Originally aired on April 20, 2021 @ 1:30 PM - 2:00 PM EDT
An interview with Michael Aylward, Head of Cloudflare's Network Partners Program and renewable energy expert, about the carbon footprint of Internet infrastructure. We'll explore how the Internet contributes to climate change and what tech companies including Cloudflare are doing to minimize this footprint.
Hello and welcome to Cloudflare TV. My name is Annika and I'm a product manager at Cloudflare. And today I'm joined with Michael and we're super excited for this conversation about sustainability and Internet infrastructure. And to start out, Michael, can you please introduce yourself? Who are you? What do you do at Cloudflare? And how did you get interested in this field and sort of your background before Cloudflare too? Yeah, totally. I handle a pretty particular part of Cloudflare's infrastructure program. So I work on our edge partner program where we partner with ISPs around the world to deliver their content as close to their customers as we can. Before I joined Cloudflare a little bit more than four years ago, I used to work in energy and climate policy and politics for like, well, since 2008, since the 2008 election. So I usually had to handle, I have a background in math and economics, so I usually had to handle sort of the more quantitative and technical parts of my office's portfolio when I was doing that. So I learned a little about renewable energy credits and things like that. And so I got kind of voluntold to work on those things at Cloudflare when Cloudflare began thinking about sustainability. Yeah, totally. And I remember the reason that I reached out to you for this segment is you were gracious enough to come and give kind of a talk to our green cloud group, which is Cloudflare's sustainability working group of employees that are really passionate about the environment and want to work to address the climate crisis. And we had this really great engaging conversation about, hey, like, how does the Internet actually contribute to the climate crisis? What is the breakdown of that? Like data centers turns out use a lot of energy, what can we do about that? So that's kind of what we want to focus on today. And I guess we could just start at really the beginning, like the sort of thesis of this conversation or the ground assumption is that Internet infrastructure consumes energy and therefore has an impact on the climate in places where that energy is not from renewable sources. So just like to start with, what does that look like? Like, what is Internet infrastructure to start with? And how does it actually consume energy? What sort of places and activities? Sorry, I was having a little honey there. I didn't get to have my tea this morning. No way. So the, I actually am not a super duper in depth expert on, like, all of the sort of granular components and places that the Internet consumes energy, but sort of at a high level, I'm like, moderately aware. So it's basically a collection of, I mean, if you think of the Internet as a system, really, really expansively, that could include our phones and computers, even though some people might say that's not part of the Internet infrastructure. You could think of that as sort of, you know, part of our footprint in terms of like the resource use. So there's some power that, you know, our phones and computers use when we read things or use the Internet otherwise. And then there's like, you know, the, I love the old saying in the industry, I guess, is that the cloud is just somebody else's computer. And so, you know, the Internet is really just a bunch of somebody else's computers and some network switches and some routers and some mechanical devices like, you know, fiber optic cables all across the world that require some power to continue transmitting information across them and processing them in the somebody else's computers in all of the data centers that comprise the Internet. So it's mostly, you know, the use of electricity to power all these things, the sort of infrastructure, the cabling, the servers, the switches, everything that goes into it, the cooling in very large data centers. I think, I can't remember if you asked this part, but basically people began worrying about the use of the sort of skyrocketing use of energy by the Internet. I actually don't know exactly when, but the sort of zeitgeist started really getting loud, like in 2008, 9, 10, 11, 12, when it became clear that the use was growing really fast. And so people were worried that the carbon footprint was going to grow really fast too. People were building data centers like crazy. And actually, don't let me jump the gun, but that was sort of like, you know, that's sort of like the context and the premise for that kind of concern, right? Sure, sure. And you mentioned, okay, so computers, that's easy to understand, right? Like I can look at my laptop in front of me and understand I need to charge it. So I plug it into my wall and I'm paying some amount for electricity at my apartment. And then that's coming from somewhere that's generating that energy. What about for folks that haven't actually been in a data center before? Could you just actually break that down? Like, what does that look like? What's a typical data center look like? And what are all the places in it that suck electricity? You mentioned cooling as one that we might not think of immediately. Yeah, hilariously, I've gotten to learn a little about these things, you know, in my work on Cloudflare's infrastructure team for the last almost three years. But I've actually never set foot myself in a data center. My boss, if he sees this, he's going to be like, well, that has to change right now. I think I probably would have already if it weren't for the pandemic over the last year, 15 months or so. But, you know, anyways, I've seen pictures. So the answer to your question is, you know, there are there basically racks and racks of different people's servers, different companies. Mostly servers are what are what take up the footprint and use the energy. And then there's some other networks, which is things that that tell devices where to route requests and which networks to to deliver or split them up on. Generally speaking, these are routers and switches. And a lot of stuff when you pack it in fairly densely. You know, it works well. It works much better at certain temperatures. And so cooling gets to be an issue for really large scale data centers. Not so much for the ones that that Cloudflare works on, but but like the ones that take up city blocks. Sure, sure. That makes sense. So, so I did just some some light googling just to double check the latest numbers on this before this call. And it looks like around the latest estimates are around 3% of electricity usage from sort of global global usage is from data centers. And that breaks down to around 2% of global greenhouse gas emissions, which sounds like a relatively small number, maybe when compared to some some other industries or sectors that take up a lot. But that is a pretty big number still. And like you mentioned, right, that that's growing. And that's the concern from people really is less, hey, 2% is a really big contributor today, but more that everything is shifting in the direction of more data, more data, more Internet usage. And then more servers and more more cooling and more energy for them over time. So can you talk about that? Like what how has this actually changed over over time? And we were talking before about this, this breakdown of sort of like usage versus efficiency versus kind of like the actual carbon footprint. Yeah, I, I found this pretty interesting. I was so the 3% number you mentioned for global electricity use from data centers. I am not aware of like, I'd love to see the data, but it's it's in that order of magnitude, it might be 50% less than that. I think the last numbers I saw were like 2%, which, you know, one percentage point, not a big deal, but 1% of global emissions, you know, percentage wise, it's a fairly large. So anyways, I've seen I've seen actually figures in like the one to 2% range. But obviously, like, you know, like you said, the the important thing there is the rate of growth. If it's going to stay at one or two or 3%, that's not that's not a big difference. But if but if it grew from 1% to 3%, which is a, you know, a 3x tripling over some whatever period of time, then that rate of growth is what we need to be concerned about. So the interesting thing that I that I learned when I looked into this over the last year, there's the data that I've seen so far. And the analysis I've seen so far says that even though global computing and and the number of data centers, the number of servers in them, the number of workloads that they actually deliver and perform. The last data that I saw was actually in late 2018 or 2019. There's sometimes data lags in these analyses. But it looked at the period between 2010 and 2019 and found that that workload, the actual, you know, work that was performed in the number of machines that were in operation, grew something like five to 6x. So like, like 500% over that nine years 2010 to 2019. But the actual electricity footprint, the amount of additional energy it used was was 6% greater. Not a 6% rate per thing but like total 6%. So how did the workload expand like 500% but the actual energy use expand only 6%? The answer is that there's a little bit of a maybe more of a lot a bit of a Moore's Law for energy efficiency in this context where, you know, all of the additional workloads were able to be expanded by finding efficiencies in the in the energy use. That's that's sort of part of the logic of these hyperscale data centers. So, the good news is that global computing has obviously exploded everybody can sort of know that even without seeing the data just how much they use the Internet, how much people sort of get from their phones and computers. The interesting piece is that counterintuitively, there's been essentially no expansion in energy use in that time, and clean energy is actually expanded in in the data center world so just anecdotally, I would bet $100 that the energy footprint of the Internet has actually dropped over the last 10 years, rather than risen, even as its footprint its use. Its sort of scope has grown by, you know, five or six x. My guess is that the total carbon footprint has dropped because it's not using any more energy and the energy and the carbon intensity of the energy it's using is lower because we're cleaning up the energy grid. The challenge that we have is going forward, Moore's law runs out at some point. So you can't just keep doubling energy efficiency every two years in perpetuity, it gets really hard when you get to those really small ones you sort of squeezing out the low hanging fruit. And so what I've seen is that the again I'm not an expert on this I'm just sort of the messenger, but what I've seen is that the experts on this are worried that in the next doubling or two, we're going to run out so that they believe that we'll have one more doubling of Internet sort of global data center capacity, where we'll be able to sort of fund that doubling energy wise with energy efficiency, we'll be able to do that and not expand energy use because we'll find energy efficiency. But beyond that, we're going to run out of luck. And so the question is, for a carbon footprint. Do we win the race, do we get, do we clean up the energy supply before we begin to really expand our energy use because right now our energy uses is not growing but it's going to grow in the future. And so the question is, by that time that it starts growing. Will the supply of energy be so clean that it doesn't matter how much energy use, because it's all clean energy. So that's sort of the the race that that we have right now in front of us. Sure, sure. That's super interesting. I wonder, are you, are you, do you know of that efficiencies that we've gained is that primarily inefficiencies in sort of like traditional energy sources or also in the adoption of renewable energy at data centers like does that also account for companies that are building entire data centers powered by things like solar, or is that sort of in its own separate category. That's in its own separate category that's just sort of gravy on top of the efficiencies the efficiencies I was talking about was just the number of kilowatts being used to perform the outputs that we have so requests served or bytes, you know transited, or, you know, just the way that workloads are measured, the question is how many watts are needed for those and the answer is the same number 6% greater number of watts are used now than were used in 2020. The question but and that's like no totally agnostic about where those watts come from the answered about where those watts come from a separate and like you said, there's been a whole bunch of clean watts that have been injected into grids where power, where data centers are really prevalent. So again like this is the thing where I don't have any, I don't know the data on this, but anecdotally my guess is that if you look at the carbon intensity of that power, it's better it's less carbon intensive now than it was in 2010, because we're making progress in greening the grid. Does that make sense. So there's two separate, there's a question of efficiency which is just watts no matter where they come from. Then there's the question of what kind of watts are those. Yeah, yeah, that makes total sense. Thanks for bringing that down. I think one, one thing that we definitely want to have spend at least a couple minutes here on is basically how can using Cloudflare help like there is some there's some good news if you are a viewer of this segment, and a user of Cloudflare CDN and maybe our edge computing products also there's a good chance that by having those capabilities sort of push to the edge or action in Cloudflare data centers as opposed to needing to go back all the way to your origin that you're saving some energy. And before, before we get back to that, though, curious sort of more agnostically not even just specifically talking about Cloudflare yet but there's a bunch of things that organizations can do to kind of reduce their energy usage from from their Internet infrastructure right like one thing is just investing in sort of the efficiency and energy use of their own applications right like if you, if you have the ability to do kind of optimizations on your own apps that make them use less energy, that's great. But then also if you there's some amount that's going to be fixed right so if you're a company, and you have you know some amount of energy use and you've managed to quantify that, what can you actually do with that information, what are some options that you can take to to sort of offset that. Yeah, it's a good question. So the easy the shorter answer is the one that you mentioned first which is like, you know, for the workloads that you do have control over. Can you make them more energy efficient and the good news about that is, is that energy efficiency for those for that category is directly correlated to other kinds of optimization so like, usually when you make something faster, it also becomes more energy efficient because speed means that less things are being used. So there's a natural sort of alignment of incentives there, say between the. Oh hello. Between, you know, the part of the company whose job it is to worry about the customer experience, maybe, or even maybe sort of the the back end workload burden. And the part of the company whose job it is to worry about the footprint that say environmental or energy footprint. The good news is that that's very aligned there's like there's not a lot of tension between those two things. Yeah. But then you asked about the question about like the fixed amount that really can't be avoided. And it really depends on basically how big the footprint is and how ambitious the company is. So if you have a relatively small footprint. You might want to take some action that that in some way accounts for your footprint. So a lot of companies do this if they're if they're sort of electricity intensive, they might want to have a gesture that they are supporting clean energy, even if they don't get to choose their energy that they use like if they don't get to choose from their supplier. And so they might decide to a very common thing in this space is to buy renewable energy credits, which are sort of a strange accounting measure, but they are eventually an indirect support of renewable energy somewhere. We wrote about these just a little bit, not not in any sort of detail in our annual sort of sustainability blog post in 2019. If people are sort of interested in a slightly more detailed description of that. And then if you have and basically if you're if you're not super power intensive, if your energy use is not really electricity heavy, but it might be traveling or something else, carbon offsets are a great way to go. Although not always awesome. Like there's a huge spectrum. There's some that are like just total sort of make work like like not very additive in terms of sustainability value. And then there are some that are on the other end of the spectrum, really great in the sense that they they perform some carbon sequestering activity that would otherwise not be performed. And so there are a whole bunch of folks who who help identify that spectrum and and find good projects. Three Degrees is a great example. We work with them and a whole bunch of other companies do too. You'll see a bunch of large companies, sustainability announcements sort of co -tagged with Three Degrees because they help a lot of companies think through these things lift in the past. I think they may even still when they decided to offset all the emissions from their drivers vehicles, they used Three Degrees to think about a strategy that could actually do that meaningfully rather than just sort of cosmetically. And then if you're a bigger energy user and power heavy, like maybe a Google or a Walmart or an aluminum manufacturer, say, you can actually build or fund either entire clean energy projects or part of them. And the reason and this didn't exist maybe 10, 15 years ago, Google sort of helped build this market. They did it for themselves and then they kind of, to their credit, open sourced the way that the way to think about analyzing these and the way to go about finding partners, syndicated buyers where you could sort of pool people who who might want to do these things together to have enough energy use to make it to make sense to go buy a whole wind farm and then do that. And thankfully, that's really been democratized. So if you have a large enough energy footprint, you can participate in one of those kinds of activities, basically buying. It's almost like adopting a wind farm or something. Sure, sure. Sponsor, sponsor a bunch of solar panels. That's awesome. OK, so I heard a couple of different things there. First category is just kind of like what is what is actually in your control and do everything that you can to minimize the energy use of the processes that are actually in control in your control? And then you pointed out that that is actually generally there's lots of other good reasons to do that anyway. Besides just the energy use rate, make things faster, use, you know, use less space, make a more efficient use of the resources that you have to help save you money on other things besides energy, all kinds of stuff. So that's awesome thing, too, is if you have some amount of energy use still that you want to do something to account for it, you can purchase Rex, you can purchase carbon offsets or you can do sort of a bigger scale project. That's either working with other organizations or if you have, you know, sort of the resources and you're producing enough energy, you can sort of take on one of these big, large projects on your own or in partnership with another organization. I remember. God, I have you to synthesize things. Yeah, to summarize, to summarize. So I remember being sort of mystified by the concept of Rex and like Rex versus carbon offsets. And like from from my perspective as like a very person coming into the space, very new, I was kind of like, you know, like forehead scrunch, like, but if I'm a company and I'm like, if I have some energy, I get that. I have some energy usage and I'm going to like pay some money to like account for that energy usage. What is actually the difference between a wreck and a carbon offset? Like what's actually happening in the world as a result of my dollars and how is that different? Would you mind actually going into a little bit more depth? Like what does a wreck do if I pay somebody for a wreck? Who am I paying and what are they doing with that money? Yeah. The way I think about this is the electricity system is is very, very similar sort of mechanically in this sense to maybe sort of a water distribution system. So imagine that you use a bunch of water somewhere, maybe in a couple of places, like in some factories and some parts of California and water suppliers, people who put water into the system to be distributed and used by a whole bunch of people, including your company and others, have varying degrees of quality in their in what they put in. Some people put in water that doesn't do anything bad to the landscape or the environment where they get it. And some people strip mine the water and put it in. The water is great. It still works well for your factory. But what it leaves behind by the producer that put it there is a big mess. The difficulty, so electricity is very much like that. Some people strip mine coal, pulverize it and basically aerate it in a giant coal steam turbine. And some people spin wind turbines and put solar panels up. So the difficulty is that just like in the water system, when the water droplets go in and they get sort of dispersed through the pipes and you take some out, some other people take some out, you can't tell which droplets came from where. They're all mixed around and they basically look the same, at least for electricity. There's no really way to tag the droplets or anything like that. So the only way that you can really tell that you can have sort of an accounting of your footprint is it's almost like a balance between inputs, injections of water or power into the delivery system and outputs, withdrawals. So you can sort of say for accounting sake that if you're responsible for inputting as much clean water into the system as you take out, then you're sort of in some way, not fully, but in a fairly strong sense, your sort of footprint of like water cleanness is pretty strong. Because you're putting in the kind of water that you want to put into that system that doesn't come from strip mine mountains. The interesting thing is that if you don't have enough water use to basically be responsible for a whole project, you can imagine that sort of like putting water into the system requires a minimum of like 50 million gallons. And like anything less than that just doesn't make economic sense. Power is like that. It doesn't make sense to put a small amount of power into the system. It economically just doesn't work. But if you still want to have a sense of contributing to clean water in the system, what you can do is the people who inject clean water into the system have an accounting system where every gallon that they put in, they can't keep track of it afterwards, but at the point of putting it in, they have a cool little serializing system where they have a serial number for that gallon. And they have a little certificate for that gallon that they keep. And they can sell those to people who want to say I supported this clean stuff. Now, sometimes they sell them to companies who are actually legally required to say that they supplied 50% of their water from clean sources. Utilities, electric utilities in the U.S. have those laws. In fact, that's by far the single biggest driving factor for why we have sort of a clean power boom in the U.S. and in the world. So most of these certificates are bought by utilities to show that they're complying with the law. But other companies, you know, still have them, too, and sometimes they have some leftover beyond the ones that they sell to people for compliance purposes. And they have those available for what's called voluntary purchases for people who want to say, you know what, I bought 10 gallons from these folks, and I retired these certificates to show that, like, nobody else can have that certificate in the future and claim that they were really responsible for clean power. Sorry, for the clean water. Because we're taking it off the market so that, like, the cleanness can't be double counted in the future. Yeah, this is accounted for, sure. So that's what a REC is. Carbon offsets come in tons of forms. They basically are, like, they're a little bit dicey. There's a whole spectrum of carbon offsets that don't have a lot of what in the industry is called additionality, in the sense where, like, the activity that they're selling a ton of carbon reduction may have happened anyways, which means that your payment for it might not have the, wouldn't have the kind of leverage or impact that you're hoping on this thing. So what you really want in a carbon offset is an activity that's otherwise not going to happen for some reason. And your contribution will somehow make it happen. There's a total spectrum of how that could happen, but some of the biggest, some of the most impactful carbon reduction strategies are actually basically stopping carbon sinks from being removed. So one of the biggest net sources of carbon pollution into the atmosphere is reducing places where carbon is sequestered. Mostly these are giant natural habitats, like rainforests and other sort of natural ecosystems that suck in a lot of carbon naturally, but that sometimes get destroyed for ranching or farming or other reasons. And if you can somehow save those from getting removed for whatever economic purpose the people in that local area were going to do anyways, and keep it as, you know, wetlands or rainforest, then that land can have a huge carbon sequestering effect. It can suck in a lot of carbon. Not all of it's created equal. Some land, some ecosystems store much more carbon than others. And obviously those have a bigger carbon impact. Sometimes it's not possible to enforce the saving of these. You might be able to save them for five more years, but then you have to ask yourself, well, what is the greater impact there? So the carbon offset world is very complicated. I don't understand much of it, but that's sort of the general, you know. Yeah, that was a super helpful overview. And it looks like we have about five seconds left. So thank you so much, Michael. Appreciate the time and the conversation. This is super educational for me. Hope our audience enjoyed it as well. And catch the rest of Green Cloud segments on Cloudflare TV this week. Thank you, Monica.