As we are familiar with it, computation is nearing its constraints. The increase in the demand for computing infrastructure is due to two factors. Firstly, it is necessary to solve increasingly complex problems with better precision and within practical timeframes.
Secondly, the surge of AI has significantly increased this demand. Major companies like Amazon, Google, or Meta are entering into agreements with energy companies to acquire Small Modular Reactors (SMR) for powering the high demands of their AI data centers. At present, in the United States, 4% of the overall electricity is utilized for AI training, with a projected increase to 12% by 2028.
If our method of computation remains unchanged, the demand for electricity in AI training is predicted to surge exponentially over the next few years. The era of sustainable computation is upon us, as quantum computing emerges to accelerate compute tasks by revolutionizing the algorithm’s physical and mathematical framework, much like GPUs did in 2007 with CUDA and the introduction of calculus-specific graphics cards..
Quantum computers are being integrated into data centers and high-performance computing environments, adhering to the well-established model of heterogeneous computing. By incorporating quantum processors into the workload-management process, they can be seen as extra accelerators within the HPC system, allowing for mixed workflows and coordination techniques similar to those used for GPUs and other specialized equipment.
Preliminary instances of this methodology are currently being investigated at the Barcelona Supercomputing Center (BSC-CNS), where quantum systems based on superconductivity have been utilized alongside the MareNostrum-5 supercomputing network to facilitate hybrid classical-quantum workflows. Quantum computing manifests itself in various ways, contingent on the physical components used to create the qubits and their operational design.
In the realm of emerging quantum computing technologies, superconducting-based analog quantum computers hold substantial promise. A significant advantage of these systems is their capability to incorporate the issue into the quantum chip’s degrees of freedom through a one-to-one correspondence, which also exhibits greater resilience to faults.
This method enables the system to mimic the inherent characteristics of the desired issue by constantly adjusting the pertinent parameters using analog control. As a result, analog quantum computing is particularly well-suited for running algorithms like quantum reservoirs, material simulations, chemistry simulations, or industrial combinatorial optimization problems, potentially surpassing digital computing by circumventing the need for discretizing continuous processes and the associated errors.
