In light of recent expert forecasts, the escalating energy consumption by power-thirsty artificial intelligence is putting a strain on the already strained US electricity system and is posing a grave threat to national efforts geared towards combating climate change. The unexpected surge in energy demands, partly propelled by the growth of data centres for AI, along with a less rapid development of renewable sources and the prolonged functioning of coal-fired plants, has caused experts to reconsider their predictions for reducing carbon emissions.
This issue was highly debated during Climate Week NYC which was conducted as a side event to the UN General Assembly last week. Here, technological corporations were given more prominence than the historically polluting fossil fuel companies. The most recent report by BloombergNEF, published this week, alerts of the slow-paced decarbonisation in the US, forecasting a mere 34% reduction in emissions from the 2005 levels by 2030.
In their new evaluation, BloombergNEF suggests that the US is veering off its national target of reducing emissions by 50-52 per cent by 2030 from 2005 levels, and achieving net zero emissions by 2050, as promised to the Paris agreement. According to Tara Narayanan, the head power analyst at BloombergNEF, the rise in AI power consumption is a significant disruption to the supply, terming it as far from ideal.
The inadequacy of grid infrastructure is showing to be a notable obstacle in transitioning to green energy – not just in the US, but globally as well. China has committed to a staggering $800bn expenditure in the next six years to counteract the stress on their energy system while shifting swiftly from coal to renewable energy.
The US, which had a stable power demand for nearly 20 years, is predicted by forecasters such as the ICF consultancy group to witness an uptick of 9% by 2028 and nearly 20% by 2033. They cite the expansion of data centres, the return of manufacturing, and electrification as the reasons. Data centres’ share in US electricity consumption is projected to rise twofold by the end of the decade, according to this year’s prediction by the Electric Power Research Institute.
The US Energy Secretary, Jennifer Granholm, remains convinced that despite a surge in power demand, the US is still capable of achieving its net zero emissions objectives. This optimism is fuelled by the green subsidies worth nearly $370bn introduced via the Inflation Reduction Act by President Biden’s government.
Granholm acknowledged the urgency of the situation to the Financial Times, stating that anticipation exists, but the initial obstacles have now been overcome and momentum is not dissipating.
Some concerns have arisen from green project developers, pointing out that the creation of renewable energy to meet unprecedented demand levels is suffering setbacks. These are due to the up to five-year duration it takes to initiate new supplies because of permitting delays and hitches in the rollout of the grid.
Sandhya Ganapathy, CEO of EDP Renewables North America, emphasised this point indicating that while the demand for renewable projects is high, they may not be created as quickly as necessary.
Simultaneously, the development of AI data centres is causing tech giants to compete to find sustainable, 24-hour power sources. A recent example saw Constellation Energy and Microsoft restarting operations at the Three Mile Island nuclear plant in Pennsylvania, the location of the US’s worst nuclear incident. This reopening was part of a 20-year agreement.
Higher electricity demand has also caused US operators to postpone coal plant closures. Power expert Akshat Kasliwal of PA Consulting diverges from Granholm’s optimism, asserting it’s increasingly difficult to envision a carbon-free US electricity system by 2035.
Pedro Pizarro, CEO of Edison International, stated that due to demand surge, gas power stations would need to stay operational to ensure stable power supply. He clarified that his company isn’t pushing to maintain the presence of gas, but stressed the industry must be resilient and affordable given the prevalence of weather extremes.
The Lawrence Berkeley National Laboratory in the US estimates that approximately 1.5 terawatts of electrical output capacity from renewable resources is currently on standby to be hooked into the grid—enough to expand the nation’s electricity infrastructure more than twofold. However, green energy projects constructed in the past year endured half a decade before their power could be channelled into the grid, and the lacking transmission infrastructure complicates moving this environmentally-friendly power from remote generating locations to high-demand areas.
According to an investigation by analytics company, the Rhodium Group, power industry emissions could potentially rise by over 56 per cent compared to the moderate emissions projection, if demand for data centres almost triples by 2035 and developers face impediments in incorporating fresh wind and solar energy infrastructure. Nevertheless, predictions of soaring demands could also be significantly scaled down if data centres manage to increase efficiency. Leaders within the tech industry suggest that embracing AI on a larger scale can decrease energy consumption by streamlining daily procedures.
Jensen Huang, CEO of Nvidia—the rapidly advancing AI chip manufacturer—asserted at the Bipartisan Policy Centre on Friday, “Even though power is required to train AI models, the models once created can perform tasks in a far more energy-conscious manner.” He further claimed, “The boost to energy efficiency and productivity derived from AI will be phenomenal”. (Source: Copyright The Financial Times Limited 2024).