The EU and Spain deploy Qilimanjaro Quantum Tech's analog quantum computer at the Barcelona Supercomputer Center
The EU is making a concerted effort to study multiple types of analog quantum computers and these computers’ use as accelerators integrated with high performance computing (HPC) infrastructure.
Key takeaways
The EU has partnered with Spain to deploy an analog quantum computer from the Spanish vendor Qilimanjaro Quantum Tech at the Barcelona Supercomputing Center. The deployment will provide Spanish and EU researchers with hybrid computing infrastructure to research how to use and develop quantum computing technologies. This is an important milestone for the EU as it increasingly works to develop sovereignty over key digital and emerging technologies and avoid reliance on US and Chinese technology giants.
Analog quantum computers are currently limited in scale, but have important benefits compared to digital quantum computers. The different computational model of analog quantum computers is especially well-suited to problems that feature continuous variables, which span a variety of applications, ranging from the life sciences to advanced industrial design and engineering. Analog quantum computers might provide the earliest instances of quantum advantage for key types of problems.
One particularly interesting use of analog quantum computers is for feature extraction as part of larger hybrid quantum-classical AI systems. Analog quantum computers are suitable for use as “reservoir computers” that leverage the natural systems dynamics of analog quantum computing to perform feature extraction in a quicker and less computationally complex manner than traditional AI model training.
The EU and Spain deploy an analog quantum computer for research purposes
The European Union (EU) and Spain deployed an analog quantum computer (AQC) at the Barcelona Supercomputing Center (BSC) and inaugurated the installation on May 28, 2026. [1] This installation represents an EU-wide effort to develop a network of hybrid high-performance computing (HPC)/quantum computing (QC) centers for use by researchers. [2] While QCs aren’t yet able to beat classical HPCs (also known as supercomputers) for problems of practical interest, this infrastructure gives researchers the ability to study both how to advance the performance of QCs and how to integrate QCs with “classical” (i.e., non-QC) HPC systems.
The BSC’s new AQC was provided by Qilimanjaro Quantum Tech, a Spanish QC maker, which also installed a digital QC (DQC) at the Center in 2025 as part of a project sponsored by the Quantum Spain program. [3] The new AQC and existing DQC will be integrated with the “MareNostrum 5” supercomputer at the Center, and the combined AQC/DQC/classical HPC infrastructure is called MareNostrum-ONA. Qilimanjaro Quantum Tech will install two future generations of AQCs at the BSC in 2026 and 2027.
European sovereignty over access to emerging technologies like QC is a key reason for the deployment of infrastructure such as MareNostrum-ONA. For example, French President Emmanuel Macron, while announcing a new round of €1 billion funding for QC on May 22, 2026 [4] stated, “In all of these questions, there’s a battle over sovereignty that is being fought and must absolutely be won... technological dependencies will more and more become industrial and strategic dependencies.”
MareNostrum-ONA is specifically the result of a project called EuroQCS-Spain joint funded in the amount of €8.5 million by the EuroHPC-JU (European High Performance Computing Joint Undertaking) and SEDIA (the Spanish State Secretariat for Digitalization and Artificial Intelligence). Prior to this new installation at the BSC, the EuroHPC-JU had also installed various types of QCs at HPC centers in Czechia, France, Germany, Italy, and Poland.
The pros and cons of analog quantum computers versus digital quantum computers
AQCs occupy a niche position in the QC industry compared to DQCs and only a handful of vendors make AQCs: in addition to Qilimanjaro Quantum Tech there is Pasqal (headquartered in France), QuEra (US), D-Wave (US), and Silicon Quantum Computing (Australia). This niche status is largely because the industry thinks there is a greater chance of scaling up DQCs (also known as “gate-based QCs” or “circuit-based QCs”) to be able to handle very large computations. DQCs can’t compute large problems today, and may never be able to do so, but if researchers can enable DQCs to leverage “Quantum Error Correction” (QEC) then large computations maybe be possible in the future.
Despite not having a similar path to using QEC, AQCs may offer more performance for computing certain types of problems at small and medium scales compared to current, uncorrected (i.e., no QEC) DQCs. This potential benefit comes from the differences in how these two types of QCs compute. AQCs compute by evolving the natural state of the AQC system smoothly toward a correct answer. In contrast, DQCs are programmed with a series of gate operations and these gate operations each introduce a small amount of error that, compounded over the course of a computation, eventually swamp the system with noise and prevent a useful answer from being obtained. While AQCs still experience some kinds of errors, they avoid gate errors because they simply don’t have gates.
More specific to the installation of the AQC in the BSC, this AQC from Qilimanjaro Quantum Tech uses a type of superconducting circuit qubit known as a “fluxonium” qubit. This contrasts with another type of superconducting circuit qubit called a “transmon” qubit that Qilimanjaro Quantum Tech uses for its DQC, including the DQC it installed last year at the BSC. Fluxonium qubits have a longer “relaxation window” compared to transmon qubits, meaning that they have a longer period to perform a computation before the computer’s qubits lose the energy needed to stay in a quantum state to return a correct answer rather than just return noise. A longer relaxation window is therefore a specific advantage of fluxonium qubits in an AQC versus transmon qubits in an uncorrected DQC. [5]
The research undertaken at BCS, and other EU HPC centers integrating AQCs into their infrastructure, may provide interesting new tools to extend the computational abilities of AQCs. One type of AQC – adiabatic QC – has been shown to be functionally equivalent to DQCs. [6] If QEC can be developed for AQCs, perhaps it can be developed for adiabatic QCs as well. However, given how hard it has been to develop QEC for AQCs, any possible adiabatic QC quantum error correction is likely to be far in the future.
In the meantime, before DQCs have the benefit of QEC to run very large computations, AQCs may provide a useful path to commercial quantum advantage – an advantage over classical computers as judged by the user – in certain use cases across optimization, simulation, and AI. As Qilimanjaro Quantum Tech’s CEO Marta Estarellas points out [7], AQCs are particularly good at “continuous variable” problems (as opposed to “discrete variable” problems). AQCs can do both, but continuous variable problems, like finding the ground state of a molecule or the optimal design of a wing using fluid dynamics calculations, inherently match the structure of the AQC computing model, reducing the overhead of embedding the problem into the system.
Reservoir computing could be a great AI use case for analog quantum computers
AQCs may provide a benefit for training AI models, especially models in which time analysis is an important component, such as in time-series forecasting. [8] This benefit is the ability to train the model faster and with less computing resources than other approaches to such training. More specifically, this training benefit is in the form of efficient “feature extraction” due to using the AQC as a “quantum reservoir computer” (QRC).
Feature extraction is the process of analyzing a dataset to pull out the important characteristics (e.g., “features”) of the data that impact the analysis of the data. A generic example that’s easy to understand is the extraction of features such as “eyes” or “arms” from a digital photo that is made up of a grid of pixels. The AI uses these features to classify the photo as representing a “person” rather than, for example, a “car”.
A QRC uses the inherent physical properties of the AQC to perform feature extraction as a dynamic physical process rather than as a computational process. To be clear, feature extraction can be performed without using reservoir computers. AI model types such as “recurrent neural networks” (RNNs) can extract features computationally. But the key idea with QRC is that using a physical process, rather than a computational process, to perform feature extraction will be quicker and require less computational resources.
Interestingly, reservoir computing can be done on many types of physical systems, not just ACQs. For example, researchers have performed actual, valid feature extraction using a bucket of water. But QRC potentially gains a key benefit from the high dimensionality inherent in quantum computing. Each additional qubit in the system doubles the number of overall dimensions processed in the system. High dimensionality is important along with the ability to mix input data together and the ability to incorporate a fading memory of the input data stream as it comes in (which is important when there is a meaningful connection between data points separated in time).
These three characteristics – high dimensionality, non-linear mixing, and temporal dynamics – theoretically enable a QRC to extract features by making a noisy data stream “separable”. A simple analogy would be to imagine a scattering of circles and squares mixed on a two-dimensional graph plot. There’s no way in this plot to draw a line cleanly separating circles from squares. But by projecting the plot into a third dimension (adding a z-axis to the original x-axis and y-axis), and “raising” the circles and squares up into this new, higher dimension, we may find that we can cleanly draw a “line” (really a plane in the case of three dimensions) between the circles and the squares. In this example, the ability to separate and cluster the circles and squares is the “feature extraction”. The ability to draw a plane between the two clusters is classification: this data element is a “circle” and that data element is a “square” because of which side of the plane the data element is located.
Research undertaken at the BSC will hopefully extend our understanding of how QRC can be optimally used for feature extraction. It’s likely that we’ll find that QRC is most useful for relatively low dimensional streaming data. Quantum computers have a “data loading” challenge, although this doesn’t quite apply to AQCs, since in the AQC case we’re not loading data into gate-based circuits, as with DCQs, we’re adjusting analog controls. Nevertheless, it’s unlikely that QRC will be used with very large classical data sets, such as video.
Why this matters
While there are real questions about the ultimate scalability of AQCs, there is reason to believe that AQCs can provide a tangible benefit over DQCs today for certain kinds of problems. This is an important consideration both from the standpoint of gaining near-term benefits from quantum computing and because the industry doesn’t know for sure that QEC of large scale DQCs will be possible. It looks like QEC is viable, but if it’s not, AQC will have a very important role to play for the quantum computing industry over the long term.
Citations
[1] Qilimanjaro Quantum Tech, “Qilimanjaro’s New Analog Quantum System Inaugurated at the Barcelona Supercomputing Center,” (May 28, 2026). https://qilimanjaro.tech/qilimanjaros-new-analog-quantum-system-inaugurated-at-the-barcelona-supercomputing-center/
[2] EuroHPC JU, “EuroQCS-Spain,” (accessed June 2, 2026). https://www.eurohpc-ju.europa.eu/eurohpc-quantum-computers/our-quantum-computers_en#euroqcs-spain
[3] Qilimanjaro Quantum Tech, “Quantum Spain Is Complete. Here’s What Researchers Are Already Doing With It,” (May 21, 2026). https://qilimanjaro.tech/quantum-spain-is-complete-heres-what-researchers-are-already-doing-with-it/
[4] France24, “France announces billion-euro boost for quantum computing” (May 5, 2026). https://www.france24.com/en/live-news/20260522-france-announces-billion-euro-boost-for-quantum-computing
[5] Nguyen, Long B., et al., “Scalable High-Performance Fluxonium Quantum Processor,” (February 5, 2022). https://arxiv.org/pdf/2201.09374
[6] Aharanov, et al., “Adiabatic Quantum Computation is Equivalent to Standard Quantum Computation,” (March 26, 2005; also listed with a date of February 1, 2008). https://arxiv.org/pdf/quant-ph/0405098
[7] EE Times, “Qilimanjaro Pushes Analog Quantum as AI Compute Demands Surge,” (May 29, 2026). https://www.eetimes.com/qilimanjaro-pushes-analog-quantum-as-ai-compute-demands-surge/
[8] te Vrugt, Michael, “An introduction to reservoir computing,” (December 12, 2024). https://arxiv.org/pdf/2412.13212v1