Abstract

The environmental footprint of digital infrastructure has become a first-order policy concern. The EU now requires sustainability reporting under CSRD, mandates data center energy disclosures under Article 12 of the Energy Efficiency Directive, and imposes transparency obligations on AI systems through the AI Act. Yet beneath these commitments lies an unresolved methodological problem: no existing standard enables the attribution of full environmental impacts across multiple impact categories and the full value chain to the specific applications and AI workloads that consume cloud infrastructure resources.

The Software Carbon Intensity standard (ISO/IEC 21031:2024) provides a useful rate-based indicator for engineering iteration but stops at greenhouse gases and cannot represent waste from overprovisioning. The Blauer Engel für Software (DE-UZ 215) sets a high bar for product properties but explicitly excludes cloud-native software. Data center reporting captures facility-level aggregates but cannot attribute responsibility downward.

NADIKI, developed by IDED, addresses this gap with cumulative, non-revocable environmental impact accounts that traverse the infrastructure stack and distinguish productive from non-productive resource use. This paper argues that NADIKI is a necessary complement — not a replacement — to existing standards.

1. The Problem: Existing Standards Leave Critical Gaps

None of the major existing instruments, individually or collectively, answers the question now most urgently posed by regulators, procurement officers, and sustainability practitioners: what is the full environmental cost of a specific application — including an AI workload — running on shared, cloud-native, multi-tenant infrastructure, and how is that cost attributable to the entity responsible for the workload?

The gap has three dimensions. First, scope of impact categories: SCI accounts only for greenhouse gas emissions; Blauer Engel measures use-phase energy but does not capture embodied emissions, abiotic depletion, water use, or end-of-life impact. Second, structural granularity: data center facility-level aggregates cannot attribute responsibility to individual tenants or AI inference services. Third, scope of software covered: Blauer Engel explicitly excludes software where more than ten percent of energy consumption occurs outside the covered platform — placing the entire universe of cloud-native software outside the scope of the only audited software ecolabel in regulatory use.

2. What Makes AI and Cloud-Native Infrastructure Particularly Challenging

• Multi-tenancy and dynamic allocation. A modern Kubernetes cluster may host hundreds of containerised workloads sharing physical resources through scheduling decisions that change second by second. Attribution requires not just measurement but a defensible allocation methodology.

• The structural role of overprovisioning. Cloud infrastructure is engineered to absorb demand spikes. At any given moment a substantial fraction of allocated capacity is doing no useful work. Any honest accounting must address the environmental cost of idle, reserved, or redundant capacity.

• The disproportionate embodied burden of AI accelerators. The embodied GHG emissions of a high-end GPU or TPU are, per unit of theoretical compute capacity, an order of magnitude higher than those of a conventional CPU. In many configurations, embodied emissions are the larger share of total lifecycle impact for an AI workload.

• Training vs. inference. AI training workloads sustain very high power draw over days or weeks. Inference workloads generate enormous aggregate volume at low marginal compute per request. A single intensity metric obscures rather than reveals their relative contributions.

• Spatial and temporal variability of energy supply. The carbon intensity of grid electricity varies by region by an order of magnitude and within a region across the day by a factor of three or more. Accounting frameworks using annual or facility-average factors wash out these effects and remove the basis for rewarding good workload placement decisions.

• Renewable energy claims. Market-based instruments — RECs, Guarantees of Origin, PPAs — allow a data center to report zero Scope 2 emissions even when the electrons flowing through the facility are generated from fossil fuels. NADIKI's commitment to physical-only renewable accounting is increasingly aligned with regulator and civil-society expectations.

3. NADIKI's Methodological Contributions

• Cumulative, non-revocable accounts. NADIKI does not produce a rate; it produces an account. Every infrastructure entity has an associated cumulative environmental impact account that monotonically increases over the entity's lifetime. Impacts, once attributed, are not reversed. This eliminates the perverse incentives created by rate-based metrics that can be gamed by inflating the denominator.

• Productive versus non-productive impact. NADIKI partitions cumulative impact into two streams: productive (resources actually performing useful work for an identified consumer) and non-productive (idle, reserved, standby, or otherwise unused capacity). This rewards right-sizing and penalises overprovisioning at the level of accounting.

• Multi-criteria impact indicators at infrastructure depth. NADIKI tracks energy consumption, GHG emissions (embodied and operational), abiotic depletion potential (ADP), water use, WEEE generation, acidification, eutrophication, and ozone depletion — computed at the infrastructure layer and propagated downward through the allocation chain.

• Full value-chain attribution. NADIKI's attribution chain begins with the data center building and extends through cooling and power infrastructure, IT racks, servers, digital resources, and finally to the applications and AI workloads. An application's environmental account is traceable, auditable, and decomposable: one can ask how much of the impact originated in the building, how much in the server's embodied emissions, and how much in operational electricity.

• Resource-type granularity and flexible allocation. NADIKI distinguishes CPU, memory, storage, and network resources. It supports bare-metal allocation for exclusive use, usage-based allocation driven by real-time monitoring data, and reservation-based allocation that attributes the cost of reserved-but-unused capacity to the entity that reserved it.

4. How NADIKI Complements — Not Replaces — SCI and Blauer Engel

Relation to SCI. NADIKI accounts can serve as inputs to SCI: the productive cumulative GHG impact of an application, divided by an appropriately defined functional unit, yields an SCI-compatible value with a defensible methodological foundation. NADIKI provides the audit trail that SCI lacks while preserving SCI's usefulness as a per-unit comparison tool for engineering teams.

Relation to Blauer Engel. For cloud-native and server-side software, NADIKI provides the production-environment accounting layer that Blauer Engel's reference-hardware lab testing cannot deliver. A future evolution of Blauer Engel toward cloud-native software would naturally rest on a NADIKI-style attribution foundation, retaining Blauer Engel's audited pass/fail rigor while substituting production-environment cumulative accounts for reference-hardware energy measurement.

Relation to data center reporting. Facility-level figures from EED Article 12 reporting become the top of NADIKI's attribution chain. The two are complementary: facility reporting gives regulators the macro view; NADIKI gives operators, tenants, and procurement officers the granular attribution that micro-accountability requires.

5. Policy Relevance

CSRD and the ESRS E-series. NADIKI's multi-criteria scope maps directly onto ESRS E1 (climate), E2 (pollution), E3 (water), and E5 (resource use and circular economy). For ICT undertakings whose principal environmental footprint resides in cloud infrastructure, NADIKI provides a methodologically defensible basis for ESRS-aligned disclosure at a granularity that aggregate facility reporting cannot achieve.

EU AI Act transparency obligations. The AI Act establishes transparency requirements for AI systems, with energy use emerging as a key compliance question for general-purpose AI models. NADIKI provides the attribution chain by which the operational and embodied environmental cost of a model can be reported with audit-grade traceability.

Public procurement. Germany's reference to Blauer Engel DE-UZ 215 in public procurement leaves cloud-hosted SaaS, AI services, and hosted productivity platforms without an analogous instrument. A NADIKI-based extension to cloud-native software would close this gap. The political demand for such an extension is visible in the German federal procurement community and in adjacent European member states.

6. Open Challenges and Roadmap

• Network infrastructure beyond the data center boundary. The wide-area network is not yet within scope. Methodological work is in progress to extend attribution across network boundaries with appropriate handling of shared transit infrastructure.

• Client devices. End-user devices are deliberately outside NADIKI's current scope. Future integration may proceed via composition with Blauer Engel's client-side methodology rather than re-implementation.

• AI training energy treatment. How to amortise training energy across subsequent inference workloads is methodologically contested. NADIKI provides the accounting machinery; the allocation policy remains a matter for community standardisation, and IDED is engaged in this discussion within relevant standards bodies.

• Verification and audit regime. A production-environment accounting layer at scale requires an audit regime adapted to high-volume, automated, continuously-updated impact accounts. Pilot work with university data center operators is informing how such a regime might be structured.

• Empirical validation of allocation coefficients. Pilot installations at university data centers are generating the dataset required for validation. The methodology will need iterative refinement as empirical evidence accumulates.

7. Conclusion

The most environmentally consequential category of contemporary software — cloud-native, multi-tenant, AI-intensive workloads running across heterogeneous and dynamically allocated infrastructure — falls structurally outside the scope of both SCI and Blauer Engel. Data center reporting captures the facility but cannot attribute downward. The result is a methodological gap precisely where the environmental impact is growing fastest, where regulatory attention is most intense, and where public procurement is most consequential.

NADIKI fills this gap by providing the missing infrastructure-accounting layer: cumulative non-revocable accounts, productive versus non-productive impact, multi-criteria indicators at infrastructure depth, value-chain attribution from building to application, resource-type granularity, and flexible allocation modes. It is designed to be composable with SCI as a source of derived rates, complementary to Blauer Engel as a cloud-native extension, and consumptive of data center reporting as input to its attribution chain.

The choice facing the sustainability community is not whether to adopt new methodology but which methodology to adopt. NADIKI is offered as a serious, evidence-based, and policy-aligned candidate: not a replacement for what exists, but the necessary missing layer that turns existing instruments into a coherent system.