Quantum Chemistry’s Honest Ledger: What the Resource Estimates Actually Say About Drug Discovery, Catalysis, and Materials Design

This article is part of my Quantum Utility Map Deep Dive series. It translates the algorithm-by-algorithm resource estimates from The Quantum Utility Ladder into business-relevant assessments for R&D leaders in pharmaceuticals, chemicals, and materials science.

The promise and the fine print

Quantum chemistry is the application that justifies the existence of fault-tolerant quantum computers. Unlike most quantum claims, this one follows from genuine mathematical hardness: simulating the quantum-mechanical behavior of molecules and materials requires computational resources that scale exponentially on classical computers but only polynomially on quantum ones. Every quantum computing roadmap, every investor pitch, every national strategy document cites chemistry as the first application that will deliver practical value.

I agree with that ranking. But it comes with fine print that most of those documents omit.

Exponential classical difficulty applies to exact simulation of strongly correlated electronic systems. In many industrially relevant chemistry problems, strong correlation is absent or confined to a small part of the molecule that classical methods can treat. Getting from “quantum computers can simulate chemistry” to “quantum computers will transform your R&D pipeline” means answering questions about the specific molecules, accuracy requirements, economic thresholds, and timelines involved.

This article works through those specifics. For each major quantum chemistry application, I assess the problem it solves, its actual resource requirements, when the hardware will exist, and its realistic economic impact. The numbers come from peer-reviewed resource estimates catalogued in The Quantum Utility Ladder. The business assessments are my own.

The FeMoco problem: five orders of magnitude and counting

No molecule has received more attention from quantum algorithm designers than the FeMo-cofactor of nitrogenase: the biological catalyst that fixes atmospheric nitrogen into ammonia at ambient temperature and pressure. Understanding how FeMoco works could revolutionize ammonia production, which consumes around 2% of total final energy consumption worldwide, or 8.6 exajoules on the IEA’s accounting, through the Haber-Bosch process (450°C, 200 atmospheres).

In 2017, Reiher et al. estimated ~111 logical qubits and 10¹⁴ to 10¹⁵ T-gates, for target accuracies of 1 and 0.1 millihartree. Assuming a logical T gate every 10 nanoseconds, they put the runs at 12 and 130 days; at one per microsecond, the same counts take about three years and 35 years. By 2021, Joonho Lee and colleagues had traded space for time, increasing the qubit requirement to 2,142 logical qubits but slashing the gate count to 5.3×10⁹ Toffoli gates using tensor hypercontraction: roughly four days at 4 million physical qubits. Rocca et al. (2024) halved that further. And in 2025, Low, Berry, Rubin et al. reported speedups of 4× to 195× over the previous state of the art for ground-state energy estimation, through spectrum amplification, while Caesura et al. (PsiQuantum and Boehringer Ingelheim) cut estimated quantum runtime by 278× using BLISS-THC factorization compiled for a photonic active-volume architecture. That factor is measured against prior quantum compilations on comparable fault-tolerant hardware, not against a classical computer, and the paper says so explicitly.

Five orders of magnitude in eight years. The algorithm stayed the same (quantum phase estimation on an electronic Hamiltonian). What changed was compilation strategy, Hamiltonian decomposition, and architectural matching.

Business assessment: FeMoco simulation at the 2021 tensor-hypercontraction level (2,142 logical qubits) is plausibly achievable on hardware projected for 2033–2035 (IBM Blue Jay targets 2,000 logical qubits). Understanding how nitrogenase fixes nitrogen does not directly produce a better industrial catalyst; it provides mechanistic insight that must then be translated through additional computational and experimental work into catalyst design. The value chain from quantum simulation to industrial impact includes several steps that quantum computing does not accelerate.

Nicole Holzmann maps those steps in an August 2026 perspective accepted by Philosophical Transactions of the Royal Society A, grouping what lies between a quantum computation and an industrial decision into six categories of consideration: the algorithmic level, system definition, embedding and interfacing, process complexity and resources, economic considerations, and decision relevance and validation. Each brings its own approximations, and nobody has traced a complete pathway through all six from a quantum-computed quantity to an industrial decision with the error propagation quantified at every stage. Her proposed first milestone for the field is that traced pathway. Her third, a systematic investigation of which modeling choices genuinely matter for decision quality, is the sharpest: if a choice does not change the final decision, effort spent improving it is wasted, whether the improvement is quantum or classical. For an R&D leader, that is the question to ask before funding a quantum pilot. Which decision changes if this number improves? How large must the improvement be before that decision flips?

The Bellonzi et al. study on homogeneous nitrogen fixation catalysts is the best available attempt to close this gap. It models the highest-utility calculation for this class of catalysts at approximately $200,000 of economic value, against a workload estimated at 139,000 QPU-hours and roughly 400,000 CPU-hours for the classical calculations. Dividing one by the other does not give an efficiency ratio: a QPU-hour and a CPU-hour measure unlike resources, with no common unit of cost, energy or elapsed time. Turning them into a factor requires a QPU price, a parallelism assumption and a repetition count that nobody has published. The $200,000 is also modeled utility for two cyanate-formation steps, not observed revenue. The paper offers a method for asking the question in the right units, which is more than most of this literature attempts. The claimed advantage is in accuracy rather than speed.

A vendor’s own timeline points later than mine. SoftBank and Quantinuum published a joint use-case study in July 2026 that indexes chemistry feasibility by spin orbitals instead of logical qubits. Under error correction they put the ceiling at roughly 10 spin orbitals on Helios today and at the same level for Sol in 2027. Apollo reaches about 100 in 2029. Only Lumos, in the 2030s, reaches 100-plus at chemical accuracy. FeMoco needs roughly 152 spin orbitals. Quantinuum has a commercial interest in the opposite answer and published the slower one anyway, which makes the document more interesting than its genre usually is. It does not make the dates more reliable: its authors call it an assumption-driven envelope rather than a forecast. FeMoco active spaces vary, so 152 spin orbitals is one model of the problem and not the definition of it.

Timeline: 2033–2036 for first industrially relevant FeMoco simulations. Impact: Mechanistic insight that accelerates catalyst design programs. Not a direct path to a new product, but a genuine R&D accelerator for organizations with the computational chemistry expertise to use the results.

Cytochrome P450: the drug metabolism bottleneck

The cytochrome P450 family of enzymes metabolizes the majority of marketed drugs, with the CYP3A4 isoform alone commonly estimated at around half. Predicting how a drug candidate will be metabolized (and whether it will produce toxic intermediates) is one of the most expensive steps in pharmaceutical development. Current computational methods (DFT, coupled cluster on small fragments) handle P450 poorly because the heme active site involves extreme spin-state fluctuations that require multi-reference quantum chemistry methods.

The figures most often quoted for P450 appear in one comparison table in Caesura et al., which collects its own result alongside earlier work, and the rows are not all the same authors’ configurations. Double factorization comes in at roughly 4,922 logical qubits and 1.9×10¹⁰ Toffolis. Tensor hypercontraction, the configuration Goings et al. (PNAS, 2022) took all the way to a surface-code compilation, is roughly 1,434 logical memory qubits and 7.8×10⁹ Toffolis, at 73 hours on about 4.6 million physical qubits. Caesura’s own BLISS-THC factorization reaches roughly 999 qubits at 1.7×10⁹ Toffolis. Its estimated runtime is 234× below the paper’s own THC reference run on a baseline architecture with comparable hardware assumptions. The factor combines 25.18× from compiling for the active-volume architecture, 8.23× from BLISS-THC and 1.12× from circuit improvements, as the paper’s Table I reports them. Every comparison in that table is quantum against quantum, with no classical baseline. And the usual rule that these methods trade width against gate count does not hold across the whole table: BLISS-THC improves on both the width and the gate count of the THC row, which is what makes it the interesting one.

Business assessment: This is one of the strongest business cases in quantum chemistry, because metabolism is one of the places drug candidates fail. A pharmaceutical company that could predict P450 metabolism accurately could drop candidates headed for metabolic failure before they reach the clinic. About nine in ten candidates that enter clinical trials fail. Drawing on analyses of trial data from 2010 to 2017, Sun and colleagues attribute 10–15% of those failures to poor drug-like properties, a category that includes metabolic stability, and about 30% to unmanageable toxicity, without saying how much of the toxicity traces to metabolites. The drug-like share was 30–40% in the 1990s. The authors credit its fall to rigorous selection criteria applied during drug optimization, and that screening is the baseline a quantum P450 calculation would have to improve on. I cannot quantify how much of either share a better active-site energy would prevent, and I am not going to put a figure on avoided trial costs without a causal model linking a computed energy to a clinical outcome.

The constraint is not qubit count. At the BLISS-THC figure of roughly 999 logical qubits, P450 fits inside a 2,000-logical-qubit machine with room to spare. The constraint is the gate budget and the physical footprint behind it. At one serial Toffoli per millisecond, 1.7×10⁹ Toffolis is about 19.7 days before any other circuit work, repetitions or architectural overhead; at 10 kHz it is under two days and at 1 MHz about 28 minutes. The real footprint problem is the other number: the Goings compilation that delivers 73 hours does so on 4.6 million physical qubits. The active space is the third constraint. The paper’s own extrapolation to a larger Compound I electronic-structure model at around 500 orbitals reaches roughly 9,000 logical qubits and on the order of 10¹² Toffolis, which is a different decade. That larger model is still the active-site system, not the whole protein with its environment and dynamics. Partial active spaces are an early-2030s application; the larger model is not.

Timeline: 2035–2038 for full P450 active-site simulation. Partial simulations (catalytic fragments, reduced active spaces) possible from 2031–2033. Impact: Potentially a large change in pharmaceutical R&D economics, through fewer clinical-trial failures on metabolism, if the computed energies change which candidates advance.

Battery materials: the degradation puzzle

Lithium-rich NMC cathode materials offer massive theoretical energy densities but suffer from voltage fade caused by structural degradation that current classical simulations cannot reliably explain. The degradation involves strongly correlated electronic states at transition-metal oxide interfaces, exactly the type of problem where quantum simulation has a genuine edge.

Initial resource estimates required more than 2,000 logical qubits and 10¹³ Toffoli gates. Recent work from Xanadu and Volkswagen put the XAS simulation at 100 workspace logical qubits for a Li₄Mn₂O cluster and under 4×10⁸ T gates for the longest single circuit. The paper’s own Table I gives the figure to keep next to it: sampling the full spectrum for that active space costs about 4.58×10¹¹ Toffolis, roughly 1,500 times the 3.11×10⁸ Toffolis of the longest single circuit. A February 2026 preprint costed RIXS spectra at roughly 414 logical qubits for a 20-orbital cluster active space, at 2.0×10¹⁰ Toffolis per shot. The width is within the logical-qubit targets for 2029–2031, but one shot needs two hundred times the 10⁸ operations IBM targets for 2029 and twenty times the billion it targets for 2033.

These are also three different calculations, not one calculation getting cheaper. A bulk cathode ground state, an XAS spectrum on a selected cluster and a RIXS spectrum have different outputs, active spaces and accuracy targets. Going from a 2,000-plus-qubit ground-state estimate to a 100-qubit XAS example is a change of task, and only the improvements inside each paper’s own baseline are algorithmic gains. Simulated spectra help researchers interpret the experiments that diagnose degradation, but they do not prevent it.

Business assessment: The economic stakes are large, given the size of the electric-vehicle and grid-storage markets, and the quantum application addresses the specific classical bottleneck. The widths, 100 to roughly 414 logical qubits, fit hardware projected for 2029–2031; the gate counts above put full spectra later. The risk is that classical AI-driven materials discovery methods may solve the degradation puzzle first; ML-based approaches to materials simulation are advancing rapidly and may identify degradation mechanisms through pattern recognition even without exact quantum simulation.

Timeline: 2029–2032 for first quantum simulations of battery degradation spectra on width; later for full spectra on gate budget. Impact: High-value if classical AI does not solve the problem first. The organization that identifies and validates solutions to Li-rich NMC voltage fade captures a multi-billion-dollar market.

Photosensitizers for cancer therapy

Photodynamic cancer therapy uses photosensitizer molecules that generate reactive oxygen species when exposed to light, destroying tumor tissue. Designing better photosensitizers requires accurate simulation of excited-state electronic structure, particularly the intersystem crossing from singlet to triplet states.

Xanadu researchers estimated that simulating BODIPY-derivative photosensitizers requires 180–350 logical qubits with Toffoli gate depths between 10⁷ and 10⁹. These are strikingly modest requirements.

Business assessment: The photosensitizer market is small compared to pharma broadly, which limits the total addressable value. The quantum application is genuine and the resource requirements are achievable on near-term hardware, but the commercial impact is niche. This application’s main value is as a proof of concept: it is a strong candidate for the first quantum-advantage drug design calculations, and the workflows and credibility it establishes will then transfer to higher-value applications.

Timeline: 2029–2031. Impact: Scientifically significant, commercially niche. Valuable as a pathfinder for broader pharmaceutical quantum applications.

The CO₂ catalyst: a climate application

Von Burg et al. estimated that a ruthenium catalyst for CO₂-to-methanol conversion requires approximately 4,000 logical qubits and ~10¹⁰ Toffoli gates, or 28 hours at 10 μs per gate. This is the kind of computation that could accelerate the design of CO₂-conversion catalysts.

Business assessment: Converting CO₂ to methanol is CO₂ utilization, which has a different addressable market from carbon capture. The 4,000-logical-qubit figure is also a 2021 estimate; Google’s compendium lists a later carbon-fixation configuration at roughly 900 logical qubits and 2×10⁸ Toffolis, so an argument that places this task beyond 2033 on the old width alone no longer holds. And as with FeMoco, the quantum simulation provides mechanistic insight that must be translated through additional steps into catalyst design. The value chain is real but long.

Timeline: uncertain, and I am no longer willing to bracket it from width. Impact: Moderate direct commercial value. Any climate benefit depends on the process energy, the hydrogen source and a lifecycle assessment, none of which the quantum calculation addresses.

The evidence for exponential quantum advantage in chemistry

Any honest assessment of quantum chemistry’s commercial potential must engage with the Lee et al. analysis published in Nature Communications in 2023, which concluded that evidence for exponential quantum advantage across generic chemical space has yet to be found. Seunghoon Lee is the first of its seventeen authors.

This is a strong result from a credible team, and it deserves serious attention. The key qualifier is “generic chemical space.” For the majority of chemical problems that the pharmaceutical and chemicals industries encounter daily, classical methods (DFT, coupled cluster, tensor network methods) work well enough. Quantum computing will not replace classical chemistry for routine calculations.

Quantum advantage remains viable in specific strongly correlated subsystems: transition-metal active sites, complexes of heavy metals, multireference excited states, extended conjugated systems with strong electron correlation. These are the exact systems catalogued in the Utility Ladder. The advantage is real but narrow: it applies to a minority of the computational chemistry calculations an R&D team runs, though that minority includes some of the highest-value calculations, where classical methods are least reliable.

These hybrid workflows are no longer theoretical. In May 2026, Cleveland Clinic, RIKEN, and IBM used a quantum-centric supercomputing workflow on a 12,635-atom protein-ligand complex, the largest heterogeneous quantum-classical electronic structure calculation to date. The atom count is the size of the embedded hybrid system, not of a quantum-simulated wavefunction. The approach uses wave function-based embedding to fragment the protein, with quantum processors handling the most strongly correlated clusters via Sample-based Quantum Diagonalization (SQD) and classical supercomputers handling everything else. The method does not yet outperform the best purely classical approaches, but the 40-fold scale increase over the prior result (achieved in just four months) demonstrates that the infrastructure for production-scale quantum chemistry workflows is being built now, alongside an IBM reference architecture for integrating quantum processors into HPC data centers.

The same architecture, applied to a different industry, shows how much of the work is still classical. In July 2026 IBM, Oak Ridge National Laboratory, and Cleveland Clinic announced the first quantum computation of fusion blanket chemistry, nine molecular configurations of FLiBe, the molten lithium beryllium fluoride salt that is the leading molten-salt candidate for breeding tritium inside a fusion reactor. An embedded wave function method carved each 21-to-23-atom cluster into one fragment per atom. Fragments below 13 orbitals went to classical computers, and only the larger ones went to an IBM Heron processor. The quantum fragments matched classical full configuration interaction to within 0.7 kcal/mol, inside the usual 1 kcal/mol threshold for chemical accuracy. The error that matters came from the embedding: fragmented and unfragmented calculations of the same clusters disagree by about 110 kcal/mol on tritium binding, roughly sixty times the thermal energy at a blanket’s 900 K operating temperature, and classical methods could already solve the benchmark clusters. Until the embedding error comes down, a fusion engineer cannot use the result.

For R&D leaders, this means keeping your classical chemistry infrastructure and adding quantum computation at specific bottlenecks. The organizations that benefit most will be those that know exactly which of their problems are quantum-amenable and have built the workflows to embed quantum simulation into their broader computational pipeline.

What R&D leaders should do now

The preparation timeline for quantum chemistry is measured in years, and the decisions that matter are being made now.

Identify your quantum-amenable problems. Review your computational chemistry portfolio for problems where classical methods produce unreliable results: strongly correlated transition-metal systems, multi-reference excited states, reaction mechanisms involving spin-state changes. These are your first quantum targets.

Invest in classical-quantum hybrid workflows. Quantum computers will not run in isolation. They will provide high-accuracy results for small, strongly correlated subsystems that are then embedded into larger classical simulations. Building the software infrastructure to shuttle data between classical and quantum solvers is a prerequisite for using quantum hardware productively when it arrives.

Develop quantum algorithm literacy in your team. Your computational chemists do not need to become quantum physicists. They do need to read a resource estimate and to understand what quantum phase estimation does and how Hamiltonian simulation works. This is a training investment that pays off whether the hardware arrives in 2030 or 2035.

Engage with quantum hardware providers. Establish relationships with IBM, Google, Quantinuum, Xanadu, PsiQuantum, or whichever providers align with your strategic needs. Early access programs, benchmark collaborations, and joint algorithm development create institutional knowledge that cannot be built on short notice.

Use the Bellonzi framework to evaluate quantum ROI. For each candidate quantum computation, estimate the computational value (what would the result be worth to your R&D program?) and the quantum workload (QPU-hours at projected hardware costs). Then assess the classical alternative: what does the equivalent classical calculation cost, and how reliable is it? If the quantum computation delivers higher accuracy at comparable or lower total cost, it belongs in your roadmap.

Build sovereign optionality. As I discuss in Quantum Sovereignty and the Utility Trap, quantum access for chemistry R&D is a strategic dependency. Avoid locking your computational pipeline into a single vendor’s platform. Design for portability across hardware providers. Consider the supply chain implications of your quantum access strategy.

The honest ledger

Here is the summary, stripped of hype.

Quantum computing will provide genuine, demonstrable advantage for a specific class of chemistry and materials problems: those involving strongly correlated electronic states where classical methods cannot achieve the required accuracy. The applications are real (FeMoco, P450, battery degradation, photosensitizers, CO₂-conversion catalysts). The resource estimates are concrete and falling, and the hardware timelines are plausible.

The advantage is narrower than the marketing suggests. Most chemistry calculations will continue to run classically. The quantum edge applies to a small minority of problems, including some of the highest-value calculations in pharmaceutical, chemical, and energy R&D. Earlier versions of this article put that minority at 5 to 10%. I have taken the number out. It was my impression, and I should not base a budget recommendation on a figure I never measured.

The economic returns are significant, and they take time to arrive because the path from quantum simulation to industrial product includes translation steps that quantum computing does not accelerate. The organizations that benefit most will be those that have invested in the workflows, the teams, and the institutional knowledge to use quantum results productively.

The preparation window is open. Hardware for the earliest applications (photosensitizers, and battery spectra on width) is projected for 2029–2031. Hardware for the grand challenges (FeMoco, P450) is projected for 2033–2038. Building the capability to use that hardware takes two to five years. If you want to be ready for 2031, you start in 2026 or 2027.

The same organizations face the cryptographic side of the quantum transition, which my book Quantum Ready and the free, open-source PQC Migration Framework cover.

For the full technical resource estimates, see The Quantum Utility Ladder. For the competitive analysis of each industry, see Quantum Computing by 2033.

From the author