Brain Trust: Microgrid Industry Thinking Through AI Data Center Demand Crisis
The energy transition crossroads of reliability and sustainability now has a major traffic jam years in the making and years from fixing.
Utility-scale power generation capacity is losing ground to meet demand from the rising forces of data center growth, electric vehicles, industrial electrification and onshoring of manufacturing.
Sometimes all good things may work out in the end, but getting there will take some deep thinking and huge commitment of resources.
Many issues are coming to a head fast, but the first challenge mentioned here may be the deepest and most immediate. The data center industry’s runaway growth is necessitating the need for at least another 21 GW of energy to match construction of facilities coming online. Research by financial firm Goldman Sachs predicts that new load could be closer to 47 GW by 2030.
This is a critically big number. One GW (1,000 MW) is comparable to the average size of a conventional nuclear power plant. The commissioning of Georgia Power’s new reactors Vogtle 3 and 4 notwithstanding, the American grid power generation industry is not building 47 new utility-scale nuclear plants anytime soon, if ever.
Hardly anyone saw it coming
And the data center expansion is not subsiding anytime soon, nor are the long-term visions for widescale transportation electrification and accelerating movement of manufacturing back to the U.S. The stunning rise of artificial intelligence technologies, for starters, is inspiring a race among tech giants, but one which can be impeded by utility-scale generation and transmission hindrances.
An irresistible force unanticipated until recently is coming right at an object considered immovable only a few years ago.
“The answer to this will be critical to the trajectory for AI power demand,” reads one line from a new research report by financial firm Goldman Sachs.
Neither the financial nor utility sectors saw this coming all that long ago. In anticipation of an ascendent triangulation featuring a renewables-dominant electricity mix, sustainability goals and flat load growth, the industry began retiring coal-fired and nuclear power plants at an accelerating rate for the past two decades.
Gas-fired combined cycle generation is being added at a record pace, due to both its reduction in carbon dioxide emissions (compared to coal) and historic availability due to the shale drilling revolution. This shift from one fossil-fuel generation to another is curbing some emissions, but hardly leads to net zero and now results in natural gas accounting for a dominant 40% of utility-scale power generation.
Grid power planners worry about resource adequacy and fuel diversity to meet the needs of a more electrified and (somewhat) decarbonized society in the future. The enormity of the challenge is forcing many industry insiders to reconsider what the energy transition portfolio will look like.
“Data centers are rapidly increasing their power demands, but they aren’t alone,” Ian Walch, director of strategic partnerships and solutions at Houston-based microgrid developer Enchanted Rock, said. “Onshoring manufacturing needs an extra 13 TWh (terawatt hours: one TWh is comparable to 1,000 GWh). By 2050, EV fleets will require 2,000 TWh per year.
“To ensure power availability doesn’t hinder the growth of these essential industries, reliable fossil-fueled generation, such as natural gas, is crucial,” Walch added.
A dilemma of utility-scale proportions
Energy has a data problem, and data centers have an energy problem. Data centers currently consume close to 2% of total electricity load, but that could rise to 10% within this decade, according to McKinsey and other research groups.
Other analysts have pointed out that the intensifying demand for energy may create regulatory inequities to satisfy the electricity hunger of artificial intelligence (AI) and new hyperscale data centers.
“Ratepayers will have to incur higher electricity rates solely to accommodate the growth of AI without receiving any direct benefits from the higher rates,” said analysts in a report from rating firm Morningstar and quoted by S&P Global.
As if on cue, utility AEP Ohio earlier this month asked regulators to allow it to set up a new billing category for data centers and cryptocurrency mining operations. AEP, like other utilities, will be required to steeply increase its capital spend to meet the needs of these customers and feed the AI boom. Under the proposed new rate structure, data centers of more than 25 MW must commit to paying at least 90% of their projected energy needs every month for 10 years, according to T&D World.
Utilities, which only five years ago were anticipating flat load growth for the decade ahead, are now caught flat-footed as a combination of historical events are pushing the acceleration of data facility construction and connected power requirements.
“It’s a challenging time,” Phil Fischer, a leader on the data centers/mission- critical solutions team at engineering and construction firm Black & Veatch, noted during a session at the recent Microgrid Knowledge 2024 Conference in Baltimore. “This was not expected, and it’s not in the RFPs (requests for proposals)” of power generation utilities, he added.
Everyone knows that the number of data server sites must grow to meet a level of demand unimaginable in quaint old Y2K.
Two workforce revolutions underway all at once
What happened? Two era-defining events in one brief module of time, that’s what.
“I take it back to COVID,” Colby Cox, managing director, Americas, for data center analytics platform DC Byte, offered during the same MGK 2024 Conference in April. “That was the genesis of when data began to shift.”
People left their offices and those workplace servers and took their business to their homes, reallocating the responsibility for all that data to cloud-based servers. This seismic shift in workplace location has not returned to the pre-pandemic office environment–and likely never will completely. The move to remote work, at-home, is the new not-so-normal driving cloud-based capacity.
And yet there’s more ominous clouds of new data evolution rushing in: The superpower revelation of AI within the data force over the past 16 to 18 months is literally changing the world of analytics, content creation and technological deployment across the economy.
So that’s what’s fueling the data center frontier’s hyperscale expansion. But what’s going to fuel the data centers themselves?
AI, with its potential for machine learning and real-time evaluation and responsive countermeasure, could revolutionize energy efficiency and dramatically assist in taking some of the load off power generation capacity that’s already stretched thin. This work would be like what smart meters and smart grid technologies have done to make office buildings more efficient.
Helpful but hardly solving the problem. The amount of energy required for AI training models alone likely swamps any benefits from its efficiency capabilities, given the pace of growth in the sector.
Processing chip colossus NVIDIA alone may ship 1.5 million AI server units per year by 2027, necessitating some 85.4 terawatt hours (TWh) annually, according to The Scientific American website quoting analysis conducted by researchers at Vrije University Amsterdam. The 85+ TWh is more energy than used by some small nations, it was noted.
First World problems? Maybe, but a mission-critical crisis, nonetheless.
What is coming to the rescue?
Renewable power installations are growing by record amounts, but even with battery storage they pose intermittency challenges. Data centers need flexible, reliable backup power, which traditionally has come from diesel-powered, on-site generator sets.
Signing off on vast amounts of diesel backup generation is problematic and politically tough in some regions. Energy transition-minded, progressive states aren’t excited about continuing that historical trend because of emissions concerns, so work is being done to facilitate other means of power generation, including nuclear power, hydrogen and biofuels.
The infrastructural timeline and costs for those answers, however, raise difficult questions.
“How do you do it reliably? How do you do it economically? How do you do it fast enough?” Fischer asked at the Microgrid Knowledge (MGK) Conference in Baltimore.
Sonya Harbaugh, who is director of utility of the future strategy at Baltimore Gas & Electric, was there at MGK, too. She stressed the point that everyone involved around power generation must be creative and collaborative at hyperscale.
“We have to think beyond the technology side,” she said. “We have to think about regulatory reforms. I think we need a lot more pilots, and I wish there was more room for innovation investment.”
Getting smaller and more focused tackles big problems
Microgrids are an answer which can solve a lot of questions about how the demands of the energy transition will be met. It’s worth noting, once again, that the major grid planners weren’t ready for what’s happening.
“Forecasts are just predictions and, as we are seeing today, what was predicted only a few years ago is wildly different from today’s reality,” said Enchanted Rock’s Walch, who also was moderator of the Microgrid Knowledge panel featuring Harbaugh, Fischer and Cox. “Instead of investing in uncertain future states, the industry should focus on reliable, scalable solutions like microgrids that can be built closer to the load and adapt to changing needs.”
Enchanted Rock is doing work with Microsoft to provide data center fuel alternatives. The proximity of microgrids to the customer is a logistical advantage which cannot be overlooked. They also offer flexibility in potential fuel resources, as natural gas, renewable natural gas, solar, battery storage and also propane can be on-site components of the system, beyond the traditional (and trusty) diesel-fired generator set. Diesel has its place, too.
Many tech and data giants such as Microsoft, Amazon and Google are reconsidering the onetime whipping post of nuclear energy, particularly small modular reactor technologies, to create zero-emissions microgrids on a macro scale. Nuclear fission energy is generated carbon-free.
Amazon Web Services recently acquired a Pennsylvania data center site, built by Cumulus, that will be directly connected to the nearby Susquehanna nuclear power plant.
Regardless of the available generation resource, data center planners, who used to be focused purely on finding locations near vibrant and savvy technology regions rich with fiber optics, are now making siting decisions based on much more elemental factors such as location, location, location.
Virginia is currently the U.S. kingpin of data centers, but that may change as the new dynamic coalesces around critical energy needs.
“We’re seeing markets disperse,” Cox pointed out. “They’re having to go where you can get power.”
The dynamics driving data centers’ revolutionary rise are easy enough to pin down, but making intelligent moves (artificial, virtual and realistic) takes precious, well-deliberated time.
Unfortunately, the pace of growth is not giving energy planners much of that most precious commodity—time.
Keep track of Microgrids and Data Centers: Subscribe to our newsletter
