Building a Photonic Quantum Computer

Introduction

Every other article in this series describes how to assemble a quantum computer from independently sourced vendor components. This one describes why you cannot do that yet for photonic quantum computing, what the subsystems look like inside the vertically integrated vendors’ machines, and where an independent integrator adds value today.

The candor is deliberate. The Quantum Open Architecture model that enables a systems integrator to combine a QuantWare QPU with a Bluefors cryostat and Qblox control electronics does not have a photonic equivalent. PsiQuantum designs its own Omega chipset and manufactures it at GlobalFoundries’ Malta, New York fab. Xanadu fabricates Aurora’s photonic circuits at its own facility and partnered with Thorlabs for custom fiber optics and EV Group for heterogeneous integration bonding. Quandela builds its quantum-dot single-photon sources in-house. You cannot buy a photonic QPU from one vendor and install it in another vendor’s optical platform the way you can with a superconducting transmon chip and a third-party cryostat.

That may change. The semiconductor fabrication model that PsiQuantum and Xanadu are pursuing, where photonic circuits are manufactured using the same processes and foundries that produce classical photonic interconnects for data centers, creates the conditions for eventual component disaggregation. But “eventual” is doing heavy lifting in that sentence. Today, the photonic quantum computing supply chain is proprietary end to end.

For the physics of how photonic qubits work, see my Quantum Computing Modalities series.

The vendor picture

Three photonic quantum computing companies have reached the stage of deploying hardware or providing cloud access in 2026. Their architectural approaches are distinct.

PsiQuantum (Palo Alto) is pursuing the most ambitious scale target in the quantum industry: million-qubit, fault-tolerant photonic quantum computers housed in purpose-built data centers. The Omega chipset, published in Nature in February 2025, integrates several critical advances: barium titanate (BTO) electro-optic switches for high-speed optical routing, superconducting materials for single-photon detection, and standard silicon photonic waveguides for qubit manipulation. Published performance: 99.98% single-qubit state preparation and measurement fidelity, 99.5% two-photon quantum interference (Hong-Ou-Mandel) visibility, 99.72% chip-to-chip quantum interconnect fidelity, 99.22% two-qubit fusion gate fidelity.

The architecture is fusion-based quantum computing (FBQC): individual photonic resource states are generated on chip, then fused together through entangling measurements to build the large-scale entangled state needed for computation. Chip-to-chip interconnects use standard telecom optical fiber, the same fiber used in conventional data centers, with no need for transduction between different physical modalities. This is a fundamental scaling advantage: while superconducting and trapped-ion systems struggle to connect separate processor modules, photonic qubits travel naturally through fiber.

PsiQuantum raised $1 billion in Series E funding in September 2025 (led by BlackRock, Temasek, Baillie Gifford, NVIDIA NVentures, Macquarie, and Qatar), reaching a $7 billion valuation. Two Quantum Compute Centers are under construction: Brisbane, Australia (backed by AU$940 million from Australian federal and Queensland state governments) and Chicago, Illinois (partnership with the State of Illinois and the City of Chicago). Both target operational status by 2027. PsiQuantum advanced to the final stage of DARPA’s Quantum Benchmarking Initiative.

PsiQuantum also introduced a new cryogenic cooling architecture that eliminates the chandelier-shaped dilution refrigerator in favor of cuboid racks fed by industrial cryoplants, operating at 2-4 K rather than the 10-20 mK required by superconducting transmon qubits. This is a qualitatively different infrastructure model: simpler, more scalable, and closer to conventional data center rack cooling than to the millikelvin environments described in the cryogenic infrastructure article.

Xanadu (Toronto) builds Aurora, described as the world’s first modular, networked photonic quantum computer with real-time error correction. Xanadu uses GKP (Gottesman-Kitaev-Preskill) bosonic qubits encoded in squeezed light states, a continuous-variable approach that differs from PsiQuantum’s single-photon, dual-rail encoding. The Borealis system has been accessible via cloud since 2022. Xanadu went public on Nasdaq and the Toronto Stock Exchange in March 2026 (ticker: XNDU), generating $302 million in gross proceeds. DARPA QBI Stage B. Canada’s Quantum Champions Program (up to CAD $23 million). Negotiations for up to CAD $390 million from the governments of Canada and Ontario for Project OPTIMISM and domestic quantum manufacturing. PennyLane, Xanadu’s open-source quantum software framework, saw 161% growth in adoption in 2025 and has become a leading platform for quantum machine learning research. Xanadu achieved a 60% reduction in optical loss in 2025, a 20-fold improvement over three years.

Quandela (Massy, near Paris) takes a different approach: quantum-dot single-photon sources that produce indistinguishable photons on demand. The Belenos system (12 photonic qubits) was inaugurated at CEA’s TGCC supercomputing center as “Lucy” in April 2026, making it the most powerful deployed photonic quantum computer. Also available on OVHcloud’s Quantum Platform (April 2026). Canopus (next generation, doubled qubits) is targeted for 2026, with a 40+ qubit machine within three years. Quandela signed an MOU with Seoul National University for semiconductor-process-based quantum computing and is developing the SPOQC (Spin-Photonic Quantum Computing) architecture for fault tolerance.

Other players: ORCA Computing (UK, time-bin encoding, boson sampling, PT-1/PT-2 systems at UK NQCC), QuiX Quantum (Netherlands, thin-film lithium niobate programmable photonic processors), Photonic Inc. (Toronto, silicon T-center color centers for distributed quantum computing and networking).

The five subsystems inside a photonic quantum computer

Despite the vertical integration, understanding the subsystems helps explain where the supply chain might eventually disaggregate and where an integrator contributes value today.

Photon source

The quantum information carrier is a single photon or a squeezed light state. Three source technologies are in commercial or near-commercial use. Spontaneous parametric down-conversion (SPDC) or spontaneous four-wave mixing (SFWM) in nonlinear crystals or waveguides generates photon pairs probabilistically (PsiQuantum, integrated into silicon photonic chips). Quantum-dot single-photon sources produce photons on demand with high indistinguishability (Quandela’s core technology, using InAs/GaAs semiconductor quantum dots). Squeezed-light sources generate continuous-variable GKP states (Xanadu’s approach, using optical parametric oscillators). Each approach has trade-offs in photon quality, generation rate, and integration difficulty. PsiQuantum’s SPDC sources are integrated directly into the silicon photonic chip, manufactured at GlobalFoundries. Quandela’s quantum dots are fabricated in-house and represent proprietary IP that would be difficult to source independently.

Photonic processor

The processor is an integrated photonic circuit (PIC) containing waveguides, beam splitters, phase shifters, and entangling fusion or interference elements. PsiQuantum’s Omega uses silicon photonic waveguides with BTO electro-optic switches. Xanadu’s Aurora uses silicon photonic circuits with Mach-Zehnder interferometers. Quandela’s systems use fiber-based or chip-based linear optical networks.

The processor chips are fabricated using semiconductor processes: PsiQuantum at GlobalFoundries, Xanadu at its own facility (with EV Group for heterogeneous integration), Quandela in-house. The fabrication leverages the same lithography, deposition, and etching tools used for classical silicon photonics, which is why photonic quantum computing advocates argue their approach will scale more naturally than other modalities. The silicon photonics industry already produces millions of chips annually for data center optical interconnects; quantum photonic circuits are a specialized application of the same manufacturing base.

Detector subsystem

This is where photonic quantum computing touches cryogenics. Single-photon detection at the fidelity required for quantum computation uses superconducting nanowire single-photon detectors (SNSPDs) operating at 0.8-4 K, or in some architectures, transition-edge sensors (TES) operating at approximately 50 mK.

SNSPDs are the dominant detector technology. A thin superconducting nanowire (typically niobium nitride, NbN, or tungsten silicide, WSi) is biased just below its critical current. When a single photon is absorbed, it creates a localized hot spot that drives the wire normal (resistive), producing a voltage pulse that is read out by room-temperature electronics. Detection efficiency exceeds 95% in commercial devices. Timing jitter is below 50 ps.

SNSPD suppliers: Photon Spot (US), Single Quantum (Netherlands), Quantum Opus (US), ID Quantique (Switzerland). These are commercially available components. Unlike photonic QPU chips, SNSPDs can be procured independently, which is why this subsystem represents the most natural entry point for independent integrator involvement.

The cryogenic infrastructure for SNSPDs is simpler than for superconducting qubits. A cryogen-free pulse-tube cooler reaching 2-4 K (Cryomech or similar) is sufficient for NbN SNSPDs. No dilution refrigerator, no helium-3, no millikelvin temperatures. PsiQuantum’s cuboid cryogenic racks use this approach. For TES detectors (higher energy resolution but slower recovery time), a small dilution refrigerator reaching 50 mK is needed, but the cooling power requirements are modest compared to a superconducting QPU installation.

Optical routing and fiber management

Photonic quantum computers involve extensive optical fiber routing between source modules, processor modules, and detector modules. PsiQuantum explicitly scales beyond a single chip using standard telecom optical fiber. This is an advantage for multi-module architectures (no transduction problem) but creates an engineering challenge in fiber management: maintaining polarization stability, minimizing coupling losses, managing hundreds to thousands of fiber connections, and ensuring mechanical stability over operational lifetimes.

The Xanadu-Thorlabs partnership specifically targets this challenge: developing stable, low-loss fiber optics that link hardware subsystems while minimizing optical drift. Reducing optical loss is critical because every percentage point of photon loss translates directly into higher physical qubit overhead for error correction.

Timing and control

Classical control electronics manage photon source triggering, phase shifter voltages, detector bias currents, and the real-time feed-forward logic required for measurement-based or fusion-based computation. The timing requirements are set by photon propagation: at the speed of light in fiber (~200,000 km/s), a 1-meter fiber path introduces roughly 5 ns of delay. Feed-forward decisions (measuring one photon and conditioning a subsequent operation on the result) must complete within these propagation windows.

Standard FPGA-based control systems (similar to those used in superconducting and trapped-ion systems) handle the timing and feed-forward logic. AWGs, DACs, and time-tagging electronics are commercially available from the same vendors that serve other modalities (Qblox, Zurich Instruments, and others make components that could serve photonic systems, though none currently markets a photonic-specific control platform).

Why independent assembly is not yet practical

The photonic QPU supply chain is vertically integrated for three interlocking reasons.

First, the photonic processor chip and the photon source are co-designed and co-fabricated. PsiQuantum’s SPDC sources are integrated into the same silicon photonic chip as the processor elements. Quandela’s quantum-dot sources are fabricated in-house using proprietary epitaxial growth processes. Separating “QPU” from “source” in the way that superconducting separates “QPU” from “cryostat” is not physically meaningful for most photonic architectures.

Second, the interfaces between subsystems are optical, not electrical. In superconducting QOA, the interfaces between QPU, cryostat, and control electronics are defined by well-understood RF/microwave connectors (SMA, SMP), standardized impedances (50 Ω), and quantifiable parameters (attenuation, insertion loss, S-parameters). In photonic systems, the interfaces are fiber-coupled optical paths where mode matching, polarization alignment, and coupling loss are system-specific and not yet standardized.

Third, the architectural approach itself is still converging. PsiQuantum uses fusion-based computation with dual-rail qubits. Xanadu uses GKP bosonic qubits with measurement-based computation. Quandela uses linear optical quantum computation with quantum-dot sources. These are not minor implementation differences; they are fundamentally different computational models that require different source technologies, different processor designs, and different error correction strategies. Standardizing component interfaces across these architectures would require the field to converge on a common computational model, which has not happened.

What an integrator contributes today

Despite the vertical integration of the QPU layer, independent integrators add value in three areas for photonic quantum computing deployments.

SNSPD cryogenics. The detector subsystem is the one component layer that an integrator can procure and install independently. Selecting the right SNSPD technology (NbN vs. WSi, single-pixel vs. array, operating temperature), specifying the cryogenic cooler (pulse-tube for 2-4 K, small DR for 50 mK TES), designing the fiber feedthrough from room temperature to the cold stage, and integrating the bias and readout electronics are tasks that require cryogenic engineering expertise but not photonic QPU expertise. For organizations deploying a vendor-supplied photonic system that requires an SNSPD upgrade or a custom detector array, an integrator provides the cryogenic interface engineering.

HPC integration. Connecting a photonic quantum computer to classical HPC infrastructure uses the same NVQLink, QRMI, and Slurm-based integration patterns as other modalities. Quandela is an NVQLink partner. The network design, API surface, authentication, and hybrid workflow orchestration are modality-agnostic engineering tasks where integration expertise applies directly.

Classical networking and security. The classical infrastructure surrounding a photonic quantum computer (power, networking, monitoring, security) is identical to any other high-value compute installation. PQC migration of the API surface, HNDL risk assessment, and physical security design follow the same patterns described in the capstone article.

Facility requirements

Photonic quantum computers have the least demanding facility requirements among millikelvin-dependent modalities, and slightly more demanding requirements than neutral-atom systems (which need no cryogenics at all).

The core photonic processor operates at room temperature on an optical table or in rack-mounted modules. Thermal stability matters: the phase shifters and interferometers are sensitive to temperature drift, and the fiber couplings shift with thermal expansion. A climate-controlled lab at 20-24°C with tight stability (±0.5°C, tighter than the ±2°C typical for superconducting) is recommended.

The SNSPD detector subsystem requires a small cryostat (pulse-tube cooler for 2-4 K operation), which adds floor space, power (5-10 kW for the cooler), and chilled-water requirements. This is substantially smaller and simpler than a dilution refrigerator for superconducting qubits.

Vibration isolation applies to the optical platform but at less stringent levels than superconducting. Fiber-based systems are more vibration-tolerant than free-space optical systems, and most commercial photonic quantum computers use fiber coupling extensively.

EMI is not a primary concern for the optical processor itself (photons are immune to electromagnetic interference), but the SNSPD readout electronics are sensitive and require proper shielding at the detector stage.

For PsiQuantum’s compute center model, the facility is purpose-built at industrial scale. The Brisbane and Chicago sites are not laboratory installations; they are data-center-class facilities designed from the ground up for photonic quantum computing. This is a different procurement model than any other modality: the facility is part of the vendor’s offering, not the customer’s problem.

Team and skills

For organizations deploying a vendor-supplied photonic system (the only current option), the customer team mirrors the neutral-atom model more than the superconducting model:

One HPC/DevOps engineer for network integration, Slurm/QRMI configuration, API surface, and user management. One to two application scientists or quantum software engineers for algorithm development using PennyLane (Xanadu), Perceval (Quandela), or vendor-specific SDKs. One cryogenic engineer if the installation includes an SNSPD subsystem that the customer manages directly (rather than under a vendor support contract). Vendor support contract for all photonic hardware maintenance.

For a cloud-access deployment (no on-premises hardware), the customer team reduces to application scientists using the vendor’s cloud SDK. No hardware operations staff required.

For organizations building photonic systems from research components (academic labs working with commercial photonic chips, custom SNSPD arrays, and lab-built optical setups), the skill set is quantum optics: photon source characterization, integrated photonics design and testing, single-photon detection, fiber optics engineering, and cryogenics for the detector subsystem. This is PhD-level experimental quantum optics, not IT operations.

Where the supply chain may open

The photonic supply chain will not remain vertically integrated permanently. The classical semiconductor industry walked an analogous path. In the 1970s and 1980s, companies like Intel and Texas Instruments designed and fabricated their own chips in their own fabs. By the 1990s, the fabless model emerged: companies like Qualcomm and NVIDIA designed chips and contracted fabrication to foundries like TSMC. Today, the fabless/foundry split is the dominant model in semiconductors. The enabling conditions were standardized design rules, mature process design kits (PDKs), and foundry services accessible to third-party designers.

Photonic quantum computing is approaching a similar inflection. Three developments could trigger disaggregation.

The silicon photonics foundry model is inherently open. GlobalFoundries, TSMC, and other foundries already offer multi-project wafer (MPW) runs for silicon photonic chips. As photonic quantum circuit designs stabilize and design rules mature, it will become possible for third parties to design and fabricate photonic QPU chips at commercial foundries without building their own fab. PsiQuantum’s Omega chipset is manufactured at GlobalFoundries using high-volume, industrially validated processes. The fabrication infrastructure exists; what is missing is the standardization of quantum-specific design rules and the availability of PDKs that let independent designers create quantum photonic circuits.

SNSPD arrays are already commercially available as independent components. As detector requirements scale (more channels, higher density, faster recovery), the SNSPD supply chain will grow to serve photonic quantum computing as a major customer alongside quantum networking and quantum key distribution.

Fiber management and optical packaging are areas where existing telecom and data center optical interconnect suppliers (II-VI/Coherent, Lumentum, Thorlabs) have relevant expertise. The Xanadu-Thorlabs partnership is an early example of this crossover. The Xanadu-EV Group partnership for heterogeneous integration bonding is another: EV Group’s bonding expertise connects multiple material platforms (silicon, lithium niobate, III-V semiconductors) onto a single chip, a process that could eventually become available to third-party photonic QPU designers.

The timeline for full photonic QOA is uncertain. My estimate: 3-5 years before an integrator can assemble a photonic quantum computer from independently sourced components with the same confidence that superconducting QOA provides today. The silicon photonics foundry ecosystem is the enabling condition; the computational model convergence is the bottleneck.

For the cost and procurement comparison across modalities, the contrast between photonic’s vertically integrated model and superconducting’s component model is one of the defining differences. For how each other modality handles the build challenge, see the superconducting, trapped-ion, neutral-atom, and silicon-spin build guides in this series.