Outcome
By the end of this lesson, you will be able to calculate a cloud workload’s carbon footprint, identify the four levers that drive it, and reason about which workloads are good candidates for carbon optimization.
| Prerequisites | T0, Foundations · M5.1, Maturity |
| Time | 9 minutes |
1. Concept
Your workloads run on electricity, and electricity is not equally clean everywhere.
Carbon intensity is how much carbon is emitted per unit of electricity, and it depends entirely on what the local grid burns: coal, gas, water, sun, wind or nuclear.
The consequence is that the same workload, unchanged, emits several times more in one region than another.
CARBON FOOTPRINT = compute_kWh × carbon_intensity (gCO2/kWh)
WHERE: compute_kWh = instance hours × power per instance type ≈ rough proxy: cost in $ × industry-typical $/kWh
carbon_intensity = depends on region's grid mix range: 30 gCO2/kWh (Norway hydro) to 700+ gCO2/kWh (coal-heavy regions)Region carbon intensity varies dramatically
REGION INTENSITY (gCO2/kWh) RELATIVE──────────────────────────────────────────────────────────────────eu-north-1 (Sweden/hydro) ~50 0.5× baselineca-central-1 (Quebec hydro) ~50 0.5×
us-west-2 (Oregon) ~120 1.2× (some hydro)us-west-1 (California) ~250 2.5× (solar-heavy but also gas)
us-east-1 (N. Virginia) ~350 3.5× (gas + coal)us-east-2 (Ohio) ~400 4× (coal-heavy)
ap-southeast-2 (Sydney) ~700 7× (coal-heavy)ap-southeast-1 (Singapore) ~500-600 5× (coal/gas)A workload in Sweden generates ~14× less carbon per kWh than the same workload in Sydney. The compute is identical; the grid is the difference.
What this means for cloud workloads
SAME workload running in different regions:
us-east-1 (Virginia, 350 gCO2/kWh): 100 m5.large × 730 hr/mo × 0.05 kW × 350g/kWh = 1,278 kg CO2/mo
eu-north-1 (Sweden, 50 gCO2/kWh): 100 m5.large × 730 hr/mo × 0.05 kW × 50g/kWh = 183 kg CO2/mo
CARBON DIFFERENCE: 1,095 kg CO2/mo (86% reduction) COST DIFFERENCE: minor (regional price variance, usually ±5%)
A team can move workloads to cleaner regions and reduce carbondramatically with minimal cost impact.The four levers
Four distinct ways to reduce a workload’s carbon footprint:
1. REGION SELECTION Move workloads to cleaner-grid regions Highest single lever (often 50-90% reduction)
2. COMPUTE-HOUR REDUCTION Schedule off when not needed (scale to zero, schedule on/off) Direct proportional reduction (same as cost saving)
3. PER-INSTANCE EFFICIENCY Use more efficient instance families (Graviton/ARM vs x86) Typically 20-40% more compute per watt Composable with the other levers
4. WORKLOAD-TIME ALIGNMENT Carbon-aware scheduling: run during clean grid hours Smaller but real impact (10-30%) for batch workloadsEach lever is independent. Stacking all four can produce 80-95% carbon reduction on flexible workloads.
Where carbon and cost align (and don’t)
CARBON LEVER COST IMPACT──────────────────────────────────────────────────────────────────Schedule off when not needed Same-direction savings (compute-hour reduction)
Graviton vs x86 instances Cheaper AND more efficient (~20% cost savings + carbon)
Region migration (clean grid) Minor cost variance; might be slightly cheaper or pricier
Time-shift to clean-grid hours Usually neutral cost (off-peak hours coincide with solar/wind peak in some regions)
Spot for batch Cheaper AND lower carbon (efficient utilization)The good news: most carbon levers also save cost. The exception: region migration, which is roughly carbon-positive and cost-neutral. The lever you pick depends on whether cost or carbon is the binding constraint.
Calculation example
WORKLOAD: 100 m5.large instances 24/7 in us-east-1
POWER DRAW (rough): m5.large ~50W (estimate; AWS does not publish per-instance wattage) 100 instances × 50W = 5 kW continuous 5 kW × 730 hr/mo = 3,650 kWh/mo
CARBON (us-east-1): 3,650 kWh × 350 gCO2/kWh = 1,277.5 kg CO2/mo Annual: 15.3 tons CO2
WHAT IF MIGRATED TO eu-north-1: 3,650 kWh × 50 gCO2/kWh = 182.5 kg CO2/mo Annual: 2.2 tons CO2
SAVINGS: 13.1 tons CO2/year (86%) COST IMPACT: minor (region price variance)86% carbon reduction. The cost impact is small enough that the decision is dominated by carbon considerations.
Hidden carbon factors
NOT JUST DIRECT COMPUTE: Cooling overhead (PUE: power usage effectiveness) Network equipment Storage (constant power for spinning disks; lower for SSD) Embodied carbon (manufacturing of hardware: long-term factor)
PUBLISHED CARBON DATA varies in what it includes: Some report compute only Some include PUE Some include lifecycle (embodied)
KNOW YOUR SOURCE.Where carbon numbers come from
PROVIDER TOOLS (account-level, monthly, lagged): AWS Customer Carbon Footprint Tool Azure Emissions Impact Dashboard Google Cloud Carbon Footprint
DO-IT-YOURSELF (workload-level): Hours x instance power estimate x region grid intensity Open-source Cloud Carbon Footprint for estimates from billing data Grid intensity from Electricity Maps or WattTimeZopNight does not measure carbon. Its cost and usage data (region, instance type, running hours) are useful inputs: carbon = energy used x grid intensity, so the hours and instance sizes ZopNight already tracks feed the energy side of the calculation.
2. Demo
A team calculating its annual carbon footprint:
TEAM Y workload assessment:
CURRENT STATE: Region: us-east-1 Instances: 100 m5.large × 730 hours/month Power: 5 kW continuous Annual kWh: 43,800 Annual carbon (us-east-1 350 gCO2/kWh): 15,330 kg = 15.3 tons CO2
OPTIMIZATION OPTIONS:
Option A: Schedule (non-prod): 60% of hours off Annual kWh: 17,520 Annual carbon: 6,130 kg = 6.1 tons CO2 REDUCTION: 60% (9.2 tons saved)
Option B: Move to eu-north-1 (Sweden): Same compute; cleaner grid (50 gCO2/kWh) Annual kWh: 43,800 Annual carbon: 2,190 kg = 2.2 tons CO2 REDUCTION: 86% (13.1 tons saved)
Option C: Combined (both): Sweden + 60% schedule Annual kWh: 17,520 Annual carbon: 876 kg = 0.9 tons CO2 REDUCTION: 94% (14.4 tons saved)
DECISION (per workload): Production workload (no schedule): Option B alone Non-prod workload: Option C combined
ANNUAL IMPACT for the team's full workload mix: Estimated: 12-15 tons CO2 reduction ESG report: meaningful improvementThe right combination of levers depends on workload class.
3. Hands-on (5 min)
Calculate one of your workload’s monthly carbon footprint:
WORKLOAD: __________REGION: __________ (look up intensity at electricitymaps.com)GRID INTENSITY: __________ gCO2/kWh
POWER DRAW estimate: Number of instances: __________ Instance type: __________ (look up power draw in cloud docs) Total power draw: __________ kW
MONTHLY HOURS: Average compute hours: __________ (730 if 24/7)
MONTHLY kWh: __________ kW × __________ hours = __________ kWh
MONTHLY CARBON: __________ kWh × __________ gCO2/kWh = __________ g CO2 = __________ kg CO2 = __________ tons CO2
ANNUAL CARBON: __________ tons CO2
LEVERS available: □ Region migration to cleaner grid □ Scheduling (if non-prod) □ Graviton migration (if eligible) □ Time-shift (if batch)
ESTIMATED reduction potential: __________ %Calculating once builds intuition. Track quarterly for trend.
4. Knowledge check
Q1
The same workload in eu-north-1 (Sweden) vs us-east-1 (Virginia):
A. Same carbon footprint
B. Sweden is ~7× cleaner due to hydroelectric grid
C. eu-north-1 is dirtier
D. Sweden is cleaner, but only marginally so
Show answer
Correct: B. Same compute, dramatically lower carbon. The cost difference is minor; the carbon difference is dramatic. Region selection is the single highest-leverage carbon decision.
Q2
A workload can be scheduled off-hours. Carbon impact:
A. Zero: carbon is fixed
B. Reduces more than proportionally
C. Reduces proportionally
D. Negative
Show answer
Correct: C. Fewer compute-hours means less electricity, which means less carbon. Scheduling is a carbon lever as well as a cost lever. The 60% schedule reduces both cost and carbon by ~60%.
Q3
Graviton (ARM) vs traditional x86 instances:
A. Same carbon
B. Higher carbon
C. Graviton is actually less efficient per watt, so the carbon saving comes entirely from the lower price rather than the hardware
D. Graviton is more compute-efficient: typically 20-40% less carbon per unit of work because ARM architecture uses less power per operation
Show answer
Correct: D. The cost savings often go together with the carbon savings.
5. Apply
Calculate your team’s annual carbon footprint using region intensity + workload size. Track quarterly. Use the result for ESG reporting and as an input to optimization prioritization.
ZopNight does not measure carbon. Use your provider’s carbon tool for account-level totals, and calculate workload-level carbon from region, instance type and running hours with public grid-intensity data.
Related lessons
- L2: Carbon-aware computing (next)
- L3: Region selection for carbon
- L4: Scheduling for carbon, not just cost
- L5: Reporting carbon to leadership
- T5.M5.1: Maturity ladder
Glossary terms touched
Carbon intensity · Carbon-aware scheduling