The main constraint on quantum computing is not competition with classical chips, but fundamental laws of physics. Qubits are so sensitive to any external influence that the slightest noise can destroy the fragile quantum state required for calculations. This is the key problem the industry has been trying to solve for decades.

The short answer

Quantum computers will not replace conventional PCs, will not appear in smartphones, and will not run standard applications. The reason lies not only in the current level of technology, but also in the very nature of qubits: they lose coherence due to thermal noise, vibrations, and electromagnetic interference. Errors accumulate exponentially as algorithm complexity grows, making long calculations meaningless without correction systems. To combat this, quantum information is distributed across multiple physical qubits, creating a stable logical qubit.

Why this happens

Imagine a coin spinning in the air: its state is undefined until it lands. A qubit works similarly, existing in a superposition of two states. Any external influence—thermal fluctuations, imperfections in materials or control signals—"peeks" at the system and destroys the superposition. This is why superconducting processors from market leaders are cooled to temperatures of 10–20 millikelvins, significantly lower than the temperature of cosmic microwave background radiation (~2.72 K). This reduces noise but requires extremely complex cryogenic infrastructure.

A physical qubit is a real system: a superconducting circuit, an ion in a trap, a photon, or an electron spin. The error rate of two-qubit operations in superconductors is typically 0.1–1%. However, progress is evident: the Quantinuum H1-1 platform achieved 99.914% accuracy on all qubit pairs, and its 98-qubit Helios processor reached 99.921%. Among superconducting systems, IQM demonstrated 99.91% accuracy for the CZ gate, while IBM, in its Egret and Heron lines, reached around 99.9%.

A logical qubit is a unit of information protected from errors by distributing the state across many physical qubits through quantum error correction (QEC). One of the most promising approaches is the surface code, which stands out for its locality of operations and a theoretical error tolerance threshold of about 1%.

Recent experiments are impressive:

  • Google Quantum AI's Willow processor demonstrated operation below the surface code threshold: as the physical qubit lattice grew, the logical error rate more than halved at each scaling step.
  • Quantinuum Helios, using 98 barium-137 ions, showed 48 error-corrected logical qubits with a ratio of approximately two physical qubits per logical qubit.
  • Researchers from Harvard, MIT, and QuEra, using a 448 neutral-atom processor, executed circuits with dozens of logical qubits—up to 96 in certain configurations.

Previously, it was believed that up to ~1000 physical qubits were needed per logical qubit, leading to estimates of a million qubits for commercially significant tasks. However, high-rate qLDPC codes could potentially reduce these costs by an order of magnitude—down to dozens of physical qubits per logical qubit.

What this means in practice

Massive fault-tolerant quantum systems remain only in roadmaps. IBM plans to build the Starling system with 200 logical qubits by 2029, capable of performing 100 million operations. But even if successfully implemented, this will be a highly specialized cloud coprocessor alongside a classical supercomputer, not a replacement for a data center. The number of physical qubits alone says little: far more important are the quality of operations, the rate of error occurrence, and the number of logical qubits the system can sustain with a low probability of failure.

Where quantum machines are truly strong

Formally proven advantages exist in specialized tasks such as boson sampling and generating random quantum states. Theoretically, speedups are justified for Shor's algorithm in factoring and discrete logarithms. A breakthrough is expected in simulating nature: pharmaceuticals (precise molecular calculations), materials science (searching for superconductors), and chemistry (creating catalysts). In optimization and finance, the advantage is not yet proven, and in everyday tasks—text processing, file storage, gaming—classical processors remain unrivaled.

What's next

If the technology is so finicky and costly, why are corporations and governments investing billions in it? In the next issue, I will break down the industry's investment logic and try to determine where the line lies between a forward-looking bet and a bubble.

My view: We stand on the threshold of transitioning from "quantum supremacy" to "quantum utility," but this path will take at least another decade. Investors and developers should focus not on the race for qubit counts, but on practical algorithms and hybrid solutions where a quantum coprocessor complements classical infrastructure rather than replacing it.