Why This Matters Now
As companies adopt AI tools, cloud platforms, and SaaS products at scale, a new and often-overlooked source of greenhouse gas emissions is growing rapidly: the energy consumed by data centres and digital infrastructure. Under AASB S2 and IFRS S2, these emissions fall squarely within Scope 3 — specifically Category 1 (Purchased Goods and Services) and, in some cases, Category 11 (Use of Sold Products).
Most businesses have no idea how to measure them. This article provides a step-by-step, audit-ready approach to quantifying Scope 3 emissions from AI and cloud services — grounded in real-world practice, not theory.
Step 1: Map Your Digital Supply Chain
Start by identifying every cloud and AI service your organisation uses. This includes IaaS/PaaS providers (AWS, Azure, GCP), SaaS platforms (Salesforce, HubSpot, Xero), AI tools (ChatGPT, Copilot, Gemini), and data storage and backup services. Create a register that captures the provider, service type, estimated usage (compute hours, storage volume, API calls), and data centre region.
Step 2: Identify the Emission Sources
Cloud and AI emissions come from three main sources: electricity consumed by servers and GPUs, embodied carbon in hardware manufacturing and replacement cycles, and cooling and ancillary energy use in data centres. For AI specifically, training large language models is extremely energy-intensive — a single GPT-4 training run has been estimated at 1,000–10,000 MWh. Inference (i.e., using the model) is less intensive per query but scales with usage.
Step 3: Source Emission Factors
The best source of emission data is the provider itself. Major cloud providers now publish sustainability reports and carbon dashboards: AWS Customer Carbon Footprint Tool, Google Cloud Carbon Footprint, and Microsoft Emissions Impact Dashboard. Where provider data is unavailable, use published benchmarks such as the IEA global average data centre PUE of 1.58, grid emission factors from the National Greenhouse Accounts Factors (DCCEEW), and academic estimates for AI model training and inference energy.
Step 4: Calculate Emissions
The basic formula is: Emissions (tCO₂e) = Energy Consumed (MWh) × Grid Emission Factor (tCO₂e/MWh) × PUE. For cloud services, if the provider gives you a direct carbon figure, use that. If not, estimate energy from usage metrics (compute hours × average server power draw) and apply the grid emission factor for the data centre region. For AI tools, estimate per-query energy use (published benchmarks range from 0.001–0.01 kWh per query for inference) and multiply by your usage volume.
Step 5: Document and Disclose
For AASB S2 / IFRS S2 compliance, you need to disclose Scope 3 emissions by category. Digital infrastructure emissions should be reported under Category 1 (Purchased Goods and Services). Your disclosure should include methodology and data sources, assumptions and limitations, year-on-year trends, and any reduction targets or initiatives. Keep an audit trail: save provider reports, document your calculation methodology, and version-control your emission factors.
Practical Tips
- Start with your largest providers — they likely account for 80%+ of your digital emissions.
- Use provider carbon dashboards where available — they are the most accurate source.
- For AI tools, track usage volume (API calls, queries) and apply published energy benchmarks.
- Review data centre locations — renewable energy grids (e.g., Norway, Iceland) have much lower emission factors than coal-heavy grids.
- Reassess annually — provider efficiency, grid mix, and your usage patterns all change.
Need Help?
Empire Carbon & Energy helps industrial and mid-market businesses quantify, report, and reduce their Scope 3 emissions — including digital infrastructure. If you need audit-ready emissions accounting or a practical decarbonisation roadmap, get in touch.