Does RSA-2048 Really Need Only 100,000 Physical Qubits? The Assumptions Behind the Claim

Bottom line

The claim that “100,000 physical qubits are enough to break RSA-2048” leaves out most of the engineering conditions that make the number possible.

More precisely, the 2026 Pinnacle Architecture preprint estimates that RSA-2048 could be factored with fewer than 100,000 physical qubits in a QLDPC-based fault-tolerant architecture, assuming a physical error rate of 10^-3, a 1-microsecond code cycle, and a 10-microsecond reaction time.

No 100,000-qubit machine has performed this calculation. The result is a conditional resource estimate built from an architecture proposal, simulated error performance, and assumptions about how the quantum and classical systems would operate.

When I encounter a qubit headline like this, I look past the count first. How accurate must the qubits be? How quickly must error-correction cycles run? How fast must the classical controller process measurements and respond? Without those details, “100,000” can easily be mistaken for a hardware roadmap.

What happened

The Pinnacle Architecture is a proposed fault-tolerant quantum-computing design based on quantum low-density parity-check, or QLDPC, codes. Its central advantage is spatial efficiency: a QLDPC block can encode multiple logical qubits rather than devoting a large separate region to each one.

The proposal combines processing units, “magic engines” that supply high-quality magic states, optional memory, and a parallelization technique called Clifford frame cleaning. The RSA-2048 estimate also uses a parallelized compilation of the factoring calculation.

The headline resource point looks like this:

Factor	Assumption or result	Why it matters
Physical error rate	10^-3	Assumed error level for physical operations
Code-cycle time	1 microsecond	Time allowed for one syndrome-extraction cycle
Classical reaction time	10 microseconds	Time allowed to process a result and adapt later measurements
Physical qubits	About 98,000	Estimated footprint for the one-month configuration
Expected runtime	One month	A long fault-tolerant computation, not instant decryption
Evidence type	Preprint, simulation, and resource estimate	Not an integrated hardware demonstration

Runtime and qubit count trade against each other. Under the same 10^-3 error rate and 1-microsecond cycle assumption, the preprint estimates about 151,000 physical qubits for a one-week computation and about 471,000 for a one-day computation.

The more accurate headline is therefore not “100,000 qubits are sufficient.” It is: the paper identifies a space-time tradeoff near 98,000 physical qubits and one month of runtime under a specific set of operating assumptions.

Evidence level

The Pinnacle Architecture paper is a preprint. It was first submitted on February 12, 2026, and revised on May 5, 2026. Its contribution is an architecture and resource analysis, not a report of RSA-2048 being factored on quantum hardware.

The authors numerically simulate memory and logical-measurement error rates for a family of QLDPC codes. The draft describes a circuit-level depolarizing noise model and maximum-likelihood decoding, then extrapolates the results to select code parameters and calculate the total resource requirement.

That analysis is useful for comparing fault-tolerant designs. It does not demonstrate that a complete physical machine would achieve the same performance. Real devices can exhibit noise that differs from an idealized model, while every quantum and classical subsystem must operate together at the required speed.

A practical way to sort the evidence is to separate three layers:

1. What the resource analysis calculates: qubit counts and runtimes under stated error and timing assumptions.
2. What the simulations support: modeled error performance for the selected QLDPC codes and logical-measurement methods.
3. What hardware still must demonstrate: integrated operation at the required scale, speed, and reliability.

Success at the first two layers does not establish the third.

Why the qubit estimate fell

The authors position Pinnacle as roughly an order-of-magnitude reduction from earlier fault-tolerant estimates. The proposed improvement does not come from assuming that individual physical qubits suddenly become dramatically better. It comes from changing how information is protected and processed.

• QLDPC blocks encode multiple logical qubits, reducing spatial overhead.
• Measurement gadgets and bridges support logical Pauli measurements.
• Magic engines are designed to supply the high-quality magic states needed for universal computation.
• Parallelization trades additional working registers for a shorter runtime while sharing part of the input memory.

This is not the same computation placed unchanged onto fewer qubits. The code, logical operations, memory access, magic-state supply, and compilation strategy all change together.

That distinction also prevents a common mistake: attaching the 100,000 figure to every quantum-hardware roadmap. A platform that cannot support the required connectivity, logical operations, timing, and control path cannot simply substitute its physical-qubit count into this estimate.

What remains uncertain

The clearest unresolved component is the real-time decoder.

The error simulations use maximum-likelihood decoding, but the paper leaves the development of a sufficiently fast practical decoder outside its analysis. A decoder may perform well mathematically and still fail the architecture’s timing requirement if it cannot process syndrome data quickly enough.

The 10-microsecond reaction time is therefore not a minor parameter. It represents an end-to-end classical path involving measurement, data transport, decoding, a control decision, and an adapted quantum operation.

The architecture would also require its processing blocks, bridges, magic engines, memory operations, and parallel-access mechanisms to work together at large scale. The preprint estimates the cost of these elements; it does not report a month-long RSA computation on an integrated machine containing approximately 100,000 physical qubits.

Other open questions include whether the projected advantage survives correlated noise, device-specific error behavior, engineering constraints, and practical decoder approximations.

What changes if the estimate holds up

If later analysis and hardware experiments preserve this reduction, the assumed scale of a cryptographically relevant quantum computer could change substantially. A fixed belief that RSA-2048 necessarily requires around a million physical qubits would no longer be a safe planning baseline.

But a smaller resource estimate does not produce an arrival date. Reducing the number of required qubits and building qubits that meet the assumed quality, connectivity, control, and error-correction requirements are different problems.

The useful decision rule is to separate resource estimation from schedule forecasting. This preprint asks what resources would be needed if its operating conditions were available. Predicting when industry will meet those conditions requires separate evidence from hardware, manufacturing, control systems, and integrated fault-tolerance experiments.

Why this does not set a new PQC deadline

NIST’s post-quantum cryptography guidance does not depend on one qubit estimate or a single forecast for a cryptographically relevant quantum computer.

NIST notes that such machines may be years or decades away, while still advising organizations to begin applying its post-quantum standards. Its three principal standards, released in 2024, are ML-KEM for key establishment and ML-DSA and SLH-DSA for digital signatures.

Migration takes time because organizations must locate vulnerable algorithms, update products and protocols, test interoperability, and manage operational effects. Under the transition timeline cited by NIST, quantum-vulnerable algorithms are to be deprecated and ultimately removed from NIST standards by 2035, with high-risk systems moving earlier.

The practical interpretation of the Pinnacle preprint is not that a 100,000-qubit RSA-breaking machine is imminent. It is that quantum resource estimates can move, so a cryptographic migration plan should not depend on one predicted qubit threshold or arrival year.

Organizations evaluating their exposure should use the current NIST Post-Quantum Cryptography project as the policy and standards reference, while treating individual architecture papers as evidence about how the technical threat model may evolve.

What to watch next

The next meaningful updates will not necessarily be records for raw physical-qubit count. Better checkpoints are:

• Whether peer review preserves the paper’s central resource estimates.
• Whether the QLDPC advantage survives correlated and device-specific noise models.
• Whether a practical decoder can approach the simulated performance within the reaction-time budget.
• At what scale the processing blocks, bridges, magic engines, and memory components are demonstrated together.
• Whether researchers publish end-to-end benchmarks that include code cycles and classical-control latency.
• Whether the proposed space-time tradeoff remains reproducible on physical hardware.

The 100,000-qubit estimate should neither be dismissed nor treated as a completed machine. The narrower, evidence-based reading is more useful: this QLDPC architecture may sharply reduce the spatial cost assigned to an RSA-2048 attack, but the decoder, control path, and integrated fault-tolerant hardware remain unproven parts of that path.

Sources

• The Pinnacle Architecture: Reducing the cost of breaking RSA-2048 to 100 000 physical qubits using quantum LDPC codes, arXiv preprint.
• Post-Quantum Cryptography, NIST Computer Security Resource Center.