Summary
- Amazon’s 2025 sustainability reporting shows total emissions rising 16% year on year.
- Purchased electricity emissions rose as the company added large amounts of data centre capacity.
- The figures link AI capacity growth to electricity demand, construction emissions, and tougher sustainability reporting.
Amazon reported a 16% rise in total greenhouse gas emissions in 2025, as rapid data centre expansion pushed more electricity and construction impact into the company’s sustainability account.
The company’s latest sustainability reporting puts total emissions at about 80.85 million metric tonnes of CO2 equivalent. Purchased electricity emissions rose sharply, with Amazon linking the increase to data centre capacity additions and higher demand from cloud and AI services.
Amazon also said it added more data centre capacity globally in 2025 than any other company, including more than 1.2GW in the fourth quarter alone. That figure turns the sustainability story into an infrastructure story. Hyperscale data centre growth is now moving at power-station scale.
Efficiency cannot hide absolute growth
Amazon continues to report progress on renewable energy procurement, efficiency, custom silicon, cooling, and lower-carbon materials. Those measures can reduce the impact of each unit of compute, but they struggle to hold absolute emissions flat when total capacity expands quickly.
That is the central tension in the AI build-out. A data centre can have a strong PUE and still add substantial electricity demand if it is large enough. A company can sign renewable energy deals and still report rising emissions if new facilities, equipment, and purchased electricity grow faster than efficiency gains.
The carbon ledger also extends beyond operations. Scope 3 emissions are affected by steel, concrete, generators, switchgear, cooling equipment, servers, chips, logistics, and construction activity. AI data centres are physical assets before they become cloud services, and their embodied impact appears long before a workload goes live.
That complicates the way hyperscale sustainability claims are judged. Annual renewable matching can support decarbonisation, but it does not always mean the facility is consuming clean power at the same hour or in the same grid region as its load. As AI demand rises, the distinction between annual matching, hourly matching, additional clean energy, and local grid carbon intensity becomes harder to ignore.
Capacity arrives before the offset
The 1.2GW fourth-quarter capacity figure is the sharpest number in the report. It suggests a scale of data centre delivery that can move corporate emissions even when efficiency programmes continue.
New capacity carries carbon before utilisation reaches maturity. Materials are procured, sites are prepared, structures are built, electrical systems are installed, and IT hardware is manufactured. Once the facility enters operation, electricity use becomes a continuing load on the energy system.
European regulators are already moving towards more granular data centre reporting. The EU Energy Efficiency Directive requires significant facilities to report energy and environmental performance data. National authorities and local planning bodies are also paying more attention to power demand, water use, heat rejection, and carbon claims.
Corporate-level sustainability reporting does not replace facility-level evidence, but it gives context for why that evidence is needed. If hyperscale expansion can move global emissions figures, local planning authorities will want clearer answers about individual sites: connection capacity, power procurement, cooling design, water use, heat reuse, and embodied carbon.
The green story gets harder
Amazon is not alone. Other hyperscalers have also reported rising electricity demand, construction emissions, or water use as AI capacity expands. The pattern suggests the sector’s sustainability challenge is no longer mainly one of intent. It is one of scale.
The operational response will need to be more specific. Power purchase agreements must show location and additionality. Cooling claims must account for water and heat rejection. Construction programmes must address materials and embodied carbon. Hardware strategies must consider refresh cycles and supply-chain emissions.
Investors will also face closer scrutiny. Financing an AI data centre platform increasingly means financing power infrastructure, construction emissions, cooling systems, and long-term electricity demand. Sustainability claims that rely on broad corporate targets will be weaker where site-level data tells a more complicated story.
Amazon’s 2025 figures do not suggest the AI build-out is slowing. They show the physical bill arriving in the accounts. The sector can keep improving efficiency, but the credibility test is moving towards absolute resource use. The carbon story now runs through megawatts, materials, grids, and the speed at which capacity is being built.
