The AI industry promises to measure everything — yet has no standard way to account for its own environmental impact. This pitch from the IPAI KI Festival 2026 argues that the absence of environmental measurement is not a technical problem, but a market failure, and introduces the NADIKI approach: real-time measurement infrastructure and a verifiable label.
The AI industry promises to measure everything — yet has no standard way to account for its own environmental impact. This pitch presentation from the IPAI KI Festival 2026 in Heilbronn argues that the absence of environmental measurement is not a technical problem, but a market failure. It introduces the NADIKI project’s approach: building the infrastructure for real-time measurement and a verifiable label.
The Paradox of AI Sustainability
AI deployment is accelerating, driven by the promise of rapid value creation. The conventional arguments for measuring environmental impact all fall short under market conditions. Moral incentives don’t drive economic behavior — that is what regulation is for. Sustainability differentiation hasn’t been shown to work for digital services, where the signaling effect is weak and the perceived impact low. And reducing environmental impact only lowers costs when there is a price on that impact — which requires carbon pricing that AI markets currently lack. Without structural incentives, measurement doesn’t happen.
From Estimation to Trustworthy Measurement
What would need to change? Three shifts are necessary: customers — including procurement in companies and public institutions — demanding an environmental impact report when buying AI products; legislation that forces the display of environmental impact or true costs per prompt; and a public reporting obligation on environmental impact by AI product companies to create accountability and enable innovation in resource efficiency.
Today’s measurement tools all share the same fundamental flaw. AI Energy Score still guesses based on model parameters. The ML CO2 Impact Calculator is built on assumptions about hardware efficiency. CodeCarbon tracks the local machine, blind to the infrastructure behind the API. Google’s widely-cited 0.24 Wh per prompt figure comes from controlled lab conditions, not billions of real production queries. The ongoing MIT debate over whether ChatGPT uses 0.3 Wh or 3 Wh per prompt illustrates that we don’t even know the order of magnitude.
NADIKI’s goal: move from estimation to reliable measurements and a trustworthy label.
The NADIKI Registrar Architecture
Behind each AI chat interface, the NADIKI Observer collects, verifies, converts, and enhances data from two layers of infrastructure. At the server level: power consumption (active and idling), CPU and GPU consumption, and manufacturing impact. At the data center level: grid CO₂ emission factor, on-site renewable production, diesel generator use, power overhead from cooling, embedded building impact, and recovered heat.
A new attribution method — cumulative impact accounts — calculates the share of environmental impact attributable to each individual chat session. As of the presentation, 4 facilities and 23 servers are connected to the system.
A Label and Certification Model
Genuine sustainability labeling requires benchmarks and comparability — infrastructure that doesn’t yet exist at scale. Rather than waiting, NADIKI proposes a label and certification model that rewards transparency and verified measurement. The goal is to create incentives for AI operators to measure, report, and be transparent, and to let them differentiate on actual measurements rather than generic green or sustainability claims. The NADIKI Verified AI Measurement seal marks providers who participate in real measurement.