The main limiter of quantum computing is not engineering competition, but fundamental physics. Qubits are so sensitive to their environment that the slightest external influence destroys their fragile quantum state. This process, known as decoherence, sets a hard ceiling on the complexity of algorithms we can run today.

The core problem: from physical qubits to logical ones

A quantum computer will not replace your laptop or run familiar programs. It is not just about the current level of development: errors accumulate exponentially as computations grow more complex. Without correction, long algorithms simply lose accuracy. The solution is to distribute quantum information across multiple physical qubits, creating a so-called "logical qubit"—a stable unit protected from errors through quantum error correction (QEC).

A physical qubit is a real system: a superconducting circuit, a trapped ion, or a photon. For superconducting qubits, the error probability per two-qubit operation ranges from 0.1% to 1%. Market leaders have already achieved 99.9% accuracy: Quantinuum demonstrated 99.914% on its H1-1 ion platform, and its flagship 98-qubit Helios reached 99.921%. Among superconductors, IQM recorded 99.91% on a CZ gate, while IBM achieved a similar level on its Egret and Heron lines.

Breakthrough in error correction

The most promising QEC method is the surface code, where physical qubits are arranged in a two-dimensional lattice. Its theoretical stability threshold is about 1% error per operation, which has already been reached. The last two years have seen a qualitative leap:

  • Google Quantum AI's Willow processor demonstrated exponential suppression of logical errors: as the lattice grows, the error rate more than halves at each scaling step.
  • Quantinuum Helios uses 98 physical qubits based on barium-137 ions and, in one mode, showed 48 logical qubits with correction—a ratio of roughly two physical qubits per logical one.
  • Researchers from Harvard, MIT, and QuEra executed circuits with dozens of logical qubits on a 448-neutral-atom processor, reaching up to 96 in certain configurations.

Previously, it was believed that up to 1,000 physical qubits were needed per logical qubit, making commercially significant tasks unattainable (estimates ran into millions of qubits). However, high-speed qLDPC codes could potentially reduce redundancy by an order of magnitude—down to tens of physical qubits per logical one.

What this means in practice

Massive fault-tolerant quantum computers remain only in roadmaps. IBM promises a Starling system with 200 logical qubits by 2029, capable of performing 100 million operations. But even if successful, it will be a highly specialized cloud coprocessor alongside a classical supercomputer, not a replacement for a data center. The number of physical qubits alone is not very informative—what matters is the quality of operations, decoherence speed, and the number of supported logical qubits with low error probability.

Where quantum advantage is real, and where it is a myth

Formally proven: on specialized tasks such as boson sampling and generating random quantum states, as well as in theoretical algorithms (Shor's for factorization).

Expected: in pharmaceuticals (molecular modeling), materials science (new superconductors and batteries), and chemistry (catalysts). Optimization and finance—so far without practical confirmation.

Absent: in everyday tasks—text processing, file storage, gaming. Classical processors remain more efficient and cheaper.

My view: the race for "raw" qubit counts is already giving way to a more mature metric—the quality of logical qubits and the efficiency of error correction. The technology is moving toward practical utility, but years, possibly decades, remain before widespread adoption. Investments in this field are a bet on a long-term breakthrough, not on quick returns.