AI, Energy & the Missing Data

AI is entering our work, tools, workflows, and routines—day by day. And with it, unsurprisingly, comes a growing environmental footprint and structural opacity. While most discussions focus on per-query stats or global emissions, what stands out today is the persistent difficulty in measuring its real impact.

Small uses, large systems #

In a recent post, Andy Masley argued that “using ChatGPT is not bad for the environment.” He reminds us that 10 queries a day account for less than 0.01 kWh—roughly 11 kg of CO₂ per year—and that newer models consume around 0.3 Wh per query.

That’s not nothing, but it’s not the main issue.

As always, the real stakes are the invisible infrastructures: power grids, cooling systems, model training, machine fabrication. Systemic impacts, rarely visible, and almost never in the spotlight.

Opaque infrastructure #

A May 2025 investigation by MIT Technology Review highlights how little we know about the energy footprint of AI systems. Companies might share aggregate metrics or sustainability goals, but rarely disclose training frequency, energy origin, or water usage for cooling.

Everything we do know points to a system that’s rapidly expanding—and nearly blind. MIT Technology Review, May 2025

This opacity has been confirmed by multiple academic works in recent years. In 2021, Lannelongue et al. proposed Green Algorithms, a framework to estimate emissions based on hardware, runtime, and local carbon intensity. In 2022, Dodge et al. showed how the same workload could produce vastly different emissions depending on time and region. In 2023, Luccioni & Hernandez-Garcia advocated for a centralized tracking standard.

The takeaway is simple: infrastructure matters.

Ecodesign without visibility #

In my work—from Kuroneko to MicroStudio—ecodesign isn’t about chasing efficiency; it’s about fairness, resilience, ecological coherence. We build systems that respect people, scale and resource boundaries.

But that approach requires understanding energy flows, material cost, and supply chains. Without visibility, design becomes a matter of style—not structure.

Friction with the present #

This isn’t new. Every major infrastructure—electricity, trains, telephony—has triggered resistance. Not because it innovated, but because it replaced before it explained.

AI moves even faster than digital tech. It reshapes value, labor, and tempo—while hiding its physical base. Systems too big to feel, too heavy to ignore.

In the end #

Either we love it or not, AI feels already inescapable. It’s becoming a routine part of our tools, platforms, and services.

That makes it even more important to understand what it runs on—not just technically, but materially. The energy, infrastructure, and resources behind these systems shape how sustainable they really are, no matter how seamless the interface appears.

Because in the end, Design isn’t about certainty. It’s about adjusting in context. And there’s a kind of optimism in that—not the blind kind, but the practiced kind. The one you build with others.