Powering AI: Insights from AMZN & GOOG Sustainability Reports

Research evidence excerpt

Powering AI: Insights from AMZN & GOOG Sustainability Reports

Barclays | Thematic Investing

Market-based Scope 2 emissions declined 3% y/y, but location-based emissions increased

37%. Google's market-based Scope 2 emissions decreased 3% y/y despite a 37% increase in

electricity consumption, as renewable power purchase agreements (PPAs), energy attribute

certificates (EACs), and limited use of granular certificates (GCs) more than offset higher

electricity demand under annual matching accounting conventions. However, under location-

based accounting—which reflects the average carbon intensity of the electric grids from which

Google physically consumes electricity—Scope 2 emissions increased 37% y/y to more than 15

million metric tons. The widening gap between market-based and location-based emissions

highlights the distinction between procuring clean-energy attributes and physically consuming

carbon-free electricity at the time and location where demand occurs.

Demand response becoming a strategic focus. Google disclosed that it has integrated ~1 GW

of demand-response capability into long-term energy contracts with multiple U.S. utilities,

highlighting growing efforts to improve grid flexibility and better align AI-driven load growth

with power system constraints. Based on our conversations with industry experts, views remain

mixed regarding the potential for data center load flexibility and the role hyperscalers can play

in demand-response programs. Proponents argue that flexible AI workloads could help shave

peak power demand, improve grid utilization, and accelerate data center deployment in power-

constrained regions. Skeptics, however, question the extent to which mission-critical AI and