Quantum Deep Learning Still Needs a Quantum Leap

TL;DR

The paper investigates whether quantum computing can meaningfully accelerate deep learning in the near to mid-term. It surveys a broad set of quantum algorithms and their fit to deep learning tasks, constructs a Quantum Economic Advantage model to forecast practical viability, and identifies three core bottlenecks: limited practical gains in standard linear-algebra subroutines due to hardware overhead, underdeveloped QRAM essential for loading classical data, and large theoretical advantages restricted to narrow, specialized cases. Across data preprocessing, hyperparameter optimization, linear algebra subroutines, and quantum neural networks, the authors show that current quantum speedups are often swamped by real-world overheads, I/O costs, and data-transfer bottlenecks. They conclude that without a quantum leap—especially in QRAM practicality and hardware efficiency—quantum deep learning is unlikely to deliver widespread practical benefits within the next 10–20 years, though targeted breakthroughs could still yield meaningful, if narrow, advantages. The work also outlines research directions and robustness checks to help steer the field toward genuinely impactful advances while acknowledging alternative hardware and algorithmic pathways that could alter the landscape.

Abstract

Quantum computing technology is advancing rapidly. Yet, even accounting for these trends, a quantum leap would be needed for quantum computers to meaningfully impact deep learning over the coming decade or two. We arrive at this conclusion based on a first-of-its-kind survey of quantum algorithms and how they match potential deep learning applications. This survey reveals three important areas where quantum computing could potentially accelerate deep learning, each of which faces a challenging roadblock to realizing its potential. First, quantum algorithms for matrix multiplication and other algorithms central to deep learning offer small theoretical improvements in the number of operations needed, but this advantage is overwhelmed on practical problem sizes by how slowly quantum computers do each operation. Second, some promising quantum algorithms depend on practical Quantum Random Access Memory (QRAM), which is underdeveloped. Finally, there are quantum algorithms that offer large theoretical advantages, but which are only applicable to special cases, limiting their practical benefits. In each of these areas, we support our arguments using quantitative forecasts of quantum advantage that build on the work by Choi et al. [2023] as well as new research on limitations and quantum hardware trends. Our analysis outlines the current scope of quantum deep learning and points to research directions that could lead to greater practical advances in the field.

Figures (9)

• Figure 1: Quantum Economic Advantage (QEA) Problem Size in a given year. Quantum computers come with large constant hardware overheads but algorithmic advantages. Therefore, problem sizes must be over a critical threshold to have an advantage on a quantum computer.
• Figure 2: QEA threshold problem size over time taking into account hardware trends in quantum-classical overhead. The green line represents the maximum theoretical size that could be achieved without any time limit. We also choose to include a time limit of 1-month to better model constraints in machine learning. The quantum advantage region illustrates problem sizes that are feasible and preferable to run on a quantum computer in comparison to a given classical algorithm.
• Figure 3: Three scenarios for quantum machine learning. Problems below the blue feasible line and above the orange economic advantage (QEA) line are likely to be impacted by quantum computing. Fig \ref{['fig:the_triptic']}a illustrates the case where a quantum algorithm has an exponential speedup over a linear classical algorithm. Fig \ref{['fig:the_triptic']}b captures the problem size necessary to see advantage with Grover's speedup. Fig \ref{['fig:the_triptic']}c shows the quantum advantage diagram for matrix multiplication, which demonstrates that while potentially feasible (with QRAM), it will not be advantageous on a quantum computer for the near future.
• Figure 4: Graph of log number of physical qubits vs time for superconducting systems. Our default model is based on the 90th percentile quantile fit as this is reflective of mainstream providers like IBM. Data sourced from ruane2025quantum.
• Figure 5: Trends in 2-qubit gate times for superconducting quantum computers. Data from ruane2025quantum.