Data Center Growth Puts the US Grid and Emissions Targets Under Pressure

Published November 11, 2026

AI is quickly becoming one of the largest new sources of electricity demand, and the scale is now coming into focus. A new report from BloombergNEF, commissioned by Bloomberg Philanthropies, projects that global data center electricity demand will almost quadruple to 2,354 terawatt-hours by 2035, equal to about 7% of worldwide electricity consumption. In the US, the shift is already visible in national data. A Lawrence Berkeley National Laboratory study for the Department of Energy found that US data centers used 176 TWh in 2023, or 4.4% of total national electricity consumption, and that their share could reach 6.7% to 12.0% of US electricity consumption by 2028. The US Energy Information Administration also expects commercial electricity consumption, a category that includes hyperscalers and cloud computing, to overtake residential use in 2027 for the first time on record. The central question is no longer whether demand will grow, but how it will be powered and what that means for the emissions goals that companies, cities and institutions have set.

Data Centers Are Turning to On-Site Gas to Get Power Faster

For developers, getting a facility energized quickly has become a top commercial priority. BNEF finds that grid constraints in the US and Europe are slowing connections for large loads, while the financial payoff of coming online fast is leading some developers to build on-site gas generation while they wait for the grid. The report puts numbers on the trend:

BNEF has tracked 181GW of announced gas-fired capacity proposed alongside data centers, with more than 90% located in the US.
If every announced on-site gas plant for US data centers were built and run under BNEF’s assumptions, the plants would emit roughly 345 million metric tons of CO2 equivalent per year.
That is 152 MtCO2e more than if the same capacity drew on average US grid power, an amount equal to 10% of US power-sector emissions in 2025.

These figures come with an important caveat. BNEF notes that the pipeline varies widely in maturity and timing, and should be read as a signal of potential scale rather than a forecast of what will be built. Even so, the report flags a longer-term concern: although many projects are meant as a bridge to a grid connection, their economics could encourage them to keep generating once online, creating a fossil-fuel lock-in risk that extends beyond annual emissions.

Faster Grid Connections Are Central to Limiting Emissions

BNEF’s conclusion for developed markets is direct. In the US and Europe, connecting new data centers to the grid is the more sustainable path, which makes faster grid connections important for limiting near-term emissions and lock-in risk, while continued power-system decarbonization remains essential. The report reviews policies in 17 markets designed to shorten connection timelines. The approaches it examines include:

Faster grid connection processes
Financial commitments from large-load developers
Flexible connection arrangements
Special zones for development
Clean-power procurement

BNEF is careful not to declare a winner. It notes that the policy landscape is still emerging, and while many measures should cut waiting times, there is limited empirical evidence so far on which will prove most effective. Mark Daly, BNEF’s head of data centers and emerging tech, framed the climate outcome as depending not only on how much power these facilities consume, but on how quickly grids can connect them and whether cleaner generation is added alongside the new demand.

Rising Demand Affects Building Owners and Energy Managers

The data center build-out is not only an issue for utilities and tech companies. Offices, campuses, hospitals and municipal buildings draw from the same grid, and the effects are already showing up in prices. EIA estimates that US residential customers will pay an average of 18.2 cents per kWh in 2026, nearly 5% more than in 2025, and expects the East Coast to see the steepest increases, with average annual growth of up to 7% over the next two years. Utilities in those regions point to higher generation fuel costs and spending to strengthen the transmission grid against extreme weather and to serve rising demand.

There is an emissions dimension as well. Organizations that report location-based Scope 2 emissions depend on how quickly their regional grid gets cleaner. If new load is met with more fossil generation, the carbon intensity of every kilowatt-hour a building buys may fall more slowly than planned. For building owners and energy managers, that raises the value of steps within their own control:

Reducing consumption through efficiency upgrades. JLL research found that light to medium retrofits covering lighting and mechanical, electrical and plumbing systems can cut energy use by 10% to 40%.
Shifting flexible loads away from peak periods to manage both cost and grid strain.
Using data and AI-driven analytics to prioritize retrofits across portfolios. JLL notes that AI is helping owners and investors make more data-driven retrofit decisions and better handle uncertainty around cost-effectiveness and payback.
Tracking grid emissions factors closely so that Scope 2 targets reflect real conditions on the local grid.

Conclusion

The BNEF report makes clear that AI growth and decarbonization are not automatically at odds, but the outcome depends on choices made now. Faster grid connections can reduce the pull toward on-site fossil generation, and continued investment in clean power can keep grid emissions falling as demand rises. Neither is guaranteed, and the evidence on which policies work best is still developing. For organizations outside the data center industry, the most reliable lever is their own energy use. Every building that consumes less, shifts load intelligently and measures its emissions accurately eases pressure on a grid that is being asked to do more than at any point in decades.