Why Qubit Count Headlines Are Misleading (Logical vs Physical Qubits)

Every few months a headline announces a new milestone in quantum computing. IBM and Atom Computing have both crossed the thousand-qubit threshold, while Google’s Willow processor takes a different tack with 105 qubits and improved error correction. The implication, especially for crypto readers worried about a quantum break, is that we’re closing in on the day Shor’s algorithm can run at scale.

Those headlines are technically accurate. The qubit count they report is a count of physical qubits, which is a different number from the logical qubits any real cryptographic attack would need.

The gap between the two numbers is currently the most important fact in quantum computing.

What a Physical Qubit Is, and What a Logical Qubit Is

A physical qubit is an actual piece of quantum hardware, a single two-state quantum system the engineers can manipulate and measure. Different companies build them differently. IBM and Google rely on superconducting circuits cooled to near absolute zero. IonQ and Quantinuum trap individual ions in electromagnetic fields. Atom Computing and QuEra hold neutral atoms in optical tweezers, while PsiQuantum and Xanadu push photons through specialized chips.

Whichever physical implementation you use, the qubits are noisy. They lose their state through unwanted interactions with their environment, a process called decoherence. They also make small errors during every gate operation. Current physical qubits across all platforms have error rates somewhere between 0.1% and 1% per gate, which sounds small but compounds disastrously across the millions of operations a working algorithm requires.

A logical qubit is what you build out of many physical qubits to fight back against those errors. The trick is quantum error correction. You spread the quantum state of one logical qubit across many physical qubits arranged in a code. The code lets the system detect and correct errors on individual physical qubits without disturbing the encoded information.

Why the Physical-to-Logical Ratio Is So Brutal

The number of physical qubits needed per logical qubit depends on two things. First, which error-correction code you use. Second, how good your physical qubits are.

The surface code is the option researchers have studied longest. Modern rotated surface codes need about 2d² − 1 physical qubits for one logical qubit at code distance d, where larger distance means stronger error suppression.

Typical noise levels in current hardware push useful computation toward code distances around 15 to 25. That works out to between about 450 and 1,250 physical qubits per logical qubit, just to keep one piece of quantum data alive long enough to be useful. Newer codes like IBM’s bivariate bicycle codes (announced in 2024) cut the overhead, but the ratio still bottoms out at hundreds of physical qubits per logical one.

And that’s the cost of one logical qubit. A useful algorithm needs many of them.

Where Current Hardware Actually Stands

In December 2024, Google’s Willow processor demonstrated a milestone called “below threshold” error correction. They built one logical qubit at code distance 7 across 105 physical qubits, and showed that the logical error rate decreased as they increased code distance. That’s one of the clearest hardware demonstrations to date that quantum error correction scales the way the theory says it should.

A real achievement, on a single logical qubit.

The largest physical qubit counts on actual machines look like this. IBM’s Heron processor runs 156 superconducting qubits with the company’s best gate fidelities to date. The earlier Condor processor hit 1,121 qubits at lower fidelity, which is why IBM has been emphasizing quality over raw count in its 2024 and 2025 roadmaps. Atom Computing crossed 1,180 neutral atoms. Quantinuum has reached 98 qubits with its Helios system while posting some of the lowest error rates in the field.

None of these has ever demonstrated more than a handful of useful logical qubits.

How Much Hardware a Cryptographic Attack Would Need

Craig Gidney and Martin Ekerå published the most-cited estimate in 2019. Their conclusion was that an attack on RSA-2048 using Shor’s algorithm would need about 20 million noisy physical qubits running for 8 hours, with error rates similar to today’s best superconducting machines. But subsequent work has trimmed that number with cleverer algorithm and code choices, and the answer is still in the millions.

20 million is four orders of magnitude above the largest current physical-qubit machines, and that’s just for RSA. ECDSA with 256-bit elliptic curves, the signature scheme Bitcoin and Ethereum use for transactions, is cheaper to attack. Recent estimates put a quantum attack on Bitcoin’s secp256k1 curve below 500,000 physical qubits. That’s still orders of magnitude beyond current hardware. Behind either figure are at least a few thousand logical qubits, none of which has yet been built.

No publicly demonstrated machine has shown anything close to the thousands of high-quality logical qubits a cryptographic attack would need.

How to Read Qubit Count Headlines

A few rules of thumb help when reading announcements about new quantum processors.

A headline that says “X-qubit processor” is reporting physical qubits, not logical. Divide by somewhere between 100 and 1,000 to get a charitable estimate of how many logical qubits the machine can support, and only if the company has the error-correction infrastructure to use them.

When a headline trumpets a “logical qubit demonstration,” check whether the company built one logical qubit or several, and at what code distance. One logical qubit at distance 3 is a science experiment. Several logical qubits sustained over many error-correction cycles starts to look like a useful processor.

Phrases like “below threshold” or “error correction scaling” point to the kind of milestone that genuinely advances the field. They mean the architecture can theoretically scale to fault tolerance. That’s separate from whether it has scaled.

Round-number announcements like “1,000 qubits” or “10,000 qubits” with no logical-qubit context are usually marketing.

The number to watch for cryptography, and therefore for the day Shor’s algorithm becomes a working attack on Bitcoin or any other ECDSA-based system, is logical qubits. The hardware exists at the physical level. The error-corrected qubits the math requires are years to decades behind that hardware.