Iceberg Quantum Pinnacle 1.2 Crevasse Decoder RSA-2048 Explained: Factoring in 3.7 Days on 72,000 Qubits

Illustrative image: a quantum computer in a dilution refrigerator — Pinnacle 1.2 is an architecture and simulation release, not a physical machine, and the announcement carries no official press visual (Wikimedia Commons / OJB_Quantum, CC BY 4.0)

TL;DR: Iceberg Quantum Pinnacle 1.2 Crevasse decoder RSA-2048 explained in one line: the Sydney startup's new QLDPC decoder ran trillions of simulated error-correction cycles at just 4 microseconds per round, validating that RSA-2048 could be factored in 3.7 days on only 72,000 physical qubits — about ten times cheaper than its own estimates from February. It is still a simulation-based estimate, not a hardware demonstration.

📌 Key Highlights

• What is new: Pinnacle 1.2, released 22 October 2026, is the first direct circuit-level simulation showing logical error rates below 10⁻¹² per logical qubit-cycle — the threshold needed to run Shor's factoring algorithm on RSA-2048.
• The enabler: Crevasse, Iceberg Quantum's new quantum LDPC decoder, with a mean decoding time of 4 microseconds per error-correction round, processing trillions of error-correction instances directly instead of extrapolating.
• Headline numbers: 48,000 physical qubits in 0.98 days at p=10⁻⁴; 72,000 qubits in 3.7 days at p=10⁻³ — roughly an order of magnitude less spacetime cost than Pinnacle 1.0.
• The honest part: these are the company's own simulation results, not a quantum computer factoring RSA. No RSA has been broken on hardware anywhere.

What Happened: Iceberg Quantum Validated Pinnacle 1.2

On 22 October 2026, the Sydney-based fault-tolerant quantum architecture startup Iceberg Quantum published Pinnacle 1.2, an update to the architecture it unveiled in February 2026. The original Pinnacle claimed RSA-2048 could be factored with under 100,000 physical qubits using quantum low-density parity-check (QLDPC) codes — an order of magnitude below the best published estimates at the time. Pinnacle 1.2 tightens those claims further: it is, in the company's words, the first direct circuit-level simulation demonstrating logical error rates low enough for RSA-2048 factoring.

The word "direct" does heavy lifting here. Most resource estimates extrapolate error rates from small simulations using statistical models. Iceberg Quantum says it simulated trillions of error-correction instances directly, and from one trillion samples per case observed just three logical errors at p=10⁻³ (implying 2 × 10⁻¹³ errors per logical qubit-cycle) and zero logical errors at p=10⁻⁴. The company also compiled a complete, verified modular exponentiation circuit — the core arithmetic step of Shor's algorithm — into Pinnacle's fault-tolerant instruction set and counted the actual operations rather than estimating them.

The result: smaller code distances now suffice for factoring ([[126,12,10]] at p=10⁻⁴ and [[254,14,16]] at p=10⁻³, down from the codes assumed in Pinnacle 1.0), cutting space overhead by about 40% and time overhead by about 30%.

Iceberg Quantum Pinnacle 1.2 Crevasse Decoder RSA-2048 Explained: The Resource Table

Here are the updated factoring resource estimates from the release, assuming a 1-microsecond code cycle time for fast hardware (atomic/photonic qubits) and 1 millisecond for slower trapped-ion systems:

Physical error rate	Cycle time	Physical qubits	Expected runtime
10−4	1 µs (atomic/photonic)	48,000	0.98 days
10−3	1 µs (atomic/photonic)	72,000	3.7 days
10−4	1 ms (trapped ion)	714,000	29 days
10−3	1 ms (trapped ion)	2.1 million	32 days
10−3 (Frozen Pinnacle)	1 µs (fixed-connectivity superconducting)	99,000	5.0 days

All of this comes with one big, repeated caveat: the numbers are calculated from the company's own compilation and simulation of its own architecture. The underlying preprint and the architecture itself have not been independently peer-reviewed.

Illustrative image: a wire-bonded superconducting quantum chip — the kind of hardware the Crevasse decoder is meant to keep error-free at scale (Wikimedia Commons / OJB_Quantum, CC BY 4.0)

Why It Matters: The RSA Countdown Keeps Shrinking

RSA-2048 underpins banking, HTTPS, VPNs and much of internet security, and Shor's 1994 algorithm can factor it in principle. For years the resource estimates made it look far away: ~20 million qubits (Gidney & Ekerå, 2019), then under a million (Gidney, 2025), then under 100,000 with the original Pinnacle (February 2026). Pinnacle 1.2 pushes the curve further down: 72,000 qubits in under four days, at error rates today's best superconducting and neutral-atom machines already approach for single qubits.

This is why the story matters beyond quantum circles: several hardware companies — PsiQuantum, Diraq and IonQ, all working with Iceberg Quantum — have public roadmaps targeting 100,000+ qubit systems within a few years. The distance between those roadmaps and the 72,000-qubit line is now a matter of engineering and scaling, not orders-of-magnitude breakthroughs. It is also why the same week brought NVIDIA JUPITER 104-qubit quantum simulation explained — magic-state techniques cutting qubit needs by 90% — and a quantum error-correction decoder that cut logical errors by 12–13% on real Google hardware data. The whole field is converging on cheaper error correction.

But keep both feet on the ground. As independent analysis stresses, nobody has broken RSA-2048 on hardware. The world's largest processors sit at roughly 1,100 noisy physical qubits with error rates around 10⁻³ — about a billion times worse than the ~10⁻¹² logical error regime cryptographic factoring demands. Pinnacle 1.2 tells us the theoretical bar is lower than we thought; the hardware still has to climb it. For organisations, the takeaway is the one governments have been issuing for years: post-quantum cryptography migration is a precaution to run now, not a fire drill later.

✅ Should you care? If you work in security or cryptography — yes, as a milestone on the "harvest now, decrypt later" timeline, not as an emergency. The estimate says a cryptographically relevant quantum computer needs roughly 70–100k good qubits, which sits inside stated hardware roadmaps of the late 2020s. If you're a curious reader, the honest summary is: the maths says it's possible with far fewer qubits than anyone thought in 2019; no machine anywhere close exists yet.

FAQs

How many qubits does it take to break RSA-2048?
The latest estimates range widely by architecture. Iceberg Quantum's Pinnacle 1.2 says 72,000 physical qubits at a 10⁻³ error rate could factor RSA-2048 in 3.7 days (or 48,000 qubits in 0.98 days at 10⁻⁴). Google's Craig Gidney estimated under a million qubits in 2025. All are simulation-based estimates; no hardware factoring of RSA-2048 has been performed.

What is Iceberg Quantum's Pinnacle 1.2?
Pinnacle 1.2 is an October 2026 update to Iceberg Quantum's fault-tolerant quantum computing architecture, first unveiled in February 2026. It is based on quantum LDPC error-correcting codes and validates, via direct circuit-level simulation, that RSA-2048 factoring is achievable with roughly ten times less spacetime overhead than the original Pinnacle estimates.

What is the Crevasse decoder?
Crevasse is Iceberg Quantum's new quantum LDPC decoder, the engine behind Pinnacle 1.2. It decodes errors in a mean of 4 microseconds per error-correction round, letting the team process trillions of simulated error-correction instances directly and demonstrate logical error rates below one error per trillion logical qubit-cycles without extrapolation.

Has a quantum computer broken RSA-2048 yet?
No. Pinnacle 1.2 is a simulation-based resource estimate from one company, not a hardware demonstration. The largest real quantum processors have around a thousand noisy physical qubits, while breaking RSA-2048 requires tens to hundreds of thousands of error-corrected qubits operating at error rates a billion times lower than today's. Keep encryption migration plans moving, but nobody is secretly close.

Takeaway: Pinnacle 1.2 is the sharpest estimate yet of how cheaply a future quantum computer could crack RSA-2048 — 72,000 qubits in 3.7 days, validated by direct simulation. It is company-reported, not peer-reviewed, and the hardware does not exist. Watch the decoders: error correction, not qubit count, is where this race is now being won.

Follow World Tech on TechByCountry for tomorrow's 6 PM update.

Sources

• Iceberg Quantum, "Pinnacle 1.2" (primary release, 22 Oct 2026) — iceberg-quantum.com/blog/releases/pinnacle-1-2/
• Quantum Computing Report, "Iceberg Quantum Validates Pinnacle 1.2 Architecture with Crevasse QLDPC Decoder to Reduce RSA-2048 Factoring Resources" (22 Oct 2026) — quantumcomputingreport.com
• postquantum.com, "No One Has Secretly Broken RSA-2048 or RSA-4096 — Here's the Science" — postquantum.com
• Intelligent CIO Europe, "Talking Point: NVIDIA's JUPITER breakthrough brings practical quantum computing a step closer" (23 Oct 2026) — intelligentcio.com