Energy-hungry artificial intelligence is consuming increasing amounts of energy from the creaking U.S. grid and threatening national efforts to combat climate change, according to the latest expert forecasts.
Unprecedented energy demand, fueled in part by the expansion of data centers to AIcombined with the slower-than-expected pace of renewable energy development and longer operating times for polluting coal-fired power plants, have prompted analysts to rethink their models for reducing greenhouse gas emissions.
The theme dominated discussions at Climate Week New York, held last week on the sidelines of the United Nations General Assembly, where tech companies were more in the spotlight than tech companies. fossil fuels at the origin of historical pollution.
BloombergNEF’s latest report this week warned of slowing U.S. progress on decarbonization, forecasting a reduction in emissions of just 34% by 2030 from 2005 levels.
The latest assessment places the U.S. trajectory even further from its national goal of reducing emissions 50 to 52 percent by 2030 from 2005 levels and reaching net zero emissions by 2050, consistent with its commitment to the Paris Agreement.
“It’s not a good thing,” said Tara Narayanan, senior energy analyst at BloombergNEF, calling AI’s increased demand for energy a “big disruption” to supply.
“It feels a lot like that moment where you’re immersed in the movie and three different plots have been developed. You don’t know if the problem is going to be resolved,” Narayanan said.
The lack of grid infrastructure is proving to be a major constraint to progress in the green energy transition, not only in the United States but around the world.
China is ready for a unprecedented spending of $800 billion the next six years to overcome strains on the energy system as it rapidly transitions from coal-fired power to renewable sources.
In the United States, electricity demand has remained virtually stable for two decades. Today, forecasters such as consulting group ICF expect an increase of 9% by 2028 and almost 20% by 2033, citing the growth of data centers, the relocation of manufacturing industry and electrification.
The Electric Power Research Institute predicted this year that data centers could double their share of U.S. electricity consumption by the end of the decade.
But Jennifer Granholm, the US energy secretary, said she believed the country could still meet its net-zero emissions targets and manage exploding electricity demand, thanks to green subsidies of nearly 370 billion dollars put in place as part of the Inflation Reduction Act by the Joe Biden administration.
“We have to be aggressive, but the momentum has started and it is not slowing down,” Granholm told the Financial Times.
Renewable project developers say producing green energy to meet historic levels of demand is hampered by the fact that it can take up to half a decade to bring new supply online due to delays in authorization and network deployment.
“The need right now is to balance that,” said Sandhya Ganapathy, managing director of EDP Renewables North America. “Unfortunately, we may not have (renewable) projects at the pace required.”
The proliferation of AI data centers has led to a race between Great technology businesses to find low-emission energy sources, 24 hours a day.
Last week, Constellation Energy and Microsoft signed a 20-year deal to reopen Three Mile Island Pennsylvania nuclear power plant, site of the nation’s most serious nuclear accident.
Expectations of increased demand for electricity have also led U.S. operators to delay the closure of coal plants. S&P Global Commodity Insights has revised his expectations Coal plant closures by 40 percent by the end of the decade, even as renewable energy increases.
“At the current rate, it’s very difficult to imagine the U.S. electricity system being carbon-free by 2035,” said Akshat Kasliwal, an energy expert at PA Consulting. “We are further away from that goal compared to where we were three or four years ago.”
Pedro Pizarro, chief executive of Edison International, a utility, said rising demand meant gas-fired power plants would also have to stay in the energy mix longer to ensure reliable supply. The gas is mainly made up of a powerful methane molecule, which retains more heat than carbon dioxide but has a shorter lifespan in the atmosphere.
“We are not a gas company. . . We don’t have a dog to try to keep gas on hand,” Pizarro said. “We, the industry, need to ensure we have a system that is reliable, resilient, in the face of more extreme weather, and as affordable as possible.”
The United States, however, has no shortage of renewable energy capacity. The Lawrence Berkeley National Laboratory estimates that nearly 1.5 terawatts of generating capacity is waiting to be connected to the grid, enough to more than double the size of the nation’s power system.
But projects built last year must wait five years before they can be connected to the grid, and a lack of transmission lines makes it difficult to transport green energy from remote production sites to demand centers.
Research firm Rhodium Group found that if demand for data centers nearly triples by 2035 and developers struggle to install new wind and solar systems, electricity sector emissions could be more than 56 percent higher than expected in its moderate emissions outlook.
However, the steep projections could also become much more moderate as data centers become more efficient, tech group executives say, and broader adoption of AI reduces energy consumption by improving operations daily.
“While it consumes power to train the models, the models created will do the job with much more energy efficiency,” Jensen Huang, chief executive of Nvidia, the fastest-growing AI chipmaker, said Friday. quickly, at the Bipartisan Policy Center. “The energy efficiency and productivity gains we will get from (AI). . . it’s going to be amazing.
Climate capital
Where climate change meets business, markets and politics. Check out the FT’s coverage here.
Are you curious about the FT’s commitments to environmental sustainability? Learn more about our science-based goals here