From pv magazine Global
Data centers come in many sizes. The largest, China Telecom’s Inner Mongolia Information Park, spans 100 hectares and consumes up to 150 MW per hour. North Virginia, in the United States, houses around 300 facilities in a grouping known as Data Center Alley, with each consuming about 10 to 50 times the energy per square meter of a typical commercial office.
Utility Dominion Energy was forced to pause grid connections for new members of Data Center Alley in 2022 and is now constructing new transmission lines to meet demand.
The United States has more than 5,000 data centers and consultant McKinsey & Company expects their power consumption to rise from a peak 17 GW/hour, in 2022, to 35 GW/hour in 2030.
Scaling up
Data centers are becoming more high density and power intensive but also more efficient.
“The hyperscale cloud providers all seem to be locked in an arms race to build out as much infrastructure as quickly as they can,” said Dan Thompson, principal research analyst at S&P Global Market Intelligence. “Some of this is high-density, high-performance, compute-type deployments, but a lot of it is also the cloud providers building out at scale. Densities in watts per square foot are rising, but I think what we’re seeing right now is just the tip of the iceberg.”
Data centers have a power usage effectiveness (PUE) ratio, which dictates how much energy is needed for computing versus other activity, such as cooling, lighting, and power losses. A PUE of 1.5 would indicate a data center requiring 500 kW of extra power for 1 MW needed for computing purposes.
S&P’s Thompson said power densities have fallen from an average 1.58 in 2020, as power density and cooling efficiency have risen. The lowest values, however, involve some trade-offs.
What’s in a data center?
Data centers house dense racks of servers containing processing, storage, and network equipment plus supporting infrastructure and cabling for power supply and power-intensive cooling, via air, water, refrigerants, or non-conductive liquids. The rise of artificial intelligence is changing data centers, which now include more specialized hardware accelerators for intensive tasks, high-performance computing infrastructure, and increased power consumption. Dedicated AI buildouts can require as much as five times more fiber optic cabling.
“The data centers we’re seeing built now are designed for PUEs of 1.3 to 1.4, so you can see some improvement there,” said Thompson. “That said, while they are designed for those PUEs, many factors could cause the building to never actually realize that PUE, depending on climate and operations. We have seen some constructions with a designed PUE of 1.15 to 1.2, however these facilities require the consumption of large volumes of clean water to reach those numbers. Given the issues around access to clean water, hyperscalers and the companies building data centers for them have tended to build slightly less efficient data centers for the sake of using very little or no water.”
The world’s technology giants are the biggest corporate power purchase agreement (PPA) buyers of renewable energy. On March 1, 2024, Microsoft and asset manager Brookfield signed a record 10.5 GW deal to deliver solar, wind, and “new or impactful carbon-free energy generation technologies” to Microsoft from 2026 to 2030.
Microsoft says its CO2 emissions are now up 30% from when it set its 2030 net-zero target, in 2020, and mainly because of data centers.
“The rise in our scope 3 emissions [from third-party, supply chain companies] primarily comes from the construction of more data centers and the associated embodied carbon in building materials as well as hardware components such as semiconductors, servers, and racks,” said Microsoft, adding that the 10.5 GW renewables PPA is on top of a 19.8 GW clean power portfolio.
Simon Maine, managing director for communications, renewable power, and transition at Brookfield, told pv magazine that the deal was eight times bigger than any previous PPA.
“We have a very large renewable power and transition business, with over $100 billion of assets in that division alone, and 30-plus-years’ experience in the sector,” said Maine. “We look to either buy assets or, more recently and more likely, buy companies. The companies will have high-quality management teams that have a full spectrum of capabilities. We have projections to install somewhere between 5 GW and 7 GW per year [to 2030]. The deal with Microsoft probably covers about 30% of that growth and that’s without factoring in further acquisitions.”
Brookfield is reported to have acquired a majority stake in India’s Leap Green Energy for $500 million, and is also said to be preparing to acquire Australian renewable energy developer WindLab, which has around 24 GW of projects in development or under construction.
Anas Papazachariou, senior PPA manager at renewables developer Cero Generation, explained how colocation can meet data center energy demand.
“A single solution where solar meets the full increase from the growing number and size of data centers is probably not optimal and I have to be honest about that,” he said. “So a lot of the offtakers are looking to create virtual portfolios whereas wind and solar, and combined batteries, are part of their portfolio because they’re actually optimizing their profiles through that basis.”
Solar-plus-storage means more expensive energy offtake agreements, but reduced risk, said Papazachariou.
Efficient clusters
“Hyperscale” data centers are clustered for efficiency. Where latency is concerned, however, many other data centers, especially those serving internet and network services, are distributed closer to population centers. These are smaller and experience more variation in demand.
Mike Bates, general manager for the Intel Energy Center of Excellence, said data centers are using workload management software that can respond to real-time energy conditions. Intel is deploying software inside data centers to manage workflows and loads, while also tracking carbon footprints of workloads for audits by companies claiming low carbon or net-zero workflows.
“One of my customers is the [internal] Intel Data Center group and we work to deploy these same solutions we’re taking outside of the market, making sure that we’ve hardened data centers for climate impacts while opening up new opportunities as well,” said Bates. “For example, our software is also able to adapt workloads for certain conditions. If I can push a workload inside the data center to the times when energy is in surplus, I can actually get paid to consume that energy.” He added that energy resiliency also includes interruptions to supply, when considering climate impacts.
Ben Levitt – associate director for the gas, power, and climate solutions North American power and renewables research team at S&P Global Commodity Insights – highlighted the cost benefits of operating data centers with flexibility.
“Data centers with flexible operations – that is, interruptible, price-responsive – cost less to supply than ones that are less flexible,” said Levitt. “Data centers that are interruptible might even be able to get a faster grid connection. In addition, and separately, it is possible that big tech may drive investment in developing and scaling the new, ‘clean firm’ technologies needed for around-the-clock clean energy for their data centers.”
Levitt said new loads will lead to new renewables investment but fossil fuel generation, and increasingly batteries, will also pick up extra demand. Ultimately, a lot will depend on local bureaucracy and permitting.
Levitt added that it is possible big technology companies will play a role in scaling new clean technology. “These efforts could accelerate the development of newer technologies that could reshape energy supply mix at a faster pace than previously considered,” he said.
This content is protected by copyright and may not be reused. If you want to cooperate with us and would like to reuse some of our content, please contact: editors@pv-magazine.com.
By submitting this form you agree to pv magazine using your data for the purposes of publishing your comment.
Your personal data will only be disclosed or otherwise transmitted to third parties for the purposes of spam filtering or if this is necessary for technical maintenance of the website. Any other transfer to third parties will not take place unless this is justified on the basis of applicable data protection regulations or if pv magazine is legally obliged to do so.
You may revoke this consent at any time with effect for the future, in which case your personal data will be deleted immediately. Otherwise, your data will be deleted if pv magazine has processed your request or the purpose of data storage is fulfilled.
Further information on data privacy can be found in our Data Protection Policy.