Quantum technologies and AI: what is realistic today, and what remains the future

Discussion about quantum computing and AI often oscillates between two extremes: utopian proclamations of revolution and outright dismissal as hype without substance. The OECD.ai platform has launched a three-part analysis attempting to find a middle path — and the first episode, authored by industry practitioners, draws a useful distinction between what works today and what remains a research goal.

The authors are Victor Gaspar, Director of Sales at Multiverse Computing — a Spanish startup specializing in quantum-inspired optimization — and Katsuyuki Hanai of Toshiba Corporation, where he leads the quantum communications business unit. An important note: the text reflects the views of the authors as industry practitioners, not a regulatory or research position of the OECD.

What is realistic and available today?

The most concrete finding of the analysis concerns tensor networks — a mathematical technique originally developed in quantum systems physics, but applicable exclusively on classical hardware. Tensor networks compress the representation of large language models by removing redundant parameters in a structured way that preserves key relationships in the data.

The result the experts cite is significant: a reduction in memory and computational requirements of 10 to 100 times compared to uncompressed models, with a moderate accuracy loss that can be recovered through brief fine-tuning. A critical distinction: this does not require quantum hardware, does not require access to quantum computers in the cloud, and is not a theoretical construct — according to the authors, startups and technology providers have already embedded tensor network techniques into commercial products.

Quantum sensors are the second category with proven near-term applications. Unlike quantum computers, which require extreme cooling and isolation from the environment, quantum sensors can operate in real-world conditions and already achieve levels of measurement precision unattainable by classical devices — measuring magnetic fields, temperature, chemical composition, or mechanical vibration. Applications range from medical diagnostics to precision agriculture and infrastructure monitoring.

Where assessment becomes critical

Quantum Machine Learning (QML) — the application of quantum algorithms to training or inference of AI models — remains the most hyped but also most problematic segment. Gaspar and Hanai identify three systemic limitations blocking a breakthrough:

First, data transfer between classical and quantum systems remains slow. A quantum processor does not receive data like a classical chip; every data input requires quantum initialization that consumes precious coherence time. This bottleneck eliminates the advantage for most ML tasks that demand massive data throughput.

Second, advantage over classical algorithms has not been demonstrated for relevant ML tasks. While quantum algorithms show theoretical speedups for certain mathematical problems, optimized classical algorithms on modern GPUs and TPUs consistently keep pace with or outperform quantum alternatives in benchmarks. The authors explicitly state that broad commercial application of QML is not expected within the next decade.

Third, the hardware specifications required for a functional QML system remain undefined. How many qubits, how much coherence, how low gate error rates — without clear targets, there is no clear development roadmap.

The hybrid approach as a bridge

The most realistic near-term scenario the authors describe is hybrid quantum-classical systems: quantum processors handle narrow, well-defined optimization problems (e.g., chemical reaction optimization, molecular simulation), while classical AI systems manage everything else. Cloud platforms today offering access to quantum computers as an API lower the barrier to entry for developers who want to experiment with these hybrid architectures.

At the policy level, the authors warn of a shortage of specialized talent that understands both quantum physics and AI, and of the need for international cooperation in developing ethical frameworks for quantum AI systems — a topic that remains peripheral to regulatory discussions for now.

The analysis continues in the next two parts of the series on the OECD.ai platform.