A sustainable compute rating starts with the ledger, not the slogan.
Carnet separates local resource pressure, customer allocation, clean energy quality, carbon completeness, and evidence confidence. Each ledger answers a different gate: expansion, procurement, financing, disclosure, or verification.
The method covers six accounts.
Completeness and granularity are different. Carnet treats resource completeness, workload allocation, and accepted evidence as separate parts of the same record.
1. Asset and compute boundary
Defines facility, cluster, workload, customer unit, geography, time period, and operating owner.
- Facility ID and operating entity
- Cluster and accelerator inventory
- Workload, API, token, or GPU-hour unit
2. Power and capacity pressure
Connects AI compute load to local electricity use, grid capacity, peak draw, and flexible-load potential.
- IT energy and total facility energy
- Contracted and available capacity
- Peak load and congestion exposure
3. Carbon and clean energy
Calculates operating carbon, hardware completeness, clean energy match, and certificate or PPA quality.
- Grid carbon intensity and market instruments
- Annual, monthly, or hourly matching
- Embodied carbon boundary status
4. Water, cooling, and heat
Measures WUE, cooling route, water-stress exposure, heat reuse, and local environmental impact.
- Water source and stress region
- Cooling system and seasonal efficiency
- Heat reuse or discharge controls
5. Allocation and claim rights
Assigns resource responsibility to customer, contract, workload, GPU-hour, API call, or reporting period.
- Allocation basis and denominator
- Customer claim language
- Double-claim and retirement checks
6. Evidence and verification
Stores source files, reviewer decisions, optimization actions, and rating changes in an evidence room.
- Meter, utility, and workload logs
- EAC, PPA, REC, or GO documents
- Verifier review status and exceptions
Rating levels make the maturity visible.
A single carbon intensity number hides whether the evidence is complete. Carnet uses levels so a compute product can move from basic reporting to verified sustainable compute without pretending every site is already at the final state.
Optimization is coordinated, then verified.
Carnet does not need to become the electricity provider, cooling contractor, or hardware vendor. The platform owns the record: baseline, recommended pathway, partner action, result, rating change, and accepted evidence.
Partner optimization routes
Green electricity procurement, EAC / REC / GO matching, PPA documentation, behind-the-meter supply, flexible load shaping, cooling upgrades, water-risk mitigation, and cloud architecture optimization.
- Partner scope and responsibility documented
- Cost, carbon, water, and claim-quality impact modeled
- Implementation evidence attached to the record
Verification outputs
Updated rating, reviewer notes, exceptions, eligible claim language, allocation table, evidence room, and a compact customer-facing sustainable compute record.
- Rating and level before / after optimization
- Claim wording matched to evidence level
- Methodology version and reviewer status preserved
Request the SCP Methodology v1.0 pack.
The full pack includes ledger definitions, rating levels, evidence templates, and optimization review rules.