When quantum actually matters (rarely)¶

Quantum computing has been proposed as a solution to virtually every computational problem known to humanity, plus several that were invented specifically to justify quantum computing research. The reality is considerably more modest. Quantum computers excel at a small number of specific problem types and are comprehensively outperformed by classical computers for nearly everything else. Determining whether your problem actually benefits from quantum computing requires understanding what quantum computers are genuinely good at, which is different from what the marketing materials claim they’re good at.

The genuine quantum applications share certain characteristics. They involve quantum mechanical systems that classical computers struggle to simulate, or they have mathematical structure that quantum algorithms can exploit more efficiently than classical ones, or they require sampling from probability distributions with complex quantum structure. Problems lacking these characteristics are better solved classically, regardless of how quantum-sounding the problem description can be made to appear in funding applications.

This is not a failing of quantum computers. It’s the reality of specialised tools. You wouldn’t use a scalpel to chop vegetables or a chainsaw for neurosurgery. Quantum computers are precision instruments for specific tasks, not general-purpose replacements for classical computing. Understanding when they matter requires distinguishing between genuine quantum advantages and expensive experiments in applying unsuitable tools to problems that don’t need them.

Drug discovery or simulating molecules¶

Quantum chemistry is the most credible near-term application for quantum computers because molecules are quantum mechanical systems and quantum computers can simulate quantum mechanics naturally. Classical computers can simulate small molecules but struggle as molecular complexity increases because the quantum state space grows exponentially with the number of particles. A quantum computer’s state space also grows exponentially, but it grows in the right way to match the quantum system being simulated.

Current drug discovery relies heavily on computational chemistry to predict how potential drug molecules will interact with target proteins. These simulations help narrow the search space before expensive laboratory testing. Classical methods work well for many molecules but become intractable for complex systems involving many electrons, strong correlations, or exotic quantum effects. This is where quantum computers might eventually help.

Algorithms like VQE can calculate molecular ground state energies for small molecules on current noisy quantum hardware. These calculations don’t yet outperform classical methods but demonstrate the approach. As quantum computers scale to larger qubit counts with lower error rates, they should be able to simulate larger molecules that classical computers struggle with. This could accelerate drug discovery by enabling simulations of complex molecular interactions that are currently too difficult to model accurately.

The timeline is uncertain. Current quantum computers can simulate simple molecules with a handful of atoms. Practically useful drug molecules have dozens or hundreds of atoms. Bridging this gap requires quantum computers with thousands of error-corrected qubits performing millions of gate operations reliably. The field is progressing but remains years or decades from routine drug discovery applications.

Even when quantum computers become capable of useful molecular simulations, they won’t replace classical computational chemistry entirely. They’ll complement it, handling the specific problems where quantum effects dominate and classical approximations fail. Most molecular simulations will remain classical because classical methods are faster, cheaper, and adequate for most purposes. Quantum computers will be specialist consultants called in for difficult cases, not general practitioners handling routine work.

Several pharmaceutical companies are investing in quantum computing research, partnering with quantum hardware providers to explore applications. This is sensible preparation for eventual quantum advantage, but current projects are research collaborations rather than production drug discovery. The potential is real but remains potential rather than deployed capability.

Optimisation problems or logistics and scheduling¶

Optimisation is finding the best solution from an enormous number of possibilities. Logistics companies need optimal delivery routes. Airlines need optimal crew scheduling. Manufacturers need optimal resource allocation. These problems are combinatorially hard, meaning the number of possible solutions grows exponentially with problem size, and finding the absolute best solution is often computationally intractable.

Quantum algorithms like QAOA and quantum annealing claim to handle optimisation problems more efficiently than classical approaches. The theory suggests quantum computers can explore solution spaces more effectively through quantum superposition and interference, potentially finding better solutions faster than classical algorithms. This would have enormous practical value because optimisation problems are ubiquitous in industry.

The practical reality is mixed. D-Wave quantum annealers have been applied to various optimisation problems with results ranging from “slightly better than classical” to “considerably worse than classical” depending on problem structure and classical algorithm choice. Academic studies comparing quantum annealing with state-of-the-art classical optimisation generally find that classical methods remain superior, though quantum annealing shows promise for specific problem types.

QAOA is more theoretically appealing but requires gate-based quantum computers with lower error rates than currently available. Small-scale demonstrations show that QAOA can find reasonable approximate solutions to toy optimisation problems, but scaling to realistic problem sizes while maintaining quantum advantage remains unproven.

The challenge is that classical optimisation algorithms have received decades of development. Techniques like simulated annealing, genetic algorithms, branch and bound, and sophisticated heuristics perform extraordinarily well on practical optimisation problems. Quantum algorithms need to outperform these mature classical methods by enough margin to justify quantum hardware costs, and this has proven difficult.

There’s also a question of problem encoding. Real optimisation problems rarely arrive in forms quantum computers can process directly. They require translation into quantum-friendly representations, which adds overhead and may eliminate potential quantum advantages. A logistics problem involving trucks, drivers, time windows, and fuel costs must be reduced to a mathematical optimisation problem, then encoded into quantum states, then decoded back into actionable logistics decisions. Each translation step adds complexity and potential for quantum advantages to disappear.

The path forward probably involves hybrid quantum-classical approaches where quantum computers handle specific optimisation subroutines within larger classical optimisation frameworks. This is less revolutionary than pure quantum optimisation but more likely to provide practical benefits once quantum hardware improves sufficiently.

Financial modelling or portfolio optimisation¶

Financial institutions are interested in quantum computing for portfolio optimisation, risk analysis, option pricing, and fraud detection. These problems involve optimisation, sampling from complex probability distributions, and processing large datasets, all of which quantum computing claims to improve.

Portfolio optimisation means selecting investments to maximise returns while managing risk, subject to various constraints. This is mathematically an optimisation problem that becomes computationally expensive for large portfolios with many assets and complex constraints. Quantum algorithms could theoretically explore portfolio configurations more efficiently than classical methods.

Monte Carlo simulation, used extensively in finance for pricing derivatives and assessing risk, requires sampling from probability distributions. Quantum algorithms for Monte Carlo simulation promise quadratic speedups over classical sampling, which would be valuable given how much computation financial institutions dedicate to Monte Carlo methods.

The practical demonstrations have been limited. Financial institutions have partnered with quantum computing companies to explore applications, but published results generally show quantum methods matching classical performance on small test problems rather than achieving significant advantages. The quantum algorithms work in principle but don’t yet outperform classical methods on realistic problems with realistic quantum hardware.

Quantum machine learning for fraud detection and pattern recognition in financial data remains largely speculative. Classical machine learning works excellently for these tasks, and quantum ML has not demonstrated advantages. The problems don’t have obvious quantum structure that quantum algorithms could exploit.

The financial sector’s interest in quantum computing is partly genuine exploration of potential advantages and partly hedging against competitors gaining quantum advantages first. If quantum computers eventually provide benefits for financial modelling, institutions want to be positioned to exploit them. Until then, the investments are modest compared to spending on classical computing infrastructure.

One significant application might be post-quantum cryptography. Financial communications rely heavily on public-key cryptography that quantum computers could eventually break. Transitioning to quantum-resistant cryptography is essential regardless of whether quantum computers provide financial modelling advantages. This is less exciting than quantum-accelerated trading algorithms but considerably more important for operational security.

Machine learning or mostly still experimental¶

Quantum machine learning has received enormous attention from researchers and disproportionate hype from the quantum computing industry. The hope is that quantum computers could accelerate machine learning by exploiting quantum effects for training, inference, or both. The reality is that classical machine learning is already spectacular and quantum ML has not demonstrated practical advantages on any real-world problems.

Quantum algorithms for ML include quantum versions of principal component analysis, support vector machines, neural networks, and various other classical ML techniques. These quantum algorithms have theoretical speedups over their classical counterparts under specific assumptions about data access, problem structure, and quantum hardware capabilities. Translating these theoretical speedups into practical advantages has proven difficult.

The data loading problem is fundamental. Quantum algorithms operate on quantum states, but most ML datasets exist as classical data. Loading classical data into quantum states requires time and quantum operations that often eliminate theoretical speedups. If you have a million classical data points and need to encode them into quantum states for processing, the encoding time alone might exceed what classical ML would take to process the data directly.

Quantum neural networks, where quantum circuits replace classical neural network layers, have been explored extensively. Small-scale experiments show they can learn simple patterns, but scaling to realistic problem sizes reveals that classical neural networks trained on GPUs are vastly more efficient. The quantum advantage disappears once you account for the overhead of quantum operations, error rates, and limited qubit counts.

Quantum kernel methods, where quantum computers calculate similarity measures between data points for use in classical ML algorithms, are among the more promising approaches. The quantum computer handles one specific subroutine within a largely classical ML pipeline. This is less ambitious than fully quantum ML but might actually provide benefits once quantum hardware improves.

The most credible quantum ML applications involve data that’s already quantum, such as analysis of quantum sensor outputs or processing quantum communication data. If your data originates from quantum systems, keeping it in quantum form and processing it quantumly might offer advantages. For classical data from classical sources, which is nearly all ML data, classical ML remains superior.

Research continues because the theoretical potential exists and because understanding quantum ML’s fundamental limitations requires exploring them practically. This is valuable scientific work. It’s not yet, and may never be, a practical replacement for classical machine learning that works brilliantly on CPUs and GPUs at scales quantum computers cannot approach.

What definitely doesn’t need quantum computing¶

Most computational problems don’t need quantum computers and won’t benefit from quantum computing even after quantum hardware improves dramatically. These problems should be solved classically because classical computers are better at them.

Customer relationship management systems don’t need quantum computing. Neither do content management systems, e-commerce platforms, social networks, video streaming services, or virtually any conventional web application. These systems process classical data with well-understood classical algorithms on reliable classical hardware. Adding quantum computing would add expense, complexity, and failure modes without providing any benefits.

Data processing pipelines, ETL workflows, data warehousing, and business intelligence systems are classical problems solved excellently by classical databases and data processing frameworks. Quantum computers don’t accelerate SQL queries, don’t improve data compression, and don’t make report generation faster.

Image recognition, natural language processing, recommendation systems, and most practical machine learning applications work superbly on classical hardware using classical algorithms. Neural networks train efficiently on GPUs, inference runs quickly on CPUs or specialised AI accelerators, and the entire software ecosystem is mature and well-supported. Quantum ML might eventually find niche applications, but suggesting quantum computers for standard ML tasks is either ignorance or salesmanship.

Accounting software, inventory management, payroll processing, document management, project management tools, and enterprise resource planning systems are thoroughly classical applications that have no connection to quantum computing. If someone proposes quantum enhancements to your accounting system, they are either confused about what quantum computers do or they’re selling expensive consulting services.

Scientific simulations that don’t involve quantum mechanics are generally classical problems. Fluid dynamics, weather forecasting, structural engineering, and most computational physics solve classical differential equations that classical computers handle effectively. Quantum computers might eventually help with specific aspects of some simulations, but the bulk of computational science will remain resolutely classical.

Web browsers, operating systems, databases, compilers, text editors, and essentially all systems software are classical programs running on classical computers for classical purposes. They will never need quantum computing because they’re not doing anything quantum. Suggesting quantum versions of these tools indicates fundamental misunderstanding of both the tools and quantum computing.

The pattern is clear. If your problem is well-solved by existing classical computers, adding quantum computing makes things worse not better. Quantum computers are specialised tools for problems with specific mathematical or physical structure that quantum algorithms exploit. Everything else remains classical, and classical computers will continue getting better at classical problems faster than quantum computers become practical alternatives.

Knowing when quantum matters¶

The heuristic is straightforward. Quantum computers matter when you’re simulating quantum systems, solving specific mathematical problems like factoring or discrete logarithms where quantum algorithms have provable advantages, or exploring quantum phenomena for research purposes. For everything else, classical computers are better and will remain better.

If your problem can be formulated as a quantum mechanical system evolving according to Schrödinger’s equation, quantum simulation might eventually help. If your problem has exponential structure matching what quantum algorithms exploit naturally, quantum computing might eventually provide advantages. If your problem is already solved adequately by classical computers, quantum computing is an expensive distraction.

The current state is that quantum computers are research instruments for exploring quantum computing, not production tools for solving business problems. In ten or twenty years, they might become useful for drug discovery, materials science, cryptography, and perhaps selected optimisation problems. They will not become general-purpose classical computer replacements.

Businesses exploring quantum computing should focus on understanding the technology, preparing for eventual post-quantum cryptography migration, and identifying whether they have problems in the narrow categories where quantum advantage might materialise. Most businesses will conclude that quantum computing is fascinating but irrelevant to their actual computational needs. This is the correct conclusion and should be reached quickly before spending substantial money on quantum experiments that were never going to provide value.

The excitement around quantum computing is partly justified by genuine scientific progress and partly generated by vendors, consultants, and researchers whose funding depends on maintaining enthusiasm. Separating genuine opportunity from manufactured hype requires scepticism, technical understanding, and willingness to conclude that quantum computing, while scientifically important, probably doesn’t matter for your specific problems. This conclusion disappoints nobody except those selling quantum computing services, which is exactly who you should disappoint if it’s the correct conclusion.