Iceberg Quantum, a Sydney-based company that designs fault-tolerant quantum computer architectures, said on October 22, 2026, that a machine of 72,000 physical qubits could factor a 2048-bit RSA integer in an expected 3.7 days. The company assumed a physical error rate of 0.1% and an error-correction cycle of 1 microsecond.
Iceberg published the figures in Pinnacle 1.2, an update to its Pinnacle architecture, as a post on its website without an accompanying paper. The company said it had cut the combined qubit-and-runtime cost of the attack by about an order of magnitude compared with its February preprint. In the first version of that paper, Iceberg estimated about 98,000 physical qubits for a run of up to a month at the same error rate. A May revision lowered that to about 94,000.
Outside researchers cannot yet reproduce the new numbers. Iceberg has not published the updated factoring circuits, a description of the decoder behind its simulations, or the hardware on which it measured that decoder’s speed.
The new estimates
Iceberg gave estimates for five configurations. At a 0.01% physical error rate and a 1-microsecond cycle, it put the requirement at 48,000 physical qubits for 0.98 days.
For superconducting chips, whose qubit connections are fixed when the chip is made, Iceberg estimated 99,000 physical qubits for 5.0 days. In a September preprint, that version, called Frozen Pinnacle, needed about 120,000 physical qubits for a one-month run.
At a 1-millisecond cycle, closer to the speed of trapped-ion and neutral-atom machines, Iceberg estimated 714,000 physical qubits for 29 days at a 0.01% error rate and 2.1 million physical qubits for 32 days at 0.1%.
Iceberg attributed the reductions from its February figures to smaller error-correcting codes, which its new simulations support, and to improved factoring circuits. The company said the underlying architecture did not change and that the smaller codes have about 40% less space overhead and about 30% less time overhead than the codes in its February estimates.
Decoder and simulations
Iceberg said the update includes the first direct circuit-level simulations in which logical error rates were low enough for RSA-2048 factoring. It credited Crevasse, a new decoder for the quantum low-density parity-check (QLDPC) codes that Pinnacle uses. A decoder is the classical system that reads error-correction measurements and works out which errors occurred.
The company reported a mean decoding time of 4 microseconds per round of error correction and said it processed trillions of simulated error-correction instances with Crevasse. It has not published the decoder’s design or its latency distribution. The 4-microsecond mean is longer than the 1-microsecond cycle assumed for the headline estimate, and Iceberg did not say how a machine would decode in real time at that pace.
According to Iceberg, most of the logical cycles in the factoring run require an error rate below 1 in a trillion per logical qubit-cycle. From 1 trillion samples of its distance-16 code, which holds 14 logical qubits, it reported three logical errors at a 0.1% physical error rate, about 2 in 10 trillion per logical qubit-cycle. Each logical cycle on that code spans 18 rounds of error correction.
Iceberg said it checked the modular exponentiation circuits from its February paper by running them on a large sample of random classical inputs. The company compared the method with one Google Quantum AI used in March for elliptic-curve circuits.
How the figure compares
In May 2025, Craig Gidney of Google Quantum AI estimated that RSA-2048 could be factored in under a week with fewer than 1 million noisy qubits, using surface codes and a 0.1% error rate.
For its 72,000-qubit figure, Iceberg used the same error rate and cycle time as Gidney but relied on QLDPC codes. In Pinnacle’s design, each physical qubit connects to others within a bounded distance, not only to its nearest neighbors, and some connections change between logical cycles. Frozen Pinnacle, with connections fixed at fabrication, needs 99,000 qubits under the same assumptions.
Iceberg said the estimates do not yet include lifted surgery, a faster method for logical measurements it presented in September, and that costs will fall further when the method is added.
No machine today approaches 72,000 physical qubits at a 0.1% error rate. Published deadlines for migrating to post-quantum cryptography are unaffected by the estimate. The European Union’s coordinated roadmap sets the end of 2030 for high-risk systems. The U.S. executive order signed in June requires post-quantum key establishment in high-value federal systems by December 31, 2030, and post-quantum signatures by December 31, 2031.
