Nineteen seconds. That is how long an IBM quantum processor took to execute a task that, according to the researchers, would require about 110 years on Frontier, one of the world’s fastest supercomputers.
The number is impressive. But the real story is more interesting than the headline, and it deserves to be told precisely.
What exactly happened
The experiment was conducted by a team led by Tigran Sedrakyan, a theoretical physicist at BlueQubit, a US-based quantum computing company. The study was published on arXiv on September 23, 2026 and revised on October 15.
The processor used was Nighthawk r2, a 120-qubit superconducting chip with square-lattice connectivity, identified in the system as ibm_phoenix.
The test is called random-circuit sampling, or RCS. In practice, the processor generated one million samples from a quantum circuit in 19 seconds of QPU time.
The exact configuration:
- 61 qubits effectively used;
- 36 cycles of depth;
- 918 native CZ gates;
- 1,000,000 samples collected in 19 seconds.
The full experiment consumed about 11 minutes of QPU time, including the fidelity estimators at every depth from 4 to 40 cycles and the final million-sample run.
Why the 110-year figure exists
The 110-year estimate is not a stopwatch someone started on Frontier. It is a mathematical projection.
The authors used a tensor-network simulation model and a bounded-fidelity rejection-sampling model. Under that framework, reproducing the same samples with comparable fidelity on Frontier would require about 1.2 × 10^27 machine operations, equivalent to more than a century of runtime.
And here is the detail that separates serious coverage from the easy headline: 110 years is an engineering estimate under stated conditions, not a law of nature.
ZME Science was direct: quantum advantage has always been a moving target. ScienceAlert recalled the same, adding that classical computing researchers are historically very good at finding shortcuts.
That is exactly what happened with Google in 2019, when Sycamore announced 200 seconds against an estimated 10,000 years. Years later, better classical algorithms drastically reduced that gap.
What makes this case different
Despite the caveats, there is something genuinely new here, and it comes in three parts.
It was done in the cloud, on commercial hardware
Nighthawk r2 is not a closed laboratory prototype. It is a commercial device available to anyone with an IBM Quantum account. The team used the standard Qiskit workflow, with no benchmark-specific calibration and no privileged access.
It was replicable by third parties
The study was conducted by researchers outside the manufacturer. That is the difference between a company testifying about itself and an independent group reproducing the result on public hardware.
Two independent methods agreed
The authors used two distinct fidelity estimators, mirror benchmarking and three- and four-patch cross-entropy benchmarking. Both agreed with each other at every measured depth, across more than two orders of magnitude of fidelity decay.
That last point is the most technically important. It is not enough for a quantum computer to be fast. You must prove the result is correct, and that is the harder part, because no classical computer can verify the answer.
The context: IBM had been building this path
This result did not appear in isolation. It fits a deliberate sequence.
On July 30, 2026, IBM and the University of Chicago announced a quantum advantage demonstration with 70 encoded logical qubits, using a structured method called doped Clifford sampling. The computation took about 15 minutes and came with a 95% statistical confidence guarantee, something no previous approach had delivered.
Jay Gambetta, director of IBM Research, said at the time: “we are now firmly in the quantum advantage era.”
IBM has publicly stated it aims to demonstrate quantum advantage by the end of 2026, and that it wants those claims tested by the wider community.
On October 7, 2026, the company advanced to Stage C of DARPA’s Quantum Benchmarking Initiative, the third stage of independent evaluation to determine whether it can build a fault-tolerant quantum computer.
The Nighthawk r2 result is, therefore, the missing piece: an external demonstration, on commercial hardware, with double verification.
What this means for businesses
Here is the connection to the ground of this blog.
The question any manager asks is unavoidable: does this change anything in my business over the next twelve months?
The honest answer is no. And it is important to say so.
The benchmark demonstrated is a sampling problem with no direct commercial application. It proves computational capability, not a solution to a logistics, chemistry or financial risk problem. The Qubit newsletter put it well: the result is real, but not yet useful.
That said, there are three lessons that already apply to any company.
1. The frontier of what is computable is moving
For decades, entire classes of problems simply had no viable solution: exact molecular simulation, large-scale combinatorial optimization, certain types of cryptography.
Each demonstration like this shrinks the perimeter of the impossible. Companies that build analytical competence today, in organized data and modeling, will be positioned to absorb the next layer when it arrives.
2. Advantage rarely arrives ready-made
The historical pattern is consistent: the technology arrives in raw form, years before it becomes applicable, and the first to benefit is not whoever has the best hardware. It is whoever has the data organized to feed it and the processes defined to use it.
Whoever treats data science applied to business as a priority today will be ahead of whoever treats it as a one-off project.
3. The bottleneck remains the same: data
A quantum computer does not fix dirty data. It does not fix poorly defined processes. It does not fix information scattered across spreadsheets and notebooks.
Before any computational leap, the foundation has to exist. That is exactly what intelligent automation does when organizing operations: connecting tools, centralizing records and turning scattered data into usable information.
And that is why Rapid Genius always starts in the same place: a solid digital structure before any advanced technology.
The caveats honest coverage requires
An article that takes this subject seriously must list the objections.
The study is a preprint. It was published on arXiv on September 23, 2026 and has not yet gone through peer review. The results are public and replicable, which reduces the risk, but it is not the same as formal validation.
The 110-year estimate depends on the model. It was calculated under a specific classical simulation model, with stated assumptions about memory and fidelity. A better classical algorithm in the future would reduce that number.
The comparison is not “quantum versus classical”. It is “quantum versus the best classical method known for this specific problem, according to the authors”. The difference is subtle and important.
The result has no immediate practical use. Random-circuit sampling is a benchmark, not an application. It is designed precisely to be hard to simulate classically, not to solve a real problem.
Why this still matters
Despite every caveat, the milestone is real, and for one specific reason.
For years, quantum advantage was a promise both grand and fragile. Every announcement came with asterisks: experimental hardware, restricted access, dubious verification, contested results.
What changed is that there is now a demonstration anyone with a cloud account can attempt to reproduce, on a commercial processor, with two independent verification methods agreeing, conducted by a group outside the manufacturer.
That is not a revolution. It is a change in category of credibility.
Conclusion: 19 seconds worth more for the method than the number
A 120-qubit IBM superconducting processor generated one million samples in 19 seconds, in a task whose classical reproduction was estimated at about 110 years on Frontier, with 1.2 × 10^27 operations.
The number is impressive and so is the caveat: it is an estimate under a specific model, in a preprint that has not yet passed peer review, and it solves a problem with no direct commercial application.
But what actually happened matters more than the headline: a team outside the manufacturer reproduced a quantum advantage result on commercial cloud hardware, verified by two independent methods that agreed with each other.
The question that remains:
If the frontier of what is computable is advancing like this, will your company have organized enough data to seize the next layer when it arrives?
Because technology, when it matures, does not wait for those still tidying the foundation.
