In December 2024, Google announced that its Willow quantum chip completed a computation in minutes that would take the world’s best classical supercomputers longer than the age of the universe to solve. That claim spread everywhere. What most coverage left out: the benchmark was specifically designed to be hard for classical computers and easy for quantum ones. It measured quantum advantage on a synthetic task, not a practical one.
This is the pattern with quantum computing coverage. The milestones are real. The implied conclusions frequently are not.
For developers and technical builders, quantum computing is worth understanding precisely because the standard framings (“exponentially faster,” “will break all encryption,” “will revolutionize AI”) mix genuine insight with significant overstatement. The useful framing is different, and it changes what you should actually pay attention to.
What a qubit is, more precisely
The usual explanation: a classical bit is 0 or 1; a qubit can be both simultaneously. That is technically accurate and practically misleading because it makes quantum computing sound like classical computing that does two things at once.
A more useful framing: a qubit holds a probability distribution over the states 0 and 1. When you measure it, that distribution collapses to a single outcome. The power is not in the measurement. It is in what you can do to the distribution before measuring.
With n qubits, you have a probability distribution over 2^n states. A quantum circuit manipulates this distribution using operations called quantum gates. With 300 qubits in superposition, the number of states that distribution spans exceeds the number of atoms in the observable universe. Classical computers cannot store or manipulate a distribution that large. Quantum computers can, because they do not store it explicitly. They evolve it.
That distinction matters. Quantum computing is not parallel processing where you try all answers and read the right one. It is about mathematically shaping a probability distribution so that when you finally measure, you are overwhelmingly likely to observe the right answer.
The three properties that make it work
Superposition gives you a distribution over exponentially many states. On its own, that is not useful. A random measurement of all states simultaneously gives you nothing actionable.
Entanglement is what creates structure within that distribution. When qubits are entangled, their states become correlated in ways that cannot be decomposed into independent components. This is what lets a quantum computer work coherently across an exponentially large state space rather than just sampling noise from it. Einstein called entanglement “spooky action at a distance.” For computation, the useful property is simply that it links qubits into a unified system rather than independent bits.
Interference is the mechanism that makes quantum algorithms actually work. Quantum gates can be designed so that probability amplitude toward wrong answers cancels out (destructive interference) while amplitude toward correct answers reinforces (constructive interference). This is what steers the computation. The hard part of designing quantum algorithms is finding circuits where this steering works for a specific problem.
These three properties together explain why quantum computers are fast at certain problems and provide no advantage on others. The speedup is not general. It is structural, and it only exists for problems with the right mathematical properties.
n
What quantum computers are actually fast at
Most articles end the quantum computing explanation before this point.
Quantum computers offer genuine, well-established speedup on a specific and limited set of problem types. Factoring large integers (Shor’s algorithm) achieves exponential speedup over the best known classical algorithms. This is why the cryptography threat is real. Searching unsorted databases (Grover’s algorithm) achieves quadratic speedup. Simulating quantum systems (molecular interactions, material properties) is the most natural use case, where the quantum hardware models quantum physics directly.
Outside these categories, the picture is much less clear. Quantum machine learning remains largely unproven; most proposed algorithms have not demonstrated convincing speedup over classical methods on practical problems. Quantum optimization shows promise in narrow cases but no general advantage. The broad claim that quantum computers will accelerate AI training or financial modeling is not supported by the current state of quantum algorithms.
For anyone building software today, the honest answer is: almost nothing in your current stack is in the category of problems quantum computers are fast at. The relevant question is whether any of the specific problem classes that quantum algorithms do accelerate appear in your domain.
For most domains, they do not. For cryptography, they do.
The actual state of the hardware
The gap between what exists and what the dangerous applications require is still substantial.
Qubits are extraordinarily sensitive to environmental noise. Temperature fluctuations, electromagnetic interference, vibration: any of these collapse the quantum state. Most quantum computers operate near absolute zero, colder than the vacuum of space, because thermal noise at room temperature makes coherent computation impossible. The time a qubit can hold its quantum state before decoherence destroys it is measured in microseconds to milliseconds for most current hardware.
Long computations require error correction. Error correction requires many physical qubits to represent each reliable logical qubit, with estimates ranging from hundreds to thousands depending on the code and target error rate. Today’s largest systems have a few thousand physical qubits. Practical attacks on RSA-2048 require millions of stable, error-corrected logical qubits.
That gap is large. But the trajectory matters. In 2019, breaking RSA-2048 was estimated to require 20 million physical qubits. By 2025, revised estimates using improved algorithms brought that to under one million. Early 2026 work using new error correction architectures suggests potentially under 100,000, though those projections remain unvalidated at scale. The reduction has come from algorithmic efficiency gains, not hardware improvements, and that kind of progress is harder to predict or assume will plateau.
n
What builders should actually track
For security and infrastructure decisions, the cryptography timeline is the only quantum computing development that is immediately actionable. NIST finalized post-quantum standards in August 2024 (ML-KEM, ML-DSA, and SLH-DSA) and published a migration roadmap with RSA and ECC deprecated by 2030. Organizations managing data with long confidentiality requirements are already in the relevant window, because harvest-now-decrypt-later attacks work on data collected today.
For everything else (AI, optimization, general software) the honest tracking strategy is to watch the class of problems quantum algorithms demonstrably solve and ask whether your specific domain has problems in that class. Not “quantum will change X industry” but “does my specific bottleneck have the mathematical structure that quantum interference can exploit?”
Most of the time, the answer is no. When the answer is yes, you will know, because the algorithm will exist and the researchers will have proven the advantage.
The Google Willow demonstration was a genuine engineering milestone. It proved that their error correction approach scales in the right direction. It did not prove that general-purpose quantum advantage is imminent, and it did not change the timeline for practical attacks on cryptography in either direction.
Quantum computing is worth following closely. The framing that makes it most useful to follow is also the most boring one: track the specific problem classes, watch the qubit estimates, and treat the broad claims with proportional skepticism.
n