Listen: Next in Tech | Episode 49: Carbon reduction in cloud

View Full Transcript

Transcript provided by Kensho

Eric Hanselman

Welcome to Next in Tech, an S&P Global Market Intelligence podcast with the world of emerging tech clips. I'm your host, Eric Hanselman, Principal Research Analyst for the 451 Research arm of S&P Global Market Intelligence. And today, we'll be discussing the carbon reduction potential of cloud with 2 returning podcast guests: Kelly Morgan, Research Director for our multi-tenant team; and principal research analyst, Dan Thompson. Welcome back to both of you.

Dan Thompson

Thanks.

Kelly Morgan

Thank you.

Eric Hanselman

One of the things that was driving a lot of this interest -- we've been doing a lot of research in this area from an ongoing basis, but there was a press pickup of this that it seemed worth highlighting really some of the key parts about this because it's something that I think, certainly, in a lot of conversations with clients and in the broad market, we tend to get this idea that cloud is a big energy consumer. But these are a series of reports that are talking about the carbon reduction potential of cloud. So what's the focus of the reports? And what's the data they're based on?

Kelly Morgan

Yes. Well, that's a great question because certainly, it comes up a lot that these cloud data centers use a lot of energy. But in fact, what we tried to do was pull apart how much energy is used by typical IT equipment and how much energy is used by typical data centers. And then we tried to compare sort of enterprise setups with a cloud setup to see if they were the same in their use of energy.

And what we found after doing surveys of enterprises, and we built a big model that kind of looked at based on the survey results, what kind of the typical enterprise usage of their IT assets was. We built this big model and determined actually that cloud servers are much more efficient in general than enterprise servers, and that's partly because they're newer and there are more of new ones in the overall kind of server farm of a typical cloud provider.

But also they're much more highly utilized, which makes them more efficient. So we actually found quite a big savings as much as a 60% efficiency improvement, just switching servers, essentially from enterprises moving to cloud, they can save up to 60% of their energy usage. And then kind of to add to that, when we looked at the data center building itself, typical cloud data centers are also much more efficient.

So although they hit the news for using a lot of electricity, they are typically way more efficient than especially older enterprise data centers that aren't fully utilized. And so we got another anywhere from 10% to 15% improvement in energy efficiency from moving to the data center as well.

So overall, we found that enterprises, if they moved a lot of their IT to cloud, they could save up to 80%, 85% of energy usage. And in some cases, even more if the cloud data centers used completely renewable carbon-free energy, you could end up saving a huge amount of your carbon usage, improving your carbon usage just by moving to cloud.

Eric Hanselman

So you've touched on 3 key points: one, that just simply the nature of equipment is more efficient. And we talk about that on the processor side, the fact that as we start rolling through a lot of these updated capabilities, silicon is getting more efficient. But on top of that, the fact that utilization is that much higher. And that's one of those things in the cloud price index, we see one of the biggest determinants of overall operating cost is efficiency.

So you got to be able to run high levels of efficiency. So cloud does that as well. And you're actually identifying the third point, which is something that Dan has talked about, which is access to a green energy grid to additionally benefit what that trends in -- not only the efficiency of what's happening.

But the content of the power itself. You got a greater propensity to be able to actually use green power if you're a big energy buyer where enterprises might not necessarily have access to that. So it sounds like there's a pretty strong underlying message here in terms of cloud migration.

Kelly Morgan

Absolutely, yes. And one that most people don't realize.

Eric Hanselman

So when we start thinking about what this -- I mean, it seems like a fairly obvious transition if you're talking about such a significant set of savings and the shift to potentially more use of green energy. And these are pretty significant benefits. Was that something that was universal? I know there are a set of regional reports that you put together. What are the regional differences across the capabilities and the savings?

Kelly Morgan

Yes. Well, I think the improvements from IT were pretty similar. We had a range from about 60% up to 67%, 68%, depending on the country. And that was according to our surveys and how often people changed out their surveys. I think one of the biggest factors is in that access to essentially carbon-free power.

So the biggest difference, I think, just from that point of view is that the European access to green power as it where is just much more available. There's more of it than in APAC. So I think -- and the U.S. is kind of in between, depending on the location. So that was one of the bigger factors.

When it came to that improvement in servers, they were pretty equivalent. I think the U.S. tended to be a little bit better. The typical enterprise virtualize their servers much more and tended to have a greater utilization of their servers, whereas the APAC and EU enterprises that we talked to were kind of similar in their usage. So that improvement from the IT equipment was about the same in the EU and APAC.

And then on the data center side, there were -- the U.S. had the most efficient data centers, at least the enterprises we talk to. And then the European ones were next and then APAC. And that's partly to just because of the climate, right? If you're in a very hot human climate, it can be really hard to have very efficient servers, very efficient data centers and service within those data centers. And so we kind of expected a lot of the APAC data centers to be less efficient.

Dan Thompson

One of the things, as I reflect on the research that we've done here, one of the things that I really appreciate about it is that we took this approach of let's squeeze out all the efficiency we can. So think here of efficiency as conservation. So we're trying to conserve the amount of energy that we're consuming.

So we did that first. So we said, let's squeeze out all the efficiency we possibly can. Let's get that energy requirement down as low as possible. And then let's look at decarbonizing that. And I think just as a general statement from a corporate strategy perspective, that's how you should always approach this. We don't want to get in a situation to just say, well, the energy is green, so I'm kind of good, right? That kind of ignores conservation, which is the other side of this coin.

So I just want to kind of underscore and articulate slightly differently, what Kelly is saying here. The idea is that we want to get that requirement, the demand number down and then from there, look at greening that. And that's one of the things I really appreciate about the way we approach this. So -- and I think it sets up as a good framework for other companies to follow.

Eric Hanselman

Well, it seems like that keeps open a lot of the various possibilities that you can potentially leverage as we've identified actually on some of the previous episodes being able to get data center efficiency crank down, it makes a real significant difference. And if we think about the transitions that are taking place in most organizations, we talk about power utilization effectiveness, PUE.

We've been in this transition for like throughout the history of data centers of grinding that down. And Kelly, to your point, there are some limits to what you can do regionally. So if you've got a Scandinavian data center, of course, your cost of cooling is going to be lower because you've got ambient air that can be used for adiabatic cooling.

If you're in hot and humid environments, you've got to work a lot harder. So it's going to limit what you can do for minimizing PUE. But the kind of advances we've made in data center efficiency are really significant we think about how we've been able to grind down a lot of those efficiencies.

Kelly Morgan

That's exactly true. But it's interesting because I think the industry and even enterprises to a certain degree, have focused so much on that data center efficiency metric and partly because it's relatively easy to at least envision, right? And there are things you can do that make immediate and somewhat sort of obvious improvements, I guess.

Whereas what a lot of people should really be thinking about is the servers themselves, right? And that's been a refrain, but that's much harder for enterprises to do to actually virtualize their servers and make sure each server is being used 80%, that's -- it's really hard to do, and it's very complicated.

And -- but that's where you would actually get a lot more gain in a lot of ways, especially as Dan was saying, when you're trying to reduce the amount of energy you're using in the first place. And so that's where the cloud part becomes kind of interesting because the cloud providers have kind of worked to do that. That's what makes their business model operate. So they've taken care of a lot of at least that tricky part of the equation.

Dan Thompson

There's also kind of this mentality of -- and I'm using air quotes here, but you can't see it, it's not my problem. So you'll have facilities managers who their responsibility is the building itself. And so they'll fixate on PUE. But to them, it's the network administrators and things like that, who are dealing with the servers. And so from their perspective, the whole utilization story is kind of not their job.

Similarly, they may look and just say, well, it's actually the utilities problem to get me green energy. I mean I'll pay for it if they can get it to me. But at the end of the day, that's kind of their problem. So looking actually on the data center floor and then outside the data center walls, there's always been somebody else's deal. Similarly, I would suggest that the multi-tenant data center industry has had a similar attitude in that it's their customers' responsibility, what's going on with the gear sitting on the floor, right?

And they have a very limited influence over that. And likewise, outside the 4 walls, if you look back 5, 10 years ago, there was definitely this kind of mentality of, well, I mean, it's the utility. What can I do? And it's only in more recent history where they've kind of realized, "Oh wait, I'm actually -- I have a lot of buying power here. I'm buying way more energy off the grid than the average enterprise. And so I can actually affect change."

Similarly, they're realizing that what happens if we incentivize our customers to actually have higher utilization rates, that helps me drive down the PUE of the facility and helps us overall from an efficiency standpoint. And so it's like there's just kind of been this awakening of, "Oh, wait, I actually can't influence this, and I should influence this."

Eric Hanselman

And is that something where it's the end consumers that are seeing that influence? You're talking about multi-tenant operators and certainly, the cloud providers have got a strong incentive. What's the extent to which you see that driven by the end consumers of those capabilities themselves?

Dan Thompson

From a power purchasing perspective, I mean, the biggest of organizations, yes, the smaller organizations who may have one data center, they may have 1 megawatt of power. I mean a megawatt of power is still a pretty -- that's still a good chunk of power, right? I mean, we're talking like neighborhood scale power. And so to the extent that even at a megawatt, like your local power board is going to pay attention to that.

And so I think you would have some pull there. But when we're talking about the major multi-tenant data center providers and the cloud providers, I mean, we're talking about tens of megawatts, if not hundreds of megawatts depending on where in the world we're talking about.

And like that definitely has the attention of the power production companies, right? Like that's a substantial load. And so like they will definitely pay attention to your request about I need a green energy option or I need to do something here from a sustainability story. How can you help?

Eric Hanselman

Well, you've identified that there are -- there's some regulatory pressure on efficiency for some operators. I think you were talking about Singapore, Perkins Liu is on talking about what China is doing in terms of managing overall power budget. So clearly, there's some shift from a regulatory perspective.

I guess the thing I was wondering is, does a lot of the ESG reporting requirements that are propping up start to influence that for the actual end consumers? And it sounds like the answer is getting that way, but we're starting to see power transparency in some facilities. And certainly, we're seeing it from the comp providers.

Dan Thompson

Yes. I mean I think one of the things that was kind of fun about this research and sort of some of the announcements that were coming out along the way, a number of the public cloud providers have come out with programs and are continuing to develop programs where they show the end user, what the "greenness" of that particular workload is.

And so the end users are now being armed with at least the data to make a decision to say, I actually want to make this as green as possible, and that's a priority for me. I'm sure user experience and costs are going to be just ahead of it at least now as it pertains to workloads, but in the future, who knows. But anyway, we're at least now armed with that data, whereas previously we wouldn't have been.

And similarly at the data center front, there are companies that are coming out with software packages that can show you the greenness of the energy that you're consuming on an hour-by-hour basis. That's another change that we've seen over the last 12 months.

Is the cloud providers anyway are saying an annualized accounting is not good enough. We want hour-by-hour accounting, and we want to work towards an hour-by-hour matching as to where our green energy is actually covering our or accounting for our demand on an hour-by-hour basis.

Eric Hanselman

Getting the same kind of granularity you expect in cloud?

Kelly Morgan

Right. And just to add to that in terms of regulatory requirements. I think you're right that companies, enterprises are increasingly looking at regulatory requirements for being sustainable or at least investor requirements for how sustainable they are. The biggest cloud providers and even, I'd say, some of the biggest enterprises have already kind of come under some of that pressure from Greenpeace and firms like that.

But also just more broadly now, a lot of investors are looking at larger enterprises and asking what their green story is, what is your sustainability, their sustainability indexes and things like that. So there's just generally a lot more pressure for enterprises to look at where they can make sustainability and efficiency improvements. And so this kind of story could then be really interesting and appealing to them.

I think another thing that you referenced was actually we've seen Singapore, Amsterdam, China, a lot of these big data center markets worried about the amount of energy use from these data centers. And I think in Amsterdam, for example, in some of these markets, the big data center providers and the big cloud firms have argued more or less successfully that, look, you're not actually paying attention to how much energy is already being used by enterprise data centers.

And in fact, if all these firms shifted to these bigger, say, leased multi-tenant or cloud data centers, you actually would end up saving energy if you shut down all these really super inefficient old enterprise facilities. So there's kind of that regulatory aspect, too, even though sometimes governments are worried about one thing. The efficiency is not always obvious and moving that electricity use from one to the other might actually make a big difference.

Dan Thompson

So it's kind of interesting, really that what users have laid out in front of them now is sort of an obvious way forward for the decarbonization of their IT assets. So the best place to go from a carbon perspective is the cloud. We know that not all workloads can go to the cloud. And so the next best thing may be a multi-tenant facility whose PUE will be lower, who's just utilization of the entire facility will be better.

Their access to green energy will be better. And then if that won't work, then perhaps now we consider a corporate data center, enterprise data center just because of perhaps proximity or maybe it needs to be in a second tier, third tier city. And they're just -- there really is no good colo option available or the colo option is actually no better than your enterprise facility.

And so I think if we can begin to think of it in terms of if our priority is greenness and I will fully can see that most enterprises are not there yet, I hope we get there. We need to get there. If that's our kind of order of operations, it would go cloud first, then multi-tenant second.

And then hopefully, whatever crumbs are left over, we can find a place to put that that's at least as green as possible for the quality of user experience as necessary, which obviously includes things like latency and just general proximity, things like this for your workers. And in any case, I think that's a great mindset to kind of carry forward and think through as companies are planning a strategy for this.

Eric Hanselman

Kelly, I want to get back to your point and Dan, certainly, from the facilities perspective, the workload placement, all those things are important. But I think you've made what is the fundamental and most important takeaway that hopefully, listeners are getting out of this, which is that it's the efficiency of the servers themselves. And that it's that aspect that is the biggest component of what you got out of the study, right?

Kelly Morgan

Absolutely, yes. And that is something that most enterprises don't recognize, I think. And even if they do, it's tricky to improve themselves. And one of the things that we looked at was all the different types of servers, and I think you've talked about that on this podcast as well from the chips and the various servers and the improvements that can be made.

So to a certain degree, even if enterprises wanted to boost that efficiency, it's really hard to attain what a lot of the cloud providers have. And it's just a lot of work and a lot of enterprises think maybe this is not the best place to put all my energy. So there is a lot of gains to be had there.

Eric Hanselman

Well, the study also looked at refresh cycles -- and the fundamental point there is that cloud operations are simply refreshing hardware at a rate that your average enterprise can't keep up. Because I think what the study was identifying was somewhere in the 4-year-ish range for a refresh cycle.

Some variation. And hey, if you've got a server that's 4 years old, that's going to be the technology marched on, efficiency improve, all of those good things happen. And now you've been significantly outpaced.

Kelly Morgan

That's exactly it. And also, the fact that cloud providers are typically buying so many servers. So even if they're refreshing them 2 to 3 years at a time, but they're also just buying brand new ones and continuing to expand. So the overall sort of number of service that they have that are the latest and most efficient, it is just going to be a lot higher than most -- almost any enterprises.

Eric Hanselman

Yes. Your average age gets driven down and it's not the sort of episodic buying that enterprises tend to do.

Kelly Morgan

Exactly.

Dan Thompson

Something I want to kind of underscore here, Eric, just to kind of get into the weeds just a little bit if we can. What's interesting about this is, as I've talked about this data to folks, a common pushback that I've heard is, well, I don't think that's true actually. Like if we look at it, Moore's Law is over with. The gains that you see in new generations of chips are just not that great.

And it's as though the person is kind of trying to approach the argument from a performance perspective or a transistor perspective, which is not actually a good argument because what -- if you see, yes, of course, the gigahertz numbers, those are not jumping leaps and bounds like they used to. The number of transistors and resistors used on chips.

That's not doubling every 12 to 18 months or whatever, that may all be true. However, if you look at the ability to do work, so the power or processing power per watt, so the amount of energy it takes to do that processing, that's what's coming down. And so if you look -- so that's just kind of a general way of thinking of the efficiency of that chip. And so these chips, again, while they may not be getting faster and faster by leaps and bounds, they are getting more efficient by leaps and bounds.

And that's one of the things that we actually charted out for this for this research project and kind of underscore that, yes, these newer generation processors do actually consume far less energy on a generation-by-generation basis. But when you start talking about over a 4-year period, it's really dramatic.

And so that's -- if some of you are listening, you're thinking you're somewhat skeptical about this and say, I don't understand how there's a 60% reduction just from changing chips. That has been really where the gains of these new chips have happened.

And it doesn't make headlines because, number one, it's a little bit kind of hard to explain. And number two, it's not sexy. It's like I'm not adding horsepower. I'm basically adding fuel economy. It's just not a sexy of a topic to talk about in the magazines and websites and whatever. But in any case, that's what's happening behind the scenes.

Eric Hanselman

Well, if you think about the -- what we get in a 150-watt, 200-watt socket, now we're just -- all you have to do is look at the number of cores. And that gets driven by smaller geometries and the silicon that's there, which result in lower voltages, which fundamentally means we're just moving fewer electrons around, which is more efficient.

Dan Thompson

That's right. That's right. Yes.

Eric Hanselman

Wow. Well, fascinating stuff. And hopefully, eye opening for a bunch of our listeners. Thank you both for the insights that all this provides. Hopefully, we'll provide some links in the show notes. Hopefully, our audience will get a chance to actually look at some of this data and see it and actually start to act on it, hopefully. I will keep our fingers crossed that it catalyzes some action.

Kelly Morgan

Absolutely. Thank you so much for having us.

Dan Thompson

Yes. Thanks for having us on again.

Eric Hanselman

And that is it for this episode of Next in Tech. Thanks for our audience, for staying with us. Join us for our next episode when Dan Kennedy is going to be back on, and we'll be discussing the results of the latest information security budgets and outlook. And we'll get Dan's perspective on what studies showed some of the data that he's been working on. I hope you'll join us then because there is always something Next in Tech.