AI to Prop Up Fossil Fuels and Slow Emissions Decline, BNEF Says

(Bloomberg) -- The world likely hit peak energy-related emissions in 2024. But the decline in the coming decade will be slowed due in large part to data center expansion powered by fossil fuels, according to the latest New Energy Outlook report from BloombergNEF. 

While renewables and storage are expected to make up more than half of the power capacity needed by 2035 to meet data center demand, almost two-thirds of additional electricity generation will come from fossil fuels like coal and gas, BNEF said Tuesday in its New Energy Outlook report. That’s because of the possibility of data centers helping to extend the life of existing coal and gas plants, the report said. 

Electricity demand globally is expected to surge on back of artificial intelligence use and data center expansion. That boom has prompted tech companies to secure round-the-clock electricity for their facilities. Some firms have turned to nuclear power, while others are eyeing natural gas as the most abundant and immediately available source of energy. 

This year is likely to be the first that the world sees a structural decline in energy-related emissions — that is, a decline driven by fundamental shifts rather than a shock like the Covid-19 pandemic or the 2008 financial crisis. BNEF projects that emissions will fall 13% by 2035, but the increase of power-hungry data centers will blunt the drop. 

Cumulative carbon emissions are expected to be 3.5 gigatons higher — equivalent of 10% of total global emissions today — over the next decade because of additional computing-driven fossil fuel power use, the report said. The US and China are forecast to be the two biggest contributors by far.

The rise of AI is also reshaping US power markets, with data center demand projected to rise from 3.5% of total electricity demand today to 8.6% by 2035, according to another BNEF report released Tuesday. Amazon.com Inc. is the biggest data center operator with close to three gigawatts of capacity and another 12 gigawatts in its pipeline, followed by Microsoft Corp., Meta Platforms Inc. and Alphabet Inc.’s Google. US companies have developed the most major AI models, indicating that most large-scale AI training takes place in the US, the report said.

For years, big tech firms have invested in clean power even before the AI and data center boom, relying on power purchase agreements from wind and solar farms. Amazon, Google, Meta and Microsoft accounted for almost 70% of such agreements in the US last year, which lets those companies build data centers connected to fossil-heavy power grids while still saying they are reducing their carbon emissions, according to the BNEF report on data centers.

The New Energy Outlook found that the global average temperature will reach 2.6C by 2100, though it is now “very close to slipping to an outcome closer to 2.7C,” the analysts wrote. The modeling is based on BNEF’s “economic transition scenario,” which models changes in the prices and deployment of technology but only includes present-day policies rather than new ones in its assumptions.  

While data centers present a “substantial opportunity in the coming decade” for power demand, estimates for future demand from such facilities is still uncertain and could change, the New Energy Outlook report added. In the US, President Donald Trump’s administration has touted the Stargate project in a bid to expand data centers, and Trump has signed executive orders to expand fossil fuel use in the service of the AI boom. But Microsoft, for one, has pulled back on data center projects around the world as it takes a harder look at its expansion plans.

(Updates paragraph 7 with new details on tech companies’ efforts to procure power.)

©2025 Bloomberg L.P.