Google Says AI Demand Is Outrunning Grid Decarbonisation
Google said its electricity demand climbed 37% in 2025 and has grown more than 250% since 2019 as AI and cloud infrastructure expanded. The company also cited 1 GW of demand-response capacity and more than 12 GW of new clean generation deals, while the public record still lacks what share of its total computing load can shift during grid stress.
Google Reports 37% Electricity Demand Growth
Google's 2026 Environmental Report says AI infrastructure is expanding faster than electricity grids are decarbonising, putting hyperscale computing demand ahead of the pace of new clean-energy infrastructure.
The report says utilities, regulators and grid operators are examining large-load interconnections, transmission constraints and the availability of firm generation for data centres.
Google's electricity demand climbed 37% during 2025, the largest annual increase in its history.
Since 2019, total electricity demand has grown more than 250% as the company expanded infrastructure for AI products and cloud services.
Renewable Matching Does Not Solve Hourly Power
Renewable-energy purchases equaled 100% of Google's annual worldwide electricity consumption for a ninth consecutive year.
Operational emissions fell 2% after additional clean-energy buying and infrastructure-efficiency work, the report said.
The report separates annual matching from Google's longer-term goal of operating on carbon-free electricity every hour of every day.
It identified interconnection queues, permitting delays, transmission constraints and shortages of firm carbon-free generation as barriers to that goal.
Ihab Osman, an independent strategist focused on data centre and mission-critical infrastructure, said annual renewable buying does not prove firm clean electricity will be available at the precise grid node and hour when data centres need it.
Flexible Load Adds A Grid Tool
Machine-learning jobs can be moved away from peak grid-stress periods while customer-facing services stay online, according to Google.
By early 2026, the company had added approximately 1 GW of demand-response capacity to long-term utility agreements in the United States.
Google said flexible computing loads can improve grid utilisation and support faster interconnection of new data centre capacity.
Google left two details outside the public record: the portion of total compute represented by 1 GW and the operating process used to move that load.
Ahmad Faruqui, an economist-at-large specialising in electricity markets and demand response, said utilities need to know what percentage of grid peak demand and data-center peak demand the 1 GW represents before judging the claim.
Google has also asked FERC for market rules that would let flexible computing demand react to live grid conditions under the framework of Order 2222.
Faruqui said public evidence on data-center load flexibility remains limited.
Clean Generation Deals Still Need Grid Delivery
Fleet-wide data centre power usage effectiveness averaged 1.09 during 2025, according to Google.
For Gemini Apps text prompts, Google cited a 33-times energy reduction and a 44-fold carbon-footprint reduction over 12 months, attributing the change to hardware, software and model improvements.
Those efficiency gains have not offset the company’s rising electricity demand.
Google said that during 2025 it secured more than 12 GW of net-new clean generation, exceeding the total from the prior two years combined.
It said AI-driven electricity demand will require advanced nuclear, geothermal, natural gas with carbon capture, energy storage and demand response.
The remaining public gaps are the share of total compute that can move during grid stress, the operating method behind its 1 GW demand-response figure, or a timetable for closing the gap between AI infrastructure growth and hourly carbon-free power.
