Which Industries Win by 2033, and How to Tell

The machine on the 2033 roadmap

IBM’s published roadmap puts a system called Blue Jay in 2033, carrying 2,000 logical qubits – the error-corrected units assembled from many imperfect physical qubits – and running on the order of one billion error-corrected operations. Google, IonQ and Quantinuum have published fault-tolerance milestones in roughly the same window. None of these machines exists yet, and hardware roadmaps slip.

Still, 2033 is close enough that boards are asking a reasonable question. If a competitor gets access to that machine, does anything change for us?

For most industries the answer is no. For a few it’s yes, and the split between the two groups has almost nothing to do with ambition, budget, or how early a company started. It follows from what the machine is good at. So rather than hand you another ranked list of sectors, we want to give you the screen we teach in our executive programs, because sector rankings go stale and the screen doesn’t.

Two questions that sort every sector

The class of the speedup

Quantum algorithms come in two broad families, and the distance between them decides everything downstream.

Exponential speedups apply to a narrow set of problems, most of them involving the simulation of quantum-mechanical systems. Classical cost grows exponentially with the size of the system being simulated. Quantum cost does not. The gap widens as the problem gets bigger, which is why the overhead of running a quantum computer eventually gets paid back.

Quadratic speedups cut the number of steps to roughly its square root. Grover’s search and quantum amplitude estimation, the sampling technique behind most quantum finance proposals, both live here. A square-root gain is genuinely useful in the abstract, and fault-tolerant machines do not run in the abstract. Each logical qubit is assembled from many physical qubits plus continuous error correction, so a fault-tolerant machine begins thousands of times slower per operation than the classical processor it’s competing with. A quadratic gain has to repay that deficit before it delivers anything, and classical hardware and solvers keep improving while it tries.

The first question, then, is which family the proposed application belongs to. Exponential, or quadratic.

The nature of the bottleneck

The second question is about the business, not the algorithm. What actually limits progress?

If the limiting step is predicting how electrons behave in a molecule, a catalyst, or a crystal, then the bottleneck is a quantum system, and a quantum computer is a native fit. If the limiting step is searching a combinatorial space, fitting a statistical model, or pricing a contract by sampling, the bottleneck is a classical computation, and quantum methods enter as one more sampling strategy competing against fifty years of tuned classical software.

Two questions, four possible combinations. Only one of them is interesting.

Nitrogen fixation, a problem that passes

Industrial ammonia synthesis runs the Haber-Bosch process at roughly 450 °C and 200 atmospheres, and it consumes something like one to two percent of world energy. The nitrogenase enzyme performs the same chemistry at ambient temperature and pressure, using an iron-molybdenum cluster called FeMo-co. Nobody can say precisely how. The electronic structure of that cluster is strongly correlated, meaning the electrons cannot be treated as moving independently, and that defeats the standard classical workhorse, density functional theory (DFT).

Markus Reiher and colleagues published the first serious quantum resource estimate for FeMo-co in 2017. The runtime it implied was measured in years, which read at the time as a polite way of saying never. Successive algorithmic work has cut those requirements by several orders of magnitude, and current estimates place the problem in the low thousands of logical qubits.

A 2024 study of homogeneous nitrogen-fixation catalysts by Bellonzi and co-authors went a step further and priced the work. It valued the highest-utility single calculation at roughly $200,000, against a quantum workload of about 139,000 QPU-hours. Argue with those numbers by all means. The structure is what a chemicals executive should recognize, because it converts a physics claim into a budget line with a value, a runtime, and an implied price per hour.

Both screen questions come out clean here. The bottleneck is a quantum system, and the speedup is exponential.

Derivative pricing, a problem that fails

Financial institutions have funded quantum research harder than almost any other sector. The most studied application, pricing derivatives by quantum amplitude estimation, is also the clearest illustration of the quadratic trap.

Chakrabarti and colleagues’ 2021 resource estimate for pricing derivatives by quantum amplitude estimation put numbers on what it would take to beat classical Monte Carlo on realistic contracts. Roughly 4,700 logical qubits. On the order of a billion T gates, the expensive non-Clifford operations that dominate the runtime of a fault-tolerant circuit. And a logical T-gate rate near 45 MHz. The qubit count is within reach of a Blue Jay-class machine. The clock rate is not, and projected fault-tolerant logical rates sit around three orders of magnitude below it.

That gap is what a quadratic speedup looks like once error-correction overhead is priced in, not an engineering detail waiting for one more hardware generation. Portfolio optimization and risk analysis meet the same wall, and so does most of what gets marketed as quantum logistics.

The rest of the map

Where the screen says yes. Pharmaceuticals qualify through molecular simulation. Goings and colleagues estimated in 2022 that simulating the full cytochrome P450 system, the enzyme family responsible for metabolizing a large share of marketed drugs, needs about 4,900 logical qubits and roughly three days of runtime. A 2,000-qubit machine won’t reach the whole enzyme, but it reaches catalytic fragments and binding pockets, which is where computational drug design actually stalls. Battery chemistry qualifies for the same reason, with voltage fade in lithium-rich cathodes as the signature target and recent algorithmic work bringing some of the relevant spectroscopic simulations into the low hundreds of logical qubits. Advanced materials qualify too, on a slower clock, because development cycles there run long enough that early-2030s capability mostly buys late-2030s position.

Where the screen says no. Finance, logistics and machine learning all fail on the speedup question, and quantum machine learning has failed publicly. Ewin Tang’s 2018 dequantization result showed that a celebrated exponential quantum speedup could be matched by a classical algorithm, and a run of similar results followed. McClean and colleagues described barren plateaus in the same period, where the training signal for variational quantum models vanishes as systems grow. What survives is quadratic at best. Meanwhile classical AI is advancing fast enough that it will likely contribute more to quantum computing this decade, through error-correction decoders and circuit optimization, than quantum computing contributes to AI.

The inversion in the marketing

Evidence for quantum advantage is strongest in the sectors that market it least, and thinnest in the sectors that market it most.

The explanation is institutional rather than scientific. Banks and technology firms run the largest quantum teams and the largest communications budgets, so they generate the most papers, panels and press. Chemists and battery engineers publish into narrower venues. Visibility tracks research spending, not physics, and an investor reading deal flow will see the inversion reproduced in pitch decks.

Access is a separate question

Passing the screen tells you quantum matters for your sector. It says nothing about whether you will be able to use one.

Fault-tolerant machines will be scarce, and the supply chain behind them is narrow at several points: dilution refrigerators from a handful of suppliers, helium-3 from a small number of state-controlled sources, control electronics concentrated in a few countries, and calibration expertise that lives in people rather than manuals. Buying an on-premises system does not resolve this. A machine that depends on a revocable service contract is a rental with extra steps. Export controls on lithography equipment and AI accelerators have already shown how quickly this kind of dependency converts into a policy instrument.

For an operating company, the practical response is optionality rather than independence. Multiple vendor relationships. Workflows written against portable abstractions instead of one provider’s proprietary stack. In-house people who understand what the hardware is doing. For an investor, a portfolio company whose entire quantum plan runs through a single foreign cloud API has a concentration risk that belongs in diligence.

What would change the answer

This screen reflects the published evidence as of early 2026, and four developments would revise it. An algorithmic breakthrough delivering a super-polynomial speedup for optimization or learning would rewrite the finance and logistics verdicts overnight, and the field is young enough that this cannot be dismissed. Faster-than-expected progress in error correction, through qLDPC codes and cheaper magic-state preparation, would pull the 2,000-logical-qubit milestone earlier than 2033. A plateau in classical AI would raise the relative value of quadratic quantum gains. And two or more vendors hitting their fault-tolerance targets early would compress every timeline here by several years.

None of these belongs in a conservative plan. All of them belong on a watch list with a named owner.

What to build before the hardware

The organizations that will use these machines well are not the ones with the biggest quantum budget today. They’re the ones whose people can already read a resource estimate and say what it implies for a specific molecule, argue with a vendor about logical versus physical qubit counts, and recognize a quadratic speedup dressed up as a revolution.

That capability takes two to five years to build, and it survives every revision to the roadmap above.

Quantum Academy’s executive and strategy programs are built for exactly this judgment: understanding what fault-tolerant machines will and won’t do, reading vendor claims against published evidence, and sequencing investment against your own sector’s screen result. You can review the programs and access options at quantumacademy.com/.

If your sector fell on the wrong side of the screen, the near-term quantum work is cryptographic rather than computational, and the migration methodology at pqcframework.org is the better starting point. For the full technical analysis behind this briefing, with algorithm-by-algorithm resource estimates and citations to the primary literature, see the original Quantum Utility Map analysis on PostQuantum.com.