Hannah Earley, co-founder and CTO of Vaire Computing, has built her startup on a single, quietly radical idea: the heat that pours out of a computer chip is not inevitable. It is a choice. Specifically, it is the consequence of a design philosophy baked into silicon since the 1960s, one that treats energy as disposable once a calculation is done. Earley's company is building chips that recover that energy and put it back to work, using a technique called reversible computing.
The scale of the problem this addresses is not trivial. According to the International Energy Agency, data centres consumed roughly 415 terawatt-hours of electricity globally in 2024, and AI workloads are the fastest-growing slice of that demand. The Lawrence Berkeley National Laboratory has modelled AI infrastructure energy use doubling or tripling inside this decade if chip architecture does not fundamentally change. Right now, the majority of that energy does not produce useful computation. It produces heat. Conventional chips are, in thermodynamic terms, extraordinarily wasteful machines.
Reversible computing is not a new theory, it traces back to physicist Rolf Landauer's work in the 1960s and Charles Bennett's refinements in the 1970s, but turning it into manufacturable silicon has defeated generations of engineers. Vaire Computing is attempting what MIT Technology Review calls a genuine commercial first: chips designed from the ground up to recover and recycle the energy they consume during computation, rather than dumping it as heat into a cooling system. Earley's background spans physics and electrical engineering, and her co-founding team has published peer-reviewed work on the underlying mechanisms. This is not vaporware with a sustainability badge on it.
For Scotland, the implications land in two places simultaneously. The Loop has tracked the case for AI server farms here, cold climate, renewable surplus, available land, district heating networks that could capture waste heat from data centres and pipe it to homes and businesses. That case is real and still compelling. But Vaire's approach points to a different part of the same problem: if the chips themselves waste less energy, the entire input side of the equation shrinks. Smaller cooling requirements, lower operating costs, a reduced renewable generation footprint per unit of AI compute. Scottish Enterprise and Highlands and Islands Enterprise have both signalled interest in attracting AI infrastructure investment to Scotland; the technology underpinning that infrastructure is now in active commercial development. The timing matters.
For Edinburgh SME owners and solopreneurs, the immediate relevance is less about chip architecture and more about what it signals for the cost trajectory of AI tools. The reason GPT-4, Claude, and Gemini-class models cost what they cost to run is, in large part, power and cooling. If reversible computing chips reach commercial scale, Vaire has not published a product timeline, but the MIT Technology Review profile indicates working prototypes, the marginal cost of an AI query drops. That means more capable AI tools, running cheaper, available to smaller businesses. The great equaliser gets a little more equal. Watch this space, and watch it from Scotland.