Data Centers and Green Energy

You can't turn on the news without hearing of this or that tech company promising "Zero Carbon" or using "100% Renewable Energy".  So, I wanted to talk to a data center expert about just how the tech companies are approaching their commitments. 

David Mytton is back to talk about what he has seen of data centers and 'green' energy.

Email us: powerbytes@cat.com
LinkedIn: https://www.linkedin.com/showcase/cat-electric-power/
Facebook: https://www.facebook.com/Caterpillar.Electric.Power/

Transcript

Lou: Intro: Good Day and welcome everyone to Power Bytes!  I am your host Lou Signorelli and Power Bytes is your destination Podcast for power generation conversations.  As always please know how much we appreciate you, our listeners.  We hope you find our topics helpful and interesting.  There are several ways for you to get in touch with the show. You can send us an email at powerbytes@cat.com, visit us at Cat Electric Power on Facebook or LinkedIn.  Please remember to subscribe wherever you listen to our show… it really helps. 

Lou: You can't turn on the news without hearing of this or that tech company promising "Zero Carbon" or using "100% Renewable Energy".  So, I wanted to talk to a data center expert about just how the tech companies are approaching their commitments.  

Lou: Joining me again is David Mytton.  You may recall David's last appearance on the show when he talked about how the Data Center industry is responding to the Data Boom.   

David is Co-founder & CEO of Console, a weekly newsletter which helps software developers find the best tools, and a recruiting platform for developers not actively looking for a new job. He was also a Research Affiliate on the data center sustainability team at Uptime Institute and recently started in the Department for Engineering Science at the University of Oxford where he is researching sustainable computing.

 David has appeared on national TV (e.g., Sky News, Channel 4 Dispatches) and been quoted in international news media (e.g., WSJ, Financial Times, Computer Weekly). 

When you have a chance, I encourage you to have a look at his blog at Davidmytton.blog 

David, welcome back!  

David: Thanks for inviting me back, Lou. 

Lou: Let's start with the claims we hear about data centers using "100% renewable energy"?

 David: This is an interesting term because it’s both technically accurate, but also very misleading. Indeed, in almost zero cases does it mean what you think it means.

 The challenge is that the demand for electricity on the grid needs to be balanced perfectly with supply. This changes throughout the day as people go about their normal activity, the weather changes, and we experience season cycles.  Grid operators deal with this through having baseload power which is always generating electricity – nuclear is the best option for this – with what’s known as dispatchable power – something like gas, which can be turned on quickly and at short notice. Other forms of power generation fill the gaps too.

 This becomes more difficult when the source of power is intermittent i.e., we don’t have much control over when it generates. Wind and solar have this characteristic because we don’t control the sun or the wind. They’re great because of their very low carbon intensity, but there can be long periods where they don’t generate any electricity, like at night or when there is calm weather.

 If the period of intermittency is short – a few hours – you can make use of batteries that were charged when the renewable power was available. But long duration battery storage is still very expensive or just too experimental to be in operation at scale. So, during those periods, grid operators will fall back on more reliable forms of power, like gas and coal.

 Which brings us back to the 100% renewables claim. If there are periods throughout the year where there is no renewable power, or insufficient renewable to be powering 100% of demand, then how can companies claim they use only 100% renewables?

 Well, most of these claims are done through accounting mechanisms.

 The first option is carbon offsets, which is supposed to net out your emissions to make you carbon neutral. There are lots of problems with carbon offset quality and using offsets doesn’t mean 100% renewable.

 The next step is to match your annual electricity consumption with equivalent purchases of renewable energy credits (RECs). RECs are financial instruments which mean that renewable energy was generated somewhere and some point. There's nothing to say the associated renewable energy was in any way linked to your demand or consumption. As a result, they’re of low cost and low quality, so are widely considered greenwashing if they’re the only component of a renewable energy strategy. RECs may be associated with much higher quality Power Purchase Agreements or green power contracts, but unbundled RECs are not a credible way to achieve sustainability goals.

 All these can be part of a portfolio of measures as we go along a journey to sustainability, but the only way to actually do what consumers expect when you say 100% renewable, is to have carbon free energy available on the grid 24/7. This is really challenging, and nobody is doing it today unless they are one of the few facilities that have a direct connection to something like a thermal or hydro power plant that can provide reliable carbon free energy.

 Lou: There is term I'm hearing more about now: Carbon Free Energy.  Is there a shift in the industry's thinking?

 David: At a high level, no. Most people are still talking about renewables, by which they mean wind and solar. Maybe hydro. But there is a small but growing cohort that is more practical in their thinking about what are we trying to achieve – carbon free energy. Wind and solar are great in many ways, but as I just discussed, their intermittency and variability are big challenges.

 Clean energy means more than just wind and solar. It can mean energy stored after being generated by clean sources of energy, it can mean nuclear, it can mean biogas or hydrogen. The key is that it is generated with low or zero carbon.

 A good example of this is how the Renewable Energy Buyers Alliance renamed in Nov 2021 to the Clean Energy Buyers Alliance. Or how the UK and Japan are working on extending the life of their nuclear power plants, or even building new ones.

 Lou: Has any of this been modeled or even attempted or is it all simply theory at this point?

 David: Google has funded two studies which published results recently. The first was from Princeton University in Nov 2021 and the second came out in Oct 2022 from TU Berlin. Both used Python software models – two different models in fact – to look at the impact on cost and carbon as the grid approached 100% carbon free energy. The goal was to see how the system responds to demand and supply, and in particular to compare cost and carbon emissions vs the existing model of using 100% annual matching.

 Both studies came to similar conclusions: Reaching 90-95% carbon free energy can be achieved with only a small cost premium compared to 100% annual matching. However, pushing closer to 100% significantly increases the cost to more than double.

 This is a positive result in many ways, but also shows there remains a challenge to hit that 100% goal. There were some big assumptions about the availability of economical long term energy storage, which I hope we will develop, but has yet to proven.

 I also found it concerning that nuclear was only properly considered in the US study. This certainly reflects European politics, particularly in Germany, but I think it’s a mistake. Nuclear needs to be a core part of the future energy grid, and it seemed like a missed opportunity to show how nuclear is a good option for the goal of carbon free energy. Europe remains under the shadow of decades of anti-nuclear activism.

 Lou: The Energy grid, no matter if you're in Europe or the USA, is complex, Do you see a role for Data Centers when it comes to helping to stabilize the grid in general?

 David: This is a fun question because over the last decade there has been so much academic literature on data center demand response. Academics seem to love the idea of how IT workloads can be moved around the world to follow the sun, or at least follow the clean energy, and always be as close as possible to the cleanest, lowest carbon energy.

 And it makes sense. We have data centers all over the world and using the cloud makes it really easy to spin up a new server in any number of global regions. Hourly fees might differ slightly between regions, but the cloud makes it really easy to access resources wherever we want.

 Lou: So why have almost no data center operators adopted demand response?

 The fundamental reason is that data center operators don’t care about the small additional revenue, especially when it’s at the expense of reliability. The fundamental purpose of a data center is to provide a highly reliable environment to run IT equipment. If the facility could suddenly lose power and either turn off entirely, or more likely, have to shift to backup power, then that increases the risk of something going wrong. The additional revenue or energy price discount from participating in demand response does not offset the requirement for reliability.

 The cloud is a bit different because you design cloud systems  assuming failure. This is why you have the concept of multiple availability zones – which are essentially independent data centers in a small geographic area. You can mitigate failure at a server or facility level by deploying across availability zones, accepting the risk of a geographic event like an earthquake within the region. Cloud providers make it cheap to move resources around within a region, so you can replicate data and build reliable services.

 But the zone granularity is insufficient for optimizing carbon – you need to move workloads over much greater distances where the weather is different, or it’s daytime vs night. The types of workloads that are most appropriate to move – those which do not need real time responses – are also most likely to be the ones using lots of data. Think training AI machine learning models, or processing analytics, or running calculations across datasets. They all run against huge volumes of data. And that is very expensive, and slow, to move around.

 This means it’s not economic to be constantly shifting huge volumes of data around the globe to benefit from cleaner energy.

 Finally, privacy becomes an issue. The US is unusual because of how big it is, yet it’s all basically a single country from a privacy perspective. There are a few state differences, but that’s nothing compared to, for example, moving data from the UK to Germany. Individual countries within Europe are too small, so the workloads have to move across countries and so hit privacy differences, even within the EU. Germany has much stricter requirements compared to the rest of the EU, for example. 

 These are all the problems academics miss. It’s theoretically possible, but practically it doesn’t currently make sense.