Category: Technology Tags: Anthropic, Claude AI, carbon emissions, artificial intelligence, sustainability, Scope 3
Businesses adopting Anthropic’s Claude and other artificial intelligence tools should account for both the operational benefits and the environmental reporting challenges associated with AI use, according to an analysis from Tunley Environmental.
The analysis, prepared by Tunley Environmental Science Team Co-Lead Dr Aaron Yeardley and Head of Marketing Ellis Clark, says AI platforms are increasingly supporting research, writing, coding, data analysis and customer service. However, organisations with net-zero targets or Scope 3 reporting obligations may have limited supplier-specific information for estimating the carbon impact of these services.
Tunley said that, as of mid-2026, Anthropic had not published a full corporate sustainability report containing audited Scope 1, Scope 2 and Scope 3 emissions. The analysis therefore uses a spend-based emissions factor as a reporting proxy rather than claiming a measured Anthropic-specific footprint.
Using a factor of 0.1177 kg CO2e per pound spent, Tunley estimates that annual spending of GBP 10,000 on Anthropic or a similar AI and software service would represent 1,177 kg CO2e, or 1.177 tonnes CO2e. At an illustrative offset price of GBP 15 per tonne, the associated cost would be GBP 17.66.
For comparison, the analysis applies publicly reported organisational emissions intensity to estimate 0.627 tonnes CO2e for the same annual spend on Microsoft Copilot and 0.513 tonnes CO2e for Google Gemini. The corresponding illustrative offset costs would be GBP 9.41 and GBP 7.70.
Tunley stressed that these calculations use different reporting approaches and should be understood as accounting estimates, not direct measurements of the emissions produced by an individual AI request.
The article also cites indicative 2025-2026 estimates of around 0.03 g CO2e for a typical Gemini text query, 0.13-0.19 g CO2e for a ChatGPT GPT-4o query and 0.2-0.4 g CO2e for standard Claude models. It notes that actual emissions can vary with model size, query complexity, output length, data-centre efficiency and the electricity mix.
According to the analysis, the direct impact of moderate text-based AI use may remain small compared with emissions from energy, transport, purchased goods and wider supply chains. The larger concern is cumulative demand as AI adoption expands and more computing-intensive applications, including advanced agents and image or video generation, become common.
Tunley recommends that organisations identify which AI tools are in use, distinguish routine text activity from more energy-intensive workloads, include AI services in Scope 3 reporting where relevant and request better sustainability data from suppliers. Where supplier-specific information is unavailable, businesses can use a transparent spend-based proxy while clearly explaining the limitations of the estimate.
The consultancy also advises companies to keep AI emissions in proportion to their overall footprint and to focus carbon-reduction work on the most material sources. Its conclusion is that businesses do not need to reject AI, but should use it thoughtfully, measure its impact where possible and communicate the methodology honestly.
Source: Contributed analysis by Dr Aaron Yeardley and Ellis Clark of Tunley Environmental, shared with Newz9 by Tunley Environmental.
Businesses adopting Anthropic’s Claude and other artificial intelligence tools should account for both the operational benefits and the environmental reporting challenges associated with AI use, according to an analysis from Tunley Environmental.
The analysis, prepared by Tunley Environmental Science Team Co-Lead Dr Aaron Yeardley and Head of Marketing Ellis Clark, says AI platforms are increasingly supporting research, writing, coding, data analysis and customer service. However, organisations with net-zero targets or Scope 3 reporting obligations may have limited supplier-specific information for estimating the carbon impact of these services.
Tunley said that, as of mid-2026, Anthropic had not published a full corporate sustainability report containing audited Scope 1, Scope 2 and Scope 3 emissions. The analysis therefore uses a spend-based emissions factor as a reporting proxy rather than claiming a measured Anthropic-specific footprint.
Using a factor of 0.1177 kg CO2e per pound spent, Tunley estimates that annual spending of GBP 10,000 on Anthropic or a similar AI and software service would represent 1,177 kg CO2e, or 1.177 tonnes CO2e. At an illustrative offset price of GBP 15 per tonne, the associated cost would be GBP 17.66.
For comparison, the analysis applies publicly reported organisational emissions intensity to estimate 0.627 tonnes CO2e for the same annual spend on Microsoft Copilot and 0.513 tonnes CO2e for Google Gemini. The corresponding illustrative offset costs would be GBP 9.41 and GBP 7.70.
Tunley stressed that these calculations use different reporting approaches and should be understood as accounting estimates, not direct measurements of the emissions produced by an individual AI request.
The article also cites indicative 2025-2026 estimates of around 0.03 g CO2e for a typical Gemini text query, 0.13-0.19 g CO2e for a ChatGPT GPT-4o query and 0.2-0.4 g CO2e for standard Claude models. It notes that actual emissions can vary with model size, query complexity, output length, data-centre efficiency and the electricity mix.
According to the analysis, the direct impact of moderate text-based AI use may remain small compared with emissions from energy, transport, purchased goods and wider supply chains. The larger concern is cumulative demand as AI adoption expands and more computing-intensive applications, including advanced agents and image or video generation, become common.
Tunley recommends that organisations identify which AI tools are in use, distinguish routine text activity from more energy-intensive workloads, include AI services in Scope 3 reporting where relevant and request better sustainability data from suppliers. Where supplier-specific information is unavailable, businesses can use a transparent spend-based proxy while clearly explaining the limitations of the estimate.
The consultancy also advises companies to keep AI emissions in proportion to their overall footprint and to focus carbon-reduction work on the most material sources. Its conclusion is that businesses do not need to reject AI, but should use it thoughtfully, measure its impact where possible and communicate the methodology honestly.
Source: Contributed analysis by Dr Aaron Yeardley and Ellis Clark of Tunley Environmental, shared with Newz9 by Tunley Environmental.
