For decades, chip designers have largely accepted waste heat as the unavoidable byproduct of computation. Vaire Computing is pursuing a different premise: some of that energy may be recoverable rather than lost.
The startup, co-founded in 2021 by CTO Hannah Earley and entrepreneur and investor Rodolfo Rosini, is building chips based on reversible computing. The company’s approach retains information that conventional circuits discard during intermediate steps of a calculation. In principle, that allows a circuit to reverse portions of the process and reclaim energy that would otherwise dissipate as heat.
That is a potentially consequential proposition as operators contend with the power demands of AI training, inference and increasingly dense data-center deployments. But it remains a hardware research and commercialization challenge—not a near-term substitute for today’s processors.
What Vaire says it has demonstrated
Vaire’s core component is a patent-pending microscopic resonator, which Earley describes as a “glorified pendulum.” The resonator is designed to store recovered energy for subsequent reuse.
Last year, the company announced a chip result in which its resonator recovered more energy than it lost, including the energy required to operate the component. That matters because reversible computing has long been theoretically compelling but difficult to realize with practical transistors and circuits. A net-positive energy-recovery result is an important proof point for the underlying device concept.
Vaire has raised more than $12 million and hired reversible-computing pioneer Michael Frank as a senior scientist. Earley’s work began in her Cambridge PhD program, where she studied the physical limits of computation and developed software intended to transform conventional programs into reversible ones.
Why conventional chips lose energy
In standard computing systems, calculations often erase information that is no longer needed. That information loss produces heat. Reversible logic instead seeks to preserve intermediate state, avoiding or reducing those irreversible operations and enabling energy recovery.
The practical implication is not simply a more efficient version of an existing CPU or accelerator. It is a different design model spanning logic, circuit components and potentially software tooling. Earley’s stated ambition is to rethink “every part of how computers are built” around reversibility.
For technology leaders, that distinction is central. An energy-recovery component can be scientifically promising while still being far from a manufacturable, programmable and economical computing platform.
The commercialization test is integration
The next hurdle is fitting this radically different architecture into familiar devices and manufacturing systems. That includes demonstrating performance and reliability under realistic workloads, compatibility with chip-design flows, manufacturability, and a compelling total cost of ownership against relentlessly improving conventional silicon.
Igor Markov, an electronic design automation researcher and former University of Michigan professor, told MIT Technology Review that Vaire has “something interesting,” while emphasizing that the company will need increasingly realistic and convincing demonstrations to win industry support.
That is the appropriate lens for customers and investors. The initial result validates a research direction; it does not yet establish a deployable processor or data-center product.
What to watch next
Vaire’s most meaningful milestones will be less about the theoretical appeal of reversible computing and more about system-level evidence:
- **Scaled prototypes** that show net energy recovery beyond a single component.
- **Useful workloads** demonstrating that reversibility does not impose unacceptable speed, area or software costs.
- **Process and packaging compatibility** with established semiconductor manufacturing.
- **A product wedge**, such as a specialized low-power application, where its architecture has a measurable advantage before attempting broad-purpose computing.
AI’s electricity footprint makes unconventional compute architectures increasingly relevant. Vaire’s work offers a reminder that efficiency gains may come not only from better models and denser chips, but from revisiting a basic assumption of digital computation: that energy spent on information erasure must be thrown away as heat.
