The energy cost of an AI query should be as visible as the price of a cloud instance. Right now, it is not, and that is a failure of design, not inevitability. Ludi Akue's recent talk on green IT makes the case that sustainability must be a core constraint in AI development, not a patch applied after deployment. We agree, and we think the practical implications for anyone building or buying AI tools are immediate.
Consider the hardware. GPU chips, the workhorses of modern AI, last only two to three years before they are swapped out. Each query consumes energy at a scale that remains hidden from the end user. Regulatory frameworks like the EU AI Act attempt to address this, but as Akue noted, enforcement mechanisms are weak. The result is a system where the environmental cost is abstracted away, and users have no reason to demand efficiency. That is where model compression and quantization enter the picture. These are not abstract research problems. They are techniques that shrink the size of a model, reduce the number of calculations per query, and extend the usable life of the hardware. For a team choosing between a bloated model and a compressed one, the decision is not just about performance. It is about whether the tool you adopt today will still be responsible to run three years from now.
This matters because the default in AI development has been to optimize for accuracy at any cost. The result is models that are enormous, expensive, and short-lived. Akue's argument flips that priority. She treats sustainability as a design constraint on par with latency or accuracy. That shift has real consequences for your workflow. If you are evaluating an AI-powered spreadsheet tool, ask whether the vendor publishes energy consumption per query. Ask whether they use quantization or novel architectures to reduce load. If they cannot answer, the cost is likely being passed to you, and to the environment, without your consent. The EU AI Act may not enforce this yet, but your procurement decisions can.
The point is not that AI must stop growing. It is that growth without constraint is waste. Akue's talk gives us a concrete path: compression, quantization, and architectural innovation. These are not future promises. They are available now. The question is whether the industry will adopt them as standard practice or wait for regulation to force the issue. For our readers, the practical next step is simple: when you evaluate your next AI tool, treat energy efficiency as a feature, not a footnote. That is how you design for longevity instead of disposal.