WASHINGTON ― Not every data center has to be hyperscale, according to Andrew Webber, founder and CEO of Digital Power Optimization.
Finding the sites and hundreds of megawatts of power these massive facilities need is “rather limited,” Webber told the audience at the U.S. Energy Association’s Energy Tech Connect Forum on Oct. 24. Rather, DPO looks to co-locate its data centers with smaller stranded or underused power projects.
“That’s kind of the point of our business,” he said. “We can flow like water in the cracks and around the energy sector and make use of assets that are otherwise undervalued. … Our view is size isn’t the only deciding factor, and in fact, the most efficient developments, the lowest-cost power may be in smaller sizes. It’s more distributed in smaller footprints all over the country … using existing infrastructure.”
DPO started out in 2022, powering cryptocurrency mining facilities from a 6-MW hydropower facility in Wisconsin. This year, it entered into a partnership with Schneider Electric to develop modular artificial intelligence data centers that will draw on up to 100 MW of power from existing wind energy installations in Texas.
Webber was one of three speakers on a panel looking at companies that have developed different, profitable models for meeting the challenges of powering the digital economy. Much of the dialogue in the power and tech industries has revolved around the tech giants ― Microsoft, Google and Amazon ― developing hyperscale facilities, said Tom Mapes, president of the nonprofit Digital Energy Council, who moderated the session.
Exact estimates from different industry analysts vary, but the general consensus is that energy demand from U.S. data centers will grow two- to threefold by 2030, accounting for anywhere from 7.5% to 9% of total electricity consumption. (See EPRI: Clean Energy, Efficiency Can Meet AI, Data Center Power Demand.)
“We’re using this kind of broad language to try to hit this generation demand issue, and it’s more nuanced than that,” Mapes said. “There are more pieces to this puzzle than just data centers, AI.”
TeraWulf, which develops both bitcoin and AI data centers, looks for “dirty sites” to clean up, said Sean Farrell, the company’s senior vice president of operations. “We’re heavily looking at coal plants, pulp and paper, and steel plants across the U.S. and outside the U.S. A lot of those were built 30 to 40 years ago.”
Located on the New York shore of Lake Ontario, TeraWulf’s Lake Mariner data center campus is built on the site of a former coal plant but is powered primarily with hydropower and nuclear. This month, the company announced a new long-term lease for the site that will allow it to expand its facilities from 500 MW to 750 MW.
CleanSpark brings bitcoin mining facilities to small towns, where it can have major positive impacts on local economies, said Chief Operating Officer Scott Garrison. The company has 26 sites in Georgia, totaling around 700 MW, which can serve as grid assets for municipal utilities or electric cooperatives.
CleanSpark owns all the servers in its facilities, so for “many of our utilities, I can shut the power off and give it back at any time,” Garrison said. “We’re building infrastructure for small, rural towns.”
Keeping its facilities small also gives the company flexibility to take power off local electric systems for shorter periods of time, he said. For example, a utility might build a substation for a new data center, which will not be at full capacity for several years. “I can sit there for two to three years, create revenue for your town and your state, and then we can move on to other places,” he said. “There are plenty of areas that have stranded power.”
Webber agreed, arguing that bitcoin mining should be viewed “as energy management infrastructure. … From the standpoint of the ability to turn it on and off, the ability to ramp it up and ramp it back down, it’s the perfect tool for energy companies to use for their own purposes and their own benefit, if only they understood it a little better.
“There’s a way to make energy companies more profitable by deploying these [facilities] in a more thoughtful way.”
Massive Capacity Waste
Farrell pointed to another benefit of TeraWulf’s model of putting data centers on the site of former coal plants with existing interconnection infrastructure: shorter times in interconnection queues.
For a project in MISO’s service territory, TeraWulf has applied for MISO’s Net Zero Interconnection Service, which allows a generation project to use excess interconnection service at a point of interconnection, he said.
But Farrell also cautioned that different kinds of data centers — for bitcoin mining, high performance computing and AI — have different power backup and interconnection needs. Grid modeling will have to incorporate different options for “how we can optimize those assets at those locations, because definitely one size does not fit all,” he said.
High-performance computing, or HPC, differs from AI in that it uses clusters of computers to process large amounts of data at super high speeds, as opposed to the sophisticated algorithms that AI uses for higher computing functions, like data analysis.
Webber said the way forward for data centers is “to try to find a pathway … without needing to change regulations or without needing to modify someone’s opinions or approvals or the regulatory overlay, because again, it’s [more] time.”
Data centers’ search for clean, dispatchable power ― like advanced nuclear ― could drive major changes, but no easy answers in the electric power industry, he said. Regulators and other decisions makers need to familiarize themselves with the different generation technologies, different types of data centers and potential impacts of both on the grid.
One example, Webber said, is that hyperscale data centers can be highly inefficient because although they run 24/7, they may not always use their full computing capacity 24/7.
“That is just an absolutely massive amount of capacity waste, infrastructure waste [and] capital waste,” he said. “If you’ve got the connection and you’ve got the power availability, make sure you’re actually using it, and that will help prevent the need to build quite so much,” he said.
The challenges surrounding data centers and their power demand power ― and their possible solutions ― are not likely to be affected by the coming election, Mapes said. “No matter who wins in a couple weeks, this conversation is only going to grow,” he said, and it needs to move out of what he sees as separate tech and energy industry silos.
As data center efficiency improves, Mapes envisions “different data centers for different opportunities and regions.”
“What we’re trying to do is … get some of these conversations up to the forefront, start talking about these now on the front end as opposed to trying to fly the plane and build it at the same time,” he said.