Map where energy and emissions arise across the AI lifecycle, from training through to inference and idle capacity.
Digital Transformation & Artificial Intelligence
Measuring and Reducing the Energy and Carbon Footprint of AI Workloads
Learn to measure the energy and carbon footprint of training and inference workloads and apply engineering and procurement levers that reduce it.
🗓️Duration5 training days
📚Content4 modules · 8 sessions
🎓On completionAccredited attendance certificate
About the programme
Course Overview
Training and running large AI models consumes measurable electricity and embodied hardware emissions, and as organisations scale generative AI from pilots to production, that consumption is starting to show up in sustainability reporting and infrastructure budgets alike. This course teaches participants to map where energy and emissions arise across the AI lifecycle, from training and fine-tuning through to inference at scale and idle accelerator capacity, and to apply carbon accounting concepts, including Scope 2 and Scope 3 categories and location-based versus market-based methods, to computing workloads. Participants instrument real training and inference jobs to measure energy consumption and attribute shared infrastructure emissions to specific models, teams or products, so trade-offs become visible alongside cost and performance. The course covers efficiency levers, including model compression, right-sized hardware, carbon-aware scheduling and cloud region choice, and procurement decisions, such as power purchase agreements and vendor efficiency commitments, that influence emissions beyond the organisation's own infrastructure. It closes with target-setting, reporting integration and governance that keeps energy and carbon performance under review alongside model quality.
Expected Learning Outcomes
Apply carbon accounting concepts, including Scope 2 and Scope 3 categories, to AI computing workloads.
Instrument training and inference workloads to measure their energy consumption and estimated emissions.
Attribute shared infrastructure emissions to specific models, teams or products for meaningful reporting.
Apply model, hardware and scheduling efficiency levers that reduce energy use without sacrificing required performance.
Assess cloud, procurement and renewable energy options for their real effect on workload emissions.
Set measurable efficiency targets and governance that keep AI energy and carbon performance under review.
Who Should Attend
Machine learning engineers responsible for training and serving infrastructure efficiency.
Sustainability and ESG teams incorporating AI workloads into carbon reporting.
Cloud infrastructure architects choosing regions, hardware and scheduling for AI workloads.
Data centre and platform operations teams managing shared AI compute capacity.
Procurement teams negotiating cloud and hardware contracts with efficiency commitments.
Chief technology officers balancing AI performance ambitions against sustainability commitments.
Course Modules
Select any module to see its sessions and points.
01Understanding the Energy and Carbon Impact of AI
2 sessions · 8 points
Session 1Where Energy and Emissions Arise Across the AI Lifecycle
- Map energy consumption across data collection, model training, fine-tuning, inference and idle infrastructure capacity.
- Distinguish the energy profile of training a large model from the energy profile of serving it at inference scale over time.
- Identify the embodied emissions in manufacturing GPUs and data centre hardware alongside the operational emissions of running them.
- Assess how model choice, context length and query volume each independently drive inference-time energy consumption.
Session 2Carbon Accounting Concepts for Computing Workloads
- Distinguish Scope 1, Scope 2 and Scope 3 emissions and identify where AI workloads typically fall within each category.
- Apply location-based and market-based accounting methods to calculate the emissions associated with a cloud workload.
- Assess the role of power usage effectiveness in translating a data centre's IT energy use into total facility energy use.
- Evaluate the limitations of self-reported cloud provider emissions figures when assembling an organisation's carbon inventory.
02Measuring AI Energy and Carbon Footprint
2 sessions · 8 points
Session 1Instrumenting Training and Inference Workloads
- Instrument training jobs to record GPU or accelerator utilisation, run duration and the energy mix of the region used.
- Deploy tools that estimate the energy consumption and carbon emissions of a specific training run or inference workload.
- Compare estimation approaches available through cloud provider dashboards against independent open-source measurement tools.
- Validate estimated figures against actual utility or cloud billing data to check measurement accuracy over time.
Session 2Attributing Emissions to Models, Teams and Products
- Build an attribution model that allocates shared infrastructure emissions to individual teams, products or business units.
- Report emissions per unit of useful output, such as per training run, per million tokens served or per active user.
- Present emissions data alongside cost and performance metrics so trade-offs are visible in the same review.
- Identify the highest-emitting workloads in a portfolio to prioritise where efficiency effort will have the greatest effect.
03Engineering for Efficiency Across the AI Stack
2 sessions · 8 points
Session 1Model, Hardware and Data Centre Efficiency Levers
- Apply model compression techniques, including distillation, pruning and quantisation, to reduce inference energy use.
- Select accelerator hardware and generation appropriate to a workload's performance needs rather than defaulting to the largest available.
- Route inference requests to smaller specialised models where they meet quality requirements instead of a larger general-purpose model.
- Assess data centre-level efficiency levers, including cooling design and hardware utilisation rates, that a cloud tenant can influence.
Session 2Workload Scheduling and Infrastructure Choices
- Schedule flexible training and batch inference workloads to run when grid electricity carbon intensity is lower.
- Choose cloud regions with a cleaner energy mix for workloads that are not latency-sensitive to a specific location.
- Right-size reserved and on-demand compute commitments to reduce idle accelerator capacity across a training fleet.
- Apply caching and batching strategies that reduce redundant inference calls for repeated or predictable queries.
04Procurement, Reporting and Governance
2 sessions · 8 points
Session 1Procurement and Renewable Energy Strategy
- Assess a cloud provider's renewable energy procurement and its actual effect on the emissions of a specific workload.
- Evaluate power purchase agreements and renewable energy certificates as complementary but distinct decarbonisation tools.
- Negotiate contractual commitments with infrastructure vendors for emissions reporting and efficiency improvement over time.
- Compare the total cost and emissions impact of on-premises, colocation and public cloud options for a sustained AI workload.
Session 2Reporting, Targets and Governance
- Set measurable efficiency and emissions targets for AI workloads consistent with the organisation's broader sustainability commitments.
- Integrate AI workload emissions into existing sustainability reporting frameworks and disclosure obligations.
- Establish governance ownership that reviews AI energy and carbon performance alongside model performance metrics.
- Communicate efficiency trade-offs to stakeholders when a lower-carbon option changes model latency, cost or capability.
What the participant receives
4 course modules
A structured syllabus
8 training sessions
across 5 days
32 detailed points
Applied, detailed content
Accredited attendance certificate
On completing the programme
Complete your registration
We will contact you within one business day to confirm.
Ready to start?
Reserve your seat and start building the skill.