The hybrid plumbing is the real story: standard Slurm scheduling, QRMI middleware, and CUDA Q plus Qiskit turn a Rydberg atom array into a scheduled resource on a Top 10 pre exascale machine.
CINECA's Leonardo supercomputer, ranked 10th on the Top500 list of the world's most powerful machines, now has a 140-qubit neutral-atom quantum processor wired into its job queue as a co-processor. The plumbing around that addition is the actual story behind Pasqal's inauguration of its SOL system at Bologna's DAMA Technopole on 2026-06-11.
The SOL machine, formally an Orion-class quantum processing unit, holds neutral atoms in arrays of optical tweezers, then entangles neighboring atoms by exciting them into Rydberg states. A user running a hybrid job submits it through standard Slurm, the same scheduler that queues classical work on Leonardo, and the quantum portion routes through QRMI, the open-source middleware that exposes the QPU to HPC environments. On top of that sit two of the most widely used quantum software development kits: NVIDIA's CUDA-Q for hybrid CPU-GPU-QPU workflows, and IBM's open-source Qiskit. That stack matters because it lets a researcher treat the Rydberg array as just another accelerator on a familiar HPC system, rather than as a separate experiment that has to be re-engineered end to end.
This is the third Pasqal system to enter the EuroHPC federated network, after Jade at France's CEA-TGCC and Ruby at Germany's FZJ-JSC, both installed under the HPCQS pilot. According to Pasqal's release, the European High Performance Computing Joint Undertaking is co-funding SOL alongside Italy's Ministry of University and Research through the Italian Research Center on HPC, Big Data and Quantum Computing (ICSC). The release positions this as part of the 2025 European Quantum Strategy's push to build out federated quantum-HPC infrastructure rather than buying standalone quantum machines. The same dateline also saw the debut of LISA, an AI-focused upgrade to Leonardo itself.
The trade press has framed the announcement as Italy's first operational neutral-atom quantum computer, which is true and which Quantum Computing Report covered on 2026-06-11 following its earlier February report on the physical delivery. National primacy is the easy framing. The harder question is what 140 Rydberg atoms wired into a pre-exascale classical machine actually buys researchers, and where the limits sit.
The honest answer: 140 qubits is enough for analog optimization problems, exploration of machine-learning kernels that map onto qubit connectivity, and quantum-chemistry prototyping that can ride a hybrid loop, sending a candidate wavefunction to the QPU, getting an energy estimate back, and feeding it into a classical optimizer running on Leonardo's CPUs and GPUs. It is not a fault-tolerant milestone. There is no claim of error correction at scale, no claim of quantum advantage on a useful workload, and no path in the release from 140 atoms to the millions of physical qubits that surface-code error correction would require for general-purpose fault tolerance. The QPU runs as an analog processor in Pasqal's current product framing, not as a gate-model machine running Shor at commercial scale.
For the user, the practical access path is Slurm plus QRMI. For a domain scientist who already submits batch jobs to Leonardo, the workflow should look familiar: write a job script, request quantum resources, get a result back. Whether that workflow survives contact with real research workloads at scale is the open question the EuroHPC federation is now in a position to answer. Pasqal chief executive Wasiq Bokhari has framed SOL as turning quantum computing "from a research promise into deployed, operational infrastructure," a claim whose value depends entirely on the workloads that get run in the next twelve months.
Two other threads are worth watching. The first is scientific throughput. ICSC and CINECA have not published benchmark numbers showing what fraction of a hybrid job runs on the QPU versus on classical hardware, what the queueing overhead looks like, and what error rates the analog processor hits in production-like conditions. Those numbers will decide whether the integration is a usable scientific instrument or a press event with a 140-qubit backdrop.
The second is commercial. Pasqal announced a business combination with Bleichroeder Acquisition Corp. II (Nasdaq: BBCQ) earlier in 2026, and an F-4 registration statement was filed with the SEC on 2026-05-26. The SOL inauguration lands in the middle of that pre-listing window, and EuroHPC-funded installations are part of the company's case for commercial scaling. The release lists more than 25 clients and partners, including Aramco, CMA CGM, OVHcloud, Thales, IBM, and Sumitomo, with more than $300 million in cumulative funding. None of that is independently audited here, and any forward-looking claim from the F-4 should be treated as a SPAC-era projection rather than a confirmed outcome.
What the inauguration actually proves is that a Rydberg neutral-atom array can be exposed as a scheduled resource on a Top-10 pre-exascale system, in a real production setting, using a real HPC-grade scheduler and open-source middleware. That is a real engineering result. The remaining question is what the workloads look like, and whether EuroHPC's federated model can produce benchmark evidence that the rest of the field accepts as scientific throughput rather than vendor marketing.