According to external sources, the latest sustainable development reports released this week by Google and the Amazon provide more intuitive data on the environmental costs of infrastructure expansion in AI. Both companies are committed to zero net emissions in the coming years, but the latest disclosures indicate that this goal is being made more difficult by the electricity and hardware requirements of AI.
Emissions continue to rise.
The report shows that Google ' s total carbon emissions increased by 25 per cent compared to the previous year, while Amazon increased by 16 per cent. Neither company in the report attributed the increase in emissions directly to AI, but both mentioned that energy consumption had increased significantly as AI was used.
According to the article, what is of real concern is not just the purchase of electricity per se, but indirect emissions that are harder to contain. Over the past few years, technology companies have managed to control to some extent the emission pressure from operating electricity by purchasing renewable energy sources. However, with the rapid expansion of the AI data centre, the marginal effects of this practice are diminishing.
Data centres as a major source of pressure
In terms of disclosure structure, the main part of the increase in emissions from Google and the Amazon comes from so-called “range 3 emissions”, i.e., product-related emissions that enterprises cannot directly control. Such emissions typically include the procurement of capital equipment, construction, transport and the use of some end products.
Google has combined to disclose some of the emissions,3 but at the same time has indicated that the use of the products sold has had less impact on emissions. This means that data centre-related construction and equipment procurement is likely to be a more important source. The article mentions that Google's range of emissions increased by 2.1 million tons of CO2 equivalent over the past year, double the baseline level of 2019.
The Amazon 3 emissions growth, mainly from capital goods and fuel and energy-related projects. In its report, the company stated that in 2025 its new global data centre capacity exceeded that of any other company and that it had added more than 1.2 Giva in the fourth quarter alone. This explains why the associated emissions are rising faster.
Chip and building materials magnification costs
According to the article, the pressure from AI came not only from room power but also from the data centre construction itself. Steel and cement are high-emission industries, and low-carbon alternatives are still difficult to meet large-scale construction needs of large technology companies.
Another pressure comes from GPU and memory chip manufacturing. Semiconductor production itself is very energy-intensive, and many advanced process factories are located in Asia, which still relies on fossil energy grids. In addition, some of the chemicals used in chip manufacturing are also high-intensity greenhouse gases, which further exacerbates the emission burden in the AI supply chain.
Net zero commitments face higher costs
According to the article, Google, Amazon and other large technology companies are not unable to meet their emission reduction targets, but the path will be more expensive. In addition to continuing to expand renewable energy procurement, they may need to invest more in low-carbon steel and cement manufacturing and to purchase more carbon removal credits to offset hard-to-cut emissions.
The core change is that AI is moving from “power procurement issues” to “infrastructure and supply chain issues” on the environmental pressures of technology companies. This means that, in future, when measuring AI costs, emissions and energy constraints will become a much more difficult part of the calculation and capital expenditure.