Silicon Valley’s scramble to build more data centres could boost US demand for natural gas by the equivalent of another New York State or California within a decade, with much of this expansion due to the energy cost of training and using artificial intelligence.
“If the world wants all [AI] workloads to be powered only by sustainable power in 2030 from currently available technologies, it will have to temper its AI initiative,” and his colleagues at S&P Global, a financial information and analytics firm based in New York, wrote in a .
According to a related from S&P Global, energy-hungry data centres could drive an increase in gas demand ranging from 85 million cubic metres per day to 170 million cubic metres per day by 2030. Those numbers are on par with states such as New York, which used about 100 million cubic metres per day in 2023, and California, which burned through 160 million cubic metres per day the same year. They would also lead to more carbon emissions.
Advertisement
“Between clean, cheap and consistent power, currently technology allows you to pick any two,” says Prabhu. “Gas is cheap and consistent – not clean.”
The report’s estimates suggest US data centres will use another 170 to 250 terawatt-hours by 2030, on top of the 170 terawatt-hours they already consume. About half of that additional power demand is likely to be met by natural gas generation, says , also at S&P Global.
Gas is a popular option for meeting the US’s increasing energy demands – utility companies are more than 200 new gas plants by 2032. However, this growth ignores how such facilities “are facing significant headwinds and scrutiny in the near term” because of vulnerability to extreme weather, volatile gas prices and regulatory limits on carbon emissions, says at the Union of Concerned żěè¶ĚĘÓƵs, a nonprofit based in Massachusetts. As an alternative, he favours renewables, describing regulatory changes underway that could enable clean energy to come online faster.
Some tech companies responsible for building the largest data centres are investing in renewables through power purchase agreements, which finance construction of renewable energy projects and commit to using such power for their data centres once the projects are completed. But until that renewable energy is available, many of the data centres may continue to rely on fossil fuels such as gas, says Prabhu.
“We think [renewable] aspirations lag technology, so gas-fired generation will be required,” says Prabhu. “Stated differently, data centres are not great for the sustainability story.”
That is a problem for tech firms, many of which have sustainability goals for achieving net-zero carbon emissions by 2030. Some are also looking to nuclear to meet their energy needs, rushing to buy power from existing nuclear plants or investing in startups developing small modular nuclear reactors, which could be cheaper and faster to build compared with traditional ones. But none of the new reactors are likely to start producing power before 2030.
In the meantime, Arbaje argues that policy-makers can prioritise renewables by implementing clean energy standards and requiring data centres to flexibly manage their workloads based on the availability of renewable power. “This new challenge does not require further reliance on fossil fuels,” he says, and such reliance would come “at the expense of public health, the climate and ratepayers’ [bill payers’] pocketbooks”.
