'Our future depends on it' — AI poses threat to clean energy mandates, affordability
ALBANY — Artificial intelligence is quickly becoming an inescapable reality of modern life. It’s being used in an increasing number of products and the technology is developing at breakneck speed.
Data centers powering AI are among the reasons electricity projections are ramping up in New York and across the country. But some argue the rise of AI is just another technocratic ruse.
“There is significant risk that the estimated load (from data centers) does not materialize or shows up in lower amounts than expected,” the credit rating agency Fitch warned last year. (Fitch is owned by the Hearst Corp., which also owns the Times Union.)
If AI does bring a data center boom, it could imperil the state’s clean energy goals and threaten affordability. Unlike other big users entering the grid, data centers generally produce very few jobs. A 1-million-square-foot, 1-gigawatt OpenAI facility in Texas will employ around 100 full-time employees once completed. Micron’s upstate New York chip fab plants, meanwhile, are supposed to create 9,000 jobs with the company.
“Data centers account for nearly half of electricity demand growth between now and 2030 (in the U.S.),” an April report from the International Energy Agency stated. “By the end of the decade, the country is set to consume more electricity for data centers than for the production of aluminum, steel, cement, chemicals and all other energy-intensive goods combined.”
Data centers consume around 4% of U.S. power. There are more than 100 data centers in New York that consume 3% of the state’s electricity. They could use “6.7 to 12% of total U.S. electricity by 2028,” the U.S. Department of Energy reported in late 2024.
There will always be something applying pressure on the grid, said Assemblyman Alex Bores, a Manhattan Democrat involved in AI policy development. The state needs to meet the moment by increasing incentives and reducing the regulatory burden of clean energy development, he said, adding, “Our future depends on it.”
Companies like OpenAI and Anthropic need massive facilities housing specialized computer chips to train popular large language models like ChatGPT. Unlike data centers used to power web searches and company software, AI centers need an enormous amount of electricity. Amazon is building a facility in Indiana requiring 2.2 gigawatts of power, enough electricity to supply around 2 million homes.
In a Niagara County village in western New York, Terawulf operates a 750 megawatt facility used for both cryptocurrency mining and AI.
All that power comes at a cost: upstate New Yorkers pay an extra $88 annually because of large-scale energy-intensive crypto mining, researchers from Berkeley and the University of Chicago found. Tax revenue from the crypto mines only made up for a small portion of the costs.
Companies bringing large loads onto the grid are required to update transmission lines and equipment to serve the project. However, they are not required to pay for new electricity generation needed to supply an increasingly strained grid.
If more wind turbines, solar fields and nuclear power plants are constructed, it will ultimately fall on New Yorkers to pay the tab. Residential energy users pay around three times more for electricity than industrial users. In the case of Terawulf, the company is paying six times less than residential customers, according to its website. Company officials declined a request for comment.
Industry gets better utility rates because they “use more electricity and can receive it at higher voltages, so supplying electricity to these customers is more efficient and less expensive,” the U.S. Energy Information Administration’s website explains. Large loads coming onto the grid are also able to negotiate rates, while everyday homeowners can’t.
The state is already falling behind a mandate requiring that 70% of electricity come from renewable sources by 2030. Trying to scale nonemitting resources at a time when demand is increasing makes reaching those goals more difficult. For reference, New York’s largest proposed solar farm will only produce 500 megawatts of electricity.
Lost bets
Stanford Physicist Amory Lovins is skeptical that an AI data center boom will ever come.
“As AI evolves, the stakes for energy policy are high,” wrote Lovins in a recent paper. “Lost bets on electricity demand could waste major investments, lock in unneeded fuel infrastructure, and derail health and environmental progress.”
The positions of Lovins and other skeptics largely hinge on a few key points.
System operators can encourage greater flexibility among industrial energy users, enabling facilities to tap the grid less during peak demand periods. Efficiency improvements have historically made computational processes much less demanding in a short period of time. Grid projections can be faulty and have led to overbuilding of energy in the past.
Throughout the early 2000s, data centers used progressively more power. Starting in 2008, internet usage continued to grow, but data center consumption slowed. Data centers got nearly 30% more efficient annually between 2007 and 2015. The kicker is, a large amount of power generation was constructed across the country to meet the demand, only for it to never come.
Now, with data centers needing more advanced units for AI development, those efficiency improvements have slowed. AI requires graphics processing units, which have tripled their power needs since 2022, according to a Deloitte report. Training OpenAI’s GPT-4 used enough energy to power San Francisco for three days, the Massachusetts Institute of Technology reported.
Even as efficiency grows, the use of AI is expanding into everyday life. AI can be used for grid planning, many companies have AI tools to assist with customer service, and airlines are exploring using the technology to set flight prices.
Fengqi You, a professor of energy systems engineering at Cornell University, equated AI advancement to the advancements in vehicles.
“The cars we are driving nowadays are much more efficient than the cars we had 100 years ago,” You said. “But look at the emissions from the transportation sector, the emissions are much more than 100 years ago because the demand is growing.”
What makes AI remarkable is the stunning pace of its adoption. It took five years for 40% of workplaces to use the internet, a benchmark generative AI met in two years, according to an International Energy Agency report. You also noted the increasing number of computers in everyday life, from glasses to watches to buildings. Odds are those will all have use cases for AI, You said.
In an interview, Lovins suggested expanding grid flexibility efforts to manage AI data center load growth as it comes. Flexibility is when a power user scales down operations during times of high demand. This isn’t possible for some industries, such as chip fab, which require stable operating conditions, but is doable for AI development and other energy-intensive technologies.
The New York Independent System Operator “assumes” many large loads entering the grid will agree to be flexible during times of high demand. NYISO manages the state’s electric supply system.
Faulty projections
Electricity demand projections have a history of being incorrect. There’s also a general lack of transparency with potential large load projects, including data centers, adding to the complexity of planning.
Between 2006 and 2023, utility planners overestimated demand by an average of 17% in 10-year projections, according to the Rocky Mountain Institute, which Lovins co-founded.
NYISO projections overestimated demand by about 5% between 2014 and 2024.
That being said, AI has come a long way in the past decade and requires unprecedented computational bandwidth.
“The trend seems to be that they’re (AI data centers) driving the load forecast over the next 10 to 20 years,” said Kevin Lanahan, a vice president at NYISO.
When a company wants to build something requiring significant power in the state, they have to enter an interconnection queue and be studied for potential impact on grid reliability. The number and scale of projects in the queue help determine future electricity demand.
There’s no responsibility to go through with a project once it’s in the queue. In New York and many other states, duplicative projects are being proposed without a clear understanding of what will actually come to fruition.
For example, Arconic, a lightweight metals company, has three data centers in the queue at the same location but with different power needs. All are for “cloud hosting and AI workloads,” according to a company filing. It’s unclear which, if any, of the centers will be constructed. The company declined an opportunity to comment.
“Conservatively, you’re seeing five to 10 times more interconnection requests than data centers actually being built,” Astrid Atkinson, a former Google senior director of software engineering, told UtilityDive.
“We’ve got the queue, (but then) how real is it, what are the plans, who’s going to pay, that all gets decided on the back end, which is really problematic,” said Mandy DeRoche, a deputy managing attorney at Earthjustice. DeRoche litigates energy cases all over the country.
NYISO only does a grid reliability study and doles out an estimate for large load users to pay for grid upgrades to meet a project’s needs. After that, the local utility enters an agreement with the company to determine electricity costs. Those numbers are often sealed from the public.
NYISO wouldn’t comment on how it weighs duplicative projects in electricity demand projection, but said that every proposal is studied for grid reliability.
Patrick Stella, a National Grid spokesman, said the company doesn’t keep track of how many proposed projects come to fruition. Stella wouldn’t comment on the specific electricity rates of data centers, but said “recent trends show a growing demand for industrial and data center developments.”
Only 15 states have enough data centers for them to have a noticeable presence. Meanwhile, other areas of consumption growth, such as home electrification, are happening nationwide.
Unlike other states, New York doesn’t have economic development incentives for data centers because of the “limited amount of jobs they create,” said Kristin Devoe, a spokeswoman for Gov. Kathy Hochul.
Despite the lack of enthusiasm, experts suggest proximity to the financial services industry and upstate’s high levels of clean energy are reasons for continued development. Data centers could use 5% of New York’s power by 2030, according to The Electric Power Research Institute.
AI for the people
Hochul’s Empire AI project highlights the complex dynamic between accelerating AI development for the public good and possibly running afoul of clean energy goals and affordability.
It’s a tightrope walk encouraging the good parts of technological advancement without inflecting the bad.
Empire AI — buoyed by over $500 million in public and private funding — brings together the state’s leading research institutions to ensure advancement benefits the public. You, who is in the consortium, said Empire AI is being used to make things like greenhouses more energy efficient. Taxpayer-funded AI initiatives will ensure the technology is a “public good,” You said. He added the project wouldn’t lead to significant data center growth.
Most AI companies are in California, but that state is nowhere near the leader in data center power demand. Many of the largest AI centers are in places like Indiana, Texas and Virginia, where they get incentives and regulatory relief.
On the other hand, Lanahan listed the governor’s AI stance as one of the reasons NYISO anticipates continued data center growth.
“New York state’s clean energy and emissions goals have always existed alongside the understanding that more electricity will be needed, not less,” Devoe said. She added that grid reliability should “never become the justification for slower economic growth.” The state “can and will” encourage economic development while pursuing clean energy mandates through an “all-of-the-above approach,” Devoe said.
Part of that approach is a Hochul plan to build a new nuclear power plant in the state.
“Empire AI is bringing together the world’s best minds to develop the next breakthrough — and we’ll power it with clean, safe, and reliable nuclear energy,” the governor said in a social media post tying AI goals and nuclear power.
The state should be focused on present energy issues like the high amount of fossil fuel sources downstate, DeRoche contended. Even though a new transmission line is being built to deliver clean energy from Canada, New York City still gets the vast majority of its electricity from emitting sources.
“I wish we would prioritize people and their future and their health and the climate by thinking through that (existing energy issues) a little bit and not planning out infrastructure for load that might not come,” DeRoche said.
Investigative Reporter
Ezra Bitterman is a Joseph T. Lyons Investigative Fellow for the Times Union. He is from Los Angeles and studied Journalism at the University of Missouri. Ezra previously reported for the St. Louis Post-Dispatch, Columbia Missourian and Euractiv. He's reachable at Ezra.Bitterman@TimesUnion.com.
Comments
Post a Comment