Consumer-Based Carbon Costs Analysis

• Consumer-based carbon costs are defined as the emissions and monetized exposures assigned to consumers using various frameworks like Scope 2 accounting, average and marginal emission factors, and market-clearing rules.
• These attribution frameworks, such as location-based and market-based methods, determine differing estimates of emissions and influence claims of carbon savings and pricing signals.
• They drive operational strategies in electricity markets and digital systems by affecting load shifting, tariff design, and ultimately the accuracy of carbon cost reporting.

Consumer-based carbon costs are the emissions and monetized carbon exposure attributed to the consumer of electricity, fuels, transport, or computation, rather than solely to the producer or generator. In electricity systems, they arise from Scope 2 accounting, carbon intensity attribution, and market-clearing rules that assign emissions from generation to loads; in broader climate-cost applications, they also denote per-unit price adders derived from social cost of greenhouse gases or from upstream obligations such as a Carbon Takeback Obligation (Maji et al., 2024, Jiang et al., 16 Jan 2025, Hanley et al., 31 Dec 2025, Jenkins et al., 2020). The concept is therefore not a single metric but a family of attribution and pricing frameworks that determine who is deemed responsible for emissions, which signal guides operational decisions, and how carbon exposure enters consumer payments, budgets, or welfare.

1. Conceptual foundations and accounting boundaries

In commercial and industrial electricity use, consumer-based carbon costs capture the emissions and potential monetized carbon exposure associated with a facility’s Scope 2 electricity use, and how those emissions co-vary with the prices the facility actually pays. For an electricity load profile $L_t$ and time-varying emission factor $EF_t$, Scope 2 emissions are written as

$E_{\text{Scope2}} = \sum_t L_t \cdot EF_t,$

while electricity cost under time-varying price $P_t$ is

$$C = \sum_t L_t \cdot P_t + C_{\text{demand}} + C_{\text{fixed}}.$$

When a carbon price $\kappa$ is applied to attributed emissions, the carbon component becomes

$$K_{\text{carbon}} = \sum_t L_t \cdot EF_t \cdot \kappa.$$

A central distinction is between average emission factors and marginal emission factors: AEFs are used for planning with inflexible or pre-existing loads, whereas MEFs are used for operational flexibility, load shifting, and siting new flexible loads. In 2023 data, MEFs are on average about 50% higher than corresponding AEFs across regions, reflecting dirtier marginal supply (Chapin et al., 13 Nov 2025).

In electricity market design, the same idea appears as a consumer-submitted carbon disutility. In the consumer-side formulations of carbon-aware market clearing, each load $i$ can submit a nonnegative carbon cost $\kappa_i$ or $c_{D,d}$, representing the cost it assigns to the emissions incurred by its electricity use. If $EF_t$0 denotes emissions allocated to load $EF_t$1, then $EF_t$2 enters social welfare directly, alongside utility from consumption and generation cost. This construction differs from producer-only carbon pricing because consumers can be heterogeneous: carbon-agnostic consumers have $EF_t$3, whereas carbon-sensitive consumers have $EF_t$4 (Jiang et al., 16 Jan 2025, Jiang et al., 1 Aug 2025).

These formulations also define the boundary between consumer-based and producer-based carbon pricing. A uniform consumer carbon cost is equivalent to a uniform carbon tax on generators. Specifically, if all consumers have the same carbon cost, the total consumer-side penalty collapses to the system-wide emissions term:

$EF_t$5

By contrast, heterogeneous consumer costs create differentiated demand-side pull for low-emission generation and can alter both dispatch and allocation across loads (Jiang et al., 16 Jan 2025).

2. Attribution frameworks for electricity consumption

A major branch of consumer-based carbon costs concerns how electricity emissions are attributed to consumers under alternative carbon-intensity frameworks. The average grid carbon intensity over sources $EF_t$6 with energy $EF_t$7 and emission factors $EF_t$8 is

$EF_t$9

Under location-based attribution, all consumers in a grid are assigned the same total generation mix:

$E_{\text{Scope2}} = \sum_t L_t \cdot EF_t,$0

Under market-based attribution, contracted carbon-free generation is first attributed to investing consumers through PPAs and RECs/EACs, and non-investing consumers receive the residual mix. If $E_{\text{Scope2}} = \sum_t L_t \cdot EF_t,$1 is contracted energy from source $E_{\text{Scope2}} = \sum_t L_t \cdot EF_t,$2 and $E_{\text{Scope2}} = \sum_t L_t \cdot EF_t,$3, the residual intensity is

$E_{\text{Scope2}} = \sum_t L_t \cdot EF_t,$4

For a consumer whose PPAs meet a fraction $E_{\text{Scope2}} = \sum_t L_t \cdot EF_t,$5 of consumption,

$E_{\text{Scope2}} = \sum_t L_t \cdot EF_t,$6

If $E_{\text{Scope2}} = \sum_t L_t \cdot EF_t,$7, then $E_{\text{Scope2}} = \sum_t L_t \cdot EF_t,$8; if $E_{\text{Scope2}} = \sum_t L_t \cdot EF_t,$9, then $P_t$0 (Maji et al., 2024).

These formulas matter because operational decisions are often driven by the carbon signal itself. Consumer emissions over time and space are

$P_t$1

Under location-based evaluation, $P_t$2; under market-based evaluation for a consumer without PPAs, $P_t$3. The divergence is not merely formal. Across 123 regions worldwide using Electricity Maps 2022 data, if all solar and wind are under PPAs, the residual mix carbon intensity $P_t$4 increases substantially relative to $P_t$5: the median increase is 11.9%, and the maximum is 194% in South Australia. California exhibits larger seasonal and diurnal differences in spring and summer daytime, when solar is abundant but contracted (Maji et al., 2024).

The practical consequence is that concurrent use of location-based optimization and market-based evaluation can materially distort reported savings. Across three state-of-the-art carbon-aware techniques, the reported overestimation can reach 55.1%, and some cases show increased emissions rather than reductions. In resource autoscaling across more than 100 regions, the discrepancy CDF shows an average discrepancy of 11.9% and a maximum of 55.1%. In temporal load shifting with $P_t$6 hour flexibility, the discrepancy CDF shows an average discrepancy of 13.7% and a maximum of 50.8%. For consumers without PPAs, market-based optimization can also yield up to 28.2% less carbon savings than location-based claims, because the residual mix is browner and less variable (Maji et al., 2024).

This attribution choice propagates directly to monetized consumer-based carbon costs. If a consumer applies an internal carbon price $P_t$7 in \$P_t$8$P_t$9%%%%4$L_t$4$EF_t$4%%%%2CI$C = \sum_t L_t \cdot P_t + C_{\text{demand}} + C_{\text{fixed}}.$310/1.551 \approx 6.44$C = \sum_t L_t \cdot P_t + C_{\text{demand}} + C_{\text{fixed}}.$4p=\$C = \sum_t L_t \cdot P_t + C_{\text{demand}} + C_{\text{fixed}}.$5/tCO2e, claimed cost avoidance is \$C = \sum_t L_t \cdot P_t + C_{\text{demand}} + C_{\text{fixed}}.$6644, a mispricing of $C = \sum_t L_t \cdot P_t + C_{\text{demand}} + C_{\text{fixed}}.$7 (Maji et al., 2024).

3. Consumer-side carbon costs in electricity market clearing

A distinct research line embeds consumer-based carbon costs directly in power-system optimization. In these models, generator $C = \sum_t L_t \cdot P_t + C_{\text{demand}} + C_{\text{fixed}}.$8 with dispatch $C = \sum_t L_t \cdot P_t + C_{\text{demand}} + C_{\text{fixed}}.$9 and emission factor $\kappa$0 produces total emissions $\kappa$1, and a nonnegative allocation decision $\kappa$2 assigns a portion of generator output to each load $\kappa$3. The allocation constraints are

$\kappa$4

Load-level assigned emissions are

$\kappa$5

Conservation follows immediately:

$\kappa$6

The market-clearing problem then maximizes

$\kappa$7

subject to DC-OPF constraints, generator and load bounds, and the allocation constraints above. Under the DC approximation this is a linear program and is convex (Jiang et al., 16 Jan 2025).

The allocation layer gives the model its consumer-specific character. The KKT stationarity condition for $\kappa$8,

$\kappa$9

implies that the market favors allocating low-emission generation to high-$K_{\text{carbon}} = \sum_t L_t \cdot EF_t \cdot \kappa.$0 loads. The 2025 equilibrium formulation extends this into a decentralized interpretation with a carbon manager and carbon-adjusted prices. For generators, the relevant price is $K_{\text{carbon}} = \sum_t L_t \cdot EF_t \cdot \kappa.$1; for consumers, it is $K_{\text{carbon}} = \sum_t L_t \cdot EF_t \cdot \kappa.$2. Theorem-level results show that generator carbon adjustments decrease with emission factors, while consumer carbon adjustments increase with decreasing carbon costs: low-emitting generators receive higher carbon-adjusted prices, and high carbon-cost consumers pay more for cleaner electricity (Jiang et al., 1 Aug 2025).

Several standard market properties are preserved. Revenue adequacy holds, with budget balance at the optimum; individual rationality holds when lower bounds on generation and demand are zero; and the standard market is recovered when all consumer carbon costs are zero. Uniform consumer carbon costs are equivalent to a generator carbon tax, but heterogeneous consumer costs are not, because they alter the allocation of clean generation across loads rather than simply penalizing emissions at source (Jiang et al., 1 Aug 2025).

Quantitatively, the effect can be large. In a simplified three-bus system, introducing consumer carbon costs decreased total generation by 28.8% and total emissions by 53.3% relative to the carbon-agnostic case. In the IEEE RTS-GMLC system, when all loads were carbon-sensitive with $K_{\text{carbon}} = \sum_t L_t \cdot EF_t \cdot \kappa.$3, total generation remained unchanged at 8550 MWh, generation cost rose about 1.5%, and total emissions fell about 4.5%; with $K_{\text{carbon}} = \sum_t L_t \cdot EF_t \cdot \kappa.$4, total generation fell 5.6%, generation cost fell 7.5%, and total emissions fell 17.5% (Jiang et al., 16 Jan 2025). A plausible implication is that consumer-based carbon costs act through two margins: generation redispatch at moderate carbon costs, and demand reduction once carbon-adjusted prices approach or exceed utility values.

4. Operational signals, optimization, and digital infrastructure

Consumer-based carbon costs are also shaped by the signals that guide operational flexibility. For commercial and industrial consumers, tariffs, demand response, and wholesale prices do not align uniformly with emissions signals. The national 2023 dataset of 1,492 C&I tariffs, hourly AEFs by balancing authority, hourly MEFs by ISO/RTO, and 138 IBDR programs across 45 states and DC shows broad misalignment of economic and emissions incentives under existing electricity tariff structures. Tariff–AEF correlation is negative on average, meaning retail tariffs tend to incentivize consumption during more carbon-intensive hours when viewed against planning signals, whereas DAM–MEF correlation is positive on average, so operational wholesale price signals can serve as a proxy for marginal emissions. DAM price variation can reach up to 60× within an ISO; tariff costs can differ by more than 100× within the same ISO (Chapin et al., 13 Nov 2025).

This distinction matters for flexible loads. For shifting $$K_{\text{carbon}} = \sum_t L_t \cdot EF_t \cdot \kappa.$$5 from hour $$K_{\text{carbon}} = \sum_t L_t \cdot EF_t \cdot \kappa.$$6 to hour $$K_{\text{carbon}} = \sum_t L_t \cdot EF_t \cdot \kappa.$$7, emissions savings and cost savings are

$$K_{\text{carbon}} = \sum_t L_t \cdot EF_t \cdot \kappa.$$8

Under positive DAM–MEF correlation, low-price hours tend to be low-MEF hours, so cost minimization and emissions minimization are aligned. Under negative tariff–AEF correlation, tariff minimization can raise Scope 2 emissions and therefore raise consumer-based carbon costs when these emissions are monetized (Chapin et al., 13 Nov 2025).

A more acute operational issue arises when carbon-aware schedulers optimize using one accounting framework and are evaluated under another. Spatial load shifting, temporal load shifting, and carbon-aware autoscaling all change behavior under location-based versus market-based signals. In the spatial case, the representative decision rule is

$$K_{\text{carbon}} = \sum_t L_t \cdot EF_t \cdot \kappa.$$9

where $i$0 is the regional carbon intensity and $i$1 is a latency proxy. Under location-based optimization and location-based evaluation, the carbon-aware load balancer reduces emissions by up to 15.4% per kWh relative to a baseline that picks the nearest data center; under market-based evaluation for the same decisions, the same strategy can yield up to 3.1% more emissions per kWh than the baseline, with a maximum discrepancy of 17.2% per kWh on the representative day (Maji et al., 2024).

The same pattern holds for temporal scheduling and autoscaling. In California, a nightly 2-hour job with $i$2 hour flexibility yields 34.8% savings under location-based evaluation but only 10.3% savings under market-based evaluation with all renewables under PPA. In Germany, location-based evaluation shows 13.7% savings, while market-based evaluation can show up to a 2.5% emissions increase for the same schedule. In resource autoscaling, location-based evaluation reports average savings of 29.2% across five AWS regions relative to a one-instance baseline, but market-based evaluation for the same schedule reports only 9.9% average savings; the average discrepancy is 19.2%, with a maximum discrepancy of 32.2% in California (Maji et al., 2024).

Digital infrastructure has also operationalized consumer-based carbon costs directly at the user/job level. In high-performance computing, Energy-Based Accounting charges users according to

$i$3

while Carbon-Based Accounting charges

$i$4

and

$i$5

These schemes replace core-hours with energy or carbon credits. Their empirical motivation is behavioral: in a survey of 316 HPC users, only 27% knew their energy use and 30% had taken steps to reduce energy; in a user study, Energy-Based Accounting reduced average energy use to 1,928 kWh versus 3,262 kWh and 3,142 kWh under the two runtime-priced baselines, while simply showing energy information did not produce a significant difference (Kamatar et al., 16 Jan 2025).

5. Sectoral incidence and price pass-through beyond load attribution

Consumer-based carbon costs also emerge through explicit pass-through channels in commodity and transport markets. In European day-ahead electricity markets with uniform marginal pricing, the zonal clearing price $i$6 incorporates the marginal unit’s carbon cost component $i$7 when fossil generators are marginal. Consumer expenditure is therefore

$i$8

The settlement modification proposed for Austria and Germany targets the resulting inframarginal rents of non-emitting generators. When $i$9, eligible non-emitting generation is remunerated at $\kappa_i$0 and the collected deduction

$\kappa_i$1

is rebated to consumers, yielding

$\kappa_i$2

Using hourly ENTSO-E data and static accounting for 2025, the estimated average expenditure reduction is about 8.5% in Austria and 4.7% in Germany. Austria’s average price falls from 104.4 to 95.5 EUR/MWh, and Germany’s from 92.8 to 88.4 EUR/MWh (Finster et al., 26 Mar 2026).

At the product level, upstream obligations translate carbon costs into explicit consumer price adders. Under the Carbon Takeback Obligation, the stored fraction follows

$\kappa_i$3

and the compliance cost per tonne of CO2 produced is

$\kappa_i$4

with %%%%84$$C = \sum_t L_t \cdot P_t + C_{\text{demand}} + C_{\text{fixed}}.$$085$$K_{\text{carbon}} = \sum_t L_t \cdot EF_t \cdot \kappa.$$186%%%%C2=\$\kappa_i$8–$\kappa_i$9/tCO2 stored. A product with emissions intensity $c_{D,d}$0 then experiences a marginal price increase

$c_{D,d}$1</p> <p>Under the paper’s low- and high-cost cases, gasoline price uplifts are approximately \$c_{D,d}$20.0302 per liter in 2030, \$c_{D,d}$30.3012 per liter in 2040, and \$c_{D,d}$41.38 per liter in 2050; natural-gas household costs are \$c_{D,d}$50.0696 per therm in 2030 and \$c_{D,d}$63.18 per therm in 2050 (Jenkins et al., 2020).

A different route from climate damages to consumer-facing carbon costs comes from social cost estimates. OPTiMEM maps carbon consumption to greenhouse-gas production, radiative forcing, ocean heat content, and NOAA weather damages, then converts social cost of greenhouse gases into per-unit consumption adders. For a consumption item with emissions intensity $c_{D,d}$7, the adder is

$c_{D,d}$8</p> <p>For 2025 under the Baseline scenario and a 300-year horizon, CO2 social cost ranges from \$c_{D,d}$996.1 per tonne at 0.435$EF_t$$00218 per tonne at 0$$EF_t$$010.066/kWh under the 0$$EF_t$021.47/gal, respectively (Hanley et al., 31 Dec 2025).

Transport applications exhibit yet another form of consumer incidence. In the European airline industry, the per-passenger marginal carbon surcharge is

$EF_t$03

with $EF_t$04 kg CO2 per kg fuel and marginal fuel burn of 2.5 kg per 1,000 km. For a 1,500 km route, this gives approximately \$EF_t$0550/tCO2, \$EF_t$0685/tCO2, \$EF_t$07140/tCO2, and \$EF_t$08205/tCO2. Counterfactual equilibrium simulations show that consumer surplus declines by 15.4% to 25.4% overall, but the burden is highly asymmetric: medium-/long-haul consumer surplus declines by 53.3% to 89.7%, while short-haul markets can exhibit positive net welfare when carbon revenue and the social value of avoided emissions are included (Chen et al., 29 Mar 2026).

6. Governance, controversies, and research frontiers

The main controversies concern attribution integrity, signal choice, and implementability. The most immediate accounting controversy is the concurrent use of location-based and market-based methods for the same emissions-reduction claim. If contracted renewables are counted both by PPA owners and by non-owners relying on undifferentiated location-based signals, green attributes are effectively double counted. This is why the literature recommends evaluating emissions and reductions under the market-based framework for consumers without PPAs, using residual mix data when available, and configuring optimization tools to match the chosen accounting framework (Maji et al., 2024).

A second controversy is methodological rather than normative: whether average or marginal emission factors should govern decisions. The evidence is explicit that AEFs are appropriate for planning with inflexible or pre-existing loads, while MEFs are appropriate for decisions that change load. This implies that retail tariffs benchmarked against AEFs and operational price signals benchmarked against MEFs are answering different questions. A common misconception is that a single “carbon intensity” suffices for all planning, siting, and operational tasks; the evidence instead indicates task-specific signal choice (Chapin et al., 13 Nov 2025).

Market-design proposals introduce their own implementation risks. In consumer-carbon-preference models, consumers could strategically inflate $EF_t$09 or $EF_t$10 to obtain cleaner allocations or favorable prices, so bid caps, verification, or bounds on admissible carbon-cost bids may be required. In the European settlement proposal, a hard threshold $EF_t$11 induces bunching incentives for eligible generators, motivating the linear-ramp variant to remove the payoff discontinuity. The baseline excludes pumped storage because its opportunity-cost-based marginal costs can create unintended distortions (Jiang et al., 1 Aug 2025, Finster et al., 26 Mar 2026).

Data availability remains a structural limitation. Residual mix data is often published with 12–18-month lags; the market-based discrepancy analysis therefore studies PPA penetration ranges and emphasizes the extreme case in which all solar and wind are under PPAs to bound discrepancies. The tariff-emissions study is limited to a single year, with proprietary DAM inputs in some regions and geospatial mapping uncertainty of about 1.85 km. The market-clearing studies rely on DC-OPF and single-period clearing, leaving AC effects, unit commitment, reserves, and contractual constraints for future work. In HPC, multi-tenant GPU attribution and full production integration remain open issues (Maji et al., 2024, Chapin et al., 13 Nov 2025, Jiang et al., 16 Jan 2025, Kamatar et al., 16 Jan 2025).

A broader policy question is whether consumer-based carbon costs should function only as an informational shadow price or as an actual payment obligation. In some models, $EF_t$12 is internalized in welfare maximization but not settled financially; in others, the point is precisely to charge users or consumers in credits, tariffs, or product premia. This suggests that “consumer-based carbon cost” is best understood as a design space spanning accounting, optimization, settlement, and finance rather than a single institutional mechanism.

The financing literature adds a final dimension: consumer contributions can be voluntary as well as mandatory. A consumer-led strategy proposes ethical or low-carbon investment redirection and a voluntary, invested levy for adaptation, anchored by the claim that required low-carbon investment is about 2–3% of world GDP, while OECD household savings exceed 5% of disposable household income. Example contribution channels include 0.5% purchase-price donations and 1–2% price premia on participating products. This does not replace regulation, tariffs, or carbon pricing, but it extends consumer-based carbon costs into the domain of climate finance and adaptation (Webster, 2018).

Taken together, these literatures show that consumer-based carbon costs are not merely a reporting convention. They determine the carbon signal seen by loads, the carbon-adjusted prices paid by consumers and generators, the direction of flexible operations, the pass-through of carbon allowances and upstream obligations into retail expenditure, and the extent to which climate damages are monetized at the point of consumption. The central technical lesson is that attribution framework, optimization signal, and settlement rule must be aligned; otherwise, carbon costs can be understated, overstated, or shifted in ways that obscure the actual emissions consequences of consumption.