Here’s something nobody in the quantum computing hype machine wants to talk about: today’s quantum computers are forgetting what they’re doing. Not slowly, either.
A new theoretical study published in Nature Physics just quantified exactly how badly noise erases quantum circuits’ memory — and the answer is brutal.
Researchers from EPFL, the Free University of Berlin, and the University of Copenhagen found that as noise accumulates through a quantum circuit, earlier operations gradually lose their impact. The system doesn’t just degrade — it actively forgets. Only the final layers of computation actually matter. Everything before that? Gone.
Think of it like setting up an elaborate domino chain only to realize that only the last ten pieces actually fall. The rest is just theater. Deep quantum circuits, the kind companies like IBM and Google have been bragging about, are effectively behaving like much shallower ones in practice.
The researchers call it a “memory fade” effect. Noise places a strict, hard limit on how deep any useful quantum circuit can be before it collapses into gibberish. And here’s the uncomfortable implication for the industry: noisy circuits may appear trainable partly because the noise has already burned away most of their complexity. The machine looks like it’s learning. Really, it’s just surviving.
IBM, Google, and a handful of well-funded startups are currently in a race they call the push toward “quantum advantage.” This paper suggests that race has a ceiling — and we’re getting close to it. Simply stacking more layers onto a quantum circuit is not going to fix this.
Progress now depends on either dramatically reducing noise at the hardware level, or finding entirely new circuit designs that can tolerate it. Neither is easy. Both are necessary.
The paper appeared in Nature Physics on April 2, 2026 (DOI: 10.1038/s41567-026-03245-z), authored by Armando Angrisani, Yihui Quek, Antonio Anna Mele, and Daniel Stilck Franša.