Quantum Sovereignty: Why France Is Racing to Protect Its Quantum Startups Before They’re Bought Out

On 19 July 2026, the Cour des Comptes used a word supreme audit institutions never use about a technology sector: predation. French quantum companies, the report said, face predation by foreign players, and the sovereignty stakes are “colossal.” The support architecture built over decades of publicly funded research cannot stop that research from being hollowed out and sold. For a body famous for grey administrative prose, this was a shout. Working out why it shouted means walking through both the physics of quantum computing and the industrial mechanics that make French quantum assets world-class and acquirable at the same time.

France did not stumble into this position. Its research base is one of the deepest anywhere, with two Nobel laureates at its foundation. Serge Haroche took the 2012 physics prize for cavity quantum electrodynamics, the experiments that showed how individual photons and atoms interact inside controlled quantum systems, a prerequisite for building quantum computers at all. Alain Aspect won in 2022 for his experimental tests of Bell inequalities, the work that turned entanglement from a philosophical argument into an exploitable physical resource. Both built their careers inside the CNRS ecosystem, and the institutions around them, ENS Paris, Institut d’Optique, Université Paris-Saclay, Inria, have been producing quantum physicists and engineers on a steady conveyor for thirty years.

Macron’s National Quantum Strategy, announced in January 2021 with €1.8 billion behind it over five years, bet that scientific capital could be turned into an industry. The plan ran across Bpifrance, the ANR, and the CEA, whose CEA-Leti lab in Grenoble is one of the best microelectronics research centers on the continent. The sum was aggressive by European standards, larger than any single member state had committed before. Set against the US and Chinese programs, it was small.

The startups that came out of this are technically serious, not vaporware. Pasqal spun out of Institut d’Optique and CNRS in 2019 and works in neutral-atom computing. Quandela, in Massy, builds photonic processors from semiconductor quantum dots. Alice & Bob came out of ENS Paris and Inria groups in 2020 and is building cat-qubit superconducting systems. Quobly is a CEA-Leti spin-out on silicon spin qubits, and C12 Quantum Electronics is chasing carbon-nanotube spin qubits. Each picks a genuinely different physical approach, each holds real IP, and each was built substantially on public money. That last fact is the whole problem. The Cour des Comptes fears that value the French taxpayer created could pass into foreign hands at prices set by startup distress rather than strategic worth.

No such thing as a quantum computer

There is no single quantum computer, which is exactly why these companies are both valuable and fragile. At least five physical implementations are in live commercial development right now, each with its own engineering bargain, and nobody knows which reaches fault tolerance first.

Superconducting qubits are the most deployed approach today, the IBM, Google, and Rigetti path. The qubit is a Josephson junction, two superconductors split by a thin insulating barrier, chilled to roughly 15 millikelvin in a dilution refrigerator, colder than interstellar space. At that temperature the junction goes quantum and can sit in a superposition of its two energy states. Gates are fast, 10 to 100 nanoseconds, and fabrication borrows from mature semiconductor lines. The bill comes due elsewhere: a dilution fridge costs hundreds of thousands of euros and eats floor space, and coherence times, the window before environmental noise wipes the quantum information, sit at 100 to 500 microseconds. IBM’s Condor processor hit 1,121 qubits in 2023. Google’s Willow chip in 2024 showed error correction running below the threshold where adding qubits starts reducing errors instead of amplifying them, a genuine landmark, though fault-tolerant universal computation is still years out.

Trapped ions take a different bargain. IonQ and Quantinuum (the Honeywell and Cambridge Quantum venture) suspend individual ions, usually ytterbium-171 or barium-133, in electromagnetic traps and drive their electronic states with lasers. The ions arrive naturally identical, so no fabrication variance. They offer all-to-all connectivity, meaning any qubit can talk to any other, unlike superconducting layouts where physical geometry constrains who talks to whom. Gate fidelities clear 99.9%, the best of any platform, and coherence runs into seconds. The penalty is speed: two-qubit gates take microseconds to milliseconds, orders of magnitude slower than superconducting, and cramming many ions into one trap gets mechanically and optically nasty.

Pasqal’s neutral atoms trap individual rubidium or cesium atoms in arrays of optical tweezers, tightly focused laser beams that pin each atom to a spot in space, with the qubit encoded in the atom’s hyperfine ground states. Gates work through the Rydberg blockade: excite an atom to a high-energy Rydberg state and its bloated electron cloud creates a dipole interaction strong enough to forbid a neighbor from being excited at the same time. Like trapped ions, every rubidium-87 atom is physically identical, so the fabrication variance that plagues superconducting qubits simply does not exist here, and coherence is exceptional at 1 to 10 seconds. Pasqal has run arrays of more than 1,000 atoms in analog mode, where the atoms collectively simulate a quantum system without individual gate-level control. Digital gate-based operation, the general-purpose kind, runs at smaller scales for now. Gates are microsecond-slow, and the optics needed to address individual atoms across a large array are demanding.

That analog lean defines what Pasqal can actually sell this year. In analog mode you engineer the atomic array to physically mimic another quantum system, a molecule, a magnetic material, an optimization landscape, and let the atoms evolve naturally into the answer. That is useful today in materials science, quantum chemistry, and combinatorial optimization, without the crushing overhead of fault-tolerant digital computation. It is a different product from what IBM or Google are selling, and it exists now rather than in a decade. The risk buried in that choice is that Pasqal wins a valuable but bounded niche while ceding the long-term universal-computing market to the Americans.

Quandela encodes quantum information in individual light particles, usually in polarization or spatial path. Photons barely interact with their environment, so decoherence is low, and they move at the speed of light, which makes them natural for quantum communication and networking. Quandela’s own contribution is using semiconductor quantum dots as deterministic single-photon sources: a dot inside a photonic cavity spits out one photon at a time, efficiently, and each photon is quantum-mechanically indistinguishable from the last, which interference-based computation demands. The catch is fundamental. Photons ignore each other, so two-qubit gates end up probabilistic instead of deterministic. Quandela has opened its systems to the cloud through the open-source Perceval framework and hooked up with European HPC centers.

Alice & Bob’s cat qubits are probably the most distinctive French bet on the board. It is still superconducting, but it uses bosonic codes, the cat code specifically, to bake asymmetric noise into the hardware. A cat qubit is a superposition of two coherent states of a microwave cavity, Schrödinger’s cat alive and dead at once. Bit-flip errors, a 0 becoming a 1, are suppressed exponentially by the encoding itself, while phase-flip errors survive. Kill one error type in hardware and a much simpler repetition code can mop up the rest, which slashes the number of physical qubits you need per logical qubit. Alice & Bob’s published estimates put their overhead at one to two orders of magnitude below a standard surface-code superconducting design. If that holds at scale it is a serious moat. And the Cour des Comptes concern is exact: that moat was dug with French public funding, the company is tiny, and one bad down round during a market correction could put it on sale for a fraction of its strategic worth.

Silicon spin qubits, the Quobly and CEA-Leti line built in collaboration with Intel, store information in the spin of single electrons trapped in silicon transistor-like structures. The pitch is manufacturing compatibility: fabrication looks almost identical to conventional CMOS, which raises the prospect of the same economies of scale that made classical chips cheap. CEA-Leti’s QUBOX platform is a real effort, and Grenoble’s status as Europe’s semiconductor capital hands it infrastructure most rivals lack. Coherence and fidelity are climbing fast, though the platform trails superconducting and trapped-ion systems in maturity.

Capital asymmetry

The global quantum market ran to roughly $1.3 billion in 2024, with 2030 projections spread between $5 billion and $12 billion and speculative long-term figures of $450 billion to $850 billion, and that enormous spread is honest uncertainty about when fault tolerance actually arrives. Near-term money lives in simulation and optimization on today’s noisy intermediate-scale quantum (NISQ) hardware. General-purpose fault-tolerant machines, the ones that unlock the full quantum advantage, are a decade out at minimum for everyone.

The United States owns the deployment story. IBM’s Quantum Network runs to more than 200 partner organizations worldwide, and its machines are the most widely accessed quantum hardware anywhere. Google runs fewer systems but has produced the field’s most important recent experimental results. IonQ trades on the NYSE, which hands it capital-market access no European startup has. Quantinuum operates the highest-fidelity systems on the commercial market. And Amazon Braket, Microsoft Azure Quantum, and Google Cloud all offer quantum access through their clouds, building platform lock-in that favors US infrastructure no matter whose hardware sits underneath.

China’s public investment is estimated above $15 billion, roughly five times the US public number and more than eight times France’s. The University of Science and Technology of China built both the Jiuzhang photonic system and the Zuchongzhi superconducting processor, each used to claim quantum advantage on specific tasks. Its Micius satellite demonstrated quantum key distribution over 1,200 kilometers back in 2017, a space capability no Western nation has matched. China’s ecosystem is largely closed to foreign acquisition, so the predation risk there runs strictly one way.

Europe is scattered across the map. The UK launched a £2.5 billion National Quantum Strategy in 2023, post-Brexit, which threatens to pull it out of EU quantum coordination. Canada has Xanadu on photonics, with the widely used PennyLane framework, and D-Wave on annealing. Germany runs strong superconducting research at Forschungszentrum Jülich, Finland’s IQM sells superconducting systems to European HPC centers, the Netherlands’ QuTech at TU Delft leads on spin qubits and quantum-internet protocols, and Barcelona Supercomputing Center is bolting quantum accelerators onto its machines. Over the top of all of it sits the EU Quantum Flagship, a €1 billion program running 2018 to 2028 that funds research consortia across the bloc.

Put the numbers next to each other and the Cour des Comptes problem is obvious. Pasqal has raised around €140 million total against firms whose parents hold cash reserves in the hundreds of billions of dollars. Google or Microsoft could buy Pasqal outright for a sum that would not dent a quarterly report, and the acquisition would carry off far more than a cap table: the team, the patents, the customer relationships, and the accumulated tacit know-how that no patent filing ever fully captures.

The cryptographic dimension

Cryptography is the most urgent pillar of the sovereignty argument, and it starts with a single 1994 result. Peter Shor showed that a powerful enough quantum computer could factor large integers in polynomial time, which breaks RSA, and solve the discrete logarithm problem, which breaks elliptic-curve cryptography. Those two algorithms sit under essentially all public-key cryptography in use today, from HTTPS and banking to email encryption, VPN tunnels, military traffic, and diplomatic cables. A fault-tolerant machine running Shor’s algorithm turns that entire stack into ruins.

The timeline is uncertain, not infinite. Intelligence agencies now plan around cryptographically relevant quantum computers appearing within 10 to 20 years. France’s ANSSI and the American NSA are already mandating migration to post-quantum cryptography, and NIST finalized its first post-quantum standards in 2024, built mostly on lattice problems believed hard for classical and quantum machines alike. The urgency has a name: harvest now, decrypt later. State actors are hoovering up encrypted traffic today, storing it, and waiting for hardware strong enough to crack it in retrospect. A classified cable intercepted in 2024 could be plaintext in 2035 if the hardware curve holds.

A nation that controls the hardware and software controls its position on that timeline. It can push its own cryptanalytic capability forward while knowing precisely when an adversary’s systems fall open, and it can build quantum key distribution, communication channels whose security rests on physics rather than on the assumption that some math problem stays hard, for its most sensitive traffic. France’s DGA funds quantum sensing and cryptography programs for exactly these reasons.

Quantum sensing is a separate field from computing and it is already paying out. Quantum gravimeters read subsurface density variations with a sensitivity classical instruments cannot touch, which matters for submarine detection, mapping underground facilities, and precision navigation. Quantum magnetometers pick up the magnetic signatures of vehicles, submarines, and electronics. Quantum-enhanced atomic clocks sharpen GPS and enable navigation with no satellite at all, and quantum inertial systems hold position without any external reference, the difference between blind and sighted for a submarine or an aircraft in a GPS-denied zone. None of this is speculative, it is being fielded now, and the underlying physics is a close cousin of the computing platforms.

Predation, and the holes in the law

The Court chose “predation” carefully. It does not mean the old-fashioned hostile takeover, a company fighting off a bidder it does not want. It means the structural trap where a French quantum firm, needing capital to survive the long crawl before commercial quantum computing matures, signs investment or acquisition terms that hand strategic assets to foreign control, frequently with no serious national-security review attached.

France carries scar tissue here. General Electric’s 2015 purchase of Alstom’s energy division moved turbine technology with nuclear-plant applications into American ownership, became a political scandal, and directly drove the strengthening of France’s foreign-investment screening regime, the IEF (Investissements Étrangers en France). The 2020 attempt to acquire Photonis, which makes night-vision optics for defense, got blocked by the government after public pressure. Those two cases are the template the Cour des Comptes is deliberately invoking.

Quantum computing was added to the sensitive sectors under IEF screening, but the Court’s point is that the mechanism does not fit quantum’s dynamics. Outright equity purchase is only one route. IP walks out through licensing deals, talent through competitive hiring, technology through research partnerships, each transaction sliding under the screening threshold on its own while together they gut a company’s competitive position. A US cloud provider dangling favorable listing terms in front of a French startup gets usage data, algorithm-performance data, and customer relationships. A foreign investor taking a minority stake below the IEF line gets board visibility and information rights. No single move triggers a review. Stacked together they amount to effective foreign control.

The EU’s 2020 FDI Screening Regulation offers a coordination layer but leaves the final call to member states and never touches most of the channels through which technology actually leaks. So the Cour des Comptes wants IEF enforcement tightened specifically around quantum IP, anchor contracts through guaranteed public procurement to give these companies revenue that does not depend on foreign capital, and licensing conditions on publicly funded research that block transfer to non-allied entities without government sign-off.

Pasqal’s joint venture with Saudi Aramco and KACST (King Abdulaziz City for Science and Technology) makes the dilemma physical. Pasqal needed money and international market access, and the Gulf partnership delivered both. Saudi Arabia is not an adversary in the way China or Russia is, but it is not a NATO ally either, and its strategic interests do not point toward European quantum sovereignty. The report drags out a question the IEF framework cannot cleanly answer: is Gulf sovereign capital predation? When should a French quantum company be allowed to take it, and what governance strings should come attached?

The governance mess, and the Airbus template

The Court’s attack on fragmented governance is structural, not rhetorical. France’s quantum ecosystem runs through the ANR for research grants, the CEA for infrastructure and spin-outs, Bpifrance for investment and loans, Inria for software, CNRS for fundamental work and more spin-outs, and several ministries on top. Every one of them has its own mandate, its own clock, its own definition of success. The consequence is that early research is lavishly funded, France’s academic output world-class on any bibliometric measure, while the crossing from lab demonstration to commercial product, the valley of death, stays lethal. Alice & Bob has demonstrated cat-qubit physics convincingly in the lab and now needs sustained capital across a 5 to 10 year runway before the technology can be commercially validated. That runway is longer than a typical VC fund’s horizon and longer than most public research programs stay patient.

The report’s call for a “European Quantum Champion” reaches for Airbus on purpose. Airbus was built in the 1970s through coordinated state backing from France, Germany, and the UK precisely because no single European country could carry the capital needed to fight Boeing alone. The analogy is imperfect, quantum carries far more technical uncertainty and far more competing approaches than commercial aviation, but the logic is identical: European fragmentation against US and Chinese consolidation is a structural handicap that markets on their own will never fix.

The EU Quantum Flagship and EuroQCI (European Quantum Communication Infrastructure) give Europe coordination frameworks, but they are research programs, not industrial-policy instruments. EuroHPC’s plan to fold quantum accelerators into European supercomputers is more concrete as a procurement lever, and GENCI, France’s national HPC agency, has already integrated Pasqal systems into national infrastructure. The Court wants a lot more of exactly that: guaranteed public procurement as an anchor that gives French quantum firms a revenue base independent of foreign capital, while government and defense institutions build their own quantum muscle in parallel.

Brain drain sharpens every other governance problem. A physicist or engineer good enough to work at Pasqal or Alice & Bob is just as employable at Google Quantum AI, IBM Research, or a well-capitalized US startup, usually at two to three times the pay. French academic salaries are not internationally competitive, and even startup compensation struggles to close the gap. The Court flags this as a threat separate from acquisition, and a nastier one: a company can stay entirely French-owned and still bleed out its best people, achieving the same effective technology transfer without a single transaction anyone can review.

What fault tolerance actually costs

Every quantum computer running today, French, American, Chinese, sits in the NISQ era, Noisy Intermediate-Scale Quantum. NISQ machines have tens to thousands of qubits but error rates far too high for the algorithms that would deliver transformative advantage, Shor’s, Grover’s, quantum chemistry at scale. On any deep circuit, errors pile up faster than the computation can advance.

Fault tolerance demands quantum error correction: you encode one logical qubit across many physical qubits so that errors on the individual physical qubits can be caught and fixed without measuring, and thereby collapsing, the logical state. The most-studied route, the surface code, needs roughly 1,000 physical qubits per logical qubit at current error rates to reach usable computation. That means IBM’s 1,121-qubit Condor holds about one logical qubit’s worth of resources under surface-code assumptions, a sobering measure of the distance still to cover.

Google’s 2024 Willow result mattered because it ran error correction below threshold: as the code distance rose, the number of physical qubits per logical qubit, the logical error rate fell exponentially instead of climbing. That exponential fall is the theoretical requirement for scalable correction, and Willow was the first convincing experimental sighting of it. It does not mean fault tolerance is around the corner, the absolute error rates are still too high for real algorithms, but it validates both the surface code and Google’s fabrication quality.

Alice & Bob’s cat qubits, if they scale as theorized, rewrite that arithmetic. Suppress bit-flip errors exponentially in hardware and a simple repetition code handles the phase-flips that remain, potentially cutting the physical-to-logical overhead by one to two orders of magnitude. A fault-tolerant machine that needs 10 to 100 physical qubits per logical qubit instead of 1,000 is a categorically different engineering problem. That is why the IP sitting in Alice & Bob’s cat-qubit patents, developed on French public money, held by a company with roughly €30 million raised in total, is precisely the asset the Cour des Comptes is losing sleep over.

Microsoft’s 2025 announcement of topological qubits built on Majorana fermions would, if it validates, open a third road to lower correction overhead. The scientific community is holding back: Majorana qubits require exotic semiconductor-superconductor hybrids whose fabrication is poorly understood, and Microsoft’s earlier claims in this exact area were retracted after peer review found errors. The 2025 results are more carefully presented, but reproducibility and scalability remain undemonstrated.

So the race to fault tolerance is genuinely open, with no approach showing a decisive edge at scale. That openness is what makes this moment strategically loaded. Whoever cracks fault-tolerant quantum computing first lands in a position like the first nations to master nuclear fission, with reach across cryptography, sensing, simulation, and optimization into every corner of the economy and every layer of national security. France has the scientific foundation to be one of those nations. What it does not yet have is the industrial and policy machinery to turn that foundation into sovereignty that lasts, and that is the gap I would bet against France closing in time. Cat-qubit physics and Nobel pedigrees will not save a company that gets sold in a down round while the ministries are still arguing over whose mandate covers it.