A conventional computer stores and processes everything as bits, each of which is either 0 or 1. A quantum computer, as explained by CSC’s experts (CSC 2026a), uses quantum bits (qubits), which differ from bits in several ways. Two of their key properties are superposition and entanglement. Superposition means that during a calculation, a qubit can be in a combination of 0 and 1. It settles into one definite value only when it is measured. 

Entanglement means that qubits can be linked so that their values depend on one another. This is what lets them describe systems whose parts cannot be treated separately. A quantum method uses these properties to arrange the calculation so that wrong answers cancel each other out and the right answer is left when the qubits are measured. This works only for problems with the right mathematical structure (CSC n.d.; CSC 2026a).

The clearest example of where this helps is a molecule. The electrons in a molecule are themselves in superposition and entangled with one another. The number of their possible arrangements grows so fast with the size of the molecule that a few dozen electrons are already too many for any supercomputer to describe exactly. 

Qubits behave in the same way as the electrons they describe, so a quantum computer can represent the molecule directly instead of approximating it. This is why the most promising uses lie where the question is how electrons arrange themselves: finding new materials for batteries, catalysts for the chemical industry, and candidate molecules for medicines (CSC 2026a).

Where a quantum computer is faster, the gain is called a quantum speed-up. As Hoefler et al. (2023) explain, it belongs to a particular method for a particular problem rather than to the machine as a whole. As the problem grows, the quantum method needs far fewer steps than the best known conventional method. For a few problems, the difference is so large that cases too big for any conventional computer come within reach. 

The simulation of molecules and materials is the main example. For many other problems, the known gain is modest. Because each step on an error-corrected quantum computer is far slower than on a conventional chip, a modest gain is unlikely to pay off in practice. A quantum computer also reads and writes data very slowly, so it suits problems with a small input and a very large number of calculations (Hoefler et al. 2023).

Optimization and quantum-enhanced AI are both named as aims for one of Finland’s next quantum computers, LUMI-IQ (CSC 2026b). They remain research goals: no clear practical benefit, such as a faster or cheaper result than a conventional computer gives, has yet been demonstrated. The companies that build the machines describe a wider field. In his VERGE keynote (Epicenter 2026), Jouni Flyktman of IQM estimated that at least half of the workloads now run on supercomputers could be moved to quantum computers and run faster. 

This goes well beyond what Hoefler et al. consider realistic, although he also said that more problem classes in which value can be shown are still needed. At the same event, Costantino Carugno of VTT gave a narrower rule: “quantum becomes useful when the bottleneck is model representation”. If more computing power were all that was needed, one would build a larger supercomputer instead (Epicenter 2026).