The Aestivation Hypothesis

The frame. Some civilisations might sleep, waiting for a colder universe where computation is cheaper. Heat is waste.

The thesis. Heat is waste, and waste is a design decision. The energy cost of computing is not an externality to be offset later. It is a choice made now, in architecture, in code, and in what we decide is worth running at all.

Landauer’s principle gives the thermodynamic floor: erasing a bit has a minimum energy cost. Real systems sit orders of magnitude above that floor, which means most of the cost is engineering, not physics. That gap is the interesting part. It is also the part that a business can act on.

The argument

• Efficiency is a survival metric. A system that spends more energy than it needs is more fragile than one that does not, whether the constraint is a planetary boundary or a cloud bill.
• Big O notation is an energy metric in disguise. An algorithm that grows quadratically does not just get slower. It consumes more, and the difference compounds with scale.
• The AI boom has made this concrete. Data centres are a visible and growing share of electricity demand, and the industry now has to answer questions about siting, water and grid capacity that it previously left to others.

A real case

The International Energy Agency estimated in 2024 that data centres used around 1.5% of global electricity, with demand projected to rise sharply as AI workloads grow. The exact number matters less than the structure: a general-purpose technology became an energy policy problem in about a decade, and most of the people building it were not trained to think about that.

Exercise

Take a script or notebook you already run. Profile it for CPU time and memory. Find the single biggest inefficiency, fix it, and measure the difference. Report the saving as a number, not a feeling.

Next

04. The Zoo Hypothesis looks at what happens when the observer is the one with power.