PARIS — France is trying to make sure the next computing revolution does not depend on a foreign software layer before the hardware has even matured.
French quantum startup Alice & Bob and the country’s state research giant CEA have launched a collaboration to develop software that lets quantum processors work alongside conventional supercomputers, initially focusing on problems in materials physics.
The objective is not to replace classical supercomputers.
It is almost the opposite.
Researchers want the conventional machine to perform most of a calculation, then hand only the parts that might eventually benefit from quantum processing to a quantum computer before integrating the result back into the wider workflow.
The more strategic part of the announcement sits in the software.
CEA and Alice & Bob plan to help develop Qaptiva, the hybrid quantum-computing platform created by French supercomputer maker Bull, as a European alternative to platforms such as Nvidia CUDA-Q. Reuters reported that executives involved in the project explicitly want to avoid the kind of software dependence they believe Nvidia’s CUDA ecosystem created during the AI boom.
That turns a relatively technical research collaboration into something much bigger:
France is already thinking about quantum-computing sovereignty before useful large-scale quantum computing has actually arrived.
The quantum computer will not do everything
One of the easiest ways to overstate this story would be to suggest France is replacing one of its supercomputers with a quantum machine.
It is not.
The project follows a hybrid computing model.
A classical high-performance computer, or HPC system, remains responsible for the parts of a problem it handles efficiently.
The quantum processing unit, or QPU, is used only for portions where quantum algorithms might eventually provide a meaningful advantage.
Nvidia describes CUDA-Q using essentially the same hybrid concept: CPUs, GPUs and quantum processors all work within one programming environment rather than quantum hardware operating as a stand-alone replacement for conventional machines.
That is increasingly where the quantum industry believes practical adoption will begin.
Not with a mysterious quantum box replacing a data centre.
But with quantum processors becoming another specialised accelerator inside the data centre.
Materials physics comes first
CEA and Alice & Bob say their initial application work will focus on materials physics.
There is a reason quantum developers repeatedly return to materials and chemistry.
Atoms and electrons behave according to quantum mechanics.
Simulating those systems precisely on classical computers can become extraordinarily difficult as the number of interacting particles increases.
A sufficiently capable fault-tolerant quantum computer could, in principle, represent some of those quantum systems more naturally.
That could eventually help researchers investigate new materials, catalysts, batteries or chemical processes.
But “eventually” is doing a lot of work in that sentence.
The current CEA-Alice & Bob programme is research and infrastructure development, not proof that a quantum computer has already beaten France’s best supercomputers on an industrial materials problem.
France is putting Alice & Bob’s hardware inside one of its major computing centres
The software project fits into a hardware plan already underway.
In June, France’s national high-performance computing agency GENCI agreed to purchase an 18-cat-qubit Alice & Bob quantum computer.
The machine will be installed at CEA’s Très Grand Centre de Calcul, or TGCC, in Bruyères-le-Châtel and connected to the Joliot-Curie supercomputer.
The system is expected to become available to users during 2027.
CEA describes it as the first acquisition of a quantum computer based on Alice & Bob’s cat-qubit technology and the first so-called early fault-tolerant quantum computing system scheduled for installation in a European computing centre.
That wording needs care.
“Early fault-tolerant” does not mean France will suddenly possess a large, fully fault-tolerant universal quantum computer next year.
The machine is an early research system intended to help develop the technologies, software and workflows needed to reach that stage.
So what exactly is a “cat qubit”?
Alice & Bob takes its name from two fictional characters commonly used in physics and cryptography — but the “cat” in its technology refers to the famous Schrödinger’s cat thought experiment.
Its hardware uses a type of superconducting quantum bit called a cat qubit.
The company’s central claim is that this design naturally suppresses one of the two major categories of quantum errors: bit flips.
That matters because errors are one of quantum computing’s biggest obstacles.
Ordinary digital bits are reliably either 0 or 1.
Quantum states are extremely fragile.
Heat, electromagnetic interference and other forms of noise can corrupt calculations long before complex algorithms finish.
The usual solution is quantum error correction, where many imperfect physical qubits are combined to create a more reliable logical qubit.
Alice & Bob’s strategy is to reduce the amount of hardware required for that process by building part of the error protection into the qubit architecture itself.
It does not eliminate all quantum errors.
The architecture strongly suppresses bit-flip errors, while another major error type — phase flips — still has to be handled through error-correction techniques.
The first machine has only 18 cat qubits
That number is another useful reality check.
Alice & Bob’s Helium system uses 18 cat qubits and was designed so those physical qubits can be used to work toward the company’s first logical qubit.
Compare 18 qubits with the millions of operations and enormous memory available inside modern classical supercomputers and it becomes obvious why this is not a replacement technology today.
The purpose of Helium is experimentation.
Researchers can study quantum error correction.
They can test how the processor communicates with HPC infrastructure.
They can develop scheduling systems and programming tools.
And they can learn how future, much larger quantum processors could be integrated into existing computing centres.
In other words, France is building the plumbing before the building is finished.
CEA already has other quantum machines
Alice & Bob will not be the first quantum company inside the TGCC ecosystem.
CEA has already installed systems from French quantum companies Quandela and Pasqal, Reuters reported.
Alice & Bob’s machine will join them in 2027.
The technologies are very different.
Pasqal develops neutral-atom quantum processors.
Quandela uses photonic quantum computing.
Alice & Bob uses superconducting cat qubits.
France is therefore not betting everything on one quantum architecture.
It is creating an environment where researchers can test multiple technologies alongside conventional HPC systems.
That matters because nobody yet knows which quantum-hardware approach — or combination of approaches — will ultimately prove most commercially useful.
Eventually, the system is supposed to connect to a European exascale machine
France’s plans extend beyond Joliot-Curie.
CEA says the Alice & Bob processor is later expected to be linked to Alice Recoque, the European exascale supercomputer being acquired through the EuroHPC programme.
The result would combine three layers of computing:
high-performance classical computing,
artificial intelligence,
and quantum processing.
That hybrid architecture is increasingly viewed as a likely future for specialised scientific computing.
GPUs did not eliminate CPUs.
AI accelerators did not eliminate conventional servers.
Quantum processors may similarly become another accelerator designed for particular classes of problems.
The software could matter more than the first quantum machine
The history of computing shows why France is focusing so heavily on software.
Powerful hardware is useful only if researchers and developers can actually program it.
Nvidia’s rise during the AI boom was built not simply on GPUs but on CUDA, the software ecosystem that made those chips accessible to developers.
Once companies built large amounts of software around CUDA, switching hardware suppliers became far more difficult.
Reuters said executives behind the CEA-Alice & Bob collaboration explicitly cited that experience when explaining why they want a competitive European quantum software stack.
That is where Qaptiva enters the story.
Qaptiva comes from Bull — and Bull now belongs to the French state
Qaptiva was developed by Bull, the historic French computing company.
And Bull’s ownership changed dramatically this year.
On March 31, the French government completed its acquisition of 100% of Bull’s advanced-computing business from Atos, in a transaction valued at up to €404 million.
Bull includes France’s high-performance computing, quantum, business-computing and AI activities.
It also operates what the company describes as Europe’s only supercomputer manufacturing plant, in Angers.
So this project now connects three strategically French assets:
CEA, the state research organisation;
Bull, the state-owned supercomputer maker;
and Alice & Bob, one of France’s leading private quantum startups.
That makes technological sovereignty central to the partnership rather than a marketing afterthought.
France wants Qaptiva to compete with CUDA-Q
Nvidia’s CUDA-Q is an open-source platform designed for hybrid quantum-classical computing.
It allows developers to combine CPUs, GPUs, quantum processors and quantum emulators through one programming environment.
Qaptiva is intended to offer similar infrastructure while remaining hardware-agnostic.
Bull and Alice & Bob announced in June that Qaptiva would be expanded to emulate cat qubits and eventually execute workloads directly on Alice & Bob processors inside HPC environments.
The new CEA collaboration gives that work a major public-sector test bed.
That could help France answer one of the most important questions in quantum computing:
Can researchers write an application once and use it across different quantum systems without being permanently tied to one hardware vendor?
But calling Nvidia the enemy would be misleading
There is a twist.
Alice & Bob is also an Nvidia collaborator.
In 2025, the French startup integrated its Dynamiqs quantum-simulation library with CUDA-Q, reporting major speed improvements from GPU acceleration.
In March 2026, it again used CUDA-Q to accelerate simulations for quantum error correction, reporting a 9.25-times speed-up compared with its CPU implementation for one decoding workflow.
So this is not a clean France-versus-Nvidia confrontation.
Alice & Bob can simultaneously use Nvidia technology and support development of an alternative European software environment.
That is common in emerging technology.
Interoperability matters precisely because researchers do not want to choose one ecosystem forever.
The real target is lock-in
That distinction may be the most important business point in the announcement.
France is not necessarily trying to eliminate CUDA-Q.
It is trying to avoid reaching a future in which there is no realistic alternative.
Once an entire industry builds its applications, training, libraries and development tools around one platform, switching becomes expensive even when another hardware provider offers better technology.
The AI industry demonstrated how powerful that effect can become.
French policymakers do not want quantum computing to reach the same stage before Europe has established its own competitive software.
Macron has put another €1 billion behind the quantum strategy
The CEA-Alice & Bob collaboration also sits inside a much larger government push.
In May, French President Emmanuel Macron announced an additional €1 billion for France’s national quantum strategy, funded through the France 2030 programme between 2026 and 2030.
That came on top of the original national quantum strategy launched in 2021.
Macron said the new funding would support quantum-computing hardware, infrastructure, software and applications, including programmes led by CEA and INRIA.
He separately announced €550 million for advanced microelectronics.
The message was unmistakable:
France increasingly views computing infrastructure the same way countries historically viewed energy, telecoms or defence technology — as something too strategically important to leave entirely to overseas suppliers.
CEA itself is not an ordinary university laboratory
CEA is one of France’s most strategically important public research organisations.
Reuters puts its annual budget at almost €6 billion.
Its work ranges from energy and advanced computing to defence technologies, and the organisation has responsibilities connected with France’s nuclear deterrent.
That background explains why “sovereign computing” carries unusual weight in this project.
The state is not merely funding an academic experiment.
It is attempting to cultivate domestic technology providers that could become strategically important if quantum computing eventually becomes useful in defence, energy, materials research or industrial simulation.
Alice & Bob is scaling rapidly too
The startup itself has moved well beyond a small university-style research team.
Alice & Bob said in April that its workforce had reached 251 employees after hiring more than 100 people in seven months.
The company has also announced plans for a roughly US$50 million quantum laboratory in Paris, funded from its recent financing round, where it plans to develop future generations of cat-qubit chips.
Its roadmap currently moves from the 18-qubit Helium system through larger Lithium and Beryllium generations toward a more ambitious fault-tolerant system around 2030.
Roadmaps in quantum computing should always be treated as targets rather than guaranteed delivery dates.
The technical obstacles remain enormous.
Even Alice & Bob says today’s quantum computers are not useful enough
One of the more refreshing pieces of messaging on the company’s own website is its blunt description of the industry:
today’s quantum computers still suffer heavily from errors.
That is important context whenever a quantum partnership generates headlines implying an imminent computing revolution.
Current quantum machines remain limited.
They are noisy.
They contain relatively few reliable logical qubits.
And for most ordinary commercial workloads, classical computing remains vastly more practical.
The race is about creating systems that eventually change that equation.
“Quantum advantage” is still something this project hopes to reach
CEA’s plan is to route work to a quantum processor only where that processor may eventually deliver an advantage.
The word eventually matters.
There is no claim in Thursday’s announcement that Alice & Bob has already demonstrated a commercially useful speed advantage over CEA’s classical supercomputers in materials physics.
The immediate objective is to build software, workflows and infrastructure capable of taking advantage of more capable hardware when it arrives.
That means Qaptiva has value even before large-scale quantum advantage exists.
Researchers can test algorithms on emulators.
They can simulate cat-qubit systems.
They can build HPC scheduling tools.
And they can develop the software interfaces required to run future hardware.
France is effectively trying to avoid waiting until quantum is useful
That is the strategy.
If governments wait until quantum processors become undeniably valuable before developing domestic infrastructure, foreign software ecosystems may already dominate.
France is instead investing before the economic winners are obvious.
It is buying early machines from French startups.
It is installing competing quantum architectures inside government-backed computing centres.
It is putting state money into hardware, software and research.
And it now owns Bull, the company supplying much of the classical supercomputing layer underneath those experiments.
That is an expensive bet.
But strategic technologies often become much more expensive to enter after standards and market leaders are established.
The Nvidia comparison is really about AI history
The reason CUDA keeps appearing in the quantum discussion is that software helped turn Nvidia’s GPU advantage into something much more durable.
Developers learned CUDA.
Libraries were optimised around CUDA.
Research frameworks depended on CUDA.
Companies invested years building applications on it.
That created switching costs even as rival chips improved.
Quantum computing is still young enough that those software loyalties have not hardened in the same way.
France wants to keep it that way.
Europe has another reason to care: AI computing demand already exceeds capacity
Europe is simultaneously spending heavily on traditional AI supercomputers.
Just weeks before Thursday’s quantum announcement, the EU awarded Bull a €387.8 million contract to build the LUMI-AI supercomputer in Finland.
Reuters reported that demand for European AI computing capacity has been running ahead of available supply.
That shows why quantum should not be viewed as a substitute for continuing classical investment.
Europe is still trying to secure enough GPUs and supercomputers for today’s AI workloads.
Quantum is a longer-term layer being built beside that infrastructure.
The real battle may be over standards rather than qubit counts
Quantum headlines often focus on who has the most qubits.
That can be misleading.
A large number of noisy physical qubits is not automatically more useful than a smaller number of higher-quality ones.
Alice & Bob itself has argued that the industry needs more consistent ways of defining and benchmarking logical qubits, because hardware approaches differ so much that raw counts can be difficult to compare.
Software has a similar problem.
Different processors use different architectures.
Different error models require different tools.
Different vendors use different programming frameworks.
The platform that succeeds in hiding much of that complexity from developers could become disproportionately valuable.
That is exactly what Qaptiva and CUDA-Q are competing to do.
And that may explain why France is moving now
The Alice & Bob computer coming to CEA in 2027 will not overturn global computing.
Eighteen cat qubits are not enough to replace an exascale machine.
Materials companies are not about to abandon classical simulation.
And Qaptiva has not displaced CUDA-Q.
But those are the wrong benchmarks for judging Thursday’s announcement.
France is trying to establish three things before large fault-tolerant machines arrive:
hardware capability, a domestic computing infrastructure and a software ecosystem capable of connecting the two.
If useful quantum computing arrives later this decade or beyond, the countries controlling those layers could have an enormous head start.
And that is why the most consequential piece of this partnership may not be Alice & Bob’s futuristic processor.
It may be the less glamorous layer sitting between the processor and the programmer.
Because the AI boom taught Europe a painful lesson:
owning powerful hardware matters — but if somebody else controls the software everyone needs to use it, true technological independence can remain out of reach.
Leave a Reply