How AI researchers could save energy by computing backwards
Writer Matt von Hippel traces reversible computing from Rolf Landauer's 1961 result that erasing a bit must dissipate heat, through Charles Bennett's proof that computation can in principle avoid that erasure, to today's renewed interest driven by AI's soaring power demand.
The piece centres on Michael Frank and Hannah Earley, whose analysis shows reversible chips running many operations in parallel could recover most of the energy conventional logic throws away as heat. It frames their startup, Vaire Computing, as the leading attempt to commercialise the idea in silicon.
The article matters because it brings a rigorous, physics-grounded explanation of reversibility to a wide technical audience precisely as data-centre energy use becomes a mainstream concern.