THE SIZE OF OUR INDUSTRY
For more than a decade, I have watched, read, and been involved in hundreds of discussions about the size of our industry. How many data centers are there? How are they classified? How much capacity is built? How much power is used? Who is using it? How efficient is it? What’s the carbon footprint? How fast is it growing? All of these are very important questions, but we never seem to have clear, defensible answers. Most are estimates based on extrapolations of data that only see part of the picture.
The last large industry debate I was involved in was in London in 2019. Christian Belady, Amber Caramella, Gary Cook, Heather Dooley, Dave Johnson, Bill Kleyman, Martin Lynch, Patrick Ohlund, George Rockett, Eddie Schutter, Noelle Walsh, and many more participated in panels at DCD as well as the iMasons’ member sumit and Advisory Council session.
The result? Data center global energy consumption estimates ranged from 1% to 10% with most saying it was below 2%. The issue is that these are still estimates, not definitive answers. On the other hand, hyperscalers like Apple, Amazon, Google, Meta, and Microsoft have accurate numbers since they build and operate their own data centers. They know the capacity, consumption, and efficiency down to the second because it matters to their bottom line.
In the beginning of 2021 Synergy claimed there are 600 hyperscale data centers, doubling since 2015. In November of 2021, Bill Kleyman reported that traditional data centers have shrunk from 50% to less than 25% of the market replaced by hyperscale data centers growth. That development continues with gigawatts of new capacity in the pipeline.
The sessions in London highlighted significant progress but still did not answer the main question. Data centers are more efficient and sustainable, and the world is more dependent upon them than ever, but we cannot quantify by how much.
In my opinion, the debate continues because we have not agreed on how to scope the digital infrastructure industry or established a taxonomy that clearly classifies how to measure it. Without these, we cannot set a baseline and show progress.
In the summer of 2021, I worked with Rob Aldrich, the chair of the Infrastructure Masons Sustainability Committee, as well as other industry experts to tackle this problem. Over the last six months we researched, discussed, debated, and ultimately aligned on a proposal.
DEFINING DIGITAL INFRASTRUCTURE
I started Infrastructure Masons in 2016 to unite the builders of the digital age. Our members represent some of the largest portfolios in the world. They build and operate the foundation of the internet of everything. Their portfolios are the engines that enable consumption of digital services world-wide. Yet if we limit the measurement to enterprise and hyperscale facilities, we are missing more than half of the global consumption. We found that the traditional definition of a data center was too narrow. We expanded the aperture to include all elements that make up the Infrastructure that delivers digital services, i.e. the IT equipment that processes and stores data, the facilities that house that equipment, and the networks that connect them all. This led to our first definition in the project.
What is Digital Infrastructure?
Digital Infrastructure is a collection of data center locations that deliver electronic services to people and machines.
We agreed that Digital infrastructure is made up of unique locations that have specific addresses just like the street address at your home or a network address on the internet. Duplication of these addresses would cause conflicts and potentially double counting of consumption. Like apps in the app store, those services constantly change with nearly unlimited permutations over time but they all consume infrastructure capacity. We kept this definition simple and concise to ensure that both technical and non-technical people could understand. Once we defined Digital Infrastructure, we needed to answer the next question:
What are Data Centers?
Data Centers are real estate locations that house IT equipment to process, store, and transmit data.
We’ve established that Digital Infrastructure is a collection of data centers and that each data center is a real estate location with a unique address. These locations house IT equipment that do one or more of these things-process data, store data, and/or transmit data. Every data center in the world is covered with this definition.
Finally, we needed to answer one more question to be able to establish a baseline of global consumption.
How are Data Centers Classified?
Data Centers are classified into three primary categories–providers, networks, and crypto.
We held five different iMasons working sessions to review and fine tune the structure. Data Center classifications and their definitions were the largest debate.
Providers include data centers that deliver services to themselves, others, or both. For example, Cloud, Hyperscale, Enterprise, Colocation, and Edge. This category was understood and agreed upon.
Networks include data centers that operate as carrier hotels, internet exchanges, fixed networks, and mobile networks. This category sparked numerous debates as carrier hotels and internet exchanges are usually intermingled with enterprise, colo, cloud and even hyperscale data centers. Our assertion is that a real estate location can have a unique address yet house multiple tenants such as a restaurant on the ground floor and apartments on the upper floors. The key is to ensure that the real estate location with a unique address accounts for all types of classes without duplication. Meaning, a data center could house both providers and networks in the same address, but consumption of power is specific to each category in that building.
Crypto includes all crypto mining and blockchain services. This is a new category that has grown rapidly over the last five years primarily due to bitcoin mining. While blockchain is the base technology, bitcoin represents over 80% of consumption today. Adding crypto to digital Infrastructure was the most controversial part of our proposal. The reason we kept it as a stand-alone category was due to growth forecasts. Crypto power consumption could surpass the other categories as services based on blockchain continue to expand. Purists do not believe crypto should be a data center category as it does not fit into the traditional definition of data center. E.g. crypto deployments do not have resiliency built in. No generators, UPS or redundancy to protect from faults. Our assertion is that Crypto sites deliver electronic services to people and machines just like providers and networks. The difference is that they do not require the same level of resilience in the data center design based on the distributed nature of blockchain.
The most important element of this Taxonomy is that every data center site will have a unique identifier-i.e. an address. This is critical as we will not be able to establish a baseline and measure progress if there is overlap or duplication. This approach already works for many other things in our society-every real estate location has a unique street address regardless of its location, size or use. The internet is based on TCP/IP which ensures that every device has a unique address in order to communicate.
ENERGY CONSUMPTION & CALCULATION
Based on the taxonomy outlined above, we can now see Digital Infrastructure in its entirety. For our work we first want to baseline power capacity and consumption. Statistica reported in 2021 that there are 7 million data centers globally. Each has a unique street address ranging from mega data center sites exceeding 1GW of capacity to micro edge deployments on the corners of streets drawing less than 1kW of power. These 7 million data centers represent 105GW of power capacity. In 2021, they consumed 594TWhs of energy globally. That represents 2.4% of global energy consumption.
Data Centers consumption breakdown for 2021:
- Networks consume 266TWhs
- Providers consume 213TWhs
- Crypto consumes 115TWhs
594TWhs of consumption translates to 67,808MW of capacity that is consumed at any given moment. To put that in perspective water processing and delivery consumes 13% of global energy. Digital Infrastructure consumes 2.4%.
Cambridge Bitcoin Electricity Consumption Index (CBECI.org) reported that bitcoin dropped from 140TWhs to 115TWhs when China banned crypto mining in Q2 of 2021. That capacity is currently migrating from China to many other regions including the US, EMEA, and LATAM. Crypto consumption will continue to increase significantly. As of the printing of this article, CBECI reported Crypto had risen back to 131TWhs as much of the China capacity has found a new home and additional capacity has been added.
Efficiency
Energy consumption is only one part of the story. We made some very conservative estimates on utilization to back into the total global capacity number. If we assume that Networks and Providers continuously utilize 60% of their built capacity and Crypto utilizes 97% of their capacity, we have an average of 65% utilization globally. That means 68GW of consumption translates to 105GW of built capacity. That leaves 37GW of capacity that is unused. We applied an average build cost of $7M/ MW for networks, $5M/MW for providers and $250k/MW for crypto. Based on these costs, 37GW of unused capacity translates to over $220B of investment dollars that are not yielding returns and more than $20B in revenue opportunities left on the table. While some capacity is needed to be ahead of demand for rapid growth categories such as cloud and social media, many agree that utilization is actually lower than our estimates leaving even more stranded capacity. Bottom line, there is a significant opportunity to increase efficiency and yield from investments.
Chris Crosby, CEO of Compass Data Centers, raised another efficiency area that should be considered. Utility transmission and over-generation. First, many large colo, hyperscale, and cloud providers invest in substations to serve their properties avoiding up to 20% of transmission losses. Second, many renewable energy sources overproduce capacity due to their variability. Technology to store this capacity cost effectively at grid scale has not been achieved yet but it is expected in the near future. Today, some providers are optimizing workloads to utilize this overgeneration as it happens. Crypto mining located at the generation source, like wind and solar farms, can ramp up becoming carbon negative during over-generation and avoid wasting that energy. Cloud and edge providers are applying carbon-aware kubernetes to dynamically move workloads to low-carbon areas to achieve the same thing.
Growth
Correlating input from data center equipment providers, colo, providers, construction firms, and edge capacity reports shows that 20,000MW of new capacity will be built in hyperscale, colo and edge by the end of 2024. That growth will be led by India, Africa, and Latin America. One of the concerns with this growth is that we will be perpetuating the low utilization model across the globe. This is not sustainable economically or ecologically.
Sustainability
On earth day in 2020, Infrastructure Masons launched a unified Sustainability Vision-Every Click Improves the Future. We believe that Digital Infrastructure contributes to society and the economy without harming the planet.
This vision was written to support sustainable business practices applying triple bottom line-people, profit, and planet. Environmental sustainability is the responsibility to conserve natural resources and protect global ecosystems to support health and wellbeing, now and in the future. Sustainability is a huge topic. For this effort, we are focusing on establishing a global baseline for data center power capacity and consumption to measure the industry’s carbon footprint. Our industry accounts for 2.4% of global power consumption, yet we are unable to show our carbon footprint-good or bad. By establishing the taxonomy for our industry, we can enable real time carbon accounting for Digital Infrastructure. This requires industry buy-in to measure embodied carbon and carbon attributed to consumption of power to deliver digital services.
We have three proposals that are enabled by the new taxonomy.
Digital Infrastructure: Source Power
Carbon tracking of source energy attributed to each unique data center location.
Every data center location has one or more power feeds to enable the data center to run. Each of those power feeds has a carbon footprint associated with the source. This is the easiest measurement as most data centers have this tracked in real time.
Digital Infrastructure: Carbon Sticker
Carbon stickers showing embodied carbon for data centers facilities, ME&P components and IT equipment housed in those data centers.
Martin Lynch, CEO of Pure Data Centers raised this brilliant idea at the iMasons End User Forum in October 2021. Each food product that you buy in a store has a nutrition facts label. That breaks down what is in that food product. I.e. serving size, calories, percent of daily value, and nutrients. We propose an equivalent “Carbon Facts sticker” that outlines embedded carbon for data center buildings, ME&P equipment, and IT equipment housed in those data centers.
The carbon facts sticker included here is for illustration purposes to show the concept of applying this to a data center. The Serving size is one data center building. Calories equate to metric tons of carbon (MTCO2e) for each element in building that data center. For example, how much embodied carbon is attributed to building the data center or manufacturing, delivering and installing a UPS, generator or switchgear. How do you attribute carbon to data center retrofit and expansion projects? This also can be applied in the same manner to any equipment in the data center. Each UPS, generator, switchgear, server, network switch, etc should have their own carbon facts sticker. The combination of the building and its components will represent the total embodied carbon footprint for that location. Additional carbon will be applied to this sticker over time as new equipment comes in and old equipment goes out and the data center is modified to deliver services over its lifetime.
Startup company Molg.ai has started to apply carbon sticker concept through their OriginMark to make a more transparent and sustainable world. CEO Rob Lawson-Shanks is planning to take this even further creating dynamic autonomous micro-factories to manufacture circular products leveraging artificial intelligence and robotics. Their OriginMark is a critical component to make this happen.
This proposal is to get people thinking and manufactures and builders to dive in. Movement is already happening as industry leaders like Schneider Electric and Vertiv are actively pursuing this concept. Due to recent feedback we are considering changing the sticker format to align with Scope 1, 2, and 3 structures to help accelerate development and adoption.
Digital Infrastructure: Real Time Tracking
Embodied carbon stickers plus source power would enable real time global tracking of the digital infrastructure industry’s carbon footprint.
Carbon stickers for the facility and devices in those facilities can now be added to the carbon associated with the power consumed to deliver those electronic services. The result is real time accounting for carbon for each data center location. But we want to take this one step further and standardize on a unit of measure that does not care about what services are being consumed in the data center. For example, every data center has network connectivity. That connectivity has associated packets that are transmitted and received in the data center. Those packets would have a fully loaded carbon tax based on embodied carbon tracked in the carbon stickers and carbon associated with source power consumed by devices in the data center. Since the digital infrastructure taxonomy ensures every data center has a unique address, the result is real time measurement of carbon for each packet in and out of every data center in the world with no double counting.
Real time carbon tracking is already being applied at a smart city level. Adam Kramer, CEO of startup company Ledger8760, has developed a platform that tracks 24/7 carbon emissions in real time. This can easily be applied to data center locations that implement the digital infrastructure taxonomy and carbon stickers.
INDUSTRY COLLABORATION
Collaboration has never been so important as we wrestle with the challenges of achieving net zero with a rapidly shrinking time frame to slow climate change. “Every click improves the future” is our vision-uniting to work together as an industry is essential if we are going to make this a reality.
The concepts outlined in this article are meant to get the creative minds in our industry thinking about how to apply these concepts. We have multiple sessions planned with iMasons members and partners to continue the deep dive into carbon accounting and lowering the impact of Digital Infrastructure. I encourage everyone to join us!
If you would like to be part of this journey of discovery, invention, and innovation, become an iMasons member at https://imasons.org/join. If you’re already a member, join the sustainability group through our member platform and join us at our upcoming events.