As artificial intelligence becomes the engine of modern corporate strategy, the race to integrate large language models (LLMs) and generative AI has moved at breakneck speed. From automating supply chain logistics to drafting complex legal documents, AI is being woven into the fabric of daily operations. However, this digital transformation is creating a significant environmental blind spot. While companies are rapidly adopting these technologies, the frameworks required to track, report, and mitigate the resulting greenhouse gas (GHG) emissions and resource depletion—such as the massive freshwater withdrawals required for data center cooling—are lagging dangerously behind.
This disconnect has prompted a call to action from industry leaders, most notably carbon management software firm Watershed. Recognizing that sustainability teams are struggling to quantify the impact of their AI procurement, Watershed has proposed a "defensible starting point" to help enterprises estimate their exposure and take accountability for their digital carbon footprint.
The AI Emissions Gap: A New Metric for a New Era
The fundamental challenge in AI sustainability is a lack of standardization. Historically, corporate emissions accounting—under protocols like the Greenhouse Gas Protocol—focused on scope 1, 2, and 3 emissions tied to physical assets, manufacturing, and travel. AI, however, is a distributed, intangible, and highly intensive energy consumer.
In a comprehensive 43-page white paper published in mid-July, Watershed advocates for a shift in focus: token-level reporting. Because AI models process data in "tokens"—the small snippets of code and text that constitute a prompt—they represent the most granular unit of consumption. Watershed proposes that companies report their AI environmental impact using a standardized metric: kilograms of carbon dioxide equivalent (CO2e) per million tokens, accompanied by corresponding electricity consumption data.
"Tokens are also the metric that many companies track for cost reasons," says the Watershed report. By aligning environmental accounting with financial accounting, companies can leverage existing engineering efforts to track sustainability metrics, effectively turning carbon management into an active, data-driven business process.
Chronology: The Evolution of AI’s Environmental Scrutiny
To understand why this accounting shift is necessary now, one must look at the rapid acceleration of AI infrastructure over the last 24 months.
- Early 2023: The global AI boom begins in earnest. Organizations rush to adopt generative AI without mature frameworks to measure the underlying energy load.
- Late 2023 – Early 2024: Industry analysts begin to flag that the energy consumption of AI training and inference (the process of using a trained model) is threatening to derail corporate net-zero targets.
- Mid-2024: The "Data Center Dilemma" takes center stage. Sustainability professionals report difficulties in obtaining granular data from AI vendors.
- July 2024: Watershed releases its white paper, providing a methodology for estimating AI-related emissions, aiming to bridge the gap between AI procurement and ESG (Environmental, Social, and Governance) reporting.
- August 2025 (Forecasted/Ongoing): Major cloud providers like Google and Microsoft begin publishing technical papers detailing the specific energy, emissions, and water footprints of their AI inference services, signaling a shift toward greater transparency.
Supporting Data: Why Granularity Matters
The data on AI emissions remains notoriously opaque. While organizations like the AI Emissions Tracker have begun rating data center companies based on their sustainability practices, these reports often provide a "high-level" view that is insufficient for the granular reporting required by modern regulatory frameworks like the CSRD (Corporate Sustainability Reporting Directive) in Europe or the SEC’s emerging climate disclosure rules.
The core of the problem lies in the split between "Frontier AI" developers and cloud service providers. Currently, companies like Anthropic and OpenAI—the architects of the models—have yet to make comprehensive emissions reporting a core pillar of their developer toolkits. Conversely, cloud giants like Amazon (AWS), Google, and Microsoft are more motivated to disclose data, largely because their own massive, publicly committed climate goals depend on the efficiency of their data centers.
The Cloud Provider Disclosure Landscape:
- Amazon (AWS): Has launched resources to help customers view cloud metrics, though these tools often fail to isolate AI-specific workloads from general cloud storage and compute.
- Google: In August 2025, the company published a technical framework detailing the energy and water impacts of its Gemini prompts, setting a high bar for industry transparency.
- Microsoft: In mid-2026, the company shared findings on how hardware optimization can lead to an 8x to 20x improvement in energy efficiency for AI inference, demonstrating the potential for "green" engineering.
Despite these advancements, the "granular metrics" needed for precise carbon accounting remain elusive for the average sustainability professional.
Official Responses and Strategic Perspectives
Industry experts argue that the time for "wait and see" has passed. John Bistline, head of science at Watershed, emphasizes that accounting is not merely a compliance exercise; it is an active management tool.
"As provider disclosure expands, companies will be able to compare AI emissions across vendors, factor emissions into procurement decisions, and track efficiency improvements year over year," Bistline explains.
The Watershed proposal encourages sustainability teams to take four key actions immediately, even in the absence of perfect data:
- Standardize the Token Metric: Begin tracking token usage across all internal and external AI projects.
- Engage Vendors: Demand transparency regarding the carbon intensity of the specific data centers hosting the models being used.
- Procure Clean Energy: Use the calculated emission estimates to justify the purchase of matching clean energy credits or to negotiate cleaner energy mixes with providers.
- Hardware Efficiency Optimization: Work with IT and engineering teams to choose models that are "right-sized" for the task, avoiding the use of massive, energy-hungry models for simple, low-stakes operations.
The Broader Implications for Corporate Sustainability
The shift toward AI-based sustainability accounting has profound implications for how corporations will function in the next decade.
1. From IT Cost to Environmental Cost
For decades, IT procurement was driven by the "three pillars": cost, speed, and reliability. Now, a fourth pillar—carbon efficiency—is being added. As companies begin to see the "kilograms of CO2 per million tokens" figure on their dashboards, it will likely lead to a cultural shift in how developers write prompts and select models. When carbon efficiency is tied to a P&L statement, engineers will naturally gravitate toward smaller, more efficient models that accomplish the same task with 10% of the energy.
2. The Freshwater Conflict
Beyond electricity, AI data centers are significant consumers of freshwater, used for cooling server infrastructure. As climate change exacerbates water scarcity, companies that do not track their AI-related water usage risk significant reputational and operational damage. The Watershed methodology serves as a warning: if you are not measuring your water-to-token ratio, you are effectively operating with an unknown liability.
3. Regulatory Readiness
With global regulators moving toward more stringent ESG disclosures, the "Wild West" of AI reporting is coming to a close. Companies that adopt a "defensible starting point"—even if that starting point relies on estimates today—will be better positioned to satisfy future audit requirements than those that claim they "could not measure" their digital footprint.
4. The Competitive Edge
Sustainability is becoming a competitive advantage. Companies that can demonstrate a low-carbon AI strategy will be more attractive to ESG-focused investors and eco-conscious customers. By integrating AI emissions into their sustainability reports, firms can tell a story of technological progress that is balanced by environmental stewardship.
Conclusion: The Urgency of Action
The narrative that AI is a "clean" technology—simply because it exists in the cloud—is being dismantled. The infrastructure supporting the AI revolution is physical, energy-intensive, and resource-heavy.
As Watershed and other pioneers in the sustainability space suggest, the path forward is not to stop AI adoption, but to make it visible. By adopting token-level reporting and pushing for greater transparency from cloud providers, organizations can move from the current era of "carbon guesswork" to an era of "carbon intelligence."
The technology is evolving faster than our ability to account for it, but the tools to bridge that gap are beginning to emerge. For sustainability professionals, the message is clear: do not wait for a perfect, industry-wide standard to be handed down from above. Start with the tokens you have, apply the metrics available, and turn your AI usage into an opportunity for environmental leadership. The future of the digital enterprise depends not just on how well we code, but on how responsibly we power the intelligence we create.
