Alice & Bob, the French Quantum Computing Company Building FTQC

Alice and Bob is a French quantum computing company building a fault-tolerant machine around an unusual hardware bet called the cat qubit. Founded in Paris in 2020, the company argues that protecting qubits against one entire class of error at the physical level can dramatically cut the number of physical qubits needed for error correction. That thesis has attracted backing from French state-linked investors and, in 2026, from NVIDIA’s venture arm, alongside a published roadmap aiming at 100 logical qubits by 2030. This profile reviews what the company does, where it came from, how its technology works, and how it sits among superconducting and error-correction rivals. Key takeaways1. Cat qubits at the core. Alice and Bob bets that hardware-level bit-flip protection slashes error-correction overhead. The approach leaves mainly phase-flip errors to be corrected by software. 2. French roots, dual presence. The company was founded in Paris in 2020 by Theau Peronnin and Raphael Lescanne. It operates from Paris and Boston. 3. Record bit-flip stability. In September 2025 the company reported bit-flip times exceeding one hour. That figure dwarfs the 430-second record it set in 2024. 4. NVIDIA-backed Series B. NVIDIA’s NVentures joined an expanded Series B round built on a 100 million euro base. Cumulative funding stands near 130 million euros. 5. A 2030 target. The roadmap aims for an early fault-tolerant machine with 100 logical qubits by 2030. The company groups its hardware into named chip families ending with Graphene. 6. Distinctive among rivals. Most superconducting peers use transmon qubits that suffer both error types equally. Alice and Bob’s bosonic encoding is a different structural choice.What Alice and Bob doesThe cat qubit bet The company is a quantum hardware company focused on a single architectural idea, the superconducting cat qubit. The company’s central claim is that a qubit can be engineered to resist one entire family of error, the bit flip, almost entirely at the physical level. By suppressing bit flips in hardware, the team argues that far fewer physical qubits are needed to build a reliable logical qubit through error correction. That overhead reduction is the commercial pitch the company has repeated since its earliest funding rounds, as covered in its brand and funding update. Why overhead matters Building a useful quantum computer is largely a problem of error correction. Standard approaches spend a large fraction of their physical qubits encoding and protecting each usable logical qubit. The company frames its mission around shrinking that ratio, and the company has described hardware-efficient designs that aim to cut the qubit count required for a fault-tolerant machine. The principle behind that overhead, and why it dominates quantum engineering, is set out in our explainer on quantum error correction. A hardware-software split The company positions cat qubits as a way to rebalance the work between hardware and software. If hardware kills off bit flips, then the error-correcting code on top only has to chase the remaining phase flips. That lets the code be simpler and smaller than the two-dimensional schemes used by qubits that suffer both error types. The company’s product line, including its quantum cloud access, is built to demonstrate this division of labour at growing scale.Origins as a Paris quantum spinoutA Paris quantum spinout Alice and Bob was founded in 2020 in Paris by Theau Peronnin and Raphael Lescanne. Peronnin serves as chief executive and Lescanne as chief technology officer. Both founders came from French academic physics, and the company is rooted in quantum optics and superconducting circuit research carried out in the country’s public laboratories. Academic foundations The cat qubit idea the company commercialises grew out of work tied to the Ecole Normale Superieure and France’s national research institutes. The founders translated laboratory results on stabilised bosonic states into a startup with a hardware roadmap. That academic lineage remains central to the company’s identity, and it has continued to publish technical results in peer-reviewed venues rather than relying only on press announcements. Paris and Boston While the company is French by origin and keeps its core laboratories in Paris, it also operates from Boston. The Boston presence gives it a foothold in the United States quantum ecosystem and closer access to American partners and customers. The dual-location structure is reflected in how the company describes itself, as a Paris and Boston based business pursuing a single fault-tolerant goal. This footprint has grown alongside its headcount, which surpassed 200 employees by April 2026.How cat qubits workWhat a cat qubit is A cat qubit encodes quantum information in the states of a superconducting microwave resonator rather than in a simple two-level circuit. The name refers to Schrodinger’s cat, because the qubit stores information in superpositions of coherent states. These bosonic encodings are deliberately engineered so that bit-flip errors become exponentially unlikely as the size of the encoded state grows. The company’s first logical qubit was built from 18 cat qubits, as detailed in our report on how cat qubits encode a logical qubit. Bit flips out, phase flips left The defining feature of the design is that it trades a balanced error profile for a lopsided one. Instead of suffering bit flips and phase flips at similar rates, a cat qubit suppresses bit flips so strongly that mainly phase flips remain. That remaining error can then be corrected with a simpler, lower-dimensional code rather than the larger two-dimensional codes that conventional qubits require. The stabilisation that keeps the cat states alive runs continuously, an autonomous process rather than a constant round of measurement and repair. How it differs from transmons Most superconducting quantum computers use transmon qubits, which are anharmonic two-level circuits that experience both error types at comparable rates. Its bosonic encoding is a structurally different choice that pushes much of the protection into the physical layer. The company argues this reduces the hardware required for a fault-tolerant machine compared with transmon-based approaches. For context on the broader transmon landscape, see our guide to superconducting quantum computing companies. How the stabilization works The protection that defines a cat qubit comes from a process called two-photon driven dissipation, which runs continuously rather than as a discrete correction step. The encoded resonator, or memory mode, is coupled to an auxiliary circuit known as a buffer mode, and that coupling is engineered so only pairs of photons are exchanged with the environment. Because photons can leave and enter only in twos, the system is pinned to a manifold spanned by the two coherent states the qubit uses. This autonomous stabilisation is what keeps the cat states alive without constant measurement, and it is the mechanism the company has built its hardware around. Why bit flips vanish exponentially A bit flip in this encoding means hopping between the two coherent states, and that transition requires the system to pass through intermediate states that the two-photon dissipation keeps suppressing. As the average photon number in the cat grows, the two states separate further in phase space, and the probability of that hop falls exponentially with that photon number. Increasing the cat size therefore buys rapidly improving bit-flip protection, which is the lever Alice and Bob pulls to reach long bit-flip lifetimes. This exponential scaling is the structural reason the company can claim such large reductions in one error channel. The price of protection The trade is not free, because suppressing bit flips so aggressively makes the other error channel worse. Phase-flip errors rise roughly in proportion to the photon number, so the same move that exponentially crushes bit flips only linearly increases phase flips. That leaves phase flips as the dominant remaining error, which an error-correcting code must still handle, though a simpler one-dimensional repetition code can do so rather than a full two-dimensional surface code. Managing that residual phase-flip rate at scale is the central engineering task the architecture leaves open.The Boson chips and the bit-flip recordsAlice and Bob’s Boson chip, the superconducting platform on which its cat qubits set record bit-flip lifetimes.The Boson chip line The company groups its hardware into named families, and the Boson line represents its first milestone, mastering a single reliable cat qubit. The fourth iteration, Boson 4, became the platform on which the company set its early stability records. Our coverage of the Boson 4 chip describes how this device demonstrated cat qubits that resisted bit flips far longer than ordinary superconducting qubits. The Boson series was the company’s proof that the core building block was reproducible. A first logical qubit Beyond a single physical qubit, the company moved toward encoding a logical qubit, the unit that error correction is meant to protect. The company reported assembling its first logical qubit from 18 cat qubits, a low count compared with the hundreds or thousands that some schemes demand. That efficiency is the practical payoff the company has promised, since fewer physical qubits per logical qubit means a smaller machine for a given capability. Bit-flip times past one hour The headline result came on September 25, 2025, when the company reported bit-flip times exceeding one hour. That figure compared against the previous record of 430 seconds, roughly seven minutes, that it had set in 2024 on the Boson 4 chip. The company noted that the new result comfortably cleared the 13-minute bit-flip threshold its 2030 device is designed to need, as our article on the hour-long bit-flip result explains. The Boson progression and external access The Boson family advanced through several iterations toward the Boson 4 chip, on which the company reported a bit-flip lifetime exceeding seven minutes, described as the longest of any superconducting qubit at the time. The company made Boson 4 available for users to reproduce its published results, offering access over Google Cloud so outside researchers could test the device. That openness, exposing the hardware to external reproduction, is part of how the company has argued its results are robust rather than one-off demonstrations. The peer-reviewed record The company has published in peer-reviewed venues rather than relying only on announcements. In May 2024 its work appeared in Nature under the title “Quantum control of a cat-qubit with bit-flip times exceeding ten seconds,” carried out with the QUANTIC team spanning Mines Paris, the Ecole Normale Superieure, and Inria. The paper reported controlling the phase of coherent superpositions while keeping macroscopic bit-flip times, a combination needed to turn a protected memory into a usable qubit. The collaboration with public laboratories reflects the academic methodology behind the company’s hardware claims. Rounds and totals The company has raised funding across several rounds since its founding, with a Series B that closed at around 100 million euros. By April 2026 the company described cumulative funding of roughly 130 million euros, alongside a workforce of more than 200 people. Our report on the 130 million euro funding position places those figures in the context of the company’s brand refresh and scale-up. NVIDIA joins the round In May 2026 the company announced that NVentures, NVIDIA’s venture capital arm, had invested in an expansion of its 100 million euro Series B. The financial size of NVIDIA’s specific stake was not disclosed. The two companies had already been collaborating on software and tooling, so the investment formalised an existing technical relationship, as covered in our piece on the expanded 100 million euro round. State support and selection French public backing has shaped the company’s trajectory beyond private rounds. It has been selected into national quantum programmes connected to France’s Plan Quantique, and the NVIDIA announcement coincided with a further state commitment to that sovereign strategy. The company was also chosen as one of several hardware developers in France’s PROQCIMA initiative, a procurement-driven push toward a large fault-tolerant prototype. These selections are programme placements rather than simple cash grants, and they underline the strategic weight France places on domestic quantum hardware. Seed and Series A The funding history starts with a seed round in 2020, the year the company was founded, which raised about 3 million euros backed by the French venture firms Elaia and Breega. That seed capital let the founders move the cat-qubit idea out of the laboratory and begin building a hardware team in Paris. The company’s Series A followed in March 2022 and raised 27 million euros, reported as roughly 30 million dollars at the time. The Series A was led by Elaia, which was already an investor, together with Bpifrance through its Digital Venture fund and Supernova Invest, with earlier backer Breega and business angels also taking part. That round, less than two years after the seed, funded the push toward the company’s first commercial cat-qubit systems. Across these rounds the investor base has stayed heavily French and state-linked, with Bpifrance recurring, which fits the company’s role inside France’s national quantum ambitions.The roadmap to 2030The 2030 target The company has published a five-stage roadmap toward what it calls an early fault-tolerant quantum computer. The endpoint is a machine with 100 logical qubits aimed at materials-science and early industrial use by 2030. The company frames this not as a universal supercomputer but as a first useful fault-tolerant system, a deliberately scoped goal tied to specific applications. Milestones by chip family The roadmap names its stages after chip families that escalate in capability. Boson represents mastering the cat qubit, achieved with the Boson series in 2024, and Helium represents building a working logical qubit below the error-correction threshold. The final stage, Graphene, is meant to deliver the 100 logical qubits that define the 2030 target. Each named family marks a concrete engineering milestone rather than a marketing label, and the September 2025 stability result was presented as evidence the early steps are on track. Partnerships and integration To reach high-performance computing centres, The company has partnered on the software and integration layer. In June 2026 it announced a deepened alliance with Bull, formerly Eviden, to bring cat-qubit processors into high-performance computing and to extend Bull’s Qaptiva platform for cat-qubit emulation and hybrid execution. That partnership built on an earlier collaboration first announced in 2023, and it complements the company’s separate software work with NVIDIA on hybrid quantum and GPU tooling.What cat qubits could doWhat an early machine would target The company frames its 2030 goal as an early fault-tolerant machine rather than a universal one, scoped to problems where a modest number of high-quality logical qubits could matter. The company points to materials science as a leading target, the kind of problem where simulating quantum systems on quantum hardware is expected to pay off first. Chemistry and certain optimization problems sit in the same category of intended use, since they map naturally onto quantum simulation and search. These are intended markets the company describes for its first useful system, not claims of solved applications today. Cloud and emulator access Even before that hardware arrives, The company has built ways for outside users to work with its technology. It made the Boson 4 chip reproducible over Google Cloud, letting researchers run the same experiments behind its published results. It also launched a logical-qubit emulator called Felis, made publicly available through the French provider OVHcloud, which lets users model the behaviour of logical qubits with tunable error rates before real fault-tolerant hardware exists. That emulator approach lets potential customers prototype algorithms against the error model the company expects to deliver. Early users and deployments On named users, the clearest verifiable case is institutional rather than a roster of named enterprise customers. France’s national high-performance computing organisation GENCI agreed to procure an on-premise system from Alice and Bob, an 18-qubit machine to be hosted at the CEA, with deployment described as a milestone for French and European research. A broad list of named commercial pilot customers is not something reliable sources confirm, so it would be wrong to claim one. The honest picture is that early access today runs through the cloud, the emulator, and public-sector procurement rather than disclosed private deployments.Where Alice and Bob fitsAmong the superconducting players The company competes in a field led by large superconducting programmes at IBM and Google, both of which build transmon-based processors and pursue their own error-correction milestones. Those companies command far larger budgets and qubit counts, but they carry the full overhead of correcting both bit and phase flips. Its wager is that a narrower, hardware-protected design can leapfrog the overhead problem rather than simply outspend it, a contrast sharpened in our overview of superconducting quantum computing companies. The error-correction race Beyond raw qubit counts, the competitive frontier is error correction, where players such as Quantinuum with trapped ions and QuEra with neutral atoms pursue different physical platforms entirely. Each architecture makes its own bet about which errors are easiest to suppress and which codes are cheapest to run. The closest parallel is D-Wave’s dual-rail qubit, another superconducting design that builds error detection into the hardware to cut the same overhead. Its bosonic route sits alongside these as a distinct answer to the same question, profiled in depth on EntangledFuture’s Alice and Bob page, namely how to reach a protected logical qubit without an unaffordable hardware bill. The shared goal across all of them is explained in our guide to quantum error correction. Why cat qubits stand out What distinguishes Alice and Bob is the decision to fix the error profile in hardware before correction begins, rather than treating all errors as equal and throwing more qubits at the problem. That choice carries risk, since the remaining phase-flip errors must still be tamed and the whole stack must scale. If the bet holds, however, the company’s roadmap implies a smaller machine for a given capability than transmon-heavy rivals, which is the entire basis of its pitch to investors and partners.The challenges aheadThe unsolved part The cat-qubit bet rests on suppressing bit flips so that only phase flips remain, but correcting those phase flips at scale is the part that is not yet demonstrated in a large machine. The architecture deliberately concentrates the remaining error into one channel, which is elegant, yet a one-dimensional code still has to keep that channel below threshold across many logical qubits. Going from a single logical qubit built on 18 cat qubits to 100 logical qubits with reliable operations between them is a large jump. That scaling, not the single-qubit lifetime, is where the hardest physics and engineering questions still sit. A smaller company in a giant field The company is far smaller than the rivals it is implicitly racing, with a few hundred employees against the deep budgets of IBM and Google. Those competitors run transmon programmes that are more mature and operate larger processors, and transmon qubits are simply better proven at scale than cat qubits today. A focused architectural advantage can compensate for size only if it holds up as systems grow, and that is not guaranteed. The company’s heavy reliance on French state-linked funding is both a support and a dependency, since its trajectory is tied to national programmes. What would have to go right For the 2030 target to land, several things must align, and each is plausible but none is certain. The phase-flip code has to scale cleanly, the logical-qubit count has to grow without runaway error, and the integration with high-performance computing through partners such as Bull, formerly Eviden, has to deliver usable systems. The September 2025 bit-flip result and the Nature publication are real evidence the early steps are working, which is the case for optimism. The case for caution is simply that quantum roadmaps across the industry have a long record of slipping, and a 2030 useful machine remains an ambitious bet rather than a settled outcome.Founded2020, ParisCEOTheau PeronninHeadquartersParis, France and Boston, USATechnologySuperconducting cat qubitsFunding raised~130 million euros cumulative (April 2026)WebsiteAlice and BobNotableBit-flip times exceeding one hour (Sept 2025); NVIDIA NVentures investment (2026)Read nextTop superconducting quantum computing companiesWhat is quantum error correctionAlice and Bob’s hour-long bit-flip resultThe Boson 4 cat-qubit chipAlice and Bob FAQWho founded Alice and Bob and when?Alice and Bob was founded in 2020 in Paris by Theau Peronnin and Raphael Lescanne. Peronnin is chief executive and Lescanne is chief technology officer. The company grew out of French academic work in quantum optics and superconducting circuits.What is a cat qubit?A cat qubit encodes quantum information in the states of a superconducting microwave resonator, using superpositions named after Schrodinger’s cat. The encoding is engineered so that bit-flip errors become exponentially unlikely. That leaves mainly phase-flip errors to be handled by error-correcting software.How do cat qubits differ from transmon qubits?Transmon qubits suffer bit flips and phase flips at similar rates, so their error correction must handle both. Cat qubits suppress bit flips in hardware, leaving a lopsided error profile dominated by phase flips. Alice and Bob argues this reduces the physical qubits needed for a fault-tolerant machine.What did Alice and Bob achieve in September 2025?On September 25, 2025, the company reported bit-flip times exceeding one hour. That result compared with a previous record of 430 seconds, about seven minutes, set in 2024 on the Boson 4 chip. The company said it cleared the 13-minute threshold its 2030 device is designed to require.How many cat qubits make up its first logical qubit?Alice and Bob reported building its first logical qubit from 18 cat qubits. That is a low count compared with schemes that need hundreds or thousands of physical qubits per logical qubit. The figure reflects the overhead reduction the company has promised.How much funding has Alice and Bob raised?By April 2026 the company described cumulative funding of roughly 130 million euros. Its Series B closed at around 100 million euros, and in May 2026 NVIDIA’s NVentures invested in an expansion of that round. The size of NVIDIA’s specific stake was not disclosed.What is the 2030 roadmap?Alice and Bob aims to deliver an early fault-tolerant quantum computer with 100 logical qubits by 2030. The roadmap runs through named chip families, ending with one called Graphene. The target is framed around early industrial and materials-science use rather than universal computing.Who are Alice and Bob’s main partners?Alice and Bob deepened an alliance with Bull, formerly Eviden, in June 2026 to integrate cat-qubit processors into high-performance computing and extend the Qaptiva platform. It also works with NVIDIA on hybrid quantum and GPU tooling. NVIDIA’s venture arm additionally invested in the company in 2026. Stay currentSee today’s quantum computing news on Quantum Zeitgeist for the latest breakthroughs in qubits, hardware, algorithms, and industry deals. Tags: