Erasing one bit of information costs at minimum $kT \ln 2$ joules. Rolf Landauer published this result in 1961. It would not be experimentally confirmed until fifty years later (Bérut et al., 2012), on a colloidal particle trapped in a double potential well. The cost is real. It is measured in heat dissipated into the environment.

The argument is thermodynamic. A bit has two possible states. To erase it is to force it into a single state. The accessible phase space is halved. The entropy of the system decreases by $k \ln 2$. The second law demands that this decrease be compensated by an at least equal increase in the entropy of the environment. The minimum compensation is $kT \ln 2$ of heat.
Bennett (1982) completes the picture. Computation itself can be made reversible, without minimum thermal dissipation. It is erasure that costs. Computing does not necessarily produce heat. Forgetting always does.
Szilard had glimpsed this in 1929. His Maxwell's demon measures the position of a particle and uses that information to extract work from a thermal reservoir. The paradox is resolved when one accounts for the cost of erasing the demon's memory. The cost exactly compensates the work extracted. The second law holds.
What happened between Szilard and Landauer is that information changed status. It went from an abstraction, something an observer knows or does not know, to a physical quantity, something that occupies a phase space, that has an entropy, whose existence and destruction are subject to the same laws as matter and energy.
Shannon (1948) had given information a measure. Landauer gave it a body.
Doctrine
A specification is information encoded in a material support. A file, a blueprint, a machining process sheet. If information is physical, the specification is too. Its erasure has a cost. Its preservation has a value.
What constrains the process passes through an arrangement of matter acting on another arrangement of matter. Whether that arrangement is a force field or a document changes nothing about the thermodynamics.
Vecteur ouvert
If destroying information has a cost, then preserving information has a thermodynamic value. Every document that survives, every specification that is not erased, every archive that persists represents entropy that was not produced.
What distinguishes, thermodynamically, an organization that preserves its specifications from one that destroys them after use?
