For Michael Frank, efficiency has always been a priority. In the 1990s, while studying artificial intelligence (AI), he realized the significant energy consumption of this technology and shifted his focus to the physical limits of computation. He was particularly interested in reversible computing, a concept introduced by IBM physicist Rolf Landauer in the 1960s. Landauer demonstrated that any process involving the deletion of information in a computer inherently produces heat, a form of energy loss dictated by the laws of thermodynamics. For instance, when a computer performs an addition, it loses information about the specific inputs, making the operation irreversible and heat-generating. Bennett, a colleague of Landauer's at IBM, proposed an alternative in 1973 called uncomputation. This method involves running calculations forward and storing the results, then running them backward to uncompute the intermediate steps. This keeps the initial information intact, thus preventing energy loss to heat. However, uncomputation requires twice the processing time, making it impractical for real-world applications. Despite these challenges, researchers continued to refine the concept. In 1989, Bennett showed that uncomputation could be achieved with less time and slightly more memory. This sparked further interest, leading to the development of prototype chips at MIT in the 1990s. Frank joined the MIT team as a doctoral student and became a leading advocate for reversible computing. However, support waned at the turn of the millennium as conventional chips were still improving exponentially, and the physical limits seemed far off. Frank's return to the field in the early 2020s was driven by the realization that traditional computing technologies were approaching their physical limitations. Smaller circuits were generating more heat, hindering further miniaturization and efficiency gains. He formed a research group at Sandia National Labs to address these issues. In 2022, Hannah Earley, a researcher at the University of Cambridge, provided a comprehensive analysis of the energy efficiency of reversible computing. She found that while reversible computers still generate some heat, especially from voltage changes in wires, they can significantly reduce overall energy consumption. By running slower but using more processors in parallel, the energy savings can outweigh the costs. This is particularly promising for AI, where computations are often distributed across multiple processors. Earley's findings have reignited interest in reversible computing. She and Frank co-founded Vaire Computing, aiming to develop a commercial version of reversible chips. The potential benefits include reduced cooling needs, allowing for denser and more efficient chip arrangements, and substantial energy savings, which are crucial for the sustainability and scalability of AI systems. Industry insiders are optimistic about the future of reversible computing. According to Torben Ægidius Mogensen, a researcher at the University of Copenhagen, seeing practical applications of reversible processors would be a significant milestone. He notes that the theoretical foundations have been well-established, and the real test will be in manufacturing and deploying these chips in real-world scenarios. Vaire Computing represents a bold step towards bridging the gap between theoretical research and practical implementation. The company's efforts could lead to a new era of energy-efficient computing, addressing the growing environmental concerns associated with the rapid expansion of AI and other data-intensive technologies. If successful, reversible computing could be a game-changer, offering a sustainable path forward in an industry desperate for innovative solutions to its power consumption challenges.