Computation as we know it is reaching its limits. This is not only due to the fact that there is a need to solve increasingly complex problems with better precision and within practical times, but also because the surge of AI has dramatically increased the demand for computing infrastructure.

Big players such as Amazon, Google, or Meta are signing agreements with energy companies to provide them with Small Modular Reactors (SMR) to power the high demands of their AI data centers. Currently, in the US, four percent of the total electricity is used for AI training, and it is expected to grow up to 12 percent in 2028.

Without a change in our computational paradigm, this demand will continue to grow exponentially in the coming years. It is time to move toward a more sustainable era of computation.

Quantum computing is expected to speed up compute workloads by redefining the physical and mathematical prism in which the algorithm is formulated, similarly to what GPUs did around 2007 with the release of CUDA and the first graphics cards conceived for calculus.

The deployment of quantum computers in data centers and HPC facilities is increasingly following the established paradigm of heterogeneous computing. By integrating quantum processors at the workload-management level, they can be treated as additional accelerators within the HPC environment, enabling hybrid workflows and orchestration mechanisms like those already used for GPUs and other specialized hardware.

Early examples of this approach are already being explored at the Barcelona Supercomputing Center (BSC-CNS), where superconducting quantum systems have been deployed alongside the MareNostrum-5 supercomputing infrastructure to enable hybrid classical-quantum workflows.

Quantum computing comes in different shapes and forms, depending on the physical hardware implementation of the qubits and their operational model. Among these, superconducting-based analog quantum computers emerge as one of the most promising technologies. One of the key strengths of such systems is their ability to directly embed the problem into the degrees of freedom of the quantum chip with a one-to-one mapping and higher tolerance to errors. This approach allows the system to reproduce the natural dynamics of the target problem through continuous analog control of the relevant parameters.

This makes analog quantum computing particularly suitable for executing algorithms such as quantum reservoirs, materials, or chemistry simulations, or combinatorial optimization industrial problems, potentially outperforming the digital counterpart by avoiding the discretization of continuous processes and the introduced errors that it implies. Fluxonium superconducting qubits are particularly well suited for analog control. Such architecture also opens the possibility of multimodal operation, enabling transitions between digital and analog control regimes within the same hardware platform, opening the door to the execution of digital-analog algorithms.

Digital quantum computers are the better choice for other use cases. Structured mathematical workloads that require precise sequences of gates, such as cryptographic problems, database search, or large-scale linear algebra, are some examples.

The best of both worlds is having access to digital and analog modalities.

The Digital-Analog Quantum Computing (DAQC) paradigm captures the advantage of both modalities within the same chip by executing algorithms that benefit from analog evolution and digital gates. The digital layer performs the preparation of data-encoding states and their measurement, while the analog layer acts as a rich reservoir providing the complexity and nonlinearities that allow the system to learn. Quantum AI is highly likely to take advantage of DAQC, especially Quantum Reservoirs and Quantum Extreme Learning Machines algorithms.

Quantum AI workloads will benefit from the integration of quantum computers in data centers and high-performance computing facilities, especially in Multimodal Quantum Data Centers, in which analog and digital quantum computers and classical CPUs/GPUs act as one orchestrated resource.

The complete AI workflow includes pre and post-processing steps involving GPU and CPU tasks, including data loading, feature extraction, mapping classical features into parameters for quantum embedding, data aggregation and visualization, among others. Another use case involving HPC-QC integration is drug design: reaction mechanisms between a drug candidate and the target can be encoded into a Hamiltonian to accurately define the molecular orbitals involved, whereas the environment can be approximated using graphical processing.

Multimodality can be accessed either on-premise, by deploying quantum devices in HPC environments, or through a Quantum-as-a-Service (QaaS) model. Quantum computing is postulated to push the limits of computation by providing acceleration to conventional data centers.

Achieving practical advantage, however, will depend not only on advances in quantum hardware but also on the effective orchestration of heterogeneous resources within hybrid computing environments. In this context, multimodal approaches that combine classical HPC resources with both analog and digital quantum processors are emerging as a key paradigm.