The qubit count just collapsed

For years the reassuring number was "millions." Breaking RSA or the elliptic-curve crypto that secures HTTPS, code signing, and Bitcoin wallets would take a quantum computer with millions of physical qubits — a machine decades away, if ever. A Caltech team, working with the freshly launched startup Oratomic, just moved the floor.

"No, they don't mean that cryptographically relevant quantum computing is 'imminent.'" — Scott Aaronson

Their estimate: roughly 26,000 qubits could crack the P-256 curve in a few days; about 100,000 could factor RSA-2048 in three months. The smallest useful run might need as few as 10,000. That is one to two orders of magnitude below the old assumption, and it lands in the same neighborhood as hardware that exists — one lab has already trapped over 6,000 atoms, and another has demonstrated past 1,000 working qubits.

The trick is almost mundane. Most error-correction schemes spend about a thousand physical qubits to protect a single logical one. Neutral atoms can be physically picked up and shuttled across the chip to wire up distant qubits on demand. That mobility lets one atom take part in many logical qubits at once, so a new code does the same job at roughly five physical qubits each instead of a thousand. The launch publicity credits an AI for finding the code; the paper itself credits the shuttling, and outside experts call the AI's role real but ordinary.

What this does not mean is that the codebreaker is coming soon. No machine has run this; these are simulated resource estimates, and the gap from 1,000 demonstrated qubits to 26,000 reliable ones is still years of engineering. What changed is the target. The number that security teams and standards bodies have been racing to outrun just shrank — and a paired result from Google cutting its own estimate tenfold the same month says the shrinking is not a fluke.