The race is no longer just about qubit counts
The latest quantum-computing headlines are easy to misread as a duel between a laboratory gate-speed record and a cryptography warning. They are better understood as two sides of the same race. Forschungszentrum Jülich, with partners at RWTH Aachen University and the University of Cologne, has reported 6.8-nanosecond single-qubit gates on a superconducting transmon processor with fidelities above 99.98% . Alice & Bob, meanwhile, has published a new resource estimate in which RSA-2048 could be factored with 22,593 physical cat qubits running for 3.14 days under specified fault-tolerant assumptions .
Neither result means a general-purpose, cryptographically relevant quantum computer has suddenly arrived. The Jülich work is a single-qubit hardware demonstration, and the Alice & Bob result is an end-to-end compilation and resource benchmark, not an actual break of RSA-2048. But together they sharpen the same question: how fast can the field convert fragile physical devices into machines that run long, useful algorithms before errors win?
Jülich’s speed record attacks leakage
The Jülich result focuses on one of the core bottlenecks in superconducting quantum processors: driving a qubit quickly without pushing it outside the two computational states that encode information. According to Quantum Computing Report, the team achieved 6.8-nanosecond operations for both X and X/2 gates, with measured fidelities of 99.981% and 99.989% and leakage below 2.0 × 10^-5 per operation . That matters because leakage errors are not ordinary bit or phase mistakes inside the computational space; they are excursions into higher energy levels that can persist and poison later error-correction cycles.
The method, called R2D, extends the established DRAG pulse-shaping approach by suppressing multi-photon transitions into non-computational states such as |2⟩ and |3⟩ . In practical terms, the researchers shaped microwave control pulses so that the intended qubit rotation still happens, while spectral components likely to excite higher levels are reduced. Quantum Computing Report says this cut unwanted leakage transitions by more than a factor of 20 compared with standard DRAG protocols .
The key engineering lesson is that speed and accuracy cannot be optimized separately. Faster gates let a processor fit more operations inside a qubit’s coherence window, which is essential for deep circuits and error correction. Yet if speed produces extra leakage, the apparent gain can evaporate. The Jülich demonstration is important precisely because it shows an attempt to push into nanosecond-scale control without letting leakage dominate the error budget.
The caveat is equally important. The reported gate was tested on a single, isolated transmon qubit; scaling the same control method across multi-qubit chips will require confronting spectator-qubit dephasing, microwave cross-talk and frequency crowding . In other words, the work is a strong hardware-control benchmark, not a finished processor architecture.
Alice & Bob turns Shor’s algorithm into a floor plan
Alice & Bob’s announcement addresses a different level of the stack: not the shape of a microwave pulse, but the size and schedule of a fault-tolerant machine able to run Shor’s algorithm against real cryptographic targets. The company says its cat-qubit architecture needs 22,593 physical qubits and 3.14 days to factor RSA-2048 in the balanced configuration, and it presents the result as 40 times fewer physical qubits than a 2025 Google Quantum AI estimate cited by the company .
The estimate is more detailed than a simple logical-qubit count multiplied by a generic overhead. Alice & Bob says the work compiles the resource-heavy parts of the algorithm down to what physical qubits do and when, producing chip layouts and schedules . Quantum Computing Report similarly describes the study as including physical routing, magic-state factory floor plans and adder operations . This is significant because cryptanalytic timelines depend not only on abstract algorithms, but also on whether the assumed machine can be scheduled without hidden bottlenecks.
The company attributes the lower count to three main ingredients: cat qubits that suppress bit-flip errors in hardware, LDPC-cat codes that reduce overhead compared with earlier repetition-code estimates, and a redesigned adder tailored for cat-qubit execution . The same announcement lists trade-offs: a faster RSA-2048 configuration would use 610,530 qubits for 1.25 hours, while a smaller footprint would use 21,188 qubits for 8.08 days . For ECDLP-256, relevant to elliptic-curve systems used in blockchain signatures and encrypted connections, Alice & Bob gives 29,588 qubits and 61.8 minutes in its balanced table .
The assumptions are the story. The headline RSA-2048 estimate assumes no bit flips during the computation, equivalent in the company’s notes to a bit-flip time of roughly one century; if bit flips occur about once an hour, Alice & Bob says the RSA-2048 requirement rises to about 45,000 qubits . Design and compilation progress can lower resource estimates, but the hardware still has to satisfy demanding physical assumptions.
A cryptography warning, not a live break
For security teams, the Alice & Bob result should not be read as “RSA-2048 is broken today.” BankInfoSecurity framed the announcement as an estimate that RSA-2048 could be factored by Shor’s algorithm using fewer than 23,000 physical qubits in a best-case scenario, while noting that the cat-qubit approach depends on qubits that almost never bit-flip . That distinction matters because headlines about “breaking encryption” can blur the gap between a benchmark and an operational attack.
Still, benchmarks influence planning. Large organizations need years to inventory cryptography, replace protocols, update hardware security modules, test interoperability and manage certificates. Alice & Bob’s release explicitly argues that U.S. and U.K. guidance calls for completing the post-quantum cryptography transition by 2035 and recommends starting now . The reason is the familiar “harvest now, decrypt later” threat: data stolen today may remain sensitive long enough to be decrypted once capable quantum machines exist.
The useful takeaway is not panic, but prioritization. Long-lived secrets, critical infrastructure, financial systems, identity platforms and signed software supply chains should move first. Migration plans should include crypto-agility, so algorithms can be swapped again if today’s post-quantum schemes face future cryptanalytic surprises.
Why the two results belong together
Jülich and Alice & Bob are not solving the same problem, but they are tightening the same feedback loop. Jülich is improving the fidelity-speed trade-off at the physical-control layer. Alice & Bob is showing how architectural choices and compilation methods can compress the physical resources required for a famous fault-tolerant workload. One result asks whether qubits can be controlled fast enough without leakage; the other asks how many protected qubits a real algorithm would need if a particular hardware model works as assumed.
That is why the quantum race is becoming more concrete. The old scoreboard was often “more qubits.” The emerging scoreboard is harsher: fast gates, low leakage, biased noise, scalable error correction, credible layouts, realistic classical control, and runtimes short enough to matter. A qubit speedrun is impressive only if it does not crash the program; a cryptographic benchmark is useful only if its assumptions survive hardware reality.
The field has not crossed the finish line. But the loading screen is getting shorter.
Comments
Be the first to comment.