Credit Agricole and Pasqal put a date on it: quantum in production banking by 2028

[Dank15]

Credit Agricole CIB and Pasqal signed a strategic partnership this week to industrialise quantum computing in capital markets, with initial production use cases targeted as early as 2028. This is not a fresh flirtation, they have been collaborating since 2019 and ran projects on counterparty credit risk and portfolio optimisation that the bank says showed quantum algorithms beating classical approaches under certain conditions

The roadmap has three tracks, quantum inspired algorithms on existing classical infrastructure for immediate gains, operational testing on Pasqal's neutral atom machines under production conditions, and eventually hybrid classical quantum deployment for problems beyond today's systems. The bank is even reorganising staff into dedicated quantum business teams, coordinators and deployment managers

The phrase doing the heavy lifting is under certain conditions. Every quantum finance result so far has been a carefully chosen problem instance where the comparison favours the quantum or quantum inspired method, and quantum inspired usually means a classical algorithm that quantum research happened to motivate. That first roadmap track requires no quantum hardware at all

Still, a major bank restructuring teams and naming a production year is different from the usual innovation lab theatre. The next phase targets monitoring capital reserve consumption against risk weighted assets, which is about as unsexy and real as banking gets. Is 2028 brave or delusional?

Crédit Agricole and Pasqal expand strategic quantum partnershipwww.eenewseurope.com

[HitmanMatt53]

Delusional for anything needing fault tolerance, brave and plausible for optimisation heuristics where you just need better answers, not provably optimal ones

[HenryThierry]

The quantum inspired track is the tell. They can declare victory in 2028 with classical algorithms and a quantum logo on the slide

[ParallelSelf34]

Bit cynical. Banks do not restructure reporting lines for logo projects, headcount reorganisation means budget owners are committed

[Rapid Crossing]

Work in quant risk at a competitor and can confirm everyone is watching this deal. If CA CIB shaves basis points off capital consumption, every treasury desk follows within a year

[ScarletWrench]

Neutral atoms for finance optimisation is a decent fit. Analog mode QUBO style problems map naturally onto atom arrays

[Undertaker_EU]

The 2019 start date matters. Seven years of joint work before naming a production date is the opposite of hype cycle behaviour

[TristanFenwick]

Meanwhile Pasqal is also SPACing onto Nasdaq at a 2 billion valuation. A flagship bank deal with a date lands awfully conveniently, no?

[Ava_75]

Every partnership announcement in every industry lands conveniently. The question is whether the technical claims hold and nobody outside the bank can audit them

[Kieron93]

That is my actual problem, none of these finance results are published in a form anyone can reproduce. Commercial secrecy and scientific credibility are pulling in opposite directions

[Matt95]

The partnership also raises the question of who captures the value if quantum becomes useful. The bank could simply buy quantum computing as a service, in which case hardware companies may end up looking more like infrastructure providers than software companies.

Alternatively, banks might develop internal expertise and proprietary algorithms that become valuable in their own right. Then the quantum processor is just another underlying platform, while the real competitive advantage sits in the algorithm, data and integration layer.

That distinction matters because the headline revenue expectations for quantum hardware companies often assume that useful quantum applications automatically translate into huge hardware demand. It might, but cloud access could also concentrate hardware into relatively few providers.

The winners may therefore include some companies nobody is currently calling quantum companies. Control electronics, orchestration software and specialised optimisation tools could quietly become very important.

[Fiend_Suplex]

The secrecy issue is probably going to become one of the defining problems for this sector. A bank has legitimate reasons not to reveal proprietary trading strategies, but without independent validation the public is being asked to trust a press release.

One possible answer is third-party auditing. Let an independent research group or technical auditor inspect the benchmark, verify the classical comparison and confirm that the claimed improvement is genuine without exposing confidential financial data.

That would create a useful middle ground between publishing everything and publishing nothing. The auditor could confirm that the result exists, the methodology is sound and the comparison is fair, while the bank keeps sensitive inputs private.

Otherwise we risk ending up with quantum demonstrations that are simultaneously commercially impressive and scientifically impossible to evaluate. That is not a great foundation for an industry trying to attract serious long-term investment.

[Adam75]

The biggest hurdle may actually be error correction and reliability. A finance application cannot casually accept an answer that is occasionally nonsense because the underlying computation wandered off into quantum wilderness.

Error mitigation can help, but it comes with its own computational and statistical costs. If you need a huge number of circuit evaluations to clean up the answer, the theoretical speedup can disappear under the weight of the overhead.

That is why I would like the partnership to publish operational metrics alongside algorithmic ones. How many runs are needed? How stable is the output? What happens when the hardware is unavailable? How much classical processing surrounds the quantum step?

Those are boring questions, which is usually a good sign. Production technology is mostly made of boring questions.

[Apogee Evan]

A production deadline is useful because it forces everyone to define what production actually means. Running a clever proof of concept on a quantum machine is one thing; putting it inside a bank's risk pipeline, with monitoring, audit trails, fallback systems and predictable latency, is something else entirely.

Imagine a derivatives desk using a quantum routine for a pricing calculation. The system cannot simply shrug and say the processor had a bad afternoon. There needs to be a classical fallback, version control, validation against known cases and enough observability for compliance teams to understand what happened.

That is where partnerships like this could become more significant than the headline qubit numbers. The hard part may turn out to be industrial integration rather than the underlying algorithm. Quantum computing becomes a real technology when IT departments can treat it as another service instead of a visiting science experiment.

[ChampionPeter16]

The phrase industrialise quantum computing is doing a lot of work in that announcement. Industrialisation means repeatability, maintenance, procurement, security, staffing and predictable costs, not just getting a processor to perform an impressive calculation once.

For a bank, availability matters enormously. If the quantum service works beautifully nine days out of ten but the tenth day happens to coincide with an important market event, that is not a production service. There needs to be redundancy and a classical fallback.

The £ and $ figures people usually associate with quantum hardware also obscure the total cost of ownership. Cooling, specialised facilities, networking, software and engineering talent all matter. A cheap quantum compute hour is not useful if integrating it costs millions.

That is why I am cautiously optimistic about a 2028 production target. It is ambitious enough to force serious engineering, but the definition of production will matter more than the date printed in the press release.

[StringTheory55]

One area I would watch closely is risk management rather than flashy trading. Risk calculations are repetitive, computationally expensive and already run through enormous classical pipelines, which makes them a plausible place for a specialised accelerator to earn its place.

A useful quantum system does not need to predict tomorrow's market better than every human trader. It might instead help a bank calculate a large set of scenarios more efficiently, allowing risk teams to run more detailed analyses within the same time budget.

The economics could become compelling if that means a bank can do an overnight calculation in an hour, or a calculation that currently takes an hour in a few minutes. But again, the relevant comparison is the whole system, including data preparation and result verification.

If quantum eventually becomes another accelerator sitting alongside CPUs and GPUs, that would be a much more believable path to production than the idea of replacing conventional computing altogether.

[Scout]

There is a slightly amusing irony in banks being early quantum adopters because finance has spent decades building extremely sophisticated classical numerical methods. If anyone is going to notice that a quantum algorithm is not actually beating a well-optimised classical one, it will probably be the people who already know every computational shortcut in the book.

That makes the bar unusually high. You cannot compare a quantum algorithm against a basic spreadsheet and declare victory. The comparison needs to include serious classical solvers, GPUs where appropriate, distributed computing and whatever optimisations practitioners would actually use in production.

On the positive side, that pressure could make banking a useful proving ground. If quantum techniques survive comparison with mature financial computing, the resulting evidence would carry much more weight than a generic laboratory demonstration.

2028 may therefore be less important as a calendar date than as a forcing function. By then, someone should be able to show exactly where the quantum component earns its keep.

[Molly32]

The reproducibility complaint in the last post is fair, but there is a tension here that is specific to finance. A bank has no incentive to publish the exact portfolio construction or pricing strategy that gives it a competitive advantage.

That does not mean everything has to remain a black box. A useful compromise would be publishing benchmark problems, methodology, error bounds and comparisons against strong classical baselines, while keeping proprietary data and production parameters confidential.

For example, a bank could say that a quantum-assisted optimiser reduced a defined computational workload by a certain percentage against a specified classical implementation, then publish enough information for independent researchers to reproduce the benchmark. That would be far more convincing than saying quantum delivered a breakthrough on an undisclosed internal dataset.

Scientific credibility and commercial secrecy do not have to be enemies. They just require some creative thinking about what gets disclosed.

[Steve]

There is a good argument that finance is exactly the wrong place to judge quantum because classical optimisation is already so mature. On the other hand, that may be why it is exactly the right place.

If a quantum system can produce a measurable advantage against the best classical techniques used by serious financial institutions, the result becomes difficult to dismiss. The benchmark is painful, but it is also meaningful.

The strongest demonstration would be something like: here is a real-world class of problem, here is the best practical classical solution, here is the quantum-assisted method, here are the total compute costs, and here is the improvement in time or quality. No mystical claims required.

Until then, I would treat 2028 as an engineering milestone rather than a prediction about the quantum revolution. If they hit it with a small production workload, great. If they hit it with a genuinely superior financial calculation, that would be a very different level of news.

[Taz]

Pasqal's neutral-atom approach makes the partnership particularly interesting because architecture matters when you start talking about industrial deployment. Different quantum platforms have different trade-offs in connectivity, control, error characteristics and scaling.

A financial institution does not necessarily care which architecture wins the academic popularity contest. It cares whether the machine can deliver a useful calculation at an acceptable cost, with enough reliability to integrate into a real workflow.

That also means the 2028 target could change shape as the hardware roadmap evolves. Maybe the first production use is modest and tightly scoped rather than some giant quantum finance platform. There is nothing wrong with that. Most technologies enter production by solving one annoying problem particularly well.

The real milestone would be a quantum workload that quietly becomes part of the bank's infrastructure and nobody thinks of it as exotic anymore.

[Leo34]

The 2028 date is what makes this interesting, although I would treat it as a target rather than a prophecy. Banking is one of the few industries where a quantum advantage does not have to mean replacing a whole classical workflow. If a quantum subroutine can improve one difficult optimisation step while the rest remains on conventional infrastructure, that is already commercially useful.

Capital markets have plenty of candidates for this kind of hybrid approach: portfolio optimisation, derivatives pricing, risk calculations and scheduling problems. The practical test is whether the quantum component reduces total cost or improves the result enough to justify the additional complexity.

That last bit is easy to overlook. If a quantum calculation saves 30 seconds but requires a specialist team, expensive hardware access and three layers of classical preprocessing, the bank has not necessarily gained anything. The production system has to beat the existing workflow, not merely demonstrate that the quantum processor can perform the calculation.

[SchrodingersCat]

A tangent, but the regulatory implications are fascinating. Once a quantum system contributes to a capital-markets decision, somebody is eventually going to ask how the result was validated and whether it can be explained or reconstructed after the fact

Banks already have extensive requirements around model governance. Introducing a probabilistic quantum component does not make those requirements disappear. If anything, it adds another layer of documentation and testing

That could actually help the technology mature. Financial institutions are very good at turning promising technology into a pile of controls, logs and approval forms. Nobody finds that exciting, but those controls are part of what allows a tool to operate at scale

So if quantum survives the compliance department, that may be a stronger signal than surviving another conference demo. Passing an auditor's checklist is the financial industry's equivalent of a boss battle.

[ForumGremlin93]

There is a danger of treating 2028 as though it means banks will suddenly be doing all their calculations on quantum computers. That is not what I would expect at all

A much more realistic scenario is a hybrid stack. Classical infrastructure handles data ingestion, preprocessing, most of the computation and validation, while a quantum processor is called for a specific subproblem where it has a demonstrated advantage

That resembles how GPUs became important. Nobody replaced the entire computer with a graphics processor. They found workloads where specialised hardware was dramatically better and built systems around that fact. Quantum could follow a similarly incremental route, although the technical challenge is obviously much greater

If the first genuine production workload is tiny but economically meaningful, I would count that as a success. The first commercial aircraft did not carry 500 passengers either