Quantum Computing Modalities: Photonic Cluster-State

(For other quantum computing modalities and architectures, see Taxonomy of Quantum Computing: Modalities & Architectures)

What It Is

Photonic Cluster-State Computing is a form of quantum computing in which information is processed using photons (particles of light) that have been prepared in a highly entangled state known as a cluster state. It falls under the modality of measurement-based quantum computing (MBQC), often called the one-way quantum computer. In this scheme, a large entangled resource state (the photonic cluster state) is generated first, and then the computation is carried out by performing a sequence of single-qubit measurements on the individual photons. The cluster state’s entanglement serves as the “fuel” for the computation, and it is gradually consumed as measurements proceed — hence the name “one-way” (the entangled resource is used up and cannot be reused). Each photon is measured in a particular basis chosen according to the algorithm’s needs, and those measurements drive the quantum computation.

This approach differs from the traditional gate-based model of quantum computing. In a gate-based (circuit) model, one applies a series of unitary quantum logic gates (such as CNOTs, Hadamards, etc.) directly to the qubits over time. In cluster-state computing, by contrast, all multi-qubit entanglement is created up front in the initial cluster state, and quantum logic operations are effected by measurements rather than by multi-qubit gates applied sequentially. Once the cluster state is prepared, the only operations needed are single-qubit measurements (plus classical feedforward as explained later); no further entangling gates are required during the computation. Photonic cluster-state computers perform quantum algorithms by measuring photons in the right bases, instead of actively shuttling photons through logic gates at runtime. The outcome of each measurement is inherently random, but the cluster-state protocol includes classical processing to adjust later measurements based on earlier results so that the overall computation yields a deterministic, correct result.

In summary, photonic cluster-state computing is a measurement-driven quantum computing approach that uses photons entangled in a cluster state as the substrate for computation. It is the leading implementation of MBQC, leveraging photons’ ease of distribution and measurement. After preparing an entangled photonic cluster (often a 2D lattice or graph of entangled photons), one performs a carefully chosen sequence of single-photon measurements. Through quantum correlations, these measurements enact a quantum circuit’s logic and produce the desired output state on some unmeasured qubits. The role of the entangled cluster is analogous to a blank quantum program: by “writing” on it with measurements, one can realize any quantum algorithm. Once all required measurements are done, the cluster state is collapsed (destroyed), and the computation is complete.

How It Differs From Photonic Quantum Computing

It is important to clarify the distinction between photonic cluster-state computing and photonic quantum computing in general. Photonic quantum computing refers broadly to any quantum computing approach that uses photons as the information carriers (qubits). This encompasses many possible architectures: not only the cluster-state (one-way) model, but also traditional gate-based schemes implemented with linear optics, analog and adiabatic photonic processors, continuous-variable quantum computing with squeezed light, and specialized photonic quantum simulators. In contrast, the photonic cluster-state model is a specific subset of photonic quantum computing: it denotes the use of the one-way measurement-based scheme (cluster states + measurements) on photonic qubits.

Not all photonic quantum computers use cluster states, though all photonic cluster-state computers are photonic quantum computers. For example, one prominent photonic approach, the Knill-Laflamme-Milburn (KLM) scheme, is gate-based: it employs linear optical circuits and ancillary photons to probabilistically implement quantum logic gates, rather than relying on a pre-prepared cluster. Another example is boson sampling devices, which send photons through a fixed interferometer; these are photonic quantum processors designed for a specific computational task but do not involve cluster states or adaptive measurements. Similarly, the photonic experiments that achieved a quantum computational advantage in 2020–2022 did so by injecting many photons into large interferometer networks for Gaussian boson sampling, without using the one-way model. Those systems fall under photonic quantum computing, but not under the cluster-state modality.

Photonic cluster-state computing specifically means that photons are used to form a cluster state and computation is done via one-way measurements on that state. Photonic implementations of the cluster model became attractive in the mid-2000s because of the difficulties in directly executing two-qubit gates with photons. Photons interact only weakly with each other, which makes deterministic logic gates challenging; early optical quantum computing proposals (like KLM) required complex tricks (teleportation-based gates with many auxiliary photons and detectors) and had very low success probabilities. The cluster-state approach offers a way around this: rather than performing gates between arbitrary photon pairs during the algorithm, one initially uses interference and measurements to create a fixed entangled resource (the cluster) and thereafter needs only single-photon operations. This simplifies the photonic hardware requirements during the computational phase. The hard part (entangling photons) is concentrated into the state preparation step. Once you have a large photonic cluster state, you can realize any quantum circuit by appropriate measurements, avoiding the need for on-demand photon-photon gates in real time.

A significant evolution in the field has been the emergence of fusion-based quantum computing (FBQC), a variant of the cluster-state approach in which small entangled resource states (such as 3- or 4-photon states) are continuously generated and then fused together by projective measurements to build up a larger computational graph. PsiQuantum’s architecture follows this FBQC model, as does work by several academic groups. FBQC retains the core idea of measurement-based computing while making the state preparation more modular and loss-tolerant, since each small resource state can be independently generated and a failed fusion attempt only destroys one small piece rather than a large pre-assembled cluster.

To summarize, general photonic quantum computing can take many forms, whereas the photonic cluster-state model refers to one specific, universal approach (the MBQC one-way approach, including its FBQC variant) implemented with photonic qubits. All photonic cluster computers are part of photonic quantum computing, but many photonic quantum computing experiments (like those involving direct linear-optical gates or boson sampling) do not involve cluster states. The cluster-state model has become the leading approach within photonics due to its suitability for photons’ strengths and weaknesses: it plays to photons’ ease of creation and measurement while sidestepping the lack of easy two-photon interactions by using entanglement + measurements as an alternative to gates.

Key Academic Papers

Photonic cluster-state computing was built on a foundation of theoretical proposals and experimental breakthroughs in the early 2000s. Some of the foundational and influential papers that introduced and developed this field include:

• Raussendorf & Briegel (2001) – “A One-Way Quantum Computer.” The seminal paper that introduced the one-way measurement-based model of quantum computing. The authors proposed the use of cluster states as a universal resource for quantum computation, showing that any quantum circuit can be executed by preparing a suitable entangled cluster and performing adaptive single-qubit measurements. This work established the theoretical framework for MBQC and identified cluster states (first defined in 2001) as the key resource for one-way computing.
• Nielsen (2004) – “Optical Quantum Computation Using Cluster States.” Michael Nielsen proposed applying the one-way model to photonic systems. His work bridged the abstract cluster-state theory with the existing Knill-Laflamme-Milburn (KLM) approach in optics. Nielsen outlined how single photons, beam splitters, phase shifters, and photon detectors (with feed-forward) could create cluster states and perform quantum computing, in principle deterministically. This was among the first proposals that linear optics combined with the cluster modality could achieve scalable quantum computing, avoiding some complexities of earlier optical gate-based schemes.
• Browne & Rudolph (2005) – “Resource-Efficient Linear Optical Quantum Computation.” Daniel Browne and Terry Rudolph (the latter a co-founder of PsiQuantum) built on Nielsen’s idea and introduced a far more efficient method to generate large photonic cluster states using linear optics. They described “fusion gates,” which are measurements that “glue together” smaller entangled states (like Bell pairs) into a larger cluster state. Their scheme required only moderate interferometric stability and achieved a higher success probability than prior approaches. This paper demonstrated universality of the cluster approach with linear optics and showed that one could scale up by fusing many small entangled photon pairs into a big cluster. The fusion-gate concept remains at the core of PsiQuantum’s architecture today.
• Walther et al. (2005) – “Experimental One-Way Quantum Computing.” This Nature paper by Philip Walther and colleagues (including Zeilinger) was the first experimental demonstration of one-way quantum computing. They used four photons entangled in a linear cluster state and performed one-qubit and two-qubit quantum gate operations via adaptive measurements, including a simple Grover’s search algorithm. Walther et al.’s experiment showed that the one-way model works in practice, albeit on a small scale: they reported successful implementation of a universal set of gates and a proof-of-principle algorithm using a photonic cluster state.
• Prevedel et al. (2007) – “Experimental Realization of Deutsch’s Algorithm in a One-Way Quantum Computer.” This experiment demonstrated a full quantum algorithm in the one-way model using photonic cluster states. The team implemented Deutsch’s algorithm on a four-qubit optical cluster state, with all possible function outcomes tested successfully. This was the first time an algorithm (beyond basic gates) was run on a photonic one-way quantum computer.
• Recent Advances (2013–2025) – In the years since, there have been many important developments across both continuous-variable and discrete-variable photonics. Yokoyama et al. (2013) and Larsen et al. (2019) (along with Asavanant et al. (2019)) demonstrated the generation of large-scale photonic cluster states in the continuous-variable domain using squeezed light, reaching thousands or more entangled modes. On the single-photon side, Economou, Lindner & Rudolph (2010) proposed methods for generating 2D cluster states using quantum dot emitters, and Gimeno-Segovia et al. (2015, 2019) developed architectures for assembling large cluster states (including a 3D cluster for fault tolerance) using linear optics and small entangled blocks. In 2023, Cogan et al. reported deterministic generation of a photonic cluster state from a quantum dot source, producing indistinguishable photons at gigahertz rates with an entanglement length of about ten photons. In 2024, Ferreira et al. (Caltech) published in Nature Physics the first deterministic generation of a 2D photonic cluster state from a single quantum emitter in the microwave domain. Building on that work, O’Sullivan et al. (ETH Zurich, 2025) published in Nature Communications the deterministic generation of a 20-qubit 2D multi-photon cluster state using coupled superconducting transmon qubits, achieving fidelities above 0.50 for up to eight qubits and nonzero localizable entanglement for states of up to 16 qubits. A 2024 paper in Nature Photonics demonstrated resource-efficient photonic quantum computation with high-dimensional cluster states, generating clusters with up to 9.28 qubits per photon at 100 Hz by entangling multiple qudits within each photon’s Hilbert space. And in June 2025, researchers at Quandela and collaborators published in Nature Communications the deterministic and reconfigurable generation of caterpillar graph states from a semiconductor quantum dot in a cavity. These advances show rapid progress toward creating bigger, more stable, and more versatile photonic cluster states.

Each of the papers above has played a key role. Raussendorf and Briegel provided the blueprint of MBQC; Nielsen, Browne, Rudolph and others adapted it to photonic hardware; and Walther, Prevedel, and subsequent experimental teams validated it on actual photons. Together, these works established photonic cluster-state computing as a real and promising modality for quantum computation.

How It Works

The one-way quantum computing model implemented by photonic cluster states operates in a way that is quite different from the circuit model. Here is an outline of how photonic cluster-state computation works:

• Cluster State Preparation: First, a collection of photons is prepared in an entangled state that has a specific graph structure — the cluster state. Typically, each photon (qubit) is initially prepared in a superposition state (like $$|+\rangle = (|0\rangle+|1\rangle)/\sqrt{2}$$), and then pairs of photons are entangled (for example, via controlled-phase (CZ) gates or equivalent entangling operations implemented by optical interference). One can think of a cluster state as a network of qubits (nodes) with entanglement bonds (edges) connecting them, often arranged in a lattice or grid. In photonic implementations, entangling operations might be done by interfering photons on beam splitters and using ancillary measurements (per the Browne-Rudolph fusion gates or related methods) to “knit” photons into the desired graph. In PsiQuantum’s fusion-based architecture, many small resource states are generated independently and then fused into the larger computational graph through projective measurements. The result of this step is a large entangled state — for instance, a 2D cluster state that looks like a mesh of photons entangled with nearest neighbors. This entangled cluster is the core resource for computation, analogous to a blank tape or a prepared quantum memory that will be processed by measurements.
• Encoding the Problem: The algorithm or computational task is encoded by choosing how and in what order to measure the qubits in the cluster state. Before measurements, the cluster does not “know” what computation it will perform; it is a universal resource. Planning a computation means deciding a sequence of measurement bases for all (or most) of the photons in the cluster, such that the effective quantum circuit induced by those measurements corresponds to the desired algorithm. For example, measuring a photon in the $$X$$ basis (i.e., $$|+\rangle/|-\rangle$$ basis) versus the $$Z$$ basis ($$|0\rangle/|1\rangle$$) will have different effects on the remaining cluster. In the one-way model, logical qubits are typically associated with certain lines of photons in the cluster, and measuring a photon in a particular basis can implement a quantum gate on the state of the unmeasured photons. At the start, the input to the computation can be encoded in some subset of the photons (e.g., by preparing those photons in states representing the input, or by entangling them with input states provided by the user).
• Single-Qubit Adaptive Measurements: Computation proceeds by measuring the photons one by one (or sometimes a few at a time). The choice of measurement basis for each photon may depend on the outcomes of earlier measurements — this is the adaptive or feed-forward aspect of MBQC. When a photon is measured, it “collapses” and is removed from the cluster, but its entanglement with the remaining photons causes a quantum gate operation to be enacted on the state of the remaining cluster (by the laws of quantum mechanics, this is equivalent to a teleportation-like effect where the state of one qubit is transferred to another with some applied operation). Because each measurement yields a random outcome, the specific operation applied can have an unintended byproduct (like an extra Pauli $$X$$ or $$Z$$ error) that depends on the outcome. The solution is to adjust subsequent measurements to account for those random flips. After measuring one photon, the classical control system records the result and uses that information to set the basis for the next photon’s measurement. This feedforward loop is repeated throughout the computation. By the end of the sequence, all the random outcomes will have been accounted for by these adjustments, ensuring the correct overall transformation has been applied to the remaining qubits.
• Entanglement as the Fuel for Gates: In the one-way model, entanglement replaces dynamic gates. The cluster state’s entangled bonds carry the quantum correlations needed for gates. When you measure a photon, you break its entanglement links with the cluster, and this act drives the computation forward. A useful way to visualize this is via quantum teleportation: measuring entangled qubits in the right basis can “teleport” quantum states from one part of the cluster to another, with certain quantum gates enacted during the teleportation depending on the measurement basis. The one-way computer performs a sequence of teleportations through the cluster state, each teleportation applying a desired gate to the logical qubits. The entangled cluster is the conduit through which quantum information flows and gets transformed. This is why a sufficiently connected cluster state can, in theory, perform any computation.
• Classical Control (Feedforward): A classical computer is an integral part of a photonic cluster-state quantum computer. It is used to perform the feedforward of measurement results. After each measurement, the outcome (a classical bit) is processed to determine how later measurements should be adjusted. This conditional logic is pre-computed classically from the known quantum circuit being implemented. In hardware, fast electronics (or sometimes optical modulators controlled by electronics) take the detector signals and apply the corresponding changes (e.g., rotating a polarization measurement basis or phase setting for the next photon). Feedforward is essential: without it, the randomness of quantum measurements would make the outcome probabilistic rather than deterministic. With feedforward, one can ensure the right gate is applied regardless of each measurement’s chance result, so that the correct final answer is obtained with near 100% probability (assuming no errors). Sakaguchi et al. (RIKEN) demonstrated high-speed feedforward control in an optical one-way quantum computing experiment, including nonlinear feedforward that allows implementation of a broader class of gates in the continuous-variable domain. This kind of real-time classical control loop is a distinctive feature of MBQC implementations.
• Output: After a series of measurements, only a few photons remain unmeasured — these carry the output quantum state. In many algorithms, you eventually measure those output qubits in the computational basis to get a classical result (for example, the answer to the problem). Because of the adaptive corrections applied during the computation, the output will be the same as if the corresponding circuit had been executed on a gate-based quantum computer. By the end, almost all photons in the cluster have been measured; the entanglement that initially permeated the cluster has been consumed to perform computations (hence “one-way”). This is in contrast to a gate-based computer where the qubits persist and are manipulated throughout. In a photonic one-way computer, once a photon is measured, it is gone from the quantum register, but its influence persists in the entangled state of the remaining photons (until they too are measured).

To illustrate with a simple example: suppose we want to perform a two-qubit gate between logical qubit A and B. In a cluster-state computer, we might have a line of entangled photons representing qubit A’s timeline and another line for qubit B. At some point, these lines will have been entangled (perhaps they share an entangled neighbor or a connecting photon between them in the cluster graph). By measuring an intermediate photon in a certain basis, one can enact an effective CNOT or CZ gate between the logical qubits. This was exactly how Walther et al. and subsequent experimental teams implemented two-qubit gates and small algorithms in their cluster experiments: the pattern of measurements and the entanglement in the cluster together realize the gate operations without directly interacting those logical photons at that moment.

In summary, photonic cluster-state computing works by first creating a multi-photon entangled web (the cluster), then “processing” that web by measuring each photon one at a time in carefully chosen ways. The quantum entanglement in the web ensures that each measurement has non-local effects on the state of the remaining photons, effectively performing gate operations. Through adaptive choice of measurement bases (feedforward of prior outcomes), one can correct for randomness and guide the computation to yield a deterministic result. The combination of quantum entanglement (for generating correlations) and classical control (for adaptively steering the measurement sequence) is what makes the one-way model work. All the “quantum difficulty” is concentrated in the cluster state: once that is prepared, the measurements (though quantum) are relatively straightforward operations. This trait is especially appealing for photonic systems, where generating a large entangled state can be done via parallel processes, and thereafter one need only detect photons (which can be done rapidly and in parallel).

Comparison to Other Modalities

Photonic cluster-state computing (the one-way MBQC model) can be contrasted with other major quantum computing modalities, chiefly the gate-based circuit model and adiabatic/annealing quantum computing. Each modality has different requirements and features in terms of scalability, fault tolerance, and practical challenges.

Gate-Based (Circuit) Model vs. One-Way (Cluster) Model

In the gate-based approach (used by most superconducting, ion trap, and semiconductor qubit platforms), algorithms are executed by applying a sequence of quantum gates on qubits, analogous to a classical circuit. This requires the qubits to be maintained coherently for the duration of the circuit and to be able to undergo two-qubit interactions on demand. In photonics, implementing two-qubit gates deterministically is notoriously difficult because photons do not naturally interact; one typically needs either special nonlinear materials or measurement-induced effects with additional photons, which are probabilistic. The one-way model offers a different route: all entangling operations are done upfront (which can be attempted many times in parallel if probabilistic), and thereafter only single-qubit measurements are needed. One-way quantum computers do not require any on-the-fly two-qubit gates, unlike circuit computers. This gives one-way computing a hardware advantage for photonics: it trades the problem of reliably performing many sequential gates for the problem of preparing a large entangled state once. The computational “steps” in one-way computing — single-qubit measurements — are relatively simple and can be fast and high-fidelity, in contrast to complex multi-qubit logic gates.

Another difference is in how algorithms are designed: gate-model algorithms are sequences of gates, whereas one-way algorithms are designed as measurement patterns on a given cluster graph. Mathematically, the two models are equivalent in power (anything doable with gates can be done with a cluster and vice versa), but the resource accounting differs. Gate-based photonic computing (e.g., the original KLM scheme) would require enormous overhead of additional photons and extremely low-loss circuits, since each two-qubit gate might only succeed a small fraction of the time, and many attempts must be buffered and coordinated. One-way photonic computing shifts this overhead to the initial state: generating a large cluster state may require many photons and entangling attempts, but once it is generated, using it is straightforward. Researchers often consider the cluster model more natural for optics, since one can entangle a large number of photons (for example, via parametric down-conversion sources or optical combiners) in parallel, whereas in a gate model you would have to interact qubits pairwise in series which is tough if each interaction is probabilistic.

From the perspective of fault tolerance, both modalities can achieve it but by different means. Gate-based computers typically employ quantum error-correcting codes (like surface codes) on physical qubits and perform syndrome measurements periodically using additional qubits and gates. One-way cluster computers can achieve fault tolerance by using special cluster states that have a trellis-like 3D structure corresponding to an error-correcting code (e.g., a 3D cluster implementing a surface code in space-time). The cluster-state model is very compatible with certain topological error correction schemes: one can build a 3D cluster state where making appropriate measurements is equivalent to performing error correction on a surface code. Photonic implementations may find this approach appealing because, instead of a fixed 2D array of qubits doing a surface code, you continuously generate a 3D entangled cluster of photons that is a fault-tolerant error-correcting code, and you consume it as you go.

In summary, the cluster vs. gate-based question in photonics is about trade-offs: cluster-state computing requires preparing a complex entangled state but then uses only measurements (with feedforward), whereas gate-based computing requires the ability to interact qubits arbitrarily throughout the algorithm. For platforms like superconductors or ions, gate-based is natural because qubits can interact via well-defined couplings; for photons, cluster-state may be more natural because interactions can be shifted to a pre-processing stage. Both are universal, but the hardware demands differ.

Adiabatic/Annealing Model vs. One-Way Model

Adiabatic quantum computing (AQC) and its practical variant, quantum annealing, is another modality where instead of logical gates or measurements, one encodes the problem into a Hamiltonian (an energy landscape) and then slowly evolves the quantum system to find the solution (usually the ground state of that Hamiltonian). The leading example is the D-Wave quantum annealer, which uses thousands of superconducting flux qubits to solve optimization problems by energy minimization. In principle, AQC is computationally equivalent to gate-based QC (any gate circuit can be encoded into an adiabatic process), but in practice current annealers are special-purpose, mainly useful for certain optimization or sampling tasks. One-way photonic computers differ from annealers in that they are digital and gate-equivalent (they can run arbitrary algorithms with the right measurements), whereas annealers are analog and typically not used for general algorithms like factoring. A quantum annealer cannot efficiently run Shor’s algorithm or many other quantum algorithms that require a sequence of logic operations. A photonic cluster-state computer could run Shor’s algorithm (in theory) because it is a universal quantum computer.

Another difference is that current annealers require heavy cryogenic analog hardware and are not error-corrected. They rely on an analog process that is somewhat resilient to certain noise, but they cannot correct arbitrary quantum errors and they lack the full capability of a universal gate quantum computer. Photonic one-way computers, by contrast, are being designed with error-correctability in mind (via cluster-based codes). In terms of use cases, annealing works for optimization problems (scheduling, route planning, some machine learning tasks), while one-way photonic computing would be aimed at the broad class of problems quantum computers can tackle, including optimization problems (via algorithms like QAOA or Grover’s algorithm) and also problems like factorization, quantum simulation of physics and chemistry, and more, which annealers cannot efficiently address.

In summary, quantum annealing is a specialized analog modality (exploiting gradual evolution to a solution), whereas cluster-state computing is a universal digital modality (exploiting entanglement and measurement). Each has different challenges: annealers face difficulty in scaling to problems beyond optimization and in ensuring the adiabatic criteria, while cluster-state devices face the difficulty of generating and maintaining large entangled states with feedforward. If one’s goal is to solve generic problems or break cryptography, an annealer will not suffice; one needs a universal quantum computer, which the cluster-state machine is. Conversely, if one is only interested in solving an optimization problem faster, a dedicated annealer might reach useful scale sooner than a general photonic quantum computer because it has simpler goals.

Current Development Status

Research and development in photonic cluster-state computing have accelerated in the past two years. As of mid-2026, the field has moved from small-scale laboratory demonstrations to the early stages of systems integration and manufacturing. Two companies — PsiQuantum and Xanadu — are building datacenter-scale facilities for photonic quantum computers, and a growing roster of smaller companies and academic groups are pushing the boundaries of photon source quality, cluster-state generation, and on-chip integration.

PsiQuantum: Omega Chipset and Fusion-Based Architecture

PsiQuantum, headquartered in Palo Alto, is the most heavily funded photonic quantum computing company, with over $2 billion in total capital including a $1 billion Series E round completed in September 2025 at a $7 billion valuation. The round was led by BlackRock, Temasek, and Baillie Gifford, with participation from NVIDIA NVentures, Macquarie, and Qatar’s sovereign wealth fund.

PsiQuantum’s approach is fusion-based quantum computing (FBQC), a variant of cluster-state MBQC in which small entangled photonic resource states are continuously generated and then fused together by projective measurements to build up a larger computational graph. In February 2025, PsiQuantum published a landmark paper in Nature detailing the Omega chipset, a quantum photonic chipset designed and fabricated on full-size silicon wafers at GlobalFoundries’ 45 nm process fab in Malta, New York. The paper represents over ninety researchers and reports a complete photonic technology stack, including single-photon sources, superconducting nanowire single-photon detectors (SNSPDs), and a next-generation optical switch based on barium titanate (BTO). The benchmarked performance metrics include 99.98% single-qubit state preparation and measurement fidelity, 99.5% two-photon quantum interference visibility (Hong-Ou-Mandel), 99.72% chip-to-chip qubit interconnect fidelity over optical fiber (demonstrated over distances up to 250 m), and 99.22% two-qubit fusion gate fidelity. All fidelities are conditional on photon detection. PsiQuantum uses dual-rail qubit encoding, where a qubit is encoded in the path of a single photon across two waveguides.

An important clarification about temperature: PsiQuantum’s Omega chips integrate superconducting single-photon detectors directly on the die, so the entire chip operates cryogenically at 2 to 4 kelvin. This is a less demanding cooling requirement than the millikelvin regime required by superconducting qubits, but it is still a meaningful infrastructure commitment. The photonic logic itself does not require ultra-low temperatures, but the detectors do, and since PsiQuantum integrates them on-chip rather than in a separate cryostat, the entire chip must be cooled. PsiQuantum’s cryogenic infrastructure uses cuboid racks fed by industrial cryoplants, an architecture that differs from the dilution-refrigerator “chandelier” model used for superconducting qubits.

PsiQuantum broke ground on two datacenter-sized Quantum Compute Centers in September 2025: one in Moreton Bay, Queensland (supported by AU$940 million from the Australian federal and Queensland state governments), and one in Chicago, Illinois. In May 2026, PsiQuantum signed a $100 million Letter of Intent with the U.S. Department of Commerce for proposed CHIPS Act incentives to support domestic manufacturing of photonic quantum computing technologies. The company has also advanced to the final stage of DARPA’s Quantum Benchmarking Initiative (QBI), alongside Microsoft, and has announced defense and aerospace partnerships with Lockheed Martin (November 2025) and Airbus (January 2026). In September 2025, PsiQuantum launched Construct, a software platform for designing, developing, and optimizing fault-tolerant quantum algorithms, which was made open-access in June 2026.

PsiQuantum’s stated goal remains a million-physical-qubit fault-tolerant quantum computer. The company has not published a detailed timeline with intermediate milestones the way IBM or Google have, but their February 2025 Nature paper and the ongoing facility construction indicate a move from component-level validation toward systems integration. Their central strategic argument is the CMOS fabrication thesis: that quantum processors manufactured on existing semiconductor production lines can potentially achieve the yield curves and cost structures of conventional chip manufacturing. That is the scale advantage needed for fault-tolerant computing.

Xanadu: Aurora, GKP Qubits, and Modular Architecture

Xanadu, based in Toronto, has taken a complementary approach, using Gottesman-Kitaev-Preskill (GKP) states — a type of bosonic qubit encoding that embeds error-correctable quantum information in superpositions of many photons — on a silicon photonics platform. In January 2025, Xanadu published in Nature the results of Aurora, a 12-qubit universal photonic quantum computer consisting of four modular and independent server racks photonically interconnected and networked together. Aurora contains 35 integrated photonic chips linked by a combined 13 km of optical fiber, all operating at room temperature (the core photonic processing, though detectors require cryogenic cooling). The published Nature results demonstrated that the main ingredients for a scalable, modular photonic quantum computer are in place: multiple smaller photonic processors functioning as one larger machine, connected through fiber-optic links.

Xanadu’s 2025 results go beyond architecture. The company demonstrated 12 logical GKP qubits with real-time error-correction decoding — a step toward fault tolerance on a photonic platform. In June 2025, Xanadu published a second paper in Nature reporting the first-ever on-chip generation of GKP states, using custom ultra-low-loss silicon nitride waveguides fabricated on 300 mm wafer platforms and photon-number-resolving detectors with detection efficiencies above 99%. GKP states are a particularly attractive qubit encoding for photonics because they enable logic gates and error correction at room temperature using deterministic linear-optical operations, rather than the probabilistic entanglement schemes that other photonic architectures require. Xanadu reported a 60% reduction in optical loss in 2025 and a 20-fold improvement over the previous three years, with loss reduction identified as the primary remaining challenge for fault tolerance.

Xanadu went public in Q1 2026 through a SPAC merger with Crane Harbor Acquisition Corp., raising $302 million in gross proceeds. The company also advanced to Stage B of DARPA’s Quantum Benchmarking Initiative and was selected for Canada’s Quantum Champions Program (CAD $23 million in additional funding). Xanadu’s architecture targets up to 100,000 physical qubits and up to 500 logical qubits in the 2029–2030 timeframe. Through fiscal 2026, the company aims to achieve device performance improvement and early fault-tolerance building blocks, with scaled error-corrected operations and early error-corrected demonstrations expected through 2028. Xanadu’s PennyLane software framework, an open-source library for quantum computing and application development, reached approximately 160,000 average monthly downloads in 2025, a 161% year-over-year increase, and it supports hardware from multiple vendors across modalities.

Other Notable Efforts

Beyond PsiQuantum and Xanadu, a growing field of companies and academic groups is developing photonic quantum computing in various forms:

• Quandela (France) uses quantum-dot single-photon sources and gate-based photonic circuits. In May 2025, Quandela unveiled Belenos, a 12-qubit photonic quantum computer that the company says delivers 4,000 times more computing power than its predecessor. The first integrated version, named Lucy, was inaugurated in April 2026 at the CEA’s Très Grand Centre de Calcul (TGCC) and coupled with the Joliot-Curie supercomputer for the EuroHPC/GENCI consortium. Quandela’s roadmap targets a doubling of qubit count in 2026 with the Canopus system and a 40+ qubit machine within three years. Unlike PsiQuantum and Xanadu, Quandela’s approach is gate-based rather than measurement-based, but its quantum-dot photon source technology and data-center-compatible form factor are relevant to the broader photonic ecosystem.
• ORCA Computing (UK) builds room-temperature photonic systems using time-bin encoded photons stored in quantum memory loops. ORCA’s PT-2 system is deployed with government and academic clients, and the company plans to deliver its PT-3 system in 2026, which it positions as a quantum advantage platform for generative AI workloads. In June 2026, ORCA partnered with Digital Realty to integrate its photonic quantum systems into the London Digital Realty Innovation Lab. ORCA joined NVIDIA’s NVQLink open reference architecture in November 2025 and demonstrated compute time reduction for AI workloads with Toyota in January 2026. While ORCA’s current systems are photonic accelerators rather than universal cluster-state computers, its investment in integrated photonics (through the 2024 acquisition of GXC’s photonics division) and its focus on data-center integration contribute to the photonic quantum computing ecosystem.
• Quantum Source (Israel) is developing a deterministic approach to photonic cluster-state generation. In September 2025, the company unveiled ORIGIN, a proprietary resource-state generator built on cavity-QED photon-atom gate technology. Rather than relying on probabilistic photon sources, ORIGIN uses single atoms trapped on a photonic chip to generate photons and enable deterministic entangling gates, creating the entangled photonic cluster states needed for measurement-based computing. Quantum Source has raised $77 million including a $50 million Series A, with first ORIGIN deliveries to selected partners planned by the end of 2026.
• Photonic Inc. (Toronto) takes a distinct approach using silicon T-centers (color centers in silicon) as a photon-matter interface for distributed quantum computing. Photonic’s architecture generates entanglement between photons and solid-state qubits, aiming at networked quantum computing rather than pure cluster-state computation.
• Academic groups continue to push the boundaries of cluster-state generation. In China, the USTC group (Jian-Wei Pan) built the Jiuzhang photonic processors that achieved quantum computational advantage in boson sampling and is investigating photonic gates and small cluster states on silicon chips for more general tasks. NTT in Japan and the University of Tokyo (Furusawa’s group) have been leaders in optical cluster states and feedforward control in the continuous-variable domain. At ETH Zurich, O’Sullivan et al. (2025) generated a 20-qubit 2D cluster state using coupled transmon qubits (in the microwave domain), demonstrating that multi-dimensional cluster states of significant size can be generated deterministically from a single device. A June 2025 paper in Nature Communications showed deterministic and reconfigurable generation of graph states from a semiconductor quantum dot in a cavity. And at USTC, Ding et al. (2025) demonstrated a high-efficiency single-photon source exceeding the loss-tolerant threshold for efficient linear optical quantum computing, published in Nature Photonics.

The State of Cluster-State Scaling

A central challenge has been to create cluster states large enough and with high enough quality to perform useful computations. Early experiments entangled on the order of 4–8 photons in small clusters or graph states. By mid-2026, the scale has expanded considerably. In the discrete-variable domain (photons as qubits), integrated photonic circuits have stabilized and interferometrically combined photons: four-photon cluster states and eight-qubit graph states have been generated on silicon photonic chips. Xanadu’s Aurora system operates 35 photonic chips connected by 13 km of fiber, and PsiQuantum’s Omega chipset has been fabricated in a high-volume semiconductor fab with yields comparable to conventional silicon photonics.

In continuous-variable photonics (where qubits are replaced by modes with continuous quantum variables), very large cluster states have been achieved: Larsen et al. and Asavanant et al. both reported entangling on the order of 10,000 light modes into large two-dimensional clusters using optical frequency combs and beam splitters. These results set records for entanglement size, though in the CV regime where error correction works differently than for qubit clusters.

On the source side, 2023–2025 has seen notable progress toward deterministic photon sources. The Cogan et al. (2023) demonstration of deterministic cluster-state photon generation from a quantum dot at gigahertz rates pointed toward a future where a single chip-scale emitter could produce continuous cluster states. Su et al. (Technion, 2024) extended this to demonstrate continuous and deterministic all-photonic cluster states from a quantum dot, with the spin of the confined hole acting as a quantum “knitting machine” that produces indistinguishable photons already entangled in a cluster state. Quantum Source’s ORIGIN engine aims to bring this concept to a commercially viable product. On the detection side, photon-number-resolving detectors have reached efficiencies above 99% (as demonstrated in Xanadu’s GKP paper), and superconducting nanowire detectors routinely achieve above 95%.

Integration is the dominant theme. The photonic quantum computing supply chain remains vertically integrated: PsiQuantum, Xanadu, and Quandela each design and fabricate their own core photonic processors. Unlike superconducting quantum computing, where one can buy a QPU from one vendor and mount it in another vendor’s cryostat, the photonic components are matched to each other at the level of waveguide geometry, photon wavelength, and timing. This may eventually change if the CMOS fabrication thesis that PsiQuantum and Xanadu are pursuing leads to standardized photonic quantum components, but for now, a buyer interested in photonic quantum computing is buying a complete system from a single vendor.

Advantages

Photonic cluster-state computing offers several compelling advantages that stem from the physical properties of photons and the nature of the one-way model:

Room-Temperature Processing (With Caveats)

Photons can be used as qubits without the millikelvin environments that superconducting circuits and some other qubit types require. Optical systems suffer virtually no thermal decoherence because photons have no charge and do not easily couple to thermal noise. The core photonic processing — photon generation, routing, interference, and manipulation — can in principle be performed at room temperature.

A precision matters here, though. While the photonic logic itself does not require ultra-cold temperatures, the detectors often do. Photonic quantum computers rely on superconducting nanowire single-photon detectors (SNSPDs) or transition-edge sensors (TES), which operate at 0.8 to 4 kelvin (SNSPDs) or approximately 50 millikelvin (TES). PsiQuantum integrates these detectors directly on its Omega chip, so the entire chip operates at 2–4 K. Xanadu’s Aurora operates its core photonic processing at room temperature with separate cryogenic detector units. The cryogenic burden is real, but at a few kelvin it is far lighter than the millikelvin environment superconducting qubits demand, and the cryogenic infrastructure required is simpler (cryocooler-based rather than dilution-refrigerator-based). For some approaches, like continuous-variable photonics with homodyne detection, the detectors themselves operate at room temperature. Overall, the thermal requirements for photonic quantum computers are significantly less demanding than for superconducting or spin qubit platforms.

Low Decoherence and High Stability

Photons interact very weakly with the environment. Once a photon is created in a given quantum state (say polarization or time-bin), it can maintain its quantum coherence over long distances and times, as long as it is not absorbed. There is no equivalent of “phase flip” noise from fluctuating fields that plagues superconducting qubits. A photon in free space or a low-loss fiber can keep its quantum state essentially unchanged for kilometers. Optical fiber losses are on the order of 0.2 dB per kilometer for telecom wavelengths, and photons do not experience “memory” effects: they either get lost or they do not; if not lost, their quantum state is nearly perfectly preserved aside from predictable phase shifts.

PsiQuantum’s Omega chipset benchmarks illustrate how high photonic fidelities can be: 99.98% state preparation and measurement fidelity, 99.5% quantum interference visibility, and 99.72% chip-to-chip interconnect fidelity. These numbers exceed what most superconducting and trapped-ion systems achieve for equivalent operations. The combination of long coherence and low operational error bodes well for eventually achieving fault tolerance: fewer error correction overheads may be needed if each photonic operation is very clean (though loss remains a main challenge, as discussed in the Disadvantages section).

Natural Networking and Distribution

Photons are mobile qubits. They inherently propagate at the speed of light and are the ideal choice for transmitting quantum information over distance. A photonic quantum processor can send qubits to another processor or receive qubits from remote sources simply using optical fiber or free-space links. There is no need for special transduction to a communication medium — the computing qubits are themselves the flying qubits. PsiQuantum’s demonstration of 99.72% interconnect fidelity over standard telecom optical fiber shows that photonic modules can communicate quantum information over meaningful distances with very low degradation. Xanadu’s Aurora interconnects four modular racks using 13 km of fiber.

This gives photonic systems a natural edge in modular and distributed architectures. Multiple photonic modules can be connected into a larger machine with minimal link overhead, and photonic quantum computers can integrate with quantum communication (QKD) infrastructure by the same mechanisms they use internally. The measurement-based model lends itself particularly well to networking: one can perform blind quantum computing or secure delegated computing by sending photons to a server.

Ultra-Fast Operations and Parallelism

Photonics operates at the speed of light. Gates and measurements in optical systems can be extremely fast (picosecond or nanosecond scale) because they are often limited only by how quickly you can modulate or detect light. Single-photon detectors can operate at up to gigahertz rates, and electro-optic modulators can adjust measurement bases on nanosecond timescales. PsiQuantum’s barium titanate (BTO) electro-optic switches are designed for the fast photon routing that FBQC requires. In addition, the cluster-state model inherently supports parallel operations: many photons in the cluster can be measured simultaneously if their measurements do not depend on each other’s outcomes. In a 2D cluster, one can often measure an entire layer of qubits in parallel, which corresponds to executing many gates at once. The combination of fast per-operation time and parallelism means high throughput.

Scalability via Modular and Mass-Manufacturable Components

Photonics benefits from the mature fabrication technologies of the telecom and silicon photonics industries. PsiQuantum’s partnership with GlobalFoundries is the clearest example: Omega chips are manufactured on full-size 300 mm silicon wafers using GlobalFoundries’ 45 nm process, the same infrastructure used for conventional semiconductor products. The chips achieve yields that match standard semiconductors, according to both companies. This mass-manufacturability means that, once a design is proven, scaling to more qubits is principally an engineering replication task rather than one requiring each qubit to be individually fabricated by hand. Xanadu uses custom silicon nitride waveguides on 300 mm wafers and has partnered with EV Group for the heterogeneous-integration bonding that joins photonic chips to detector arrays, and with Thorlabs for custom fiber optics.

Because photons do not directly interact, having more photons in the system does not necessarily complicate control in the way adding more superconducting qubits does (where crosstalk and frequency collisions become an issue). In a photonic cluster, qubits interact only via the predefined entanglement connections. This can make architectural scaling more straightforward. Both PsiQuantum and Xanadu envision modular architectures where additional racks or modules can be added and connected by fiber — Xanadu’s Aurora paper explicitly demonstrated this four-rack modular concept.

Compatibility with Fault-Tolerant Schemes

The structure of cluster-state computing is well-suited to implementing certain error-correcting codes, especially topological codes. A prime candidate for fault tolerance is to use a large three-dimensional cluster state that encodes a surface code in its entanglement structure. Photonic cluster states can be built to embody this 3D structure, and once built, error correction is performed by measuring the cluster in particular ways to detect and correct errors. Since photons naturally have low error rates for bit-flip and phase-flip operations, the main errors to correct are losses. Topological codes can be adapted to handle loss (erasure errors) up to a threshold (several percent). Fusion-based architectures like PsiQuantum’s explicitly design the architecture to tolerate a certain loss rate while still creating a logical cluster suitable for fault-tolerant computation.

Xanadu’s GKP encoding offers a complementary route to fault tolerance. GKP states encode a logical qubit in superpositions of many photons in a way that makes the qubit inherently resilient to photon loss and noise, and logical gates can be implemented deterministically using linear optics and homodyne measurements — no probabilistic entangling operations required at computation time. The June 2025 Nature paper showing on-chip GKP generation, combined with the real-time error-correction decoding already demonstrated on Aurora’s 12 logical GKP qubits, suggests that GKP-based fault tolerance is moving from theory to practice.

Disadvantages

Despite its promising features, photonic cluster-state computing faces several significant challenges. These include fundamental issues with photonics as a platform and practical engineering difficulties in creating and handling large cluster states:

Probabilistic Entanglement and Photon Sources

A core challenge in optical quantum computing is that, with current technology, generating entanglement between photons is often probabilistic. In linear optics, two photons entering a device do not deterministically entangle; one relies on measurement-induced entanglement (such as fusion gates or post-selected interference outcomes). The Browne-Rudolph fusion gate succeeds only 50% of the time. Spontaneous parametric down-conversion (SPDC), the workhorse for creating entangled photon pairs, produces photons at random times — it is probabilistic whether you get a pair in a given pump pulse. This means scaling up requires massive parallelism or multiplexing to compensate for low success probabilities, adding complexity.

However, this challenge is being addressed on multiple fronts. Quantum dot sources have demonstrated deterministic generation of indistinguishable entangled photons at gigahertz rates (Cogan et al., 2023), and Quantum Source’s ORIGIN engine aims to commercialize deterministic resource-state generation using cavity-QED photon-atom gates. PsiQuantum’s fusion-based architecture is designed to work with probabilistic sources: by continuously generating and fusing small resource states, the architecture tolerates fusion failures without destroying the entire cluster. PsiQuantum’s February 2025 Nature paper achieved 99.22% two-qubit fusion gate fidelity, though improving this further and increasing the success probability of each fusion attempt remain active areas of work. Ding et al.’s 2025 Nature Photonics paper demonstrated a high-efficiency single-photon source exceeding the loss-tolerant threshold for efficient linear optical quantum computing, another step toward practical photonic sources.

Photon Loss

Loss is the dominant error mode in optical quantum systems. A lost photon means a lost qubit from the cluster, an error that can be more damaging than a simple bit-flip because it breaks the entanglement bonds to that photon, potentially fragmenting the cluster. Loss accumulates with system size: even 0.5% loss per optical component becomes significant when a photon passes through dozens of components. Detector inefficiency is another form of loss (not clicking on a photon is equivalent to that photon being lost to the computation).

The cluster-state model can tolerate some loss if using certain codes, but only up to a threshold of a few percent. Xanadu explicitly identified optical loss as the primary remaining challenge on its path to fault tolerance, and the company’s 60% loss reduction in 2025 (20-fold improvement over three years) indicates the trajectory of improvement. PsiQuantum’s Omega chipset uses ultra-low-loss waveguides and high-efficiency SNSPDs integrated on-chip to minimize the total loss budget. New waveguide fabrication methods yield losses below 0.1 dB/m, and novel detector technologies approach 99.9% efficiency.

Approaches to mitigate loss include: building in redundancy (making the cluster state larger than needed so that some photon loss can be tolerated via error correction), heralding (detecting early if a photon is lost and attempting to replace it or adapt the computation), and the fusion-based architecture itself, where small resource states are generated independently and a failed fusion destroys only one small piece rather than a large pre-assembled cluster. Loss remains one of the hardest disadvantages for photonic quantum computing, but the gap between current loss rates and fault-tolerance thresholds is narrowing.

Complexity of Large-Scale Cluster Creation

Even if loss and source issues are managed, the task of creating a large cluster state (say with millions of entangled photons) is an enormous orchestration problem. It is not enough to have good sources and low loss; one also needs to synchronize photons, ensure they interfere coherently, and manage a potentially massive physical apparatus.

The spatial approach (many photons generated simultaneously from many sources and interfered across a large optical network) requires stabilizing a large interferometer and keeping phase stability across all paths. Any vibration or drift can break the delicate interference conditions. The temporal approach (fewer components recycled over time by storing photons in delay lines) reduces hardware count but introduces the need for photon storage and long delay lines with their own loss and coherence challenges. In both cases, control complexity grows: hundreds of phase actuators may need real-time adjustment, thousands of time slots must be managed, and signals must be routed to detectors in sync.

Photonic engineers are countering this by pushing integration (putting as much as possible on chips to reduce drift and size), using automated calibration algorithms to stabilize phases, and using modular fusion-based architectures that break the large cluster generation into many small, independent, parallelizable tasks. PsiQuantum’s factory-scale fab approach and Xanadu’s four-rack Aurora system both reflect this philosophy: build many small things well and connect them, rather than trying to construct one monolithic entangled state.

Detection and Feedforward Latency

Another practical challenge is the requirement of ultrafast detection and feedforward for one-way computing. In cluster-state computation, once you measure a photon, you might need to adjust the basis of another measurement that could be happening very soon after. If photons are spaced by a few nanoseconds in a pulsed system, the single-photon detector must produce a signal, and the control logic must compute the new setting and apply it to a modulator, all within a few nanoseconds. This is demanding for electronics. In many experiments, researchers delay the next photons (through fiber delay lines) to allow time for feedforward, but this adds loss and complexity. The RIKEN group implemented feedforward in the optical domain for some continuous-variable operations, and PsiQuantum’s BTO electro-optic switches are designed for nanosecond-scale modulation. But the system-level integration of fast detectors, fast classical logic, and fast modulators remains a challenge at scale.

In summary, the key challenges for photonic cluster-state computing are photon source quality, photon loss, the complexity of orchestrating large entangled states, and feedforward latency. None of these challenges are viewed as insurmountable: each has a research roadmap and is seeing measurable year-over-year improvement. The 2025–2026 results from PsiQuantum, Xanadu, Quantum Source, and academic groups show that the gap between current capabilities and the thresholds needed for fault-tolerant operation is closing.

Impact on Cybersecurity

The advent of large-scale photonic cluster-state quantum computers (or any quantum computers) has profound implications for cybersecurity, both positive and negative. Photonic systems in particular tie in naturally with quantum communication and cryptography. As I have written extensively, the question of when these implications become urgent is now beside the point: regulators, insurers, investors, and clients are setting their own quantum deadlines, and those ecosystem-driven deadlines matter more than predictions about when a cryptographically relevant quantum computer (CRQC) will arrive.

Enhancing Quantum Cryptography (QKD and Beyond)

Photonic cluster-state computers could enhance quantum cryptography techniques such as Quantum Key Distribution (QKD). Since photonic cluster-state devices use photons and can produce complex entangled states, they can serve as advanced QKD transmitters/receivers or as entanglement swapping nodes in quantum networks. A photonic quantum processor can easily send qubits over fiber and integrate cryptographic key exchange with quantum computation. The modular, networked architecture demonstrated by Xanadu’s Aurora (multiple racks connected by fiber) is a prototype for exactly this kind of quantum-networked infrastructure. Cluster states have been proposed for quantum repeaters that could extend the range of entanglement distribution by dividing the channel into segments and connecting them with error correction.

Threat to Classical Cryptography

A full-scale quantum computer (photonic or otherwise) poses a serious threat to many classical cryptographic systems in use today. Most current public-key encryption (RSA, Diffie-Hellman, elliptic curve cryptography) relies on mathematical problems that a quantum computer can solve exponentially faster using Shor’s algorithm. A photonic cluster-state quantum computer with sufficient qubits and low error rates could run Shor’s algorithm to factor large RSA moduli or break elliptic curve crypto, rendering essentially all internet communications that rely on those schemes insecure. This threat extends beyond encryption: Trust Now, Forge Later (the authentication analog of the better-known Harvest Now, Decrypt Later threat) means that even digital signatures are at risk, since a future quantum computer could forge signatures on documents and software updates retroactively.

The timeline is uncertain, and photonic cluster-state computers are among the contenders to reach CRQC scale. The resource estimates for breaking RSA-2048 have dropped from 20 million physical qubits (Gidney & Ekerå, 2021) to approximately 4 million physical qubits (Gidney, 2025), and further algorithmic improvements may compress the requirements further. PsiQuantum’s goal of a million physical qubits is within an order of magnitude of these estimates. As I track in my CRQC Quantum Capability Framework, the path from where photonic quantum computers are today to a CRQC is long but the engineering trajectory is measurable.

This is why NIST has already finalized post-quantum cryptographic standards — ML-KEM (formerly CRYSTALS-Kyber), ML-DSA (formerly CRYSTALS-Dilithium), SLH-DSA (formerly SPHINCS+), and FN-DSA (formerly FALCON) — and is urging adoption. Government agencies and standards bodies worldwide have set concrete deadlines for PQC migration, and the urgency of those deadlines is independent of any specific quantum computing modality.

Blind Quantum Computing (Secure Delegation)

Photonic cluster-state computing offers a unique security application: Blind Quantum Computing (BQC), a method for a client to delegate a quantum computation to a quantum server without revealing the input, output, or algorithm to the server. This is important if quantum computing is provided as a cloud service (which is likely, given the complexity and cost of quantum hardware). The cluster-state model has a built-in way to achieve BQC: a client can prepare single photons in certain encoded states (e.g., randomly rotated qubits) and send them to the server, who incorporates them into its large cluster state and performs the measurements instructed by the client. Because the qubits were encoded with the client’s secret random rotations, the server’s measurement results are encrypted and the server cannot interpret them.

The one-way model makes it easier to implement blind computing because measurements and their needed bases are the primary actions. A client can use the Broadbent-Fitzsimons-Kashefi protocol (2009) where the client’s random choices of measurement angles hide the true computation from the server. Photonic systems are well-suited to this because they can transmit qubits from client to server via fiber and because cluster-state computers naturally align with the protocol. The first demonstration of blind quantum computing was done with a photonic setup in 2012 using a four-photon cluster state. With full-scale photonic cluster computers, companies or individuals could send quantum-encrypted tasks to a quantum cloud and get back the result without the provider ever being able to read the data or know what algorithm was run.

Post-Quantum and Quantum-Resistant Measures

In preparation for quantum computers, the cybersecurity community is working on two fronts: deploying post-quantum cryptography (PQC) which runs on classical hardware but is hard for quantum computers, and developing quantum cryptography (like QKD) for scenarios requiring information-theoretic security. Photonic cluster-state computers could be used to analyze the quantum hardness of proposed PQC schemes (by attempting to solve underlying problems using quantum algorithms, thus identifying weak algorithms before they are widely adopted). Because photonic quantum computers can interconnect through optical links, they could be part of a hybrid infrastructure where classical PQC and QKD are both used.

The transition to PQC is a pressing issue. NIST has finalized its first PQC standards, and organizations should be acting now. For those starting their migration, my open-source PQC Migration Framework provides a structured approach, and Practical Steps to Quantum Readiness outlines concrete next steps.

The impact of photonic cluster-state computing on cybersecurity cuts both ways. On the positive side, it will bolster quantum cryptographic methods, enabling secure key exchange over long distances, secure delegation of computation via blind quantum computing, and integration into the emerging quantum internet. On the negative side, it accelerates the threat to traditional cryptography; once such a quantum computer is operational, classical RSA/ECC-based encryption and certain hashing or discrete-log-based schemes will no longer be safe. This necessitates the urgent adoption of post-quantum cryptographic algorithms in all sectors before quantum computers reach that level. The mere prospect of a photonic quantum computer in the near future means that even today’s data might be vulnerable in the future if recorded now. Organizations are recommended to start transitioning to PQC now, and governments are investing in standards and migration plans accordingly.

Future Outlook

The future of photonic cluster-state computing is promising, but key milestones and breakthroughs are needed before it reaches commercial viability and widespread use.

Timeline to a Fault-Tolerant Photonic Quantum Computer

As of mid-2026, photonic approaches are in a tight global race with superconducting qubits, ion traps, neutral atoms, and silicon spin qubits. The timelines that companies have stated vary. PsiQuantum aims to build utility-scale quantum computing sites in Brisbane and Chicago in the coming years, with a million-physical-qubit target but no publicly specified date for that milestone. Xanadu’s public filings target up to 100,000 physical qubits and up to 500 logical qubits in the 2029–2030 timeframe, with early fault-tolerance building blocks expected through 2026 and scaled error-corrected operations through 2028. Personally, I estimate that a CRQC (from any modality, not just photonics) could arrive within the next several years; my reasoning is detailed in my Q-Day predictions analysis and in the CRQC Readiness Benchmark (Q-Day Estimator).

A reasonable staging of what to expect:

• Near-term (2026–2028): Demonstrations of larger photonic cluster states, the first implementations of quantum error correction codes on a photonic platform (Xanadu is targeting early error-corrected demonstrations in this window), and continued improvement in GKP state quality, fusion gate fidelity, and optical loss reduction. PsiQuantum’s Brisbane and Chicago compute centers will begin commissioning hardware. Xanadu aims to produce early error-corrected qubits. Quandela targets 24 qubits with Canopus and then 40+ qubits within three years. Quantum Source plans first ORIGIN deliveries.
• Mid-term (late 2020s): If current trajectories hold, photonic machines with on the order of several hundred logical qubits could become operational. This might be enough to perform some specialized tasks beyond classical ability (certain chemistry simulations or optimization problems) in a fault-tolerant manner. Xanadu’s 2029–2030 target of up to 500 logical qubits falls in this window. PsiQuantum’s compute centers should be hosting progressively larger systems. Whether this is when a CRQC arrives depends on how fast loss reduction, source quality, and system integration advance, and on whether algorithmic improvements further reduce the qubit requirements for cryptanalysis.
• Long-term (2030s and beyond): Photonic cluster-state computers, if successful, will be scaled up further. Because photonic machines can be networked, we may see distributed quantum computing where multiple photonic quantum computers at different locations link to act as one larger machine. The vision is a fault-tolerant quantum internet where photonic cluster-state quantum computers serve as both the computing nodes and the communication channels.

Expected Breakthroughs Required

Several key breakthroughs are needed to realize this future:

• Deterministic single-photon sources at scale: On-demand sources that produce indistinguishable single photons with efficiency above 99% and multi-photon event probability below 10⁻⁶ at GHz rates. Quantum dot emitters coupled to cavities are the leading candidates. Cogan et al. (2023) showed deterministic cluster-state generation at GHz rates from a quantum dot, Ding et al. (2025) demonstrated a source above the loss-tolerant threshold, and Quantum Source’s ORIGIN aims to commercialize cavity-QED-based deterministic generation. A breakthrough would be achieving a >99.9%-efficient source of indistinguishable photons at GHz rates on-chip with high yield.
• Further loss reduction: Xanadu’s 60% loss reduction in a single year shows the pace of improvement, but optical loss must drop further before fault-tolerant operation is practical. Sub-0.1 dB/m waveguide loss, above-99.5% detector efficiency, and improved fiber-to-chip coupling are all active targets. New materials (lithium niobate, aluminum nitride, silicon nitride) and 3D photonic integration may help.
• Quantum error correction demonstrations: A demonstration that an error-corrected logical qubit on a photonic platform survives longer than the underlying physical qubits would be a watershed moment. Xanadu’s real-time error-correction decoding on 12 logical GKP qubits is a step in this direction, but the goal is a code-distance-sufficient logical qubit with below-threshold error rates. PsiQuantum’s FBQC architecture is designed for this, but the demonstration has not yet been published.
• Better feedforward and control electronics: Ultrafast logic (perhaps implemented in microwave photonics or highly optimized FPGAs/ASICs) that can handle GHz-clocked operations. PsiQuantum’s BTO switches and Xanadu’s programmable modulators are tackling the hardware side, but the integration of photonics with CMOS electronics for real-time decision-making at GHz rates remains a systems engineering challenge.
• Modular networking at scale: Xanadu’s Aurora demonstrated a four-rack networked machine; the next step is scaling this to tens or hundreds of modules with sustained high fidelity across all inter-module links. PsiQuantum’s 99.72% interconnect fidelity at 42 m (and up to 250 m) provides a starting point, but system-level integration at datacenter scale, with potentially thousands of modules, requires significant further engineering.

Commercial Viability and Applications

Once a fault-tolerant photonic cluster-state computer is built, likely initial applications include:

• Cryptography and Security: Breaking classical crypto or running new quantum-secure protocols. PsiQuantum’s defense partnerships (Lockheed Martin, Airbus) and DARPA QBI involvement signal this direction.
• Chemistry and Materials Science: Quantum simulation of molecules to discover new drugs, catalysts, or materials. A photonic quantum computer with a few hundred logical qubits could surpass classical supercomputers for simulating complex chemical systems. This is considered one of the first useful applications of quantum computers.
• Optimization and Finance: Solving certain hard optimization problems faster using quantum algorithms. ORCA Computing’s focus on accelerating AI workloads with photonic quantum systems, and Quandela’s work on HPC-quantum hybrid workflows for structural mechanics and meteorology, are early explorations of this space.
• Quantum as a Service: Cloud-accessible quantum computing. Quandela’s Belenos is already available via the cloud to 1,200+ researchers across 30 countries, and OVHcloud’s Quantum Platform offers it as Quantum-as-a-Service. Xanadu’s Borealis has been available on the cloud since 2022. As these systems scale, blind quantum computing protocols could allow users to run computations without the cloud provider seeing their data.

Role in Quantum Networks and Hybrid Architectures

Photonic cluster-state computers are likely to be central nodes in the emerging quantum internet. They will likely work in tandem with other types of quantum devices:

• Hybrid systems: A quantum data center might combine memory nodes consisting of matter qubits (NV centers, ions, or silicon spins) with photonic cluster states that serve as the flying qubits connecting them. Photonic processors could handle communication and some fast processing, while matter qubits store information or perform particular high-fidelity operations. Photonic Inc.’s silicon T-center approach is designed for exactly this kind of photon-matter networking.
• Scaling by networking: Rather than building a single monolithic million-qubit machine, an alternative approach to scale is to network many smaller quantum computers. Photonics is the only practical medium for this because it is the quantum medium for communication. Both PsiQuantum and Xanadu envision modular architectures where racks are connected by fiber. This approach might overcome fabrication yield issues — ten chips with 100,000 components each, networked together, may be easier to build than one chip with a million components.
• HPC integration: NVIDIA’s NVQLink architecture for quantum-classical integration now has photonic partners: Xanadu, ORCA Computing, and Quandela all participate. The integration of photonic quantum processors into HPC environments (with GPU-accelerated classical co-processing for error decoding, circuit compilation, and hybrid algorithms) is already underway.

The outlook for photonic cluster-state computing has strengthened considerably since 2023. The February 2025 PsiQuantum Nature paper, the January 2025 Xanadu Aurora Nature paper, the June 2025 on-chip GKP demonstration, and the academic progress in deterministic cluster-state sources have collectively moved the field from theoretical promise to early systems engineering. The challenges remain formidable — optical loss, source quality, and feedforward latency all need further improvement before a fault-tolerant machine is operational — but the year-over-year improvement rates are encouraging. With over $3 billion in private capital and hundreds of millions in government funding flowing into photonic quantum computing, and with datacenter-scale facilities now under construction, the question is no longer whether the photonic cluster-state approach can work in principle, but whether it can be engineered to work at the scale and fidelity required before competing modalities get there first.

(This article was updated in June 2026 with the latest developments from PsiQuantum, Xanadu, Quandela, ORCA Computing, Quantum Source, and academic research groups.)