The Next Scope 3 Blind Spot Could Be AI

Resource-constrained sustainability teams are increasingly turning to AI to handle growing volumes of data and to streamline carbon accounting and sustainability reporting. The Verdantix 2026 global corporate survey finds that 32% of firms already use AI in sustainability reporting, while 65% believe that agentic AI will transform sustainability data management. But as adoption grows, so does scrutiny of the environmental footprint of AI. Sustainability teams increasingly need to understand how AI's energy demand, infrastructure requirements and supplier dependencies affect emissions inventories.

Measuring AI emissions remains challenging

Watershed's recent whitepaper, Estimating GHG Emissions from AI Use, proposes a framework for estimating emissions from model inference, training, data centre overheads and embodied hardware emissions. The difficulty is that workloads vary, reporting boundaries differ and key data often sit with model providers. Published estimates can therefore differ by orders of magnitude, limiting useful benchmarking and target-setting.

Watershed argues for moving from spend-based estimates to activity-based accounting, using metrics such as token volumes, electricity use and provider-specific emission factors. It also calls for vendors to disclose model-level energy intensity, training emissions, serving-region data and infrastructure efficiency metrics, without which firms cannot credibly account for AI emissions.

The transparency gap around AI emissions

A Verdantix January 2026 Market Insight report found that none of the major hyperscalers disclose a consolidated figure for AI-specific emissions from the LLMs they develop. Instead, these emissions are folded into broader categories, such as cloud services, data centre operations or Scope 3 capital goods, making AI-specific emissions difficult to quantify. This lack of visibility is becoming more consequential as AI-related energy demand rises quickly: in its 2026 Environmental Report, Google reported a 37% increase in electricity consumption and acknowledged that AI infrastructure is expanding faster than grids decarbonize. The International Energy Agency (IEA) estimates that global data centre electricity consumption could more than double by 2030, reaching roughly 945TWh annually, with AI as the primary driver.

Policymakers are also focusing on the transparency gap. In June 2026, UN Secretary-General António Guterres launched the AI Environmental Transparency Initiative, calling on major AI firms to disclose the carbon, water and land-use impacts of their systems and data centres. The initiative reflects growing recognition that improved disclosure is a prerequisite for governance and accountability.

AI could become a Scope 3 reporting blind spot

The issue also has significant implications for Scope 3 reporting. For most organizations, emissions from third-party AI services, foundation models and AI-enabled software are likely to sit within Scope 3 Category 1 (Purchased Goods and Services). Yet without standardized provider data, firms have limited ability to quantify these emissions. As AI becomes embedded across enterprise software and customer-facing applications, more of organizations' digital footprints could sit within a reporting blind spot.

The question is no longer whether AI emissions should be accounted for, but whether the data will be good enough to do so credibly. Frameworks such as Watershed's help move the market beyond rough estimates, but without provider-level disclosure, firms will keep reporting AI emissions through proxies rather than evidence.

If you want to learn more about the use of AI in software and enterprise, read Verdantix Market Trends: Enterprise AI Agent Adoption, and for use cases of AI for sustainability, see Verdantix AI Applied Radar: AI Applied To Sustainability Management.