to completely transform the way that you use data center capacity. The same way that EMC, storage area networks, and network attached storage have changed the way you thought about storage, we're seeing the exact same transformation occurring in the data center.

Why? Why are we seeing this happen in our view? Digitization. This is in the context of giving this presentation and ultimately preaching to the choir. You see this every day. This is why we're here. This show a year ago was 500 attendees. This year it's 1,000 attendees. Enterprise infrastructure, physical infrastructure, energy infrastructure, compute, storage, network are all growing at absolutely precipitous rates. The way that we look at this, at least the way that our organization looks at this, is to take a step back and really consider what's happening. We can look at this trend and then we can find a solution.

I'm going to attempt to walk you down the same garden path that we walked down internally that took us to where we are today with our modular solution and where we think the data center is going and what we're implementing ourselves. We're no longer building data centers traditionally. All of our data center build out now, we have a 20-megawatt build out in New Jersey right now that's just coming to completion. We have a 20-megawatt build out in Arizona that's coming to completion. They're all built out using our technology and in a modular footprint.

The first Industrial Revolution is where it started. We took every economic process on the earth, transportation, food, communications, all of these things that were done in a highly analog way and we mechanized them all. This was 125, 150 years ago. We went from analog processes. If we wanted to communicate, I walked across the street, I walked down the street, I came to your door and knocked on your door. If you weren't home, I went back. Cycle times were very long. Energy densities were very low.

I'm going to keep bringing up this concept of energy density. If you think about it on a transactional basis, how much energy was consumed to conduct that transaction? My desire to communicate with my neighbor is a transaction. It's a communication transaction. I can do that physically. We did pre-Industrial Revolution. We walked over. We talked to them. We got on a horse potentially. We rode in a wagon or whatever the particular outcome was. That's how we conducted that transaction.

This has been played out through every other economic process, like I said, energy delivery, transportation, banking, essentially trading. One of the largest financial services, one of the largest consumers of data center and infrastructure didn't exist essentially at this point in time because there wasn't communications capable of facilitating trading activity. If you wanted everyone in the world to be able to trade, they all had to communicate. The cycle time for that transaction was so short.

In the first Industrial Revolution, we took everything we did analog to a mechanical footprint. What happened in that process? What we did was we shortened cycle times. That's another concept I'm going to talk about as we go through this. We took what took a day, a week, or a month and we made it a minute, an hour, five hours, or a day. We shortened things exponentially. We reduced the cycle time of transactions. That communications between my neighbor and I that took half a day now took, with a telephone or telegraph, hours or minutes. Across-country communication took a day instead of months. We shortened the cycle times.

The other thing that's happened in that process is we spawned an entire energy industry at the same time. The oil industry only came into existence because of the transactional speed that we were conducting mechanical transactions now. It was all of the industrial activities, steam engines, all of these energy-dense devices were invented to support this mechanical revolution. The second we were able to do something quicker, we did it more often. I always say it's like the Asteroids effect. It's something we coined internally. We talked about if simply we were all satisfied with playing Asteroids, we wouldn't really need Xboxes or PS3. At the end of the day, you don't need the most powerful math co-processor on planet earth to run Asteroids. But we're not satisfied with that. The second we can play 3D, fully involved, digital surround sound with your neighbors or people across the world online and Internet driven, we do it and we do it more often.

The point of it is we shorten cycle times, we increased the amount of energy, and we spawned essentially an entire energy industry around the mechanization. Automobiles, trains, cars, planes, everything that's happened up until, let's call it, 20, 30, 35 years ago.

What happened next? We're in the middle of the second Industrial Revolution. No one's even sitting around talking about it. We're all talking about it because we know it's going on. But most people don't recognize the fact that this is an absolute transformation. Every single thing that changed during the first Industrial Revolution is changing again but exponentially. The same way the cycle times shortened from analog communications to mechanical communications, now we're going from mechanical communications to digital communications. These cycle times are going down exponentially.

What's conversely happening is the energy being consumed per transaction is going up exponentially. The gross energy consumption's going up by factors. The reason you see that is the same effect. If I can transact a trade in milliseconds or microseconds, I don't just do one trade per day. I do a million per second. I do a hundred million per day. In fact, a hedge fund in Connecticut, I believe, today is responsible for approximately 15% to 20% of the daily volume on the NASDAQ. That's one trader from a transactional perspective. What we've done is we have driven all of these things along.

What's interesting for us is the reason people don't pay attention to this revolution is it's only really impacting a small slice of the entire economy. Everyone's getting the benefit of it. We all are at home and at work being able to bank all the time. Banking is one of my favorite examples of this entire outcome. Banking hours was the epitome of "I don't work a lot. I come to work at 9. I go home at 3." There are bank holidays. Think about what banking is now in the current digital revolution. You can take a picture of a check in your hotel room in Singapore on Sunday at 2 a.m. and deposit that check in your bank account wherever you want it anywhere in North America, anywhere in the world. It went from you can only do that transaction during a given period of hours during a week at a physical location to anywhere at any time. The compounding effect of this is that we're seeing this on the infrastructure side.

The ultimate outcome of this and where we're going is that we're on a track to computers being able to compute more capacity than all of the human brains on planet earth combined. We're just on a track. Starting in 1900 and all the way extended through 2100, in fact those numbers are probably by a very conservative measure of what's actually happening from a timing perspective. Look at the cost. Calculations per second per thousand dollars. This is actually the appropriate expression of Moore's Law. Moore's Law actually didn't deal with computational power. It didn't deal with any of these things that it basically is associated with. Moore's Law had to do with the cost to produce a chipset for a given amount of processing capacity. This is the ultimate expression of it.

If you look at classic economics and the human condition, once something becomes essentially costless, we'll use it infinitely. It's inverse of each other. Think about the way your users treat your resources in your organization. Until you start allocating the costs of a given resource in your data center pool, everyone just uses as much as you would give them. Storage, CPU, network, they'd be downloading videos, storing all their home pictures on their computer at work. They infinitely use these resources.

Where we're going and the track we're on, where this gets interesting, and where we fit in, at least we believe we fit in, is if we're going to replicate the amount of compute that's in all of our heads globally in computers, computers use energy just like humans do. Interestingly enough, it's very similar. We're actually a persistent machine. Think about it. Our brain turns on once in life and turns off once in life unless we get really lucky and it turns off once and turns back on. You're born. You die. In between, you have a machine that sustains that stateful existence called your body. It consumes energy, converts that energy into glycogen, glycogen gets used by the neurons to process transactions.

Similarly, data centers fulfill the same outcome. It's an energy machine that maintains a persistent state in a stateful environment to produce processing cycles. It turns on once. Hopefully, it turns off once. Hopefully, that's what we all endeavor to do. At the end of the day, the data center infrastructure globally is going to have to be at least equivalent to the energy infrastructure for powering human brains. It's going to have to be equally delivered.

The math. This is where it starts to get really interesting. This is where it becomes relevant to you. These next four or five charts tell the entire story. This is a logarithmically charted graph. Essentially, it's just steep. They smoothed it. From the ENIAC in 1940, one computer, one application, to where we're at today. Computation per kilowatt hour is on the Y axis. We have time across one and computational per kilowatt hour. The point of it is that since the ENIAC computer, computations per kilowatt hour have doubled every 18 months. The point of this is if we build the resource, people use it, and they use it more than they used it before. It's this compounding effective usage.

Digital universe 2009 to 2020, we're going to see a growth factor of 44 in the amount of data stored globally from essentially 0.8 zetabytes today, which is a trillion gigabytes, to 35. Web timeline and number of websites. 1990 was zero to 150 million calculated today. This is just through 2008 too. Internet users in the world. This is users. Let's not talk about devices. I always love show and tell so I brought all my devices that I carry with me every day. IP address, IP address, ability to get me five IP addresses so I can connect all my friends to you if my own devices aren't sucking it all up, IP address, IP address. One user, me, represents 10 devices. In this room, how many devices do we have? I'm fortunate because I'm looking around and I'm not seeing too many people staring at their computer. Earlier, I was watching the row in front me, there was a guy on Facebook. There was a guy on Twitter. The point of it is that we sit and do this all the time.

Data creation is rising at 60% per year. By 2012, data creation will be 2X where we can store this. This is from Intel's tera storage project. Walmart adds a billion rows per minute to its POS database. Per minute. A human genome sequence fully sequenced is two petabytes of storage. A gentleman that I have the fortune to know who's on my advisory board recently sold a company called Abraxis Healthcare. He's endeavoring right now to sequence every human on planet earth. That's what he wants to do. He's a billionaire so who knows. He has a good shot at it if anyone does. The point of it is take how many billion people on planet earth times two petabytes per person. This isn't that far off. Just alone the size of that data. The Hadron Collider can generate terabytes per second of data.

Where does all this rubber hit the road and why do we care? We have to spend money to facilitate this. It's now becoming meaningful. What this chart proves is that it's going to be so meaningful you're not going to be able to escape it. The CFO, the board of directors, the CEO, the CIO, the people that you have to go, or you if you're the person that has to sign the check, is not going to be able to afford to build data centers the way they've been traditionally built. There are a couple simple outcomes.

This is simply a one megawatt, usable IT user with a tier three, Uptime Institute, tier three enterprise class data center requirement. I've basically taken this chart from 2011 to 2051, so a 40-year life cycle. A couple basic assumptions are at the bottom. 8% annual reduction aggregate construction costs. What's interesting about 8% reduction in annual construction costs is it's never been accomplished. In the history of the world, construction costs have never gone down over any 10-year cycle, ever. The point of it is this basic assumption is false to start. We're assuming it to put the best light possible on the topic here. Essentially, incremental unit costs for construction of data centers is going to go down 8% per year forever.

We're going to see a 20% annual growth rate in IT usage of data center capacity, meaning megawatts of connected devices. The point of it is IT is growing at 20%. If you compound all of these other things that we've looked at here, none of these are 20% growers. From 1995 to 2005, we put 800 million users on the Internet. From 2005 to 2010, we put another 800 million users on the Internet. These aren't growing 20% per year. This is growing by factors. Same with every single measure we see here. What we all know is that every single device that does this has to be plugged into an outlet.

The concept that somehow efficiency is going to solve this problem just doesn't work. Here's why. Servers and processors can get more efficient. They're not actually at the end of the day when you consider it against scalar performance. There's an Intel report, energy per instruction. Google it and look it up. It's really interesting. Basically, it says from the i486 chip to the Cedar Mill chip, on an energy per instruction basis, so the amount of energy consumed per instruction across the chip, they charted it against the scalar performance.

This is back to the Asteroids concept. If you wanted to conduct a transaction at the same pace you did with the i486 chip, certainly it's more efficient than it was. No one wants that. The Cedar Mill chip is 8 times the scale of performance of the 486 chip and 37 times the power of consumption per instruction. Again, in the light most favorable to the analysis here, we've chosen to assume that IT demand for energy is only going to go up 20% when IT itself is expanding massively beyond that. The implication is when you look at energy feeding these systems doing work, it should be going up exponentially but we're just going to assume 20%. We assume we start out $15 million per megawatt, total cost of ownership per IT megawatt of data center. That's a pretty fair number. Uptime Institute estimates somewhere in the neighborhood. I think if you really dig into data center costs in most organizations, that TCO concept gets wildly ignored. They like to pick and choose what portions of it actually ends up being the data center infrastructure.

What I'm talking about here with $15 million is up to the point where you can put IT equipment in the data center, fully networked, all your infrastructure in place but no CPU, no storage, and no network devices. There are only three things that go in data centers. I always joke that everyone talks about data centers do this, this, and this. There are only three devices that go in data centers. They are storage devices, network devices, and computer devices. Then you have some ancillary things, KBMs and these sorts of things. They're all going away at this point in time. The point of this is those are the three things that go in a data center.

When you look at this $15 million, again, the green line going here from left to right is dollars per megawatt, the cost to produce per megawatt. The line going to the sky is the annual spend over the next 40 years to keep up with this 20% growth rate in your megawatts of IT load growing. That's assuming it goes down every year, the cost goes down 8% every year. Where I think we are in this analysis, do you notice how the purple line actually kicks down to start? That's because you build the data center and then you just start incrementally adding capacity as it's growing very slowly. You can fake it and fix it as you go. Then all of a sudden you hit this inflection point where I need an entirely new data center.

The cycle time to build a data center from greenfield is three years or 24 months. Let's say you're really good at it as an enterprise and it takes 30 or 24 months. In that cycle time of two years, your IT stack has grown another compounded 40%. A gentleman, our COO was at a large healthcare provider. Started greenfield data center project. By the time they got done, it wasn't big enough. You can't keep up. It's impossible. The outcome here is snowflakes, facilities-based data centers, are simply too expensive and take too long to deliver to meet the demands of this revolution or the demands of what we do for a living, which is managing enterprise IT and infrastructure.

The objective. What we did is we said, "Look, we've got this problem." I'll tell you why we identified this problem. We run massive data centers that are provided as a service to customers. We built 11 megawatt IT usable data center in Scottsdale, Arizona. Filled it up in about 18 months. The concept then was let's build them bigger. The way we're going to get efficiency, we're going to drive this cost per megawatt down. We're going to be able to deliver a better value point to our customers. We're going to be able to deliver these in the size that we need so we can actually have a sustainable business model. Built the seventh largest data center in the world according to Data Center Knowledge, IO Phoenix. It was 538,000 square feet, 35 megawatts of UPS online. Filled it up in 14 months from the day we opened it.

Again, we're looking at the same exact problem. That curve there is my need to raise capital to be able to build data centers to provide it to customers like you. I'm going, "This doesn't work." Over time, I can't drive the costs down enough to make this work and sustain my customers. My existing customer base, all of us, grows at 30% per year compounding. Right now, I'm adding anywhere between three to seven megawatts every six months just to maintain existing customer demand. You start multiplying that by aggregate, 10, 12, whatever the number is, per megawatt. These become enormous numbers.

The point of it is, is we said, "Look, we've got to solve this problem. How are we going to solve it?" We have to look at what the objective is in the first place. Unfortunately, the data center has not been controlled by IT. It's been controlled by facilities or some offshoot of the facilities team, which made sense at the time. It doesn't make sense anymore. The only people that know what actually goes in the data center, what it's being used for are the people running the applications and managing the infrastructure. They have to be able to control how it's powered. The only way we're going to solve this cost problem is by integrating the user, the application, with the device delivering the energy.

Ultimately, the objective, in our view, is the data center is IT and IT is always on. That's the objective. The objective of the data center isn't to have big generators, big chillers, all this cool stuff, and raised flooring. It's to keep the applications on all the time. I wouldn't care. If I was a CIO of pick a Fortune 500 or Walmart who adds a billion rows per minute, I need those billion rows to fill up every minute, I don't care how you get my applications to stay on all the time. Just tell me they're on all the time.

In the context of looking at what the problem is, we devised what we believe to be a solution - standardized hardware. The first thing we came across, and this is literally taking the easy part, is that you can't drive efficiency and production building something different every time. You can never layer a management tool on to it if it's different every time. In this room, there are probably 100 data centers being managed globally. How many of them have the same infrastructure? How many of them have the same management software? Heck, half of them are probably managed by old building management systems. The same thing managing the air conditioner in this room has been reprogrammed to manage your data center. This room generates, on a good year, what, a million dollars in revenue, two million dollars in revenue? That same BMS system that's managing this room is managing a data center for a Fortune 5 bank that generates how many billions of dollars of revenue per year? The solution hasn't met the need.

We endeavor to create standardized hardware. Standardized hardware means manufactured. If we manufacture, we can push unit costs down over time. Every manufactured process delivers that over time. Supply chain enhancements, innovation cycle, compounding intellectual property, taking feedback from you and putting it into the solution so we can make it better every time we build it, using it in our own delivery footprint so we can take that and reuse our own intellectual property, our own knowledge of operating these and make them better the next time.

Second way we have to accomplish this because simply manufacturing the same thing over and over again doesn't even get us halfway there to where this needs to be from a cost per megawatt. It gets us, let's call it 5%, 7%, 10%. The next is because now we have standardized hardware, we can now optimize this with software. Because we have standardized hardware, standardized software means we now can take that standardized software and interface it to the application.

Again, the ultimate goal here is to keep the application alive all the time, not to keep the data center alive all the time. Who cares? The data center can fall off the end of the earth as long as the application stays alive. At the end of the day, because we've standardized all these things, the outcome is improved operational sustainability at every level. It's cheaper to deploy, cycle times are shorter, just-in-time delivery. Literally we can manufacture 800 of these modules per year. Each module is approximately 200 kw of IT, 20 racks right now as we describe it today. You can buy it when you need it. You don't go to try to buy 10 years of storage and guess.

If any of us were good at guessing what IT looked like 10 years from now, we should all be buying stocks, not running IT systems. At the end of the day, we have no idea. We've been given, frankly, the most impossible task ever which is go build a data center and make sure it works for the next 10 years. Is anyone going to tell you that they're not going to buy another company, they're not going to sell a company? Is the CEO going to sit there and tell you what the company's going to look like 10 years from now? Absolutely not.

What do we call this? We call it digital energy technology. What does it look like? This is what it looks like right now. These are modules in production in our manufacturing facility. These are them being wrapped for delivery. This is them being put on a truck. It includes every component you need in the data center. Right now, what the data center is defined as a chiller, generator, UPS, energy management, energy storage, energy delivery, energy recovery, cooling systems. All of this ships together. It's one integrated system. What it looks like installed.

What it looks like inside. Full four-foot cold aisle, three-foot hot aisle, exactly the same as your current environment. Dashes, fire suppression, leak detection, fully integrated automation layer that's then integrated to our operating system that manages it. This is the operating system. Touch-panel screens in each one of the modules. iPad app, iPhone app, web enabled if you want it to be. All of these things, because we have standardized hardware, every single one of them comes out of the factory looking the same. We now have built an operating system that can manage that. You can see your entire global data center environment in one shot.

We've integrated this into VMware's virtualization layer and to the other virtualization layers, so now hardware brokering is actually something that can be done from the data center level. This module's in yellow state. This module's in red state. You now have the ability to tier your data center as well. You can deploy a system and say, "This module is a priority 100. This module's a priority 1." If the system can't sustain it, it shuts the one off and keeps the other one on. Cost of ownership now is the same.

I was listening to the previous session. They were talking about hardware virtualization. What's interesting about hardware virtualization is the two barriers I took away from the conversation to desktop virtualization is data center capacity. Now you're taking every desktop computer in your environment and putting it in the data center. Two is proximity to the users. There's actually a latency impact.

Here's proof in the pudding. IO New Jersey, we took possession of an empty warehouse, a very big empty warehouse, 836,000 square feet, on April 1st. Today it's fully functional. Core power is 7.8 megawatts. Critical is 4 megawatts. IT load is 3.2-ish depending on the redundancy level you want to run at. Fully deployed, started commissioning today. Two months-ish from the day we took possession of the site to the time the data center's deployed. The real advantage of it, we have 112 of these modules currently in production today for customers. We can produce 800 of them a year out of one manufacturing site we have today. At the end of the day, what it allows you to do is start treating the data center procurement cycle the way you treat every other part of the data center infrastructure.

The transformation that is occurring here is that we have to quit thinking about the data center as a separate piece of the pie and pull it into IT. This isn't just another box you buy. We have customers right now that are pre-populating these in the factory. We have a customer customization zone. Then they're shipping them already pre-populated both to their sites and our sites where we host them. You can purchase it if you want to run it at your data center. You can do data center augmentation. You can take an existing data center that's not properly allocated. You have too much generator, too much chiller, too much UPS and not enough floor space. You have too much floor space and not enough UPS. You can now solve these problems in existing data centers and extend the useful life of them.

There's another entire elephant in the room no one wants to talk about with regard to data centers. What's the accounting department booked your data center as from useful life? Ten years? Twenty years? Let's say that some aggressive accountant decided that it was a real estate asset and booked it for a 27- year useful life. I think the IRS allows 30 years. Who wants to deliver the message to the Board that your auditors just made you take an impairment charge of $100 million because your data centers are no longer useful? No one does.

The point of it is this thing's designed and the cost point on this is designed to function on a 10-year useful life, three IT cycles. It can be in-place upgraded. All the infrastructure in it is modular. Air handling infrastructure is modular. A single module can be upgraded from 200 kw to 600 kw with in-place upgrades while it's running. You have to take one side down at a time. Concurrently maintainable. All of the attributes you have in your existing data center footprint but delivered in a fashion where you can start treating it like the traditional IT process.

At the end of the day, that wraps it up a little bit. I'd love to answer questions about the product, about what we're doing, about how we see this transformation occurring. I think the most important point from our perspective is you have the ability to come see it. This isn't vaporware. If you'd like, we'd love to schedule a time for you to come to Phoenix, come up to New Jersey. We're in Edison, right at the 287 and the 95. Schedule a tour, time to come see the product. We'll take you through our factory, walk you through the OS. We can set up the OS on an iPad, take it to your office on a console, show you existing running systems. At Gartner Data Center in December, we're actually going to have a small sub-system up and operational at a site in Las Vegas. We'll be able to actually host an event where you can come out and see a system running. We'll bring it in. We'll show you the setup process and deliver it.

The neat thing about this is this is happening, right now, real time. We've made the decision that we're never going to build another square foot of traditional data center. Data centers shouldn't be measured in square footage anyway. We all learned this lesson the hard way. Let's measure it. It's 100,000 square feet. The CFO says, "Well, heck, it's only got 50,000 square feet of stuff in it. Where's the other half?" The other half doesn't matter. What you really count is kilowatts or megawatts plugged into the back of those servers.

Any questions? I'd love to answer questions. My team, any questions? Yes?

Participant: I guess the greatest question, as I think about a data center and the modularity of it, out of the area of connectivity ... [inaudible]

George: It's a good point. We kind of look at the IT stack. Obviously, network connectivity is at the top of the stack. Then you have the data center that sits in the middle. Then we see at the bottom you have digital energy infrastructure which is how you power that stack. Then network's how you connect it. Our view is that until wireless infrastructure reaches a certain point for your distributed content distribution, obviously fiber optic connectivity's the only result at this point in time that works and delivers. Network is still going to have to be brought to the site by the carriers at the end of the day. Now what you're going to be able to do is leverage campuses. You can also leverage more effectively where their network exists today. Our view would be that over time what's going to happen with the network is that you're going to take the modular data center to where the network capacity sits and where it makes sense from a proximity perspective.

We all have data center needs. We break it into four buckets. Zone 1 is your network-centric applications. One's a latency truly does matter on. Need to be within 500 to 800 miles of whatever you're communicating with. If you think about it in the classic data center providers, this is what Equinex does for people. They provide these network-centric colocation sites where latency really matters. Then you have your Zone 2 and Zone 3, which are city-based applications and/or regional-based applications that are sitting in an office building because of land requirements or man requirements. Then you have your global data centers that are going to sit somewhere. Someone's always going to be a long ways from them. You try to put them in the best spot possible.

I think this follows the same context. Now you just simply take the data center to where the network is. I think that what changes now is rather than having to pull the network to you, you can take the data center to the network, which I think at the end of the day is what's going to end up happening. These carriers can't afford to keep pulling fiber everywhere you go. You already see it. Not to mention anyone by name. They fight you tooth and nail. Then they want to charge you $10 million so they can then charge you $500,000 a month. It doesn't really function. It's much more convenient to be able to say, "Tell me where you network is. I'll pour a slab."

The only thing is most of your data centers today have unbuilt capacity in them. A lot of them have raw space where a product like this can be delivered. We're doing this for customers right now where we're taking the product, delivering it inside an existing shell so the network assets, operational assets all are sitting right there already.

The second point behind network is people. Everyone forgets you have to have people there to run these data centers. A lot of the hardware vendors espouse this view that if you go to a modular footprint, everything's going to be their product so you don't need people anymore to manage it. We all know that's never going to happen. You walk a data center, there are hundreds of vendors of products sitting, there are hundreds of applications. There are applications from 50 years ago, stuff on mainframes and running COBOL.

Any other questions? Yes, sir?

Participant: How many rows of racks are in the module?

George: We have two designs for the module. The module that's in production today has one row inside of it, which is up to 20 racks in one row. We have a second design, which is our service provider design which allows you to get three rows in two modules. You essentially take two modules and put them together. That just is a density. If you're above 5 kw a rack, you can't utilize the three row version because your densities are too high, again, counting kw. The product scales immensely from a kw perspective from 5 kw up. In fact, this is one of the interesting things about a modular product is it becomes more economically efficient as you scale the densities up because your investment in that individual module gets spread across more kw rather than less. The point of where it starts is at 2 to 2.5 kw per rack. That's about the bare minimum that this product makes any sense for anybody. I don't think anybody really runs below that threshold anymore. That's in the service provider framework, which we get essentially 50 cabinets across 200 kw.

Then you start going into the dense environment, which is our enterprise class product, the Kepler product. That's where you're getting 20 racks across 100, which is 5 kw, and then you can go up to 600. You can stretch that to 30 kw per rack and in a sustainable environment as well. Another note of this is that the product's designed UL listings and process. From a local municipality perspective, they don't really get involved because it's UL labeled. It's ADA and OSHA listing and process as well. It's in a work space you can actually put people in and not have your HR department freak out. We can temperature control through the redundancy levels so that you never have 140 degree hot aisle even if you're running at very, very high densities. It reduces the redundancy when you're doing that, but it's only when people would be inside the modules. You have the ability to manage these sorts of things.

Participant: Are those modules strapped together somehow?

George: Actually, ironically they simply sit right next to each other. If you think about it, that's a single module. That's about 42 feet long. Right there you see it. You push them right up next to each other and set them down. When you have eight or more modules together, we have a seismic bracing that you attach which actually make the seismic system. It's the first fully seismically rated data center system. You can put this in a Zone 4 environment and theoretically it will survive. We haven't tested yet. The point is they literally just sit there. All you need's a slab on grade, essentially your typical warehouse-style slab works just fine. The advantage we have here in the United States is there's about half a billion square feet of empty industrial right now that leases for like $0.60 per square foot per month. It's $6, $7 typically. There are lots of data centers waiting to happen. The power densities are such that you can utilize standard industrial distribution.

Participant: To get from one of those modules to the other, you have to come out a door ...

George: No. There are pass-throughs engineered in the side. They have different sizes, any requirement from 12- to 18- to 24-inch pass- throughs, 6 to 12 inches deep. From a cabling perspective, you can do all the cabling. All the power distribution happens over the top. All the coolant management happens on the outside below the raised floor. If you notice, in this, that raised floor is at four feet. You can see underneath the access panels, right under the doors there, is where all of the modular infrastructure is installed. All of your support and operational staff that manages this, we train your staff, IO-certified engineer, IO-certified technician, IO-certified designer just like a Cisco model. All of this is really good from a security perspective and SaaS management, HIPAA, and PCI, and everything else.

When your service staff is maintaining it, they don't have to go into the IT space. All of the service maintenance for the module can be done in that subfloor. It's a full four foot high. Air handler, maintenance, fan maintenance, all of the things that you need to do. You can essentially do all of that management without ever granting access to your service personnel or vendors to the actual IT space.

You can see there the power distribution across the top. That's all modular as well. You can add another module without taking the system offline. The beauty of it is, again, this is the kind of change in thinking this has to drive. I don't have to go build a 3 megawatt data center or a 10 megawatt data center. I can go build a megawatt or 800 kw. Then when I need another 200 kw, I just buy the next module. I buy the next module. I buy the next module. Yes?

Participant: [inaudible]

George: That's correct. It's modular in the unit. How this system works is there's a power pack you purchase or we deploy depending on whether it's in our data centers as a service business segment where we run these various large sites like Phoenix and New Jersey and you can just rent the capacity there, or where you buy it. The power pack is made up of energy storage, which is a UPS system. We give you the option of energy storage types. If you want flywheel, VRLA, or wet cell, you can choose those products. We actually are multi-vendor independent on the energy storage. You can use whatever UPS vendor's currently on your spec. We deploy that. UL list it, deliver it. That power pack is then tied to a generator and a chiller that matches capacities exactly so that you have this integrated system that's fully tested and deployed using our OS. At the end of the day, you have a fully embedded power pack.

Depending on your redundancy levels, how many power packs then you install. If you want to run a 2N, you put two and then you only use half the IT load. If you want to run it N plus 1, you could go three power packs, one's redundant and then cross tie. You have the ability to do a lot of active management from that perspective. The OS gives you a redundancy view. You can see what every module's running at from a redundancy level based on the current demand. You can actually now actively manage redundancy across the system rather than having to guess at it or deem something for the entire site.

I was in a very nice data center. I have the fortune of visiting my customers' data centers. They have very nice data centers. They've spent a lot of money on these things. Do you know what I see over and over again? I sit at the main service entry section and the main critical power section. Most of these sites are running at 20% to 25% utilization. If it's a 2N or 2N plus 1 environment, it never will run past 50% utilization. This is massively inefficient. Capital, equipment, energy consumption, everything about it is inefficient. The worst part of it is then you go inside the data center and you find that half the stuff sitting in that data center doesn't actually need any of this redundancy. I always love it when I see rows of powder horns inside a Tier 3 or Tier 4 data center. They're off most of the time anyway.

The point of it is that this gives you the ability to tier by 200 kw chunks what's important and what's not. You can spend the heavy iron. Our view over time, our roadmap actually, to drive the cost down to where it needs to be to keep that line that I showed that goes to the sky flat enough that we can sustain it, we've got to get rid of a lot of this stuff. Generators, static UPSs, these things have to go. They're just too expensive, complicated, and hard to manage. At the end of the day, they're just too expensive for the system.

In a classic engineering sense, you take and break this thing apart and throw it on the ground, everything, and you start counting the cost of each individual component. Some of these things have to go away. Full-rated energy production on site either has to be the prime energy source for the system over time or it has to be just for the devices that actually need that type of sustainability.

We see what's going to happen is that the power packs will actually start degrading the amount of energy generation necessary on site. Through the OS and as you see virtualization becoming a more powerful tool like with long-distance vMotion, all of your WinTel apps you can just vMotion out of the box to one that's still running. Then your prime rated energy production on site where you have these 72 hour requirements or multi-day or infinite requirements can just be for the systems that need that.

If you go through your data center and really sliced and diced every application and put them in the matrix, which we all do, not many of them truly need this kind of Tier 4 sustainability. This gives you the tools. The problem is we don't have the tools to do that. Before this and before the standardized hardware, we had no idea what's using what, what the redundancy level in given parts of the data center is.

Caterpillar's a huge partner of ours. They've provided a significant amount of capital to us. You're going to see yellow engines outside of our data centers, for the time being. When we get rid of them, you won't.

Yes, sir?

Participant: Do you have to have your mobile power to this, or can they sit outside?

George: Actually they're a NEMA 3R. We can either apply a NEMA 3R or NEMA 4 enclosure rating to the system. It can be set outside and deployed in an external environment. If you're in a subtropic, tropical, or arctic environment, there are some modifications that have to be done to the system to accommodate for that particular climatic condition. For the most part, the system can be deployed outside.

It creates some interesting operational issues. You have to think about dust control. All these things of having a system sitting out in the environment, you have to start thinking about.

You can't stack them. Physically the box is designed to be stacked, but just operationally it doesn't work. How do you get the equipment up to it, to the second level? How do you get people up there? How do you manage? It just becomes overly cumbersome.

Yes?

Participant: When did you deploy the first one and how many do you have?

George: As of today, we're on the third generation of the product internally. We've been using it for about a year and a half now in various forms of prototype. Out of our organization, almost a full third of our company are engineers. It's a good thing and bad thing depending if you're a sales guy or if you're a technician. We take this very seriously. We won't put something out there we're not willing ... One, we have to run it and maintain a service level agreement. Two, we wouldn't put something out there with our brand on it that we wouldn't think would be able to sustain for our customers. We're on the third generation internally. We've had modules deployed since mid- February of this year. As of today, we have approximately, I think, a little over 35 modules in production. We have 112 currently in manufacturing process for delivery this year. We have a backlog of about 300 modules going in the next year. At 200 kw a piece, it's a significant number. We'll ship anywhere in the world. We'll support the product globally.

Participant: So, 35 in production?

George: Yeah. You can come out, like I said, to our New Jersey facility and our Phoenix facility. You can come see them operational. Plus they're running at customer locations as well.

All right. I thank you very much for your time. I know everyone's busy. I know everyone wants to get home. I appreciate you taking the time. Thank you.