For six years, 48 qubits was the wall. No classical supercomputer on Earth could perform a universal, exact simulation of a quantum machine beyond that threshold, a limit set in 2019 by Japan’s now-retired K computer. In June 2026, researchers at Germany’s Forschungszentrum Jülich announced they had blown past it, simulating a full 50-qubit quantum system on JUPITER, Europe’s first exascale supercomputer.
The achievement, described in a peer-reviewed paper accepted by Future Generation Computer Systems, required roughly 2 petabytes of memory. To put that in perspective, 2 petabytes is about 2 million gigabytes, enough to store roughly 40,000 Blu-ray discs’ worth of data. All of it was needed just to hold the complete quantum state of a 50-qubit system in memory at once.
Why two extra qubits are a massive leap
In quantum computing, every additional qubit doubles the amount of information a simulator must track. Going from 48 to 50 qubits does not sound dramatic, but it means the classical machine must handle four times as much data. That exponential scaling is precisely why quantum computers are expected to eventually outrun their classical counterparts, and why pushing the classical ceiling even slightly higher is a significant engineering feat.
The previous record was documented in a 2019 paper published in Computer Physics Communications. That study used an earlier version of the same Jülich simulator family, running on Japan’s K computer, and topped out at 48 qubits. No group had publicly surpassed that mark for universal, exact simulation in the years since.
How JUPITER pulled it off
JUPITER, housed at the Jülich Supercomputing Centre, is built around thousands of NVIDIA GH200 Grace Hopper Superchip nodes that blend CPU and GPU processing power. The Jülich team developed a purpose-built simulator called JUQCS-50 to exploit that hybrid architecture, splitting the enormous 50-qubit state vector across thousands of nodes while overlapping communication with computation to keep the GPUs continuously fed with data.
According to the accompanying preprint, JUQCS-50 maintains bitwise-exact amplitudes for every single basis state rather than relying on lossy compression or problem-specific shortcuts. That distinction is critical: it is what qualifies the run as a universal, exact simulation, meaning the system can handle any arbitrary quantum circuit, not just ones tailored to make classical simulation easier.
Prof. Kristel Michielsen, a lead researcher on the project, confirmed the approximately 2-petabyte memory requirement in the institutional release distributed by Forschungszentrum Jülich. The paper also documents strong scaling tests at lower qubit counts, showing near-linear speedup as more GH200 nodes are added, evidence that the simulator’s design genuinely harnesses JUPITER’s exascale resources rather than simply brute-forcing the problem.
What has not been confirmed yet
The result currently rests entirely on the Jülich group’s own publication chain. Peer review provides a layer of scrutiny, but no independent team has yet replicated the 50-qubit simulation on a different exascale system. As of early July 2026, no such attempt has been publicly announced, though centers operating comparable machines in the United States and Asia could potentially try.
Some technical details also remain outside the public record. The raw node-hour counts and memory traces from the actual JUPITER run are referenced in the preprint but have not been released as open data. Without those logs, outside researchers cannot fully verify the wall-clock performance or reproduce the exact scaling curve. Specific compiler flags, partition sizes, and GPU memory allocation parameters are described in the press summary but not fully detailed in the journal paper, a gap that matters because small configuration choices at this scale can determine whether a simulation completes or crashes.
There is also an important distinction readers should keep in mind. Other research groups have simulated larger qubit counts using approximate or circuit-specific methods, such as tensor-network techniques applied to Google’s Sycamore circuits. Those efforts solve a different and narrower problem. The 50-qubit record applies specifically to universal, exact simulation, and the two categories should not be conflated when comparing headline numbers.
What this means for the quantum vs. classical race
When Google published its quantum supremacy experiment in 2019, the central question was whether any classical supercomputer could match the output of a roughly 50-qubit quantum processor running random circuits. Classical simulation limits were, and remain, the yardstick against which quantum advantage claims are measured.
Pushing that classical frontier from 48 to 50 qubits tightens the contest. It means conventional machines can now brute-force their way through problems that were recently considered out of reach, raising the bar that near-term quantum devices from Google, IBM, and others must clear to demonstrate genuine advantage on unstructured tasks.
But the result cuts both ways. The sheer cost of the JUPITER run, in terms of energy, specialized hardware, and engineering effort, far exceeds what a comparable quantum processor would need to execute the same circuit, assuming similar error rates. Classical simulation is catching up, but it is catching up expensively.
How much room JUPITER has left is another open question. The 2-petabyte memory footprint suggests only a slim margin remains before the system would exhaust its addressable space for a single job. Scaling to 51 or 52 qubits with the same exact method could demand either substantially more memory per node or aggressive compression techniques that would sacrifice the “exact” label. Until someone publishes a concrete attempt at higher counts, the community can only estimate these limits from theoretical scaling curves.
Why the next two qubits will be even harder to reach
The Jülich result does not erase the promise of quantum computing. It does, however, redraw the line that quantum hardware must cross to prove it can do something no classical machine can replicate. For researchers designing quantum algorithms, it provides a sharper benchmark. For engineers building quantum processors, it raises the stakes: the classical competition just got two qubits harder to beat.
And for anyone watching the broader technology landscape, classical and quantum computing are locked in a dynamic, iterative race. Each time one side advances, it forces the other to push further. The 50-qubit simulation is the latest move in that contest, and it will not be the last.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
