Will quantum computers revolutionize drug discovery?
No, not yet, and "revolutionize" oversells it. As of August 2026, real quantum hardware has only simulated molecules of a few atoms. Any laptop does those sums exactly in milliseconds. Truly useful chemistry needs error-corrected machines that exist only on company roadmaps. Even then, quantum computers would speed up just one step: working out how electrons behave in hard molecules. Most of the slow parts of making a drug are elsewhere.
Yes in principle — its cost has been estimated since 2017 Quantum phase estimation can in principle compute the lowest energy (the ground-state energy) of molecules whose electrons are strongly linked, to chemical accuracy. For some of these molecules, ordinary methods fail. Reiher and colleagues made the first serious cost estimate for a molecule of real industrial interest. It is a costed theory, not a demonstration: the machine it needs does not exist.
Yes — for molecules of a few atoms only Experiments using the variational quantum eigensolver (VQE) have computed the lowest energies of textbook molecules on real devices. These are engineering milestones. An exact calculation on a laptop gives the same numbers in milliseconds. No molecule beyond the reach of ordinary computers has ever been simulated on quantum hardware.
Disputed — ordinary chemistry methods keep moving the goalposts A study by Google, QSimulate and Boehringer Ingelheim asked where the line between ordinary and quantum advantage really sits for a real drug-related enzyme. The honest finding: today's ordinary methods handle it. And the point where quantum would win keeps moving further away as ordinary algorithms improve. Do drug-relevant molecules that are hard for ordinary computers but cheap for quantum ones exist in useful numbers? That is an open, contested question.
"Revolutionize the pipeline"? Disputed even by quantum researchers A 2024 Nature Physics perspective, co-written by quantum-industry researchers (Google, Boehringer Ingelheim, universities), sets out where quantum computers could plausibly help. The list is narrow: specific electron-structure sub-problems, such as metal centres and covalent inhibitors. Most of drug discovery is not electron structure. Finding targets, ADMET (how the body absorbs, spreads, breaks down, clears and reacts to a drug), formulation and clinical trials gain nothing from a quantum computer.
Useful machines promised from about 2029 — treat as a claim IBM's fault-tolerance roadmap is the most concrete public plan. Even its 2029 target falls short of published drug-chemistry cost estimates, unless algorithms improve further. Every company's roadmap dates have slipped before. Until logical qubits are shown at scale, dates are claims, not capabilities.
What could a quantum computer actually do for drug discovery?
Why care? Making a new medicine takes many years and a lot of money. If computers could predict how a drug molecule behaves, fewer lab tests would fail. So people ask whether quantum computers could help.
They could help with one specific job: electronic structure. That means working out where a molecule's electrons sit and how much energy they have. Electrons decide how atoms bond and react. In some molecules the electrons are strongly correlated, which means each one's behaviour depends tightly on all the others. Think of a crowded dance floor where every dancer reacts to every other dancer. Unlike dancers, electrons have no fixed spots. The math tracks the chance of every possible arrangement. Ordinary methods either fail on these molecules or cost too much.
For most drug-like molecules, ordinary methods (DFT, coupled cluster, DMRG) are already accurate enough. The interesting cases are a minority. They include metal atoms at the heart of enzymes (transition-metal active sites), bonds breaking, and some light-driven chemistry.
The quantum method is quantum phase estimation. It stores the molecule's electrons in qubits. Then it lets them change under the molecule's own energy rules, called its Hamiltonian. Finally it reads out the lowest possible energy, the ground-state energy, from an interference pattern between amplitudes. An amplitude is a number that says how strongly a qubit leans toward each result.
There is no "trying all arrangements at once". The gain is storage. The full description of the electrons, the wavefunction, grows exponentially with molecule size. The number of qubits grows only polynomially, which is far slower. Here is the scale: writing down the state of just 50 qubits on a normal computer takes 2^50 numbers. That is about a million billion. Interference then pulls out one number from all that: the energy.
Everything published so far says this needs error-corrected machines. One useful energy calculation is estimated at around 10^9–10^10 error-corrected gate operations. That is one to ten billion steps with no uncorrected mistake. See the assessments above and why error correction changes the arithmetic.
What can today's hardware do?
Molecules of a few atoms, with visible error bars. The best results on real devices are VQE-style calculations of H2, LiH, BeH2 and molecules of a similar size. VQE, the variational quantum eigensolver, works like this: a quantum circuit makes a guess, and an ordinary computer tunes it, over and over. These runs use a handful of qubits. They also need error mitigation, clean-up tricks applied after the run. Even then, they only match numbers an ordinary computer gets exactly and instantly.
Drug companies do run quantum pilot projects. Papers sometimes report "quantum-enhanced" screening hits. Read those carefully. In every published case so far, the quantum processor handled a small piece that an ordinary method could have handled as well or better. That is learning by doing, like a student driver in an empty parking lot. It is practice, not capability.
You can feel the real scale yourself. Build a two-qubit VQE ansatz, a starting circuit with adjustable dials, in the lab. Then compare it with what current QPUs offer.
What would have to change for the answer to change?
Three things, all of them, in order:
- Fault tolerance at scale. Fault tolerance means the machine fixes its own errors as it runs. The need is hundreds of logical qubits running billions of steps. A logical qubit is one reliable qubit built from many error-prone ones. Company roadmaps put this at 2029 and beyond. None has shown it.
- Ordinary chemistry methods standing still. They haven't. Each time quantum cost estimates improve, classical methods (DMRG, machine-learned force fields, neural wavefunctions) improve too. The cytochrome P450 study shows how far they already reach.
- A proven win on a real molecule. This means one checked case where a quantum computer gives a chemically useful number that the best ordinary method could not. That would change this page overnight. It has not happened.
Even if all three happen, the likely result is "a powerful new tool for hard electron problems". It is not a revolutionized pipeline. Related: can quantum computers fold proteins?
Classifications follow the QPU137 editorial policy: every applied label carries a date, source, scale, and hardware, or it does not render. Found an error? Report it.