Blog · October 11, 2026

When will quantum computers help make medicines?

The chemistry calculations that could help with drug design need about 1,000 to 1,500 error-corrected qubits. The largest demonstrations have about 70. What's been done so far, and where quantum computing would fit in making a drug.

A heme molecule drawn in white dots: a flat ring of carbon and nitrogen atoms around a central iron atom, with an oxygen above and a sulfur below.

Not soon, and when they do, it will be for a narrow part of the work. The calculations people hope to run need about 1,000 to 1,500 error-corrected qubits working for hours. The largest error-corrected demonstrations so far have about 70, and no quantum computer has yet done a useful chemistry calculation better than an ordinary computer can. No medicine has been developed with one.

What a quantum computer would calculate

How a drug behaves depends partly on how electrons arrange themselves in its molecules and in the proteins it touches. Those electrons follow quantum rules, and the exact calculation grows so fast with the number of electrons that ordinary computers use approximations. For most molecules the approximations work well. For some, especially ones built around metal atoms like iron, they break down.

A quantum computer could, in principle, do these calculations without the shortcuts, because its qubits follow the same rules as the electrons. The standard test case for drugs is cytochrome P450, a family of liver enzymes with an iron atom at the centre that breaks down many common drugs. Predicting exactly how they act on a new molecule is still hard for ordinary computers.

Needed and built

Estimates are counted in logical qubits, the reliable qubits that error correction builds out of many physical ones.

Needed: P450 drug enzyme, simplified model

about 1,150 logical qubits, roughly a billion operations · Google, 2025

Needed: FeMoco, the enzyme that makes fertilizer in nature

about 1,100 to 1,500 logical qubits, hours of running · Google, 2025

Achieved: largest logical-qubit demonstration (not chemistry)

about 70 logical qubits · IBM and University of Chicago, 2026

Achieved: logical qubits used in a chemistry calculation

2 logical qubits, on an iron catalyst · Microsoft and Quantinuum, 2024

Filled: needed. Dashed: achieved. Full width is 1,500 logical qubits.

On today’s standard designs, those 1,000 or so logical qubits take about 3 to 5 million physical qubits. The largest machine that runs full computations has 1,121. Hardware companies with newer designs claim far fewer: Alice & Bob says about 99,000 of its “cat” qubits, but that hardware hasn’t been built at scale and the estimate hasn’t been peer-reviewed.

Both sides keep improving

The estimates have come down a long way. In 2017, the first detailed estimate for FeMoco needed around a million billion operations. By 2025, better methods had cut that to a few hundred million, roughly a million times fewer. The number of physical qubits has stayed in the millions, because that is set by how error correction works.

Ordinary computers have improved too. In January 2026, a team led by Garnet Chan at Caltech calculated the standard FeMoco model to the accuracy chemists need, using classical methods alone. FeMoco had been the field’s main example of a problem only a quantum computer could solve. A 2023 study by researchers at Caltech, Google and elsewhere found no evidence yet of an exponential quantum advantage across chemistry in general. P450 is still a harder case: one standard classical method fails on it, and another becomes impractical as the model grows. It remains a real target.

What has been done on real hardware

Each of these is a genuine result. None has yet beaten an ordinary computer on a useful chemistry problem.

  • Microsoft and Quantinuum, 2024. Used 2 error-corrected qubits to calculate part of an iron catalyst’s chemistry accurately. The companies noted that an ordinary computer can get the same answer, and that a useful advantage would need around 100 logical qubits.
  • IBM and RIKEN, 2024 to 2025. Combined a 77-qubit chip with the Fugaku supercomputer to study iron-sulfur clusters found in proteins. The results matched good classical methods without surpassing them.
  • IonQ with AstraZeneca, 2025. Reported a twentyfold speedup on a step of a drug-making reaction. The comparison was with IonQ’s own earlier runs, not with ordinary computers.
  • Google, 2025. Reported a calculation 13,000 times faster than a supercomputer. That was a physics measurement. The molecule experiment in the same work was a proof of principle with no speed claim.

Where it would fit in making a drug

A new medicine takes 10 to 15 years from discovery to approval, and the median cost of bringing one to market in the US was about $1 billion in a 2020 study. Around 90% of drugs that enter human trials fail, mostly because they don’t work well enough in people or cause harm.

A quantum computer would help with one kind of question: exactly how a molecule interacts at the level of its electrons. That can help when choosing between candidate molecules, predicting how the liver will break one down, or designing how to manufacture it. It doesn’t predict how a drug behaves in a whole human body, which is where most candidates fail.

What to make of it

Chemistry is still the most credible long-term use for quantum computers, because molecules follow the same rules qubits do. Drug discovery is one part of that, not the main one. The best-studied targets are catalysts and materials, such as FeMoco for fertilizer and ruthenium compounds for capturing carbon dioxide, so the first useful results are more likely there.

When a medicine eventually benefits, the quantum calculation will be one step among hundreds, run alongside ordinary computers. A drug made by a quantum computer is not likely to be how it happens.

Sources