For two decades, quantum computing has been stuck in a frustrating loop: build more qubits, watch errors multiply, try to fix the errors, repeat. The fundamental problem has been that quantum bits are extraordinarily fragile. They lose their quantum properties—a process called decoherence—within microseconds. Every operation introduces noise. Scale up the system and the noise scales faster.

That dynamic appears to be shifting. In late 2025 and early 2026, three separate research groups demonstrated quantum error correction below the threshold where adding more qubits actually improves reliability rather than degrading it. This is not a marketing milestone. It’s a physics milestone, and it changes the trajectory of practical quantum computing.

What the Threshold Means

Quantum error correction works by encoding a single “logical qubit” across many physical qubits. If some physical qubits accumulate errors, the redundancy allows the system to detect and correct those errors—similar in principle to error-correcting codes in classical computing, but vastly more complex.

The critical threshold is the point where the error rate per physical qubit is low enough that adding more physical qubits to the error correction code actually reduces the logical error rate. Below this threshold, bigger codes mean better performance. Above it, bigger codes just mean more errors.

Google’s Willow chip demonstrated this threshold crossing with their surface code implementation in late 2024, showing that going from a distance-3 to distance-5 to distance-7 code improved logical qubit performance each time. In early 2026, both IBM and a collaboration between Harvard and QuEra pushed further, demonstrating below-threshold operation in different qubit architectures—superconducting circuits and neutral atoms respectively.

The fact that three different hardware platforms have crossed this threshold suggests it’s a genuine capability rather than a one-off result.

The Gap Between Threshold and Useful Computation

Crossing the error correction threshold is necessary but not sufficient for practical quantum computing. The logical error rates achieved so far—roughly one error per thousand to ten thousand operations—are still orders of magnitude too high for the algorithms that would provide quantum advantage.

Shor’s algorithm for factoring large numbers, the famous example that threatens current encryption, requires logical error rates below one in a billion. Running it on a 2048-bit RSA key would need roughly 20 million physical qubits with current error rates. Today’s largest quantum processors have around 1,000 qubits.

More near-term applications like quantum chemistry simulations for drug discovery or materials science need logical error rates around one in a million—still a thousand-fold improvement from where we are.

But here’s the important part: the improvement trajectory is now clear. Below the threshold, every incremental improvement in physical qubit quality translates to exponential improvement in logical qubit performance. The path from where we are to where we need to be is steep but visible.

What This Means for Different Industries

Cryptography and security: The timeline for quantum computers breaking current encryption has moved forward, but it’s still measured in years, not months. NIST’s post-quantum cryptography standards, finalised in 2024, need to be adopted broadly. If your organisation hasn’t started evaluating post-quantum migration, the clock is ticking faster than it was a year ago.

Pharmaceuticals: Quantum simulation of molecular interactions is likely the first commercially valuable application. Companies like Roche and Merck have active quantum computing programs exploring protein folding and drug interaction modelling. Useful results from these programs are plausible within 5-8 years given current trajectories.

Finance: Portfolio optimisation and risk modelling under quantum speedup remain theoretical advantages. The error correction milestones move the timeline closer but don’t change it dramatically—expect 7-10 years before quantum computers meaningfully outperform classical approaches for financial modelling.

Machine learning: Quantum machine learning remains the most uncertain application area. Theoretical speedups exist for certain problems, but practical demonstrations haven’t yet shown clear advantage over classical approaches running on modern GPUs.

The Hardware Competition

Three qubit technologies are now credible paths to fault-tolerant quantum computing:

Superconducting qubits (Google, IBM) have the most mature ecosystem but face fundamental scaling challenges with wiring and cooling at extreme scale.

Trapped ions (IonQ, Quantinuum) offer higher individual qubit quality but slower operation speeds. Quantinuum’s recent results with logical qubit circuits are particularly notable.

Neutral atoms (QuEra, Pasqal, Atom Computing) are the dark horse. They can scale to thousands of qubits relatively easily and showed strong error correction results in 2025-2026 demonstrations. Their main limitation has been operation speed, which is improving.

The Honest Assessment

We are closer to useful quantum computing than at any previous point. The error correction threshold crossings are real, meaningful milestones. The physics now works in our favour as we scale up rather than against us.

But “closer” doesn’t mean “close.” Practical quantum advantage for real-world problems is still 5-10 years out for the most promising applications, and longer for others. Companies should be tracking developments, starting pilot programs with quantum computing providers, and preparing for post-quantum cryptography migration.

The right response is not hype. It’s not dismissal either. It’s informed preparation for a technology transition that just became significantly more certain to arrive.