Reversible computing: the chip that recycles its own heat
Vaire Computing says its chip recovered more energy than it lost. What reversible computing actually is, why it is decades old, and what it means for anyone in Sri Lanka paying for compute.
Reversible computing has been a textbook curiosity for half a century, and this week it got a face. Hannah Earley, 31, cofounder and CTO of Vaire Computing, is on MIT Technology Review's 2026 Innovators Under 35 list for building chips that recover energy conventional processors throw away as heat.
Eshan Raul's profile, This founder is teaching chips how to recycle (their energy), covers her path from a nine-year-old programmer to a Cambridge PhD to a startup that has raised more than $12 million. I want to do something different: explain what the chip actually does, be honest about how early it is, and work out why a builder in Sri Lanka should care about a lab result in Iowa.
🔍 Why every chip you own is a heater
Here is the physics in short. Every time a processor erases a bit, it pushes that bit's state out into the environment as heat. Textbook thermodynamics, the Landauer limit, puts a floor on what erasure costs. Real chips sit far above that floor, but the direction is the same: computation ends in heat.
Earley's framing, as reported in the article, is that this waste is a design choice, not a law. Her analogy is a car racing through a city and braking hard at every intersection. Conventional logic throws away its momentum at every gate. Reversible logic keeps it.
| Approach | What happens to intermediate results | Where the energy goes |
|---|---|---|
| Conventional CMOS | Erased at every step | Dissipated as heat |
| Reversible computing | Retained so the computation can run backwards | Recovered and reused |
Bottom line: the goal is not a faster chip. It is a chip that finishes the same calculation and hands most of the energy back.
⚡ The "glorified pendulum" that makes it work
The idea of reversible computing was proposed more than 50 years ago. The hard part was never the theory. It was building a physical component that could catch recovered energy and store it for the next cycle without burning more than it saved.
Vaire's answer is a patent-pending resonator. Earley calls it "a glorified pendulum": a microscopic oscillator that stores energy on one swing and gives it back on the next. According to the article, the company announced last year that a chip carrying this resonator recovered more energy than it lost, even after accounting for the power needed to run the resonator itself.
A few things I want to be clear about, because the article is careful and I should be too:
- No efficiency percentage is published. "More than it lost" is a net-positive result, not a headline number.
- No product, part number, or ship date is given.
- One chip is described. That is a proof of concept, not a product line.
The team is not guessing. Michael Frank, whose 1999 thesis is a foundational reversible-computing text, joined Vaire as senior scientist. Cofounder Rodolfo Rosini is a technology entrepreneur and investor. The company was founded in 2021, the year Earley finished her PhD.
📊 How far from your laptop this really is
The article quotes Igor Markov, a University of Michigan researcher in electronic design automation. His view is that Vaire clearly has "something interesting," and that it will need a series of increasingly realistic demonstrations before the industry backs it. I think that is exactly right, and here is what those demonstrations would have to clear.
| Hurdle | Why it is hard | Status per the article |
|---|---|---|
| Net energy recovery in one component | Resonator losses must be lower than savings | Achieved (announced 2025) |
| Scaling to a full logic block | Every gate must be reversible, not just one | Not described |
| Fitting into existing manufacturing | Fabs and toolchains assume conventional CMOS | Named as a major obstacle |
| Software that can exploit it | Compilers must keep intermediate state around | Not described |
Reversible computing means rebuilding the computer from the ground up, which is why it stayed in papers for five decades. What changed is that someone now has silicon, funding and a pioneer of the field on payroll.
Reality check: if you are planning a hardware purchase for 2027, nothing in this story changes it. If you are choosing what to study, it might.
💰 Why a Sri Lankan builder should care anyway
Electricity is the quiet tax on everything we run. Anyone who has watched a CEB bill after a month of overnight GPU training knows compute cost in Sri Lanka is as much a power problem as a hardware problem. The same holds at every scale above us: new AI data centres are limited by megawatts, not chip supply.
That is why a chip that recycles energy matters even to someone who will never buy one:
- Energy per operation is becoming the metric that decides where AI runs. A country with expensive, sometimes unreliable power is on the wrong end of that curve today. Any technology that cuts joules per calculation moves the line in our favour.
- Edge devices win first. The article names laptops and phones alongside data centres. A device that runs cooler runs longer on the same battery, which is a bigger deal in Anuradhapura than in Amsterdam.
- The skill gap is smaller than you think. Earley did a computational biology PhD, not chip design. Her own line in the piece, after her first design breakthrough, was "I'm not completely out of my depth." The barrier to a hard problem is often confidence, not credentials.
If you want to put a number on your own compute footprint before any of this arrives, our AI Energy & Carbon Footprint Calculator estimates the electricity behind a given volume of model calls, and the Appliance Electricity Cost Calculator turns a GPU's wattage into a monthly rupee figure on current CEB tariffs.
🛠️ What you can actually do with this idea today
You cannot buy a reversible chip. You can borrow the mindset. The core insight is that throwing away state costs energy, and that applies well above the transistor level.
- Cache aggressively. Recomputing a result you already had is the software version of braking at every intersection. Prompt caching, memoised builds and materialised views are all "reversible" in spirit.
- Measure before optimising. Earley's team got attention because they measured net recovery, not because they claimed it.
- Read the old papers. Frank's 1999 thesis is a free starting point, and it needs undergraduate physics, not a fab.
💡 What this means for you
Key takeaway: Vaire Computing has shown, in one component, that a chip can recover more energy than it spends. That is a real result and a very early one. The direction of travel matters more than the timeline: the industry is starting to treat energy per operation as the number to beat, and that trend favours anyone building for expensive-power markets like ours.
For students, this is a signal that hardware and physics are back in demand, in a field small enough for a newcomer to matter. For working engineers, efficiency is measurable now, with tools you already have. For anyone paying a power bill to run compute in Sri Lanka, it is a reason to keep watching, and no reason to wait.
Original source
This founder is teaching chips how to recycle (their energy)AI-assisted draft, reviewed and approved by Induwara Ashinsana before publishing. Sources are linked inline; if something here is wrong, tell me and it gets corrected.