Jun. 15—WATERFORD — Thomas Quinn, president of NE Edge, says there’s more than enough energy in the regional electric grid to support the company’s proposed 1.2 million square-foot data center that would be built on 55 acres of the Millstone Power Station.
The new facility is needed to support the growing use of artificial intelligence, Quinn said during an interview last month.
The project’s primary source of power would be 300 megawatts of energy supplied directly from Millstone’s two operating nuclear reactors. Two data center buildings would house thousands of graphics processing units (GPUs), pieces of hardware that provide computing power for AI.
Industry analysts confirmed there’s enough energy in the New England grid now to support the project, but future needs are uncertain due to the growing demand for AI systems, and electric vehicles, heating and air conditioning.
The town’s RTM and First Selectman Rob Brule in early 2023 unanimously agreed to enter into a host fee agreement with NE Edge that signaled the town was open to potentially hosting a data center in town. In addition to setting a payment to the town of $231 million over 30 years, it also mandated requirements on sound and environmental guidelines NE Edge must uphold for the project.
The project still needs a ruling from the Connecticut Siting Council before it can proceed through the town’s boards and commissions for approval or denial.
The impact of the project on the amount of available electricity and its effect on electric ratepayers have remained as chief concerns of those who oppose it. Other concerns have included noise, the environment and quality of life, and the fact that NE Edge has yet to develop a data center.
NE Edge did propose a data center in Groton in early 2022, but the Town Council, yielding to residents’ concerns about noise, environment and a possible decrease in property values, ultimately voted to cease negotiations with the company. The town then imposed a yearlong moratorium to establish zoning regulations that restrict data centers.
Prior to the restrictions, Groton had signed a host fee agreement with a different developer, Gotspace Data LLC, of which Quinn used to be president. But he and his partner, current Gotspace CEO Nicholas Fiorillo, had a disagreement, and Quinn started NE Edge and proposed his own project.
After Groton changed its zoning laws, Fiorillo filed a lawsuit in New London Superior Court against the town claiming it had derailed the Gotspace project. Gotspace had already filed a previous lawsuit against NE Edge, along with multiple other parties, in which it had accused NE Edge of stealing its intellectual property, real estate purchase contracts and rights to build data centers in the area. Quinn said he was never served in any of the lawsuits.
The case, he said, was “dismissed in Massachusetts, then refiled the next day in Rhode Island the next day and dismissed with prejudice in Rhode Island.”
With prejudice means the court deemed the lawsuit can’t be refiled.
“It can’t be brought back, it’s over,” he added. “It was simply a tactic to leverage the good work of the team.”
Part of Groton residents’ concerns pertained to the use of diesel generators to back up the Groton project, which would have been supplied primarily by the grid.
But Quinn is confident this project will go differently. The location of Millstone offers several benefits, he said, one being that the project will not require those generators, since its backup energy would come from the grid, which he says has a surplus of energy.
Growing energy demand
Matthew Kakley, media relations supervisor for ISO New England, which manages the grid for Connecticut and the five other New England states, said the overall system capacity of the New England grid, or how much energy it can produce on a given day, is about 30,000 megawatts between base load power from nuclear and natural gas, and shifting power from renewables like wind and hydroelectric energy.
Base load power refers to the baseline amount of power needed to serve consumers throughout the day, Millstone’s reactors generate around 2,100 guaranteed megawatts, unless in the event of a shutdown, to that base load.
Kakley added the NE Edge projected 300 megawatt usage of that power is “not an insignificant number.” It’s about the size of a small power plant, he said.
But he said he believes that in the short to medium term, the state has enough energy to support the project.
However, he added that ISO New England, which prepares a long-term forecast, projects an increase in electricity use over the next decade of about 17%, driven largely by electrification of vehicles and heating.
“And so when you look beyond that ― this forecast doesn’t go that far ― but we are looking at electricity use in New England as much as doubling by 2050,” he said.
Quinn, trying to allay energy concerns, pointed to generation projects he expects to come online and said data centers are necessary for the country’s future, where artificial intelligence will be increasingly involved in people’s daily lives.
Quinn cited a monthly report prepared by ISO New England for its stakeholders, pointing to proposed power projects that would generate about 42,000 additional megawatts, about half of which are expected to come online before the data center begins operating.
But Kakley said that number is a little misleading, because those are proposals, and its not likely that all of them will be approved. Quinn says that number depends a lot on the wind,
As for the specific impact of data centers, Kakley said it’s something ISO New England hasn’t explored a lot, “other than that it’s a focus of the states.”
“It’s something that we would kind of keep an eye on, and be a resource for the New England states,” he said. “But we don’t have any kind of specific data for data centers or AI-driven demand, other than that we are following the trends, and that there is a lot of attention being paid on that.”
He also pointed out that NE Edge is prepared to pay, along with the $231 million to Waterford over 30 years: more than $1 billion in public benefit and federally mandated congestion charges ― two charges that appear on state ratepayers’ electric bills that contribute to the Millstone clean energy contract, energy assistance and the state’s renewable energy programs ― along with $63 million contributed separately to the state’s energy assistance program.
Quinn said the NE Edge project would not receive its energy from the grid unless in the cases of shutdowns to Millstone reactors. Each of Millstone’s two reactors shuts down every 18 months for scheduled refueling and maintenance. The reactors have also occasionally had unscheduled shutdowns.
Quinn said the data center, through a private agreement with Millstone owner Dominion Energy, has agreed to pay more for the electricity it gets from the power plant than if it bought it from the grid.
Millstone Site Vice President Michael O’Connor said in an April interview that the project would benefit the power station by giving it rental income, as well as a guaranteed buyer of electricity from the Millstone plant.
So if NE Edge is going to pay more than market price for the electricity the data center will get from Millstone, how is the project still economically viable for the company?
Quinn offers that it’s because the location provides the company with a constant direct energy-supplier, which will allow it to save on the distribution charges it would have to pay if it purchased its energy through the grid. And there is an existing fiber optic network on the site that would help transport data to and from the center.
“We don’t have to tear up roads,” Quinn said. “That’s really, really important because otherwise, can you imagine tearing up all the roads up I-95?”
AI here to stay
Professor Omer Khan, a computer architecture and parallel computing specialist in the University of Connecticut’s Electrical and Computer Engineering department, and two other UConn professors, agreed AI is not going away anytime soon.
“AI, if you think about it, it’s basically becoming complex,” Khan said. “We use it with everything. From medicine, insurance, farming ― everything we do. Mobile computing. All of our phones. Transportation. Internet. It’s need is growing, and all of us are a part of it.”
Khan, along with Laurent Michel and Derek Aguiar, two professors in the university’s school of computing, explained that AI requires “massive amounts” of energy to power hardware that stores data for it to train on, and answer questions.
Popular AI chatbot ChatGPT, for example, has the ability to answer questions posed to it by users.
For the AI-based chatbot to formulate a response to a question, it must associate words that commonly go together to produce the most likely desired answer, Aguiar said.
But to give that response, Michel said, it needs to have analyzed “absolutely massive” collections of data in an infrequent but intensive process known as training, Michel said.
“And that’s what takes up all the energy,” he said.
But energy consumption continues well beyond training, as AI searches for answers to people’s questions.
“If you have millions of people asking questions of ChatGPT, and those questions are routed at the same time to the data center, it means that they need to have the ability to answer one million questions within one second,” he said. “And that’s what consumes energy. It’s lots of questions coming in.”
Khan said the size of the collective data now being analyzed by AI models has crept into the exabytes, an amount of data equivalent to 1 billion gigabytes. And the more AI is used, that data continues to grow.
“Actually, with the ChatGPT data, the data growth has skyrocketed in the last year,” he said.
Michel said he believes the appetite for the industry to put more AI into various products is not going away.
“They’re talking about using this kind of technology to run human resources departments,” he added. “To run call centers, to run tons of different things.”
Michel agreed that more people using AI is driving the need for data centers to carry out AI processes.
The push toward AI is also demonstrated in initiatives being taken by neighboring Massachusetts to position itself as a global leader in artificial intelligence technology.
Massachusetts Gov. Maura Healey in February established a task force that will make recommendations to the state for how it can help businesses integrate AI, and she said she would be seeking $100 million to fund that task force.
Healey had previously unveiled the FutureTech Act, a $1.23 billion plan to modernize informational technology systems, according to a news release.
Quinn, citing the latest report from the United States Energy Information Administration on Connecticut’s 2022 energy usage, pointed out that the state in that year had generated 15.2 million Megawatthours more than what it had sold to Connecticut electric customers. Megawatt hours is a unit of measure that takes the amount of energy produced at a given moment and multiplies it out over an hour.
Quinn said that left the 15.2 million, equivalent to 35.5% of the state’s energy that year, to be sold out of state. Quinn said the Massachusetts initiatives will further increase the demand there for Connecticut’s excess electricity.
Because Connecticut and Massachusetts use the same electric grid, Quinn argues it would be better to keep that energy in the state, paying in-state electricity charges while assisting Connecticut’s AI progress, instead of sending it to usher in the AI future in Massachusetts.
Future data demands
Worldwide, the data center phenomenon is growing, Khan said, and its going to need a lot of power.
“Right now, my understanding is these data centers, globally, are very soon going to consume about 7% of our power generation in the world,” said Khan. “I mean, that’s a big number in terms of just data centers consuming that much. So can we reduce that?”
Khan said as AI continues to grow, efforts should be made to reduce the amount of data it requires to operate, and thus, its power consumption. Those are the “hard questions,” he said users and consumers of artificial intelligence should be asking of tech companies and power providers.
“I think we can reduce the amount of data we need to train a model and operate a model,” Khan said. “Then we’d get to the root cause of the problem, which is making us invest so much money in a hyperscale data center.”
“And if there are ways that we can incentivize using less data, I think, that will be the root cause of the problem that we have,” he said.
Signup bonus from
