Quantum Computing’s Next Test Is the Factory Floor
Xanadu is taking key photonic quantum components to GlobalFoundries’ 300 mm line in Malta, New York. The agreement does not mean a useful fault-tolerant quantum computer has arrived. It means one of the field’s hardest questions—whether delicate quantum hardware can be made repeatably at semiconductor scale—is moving from the laboratory toward a U.S. production line.
On October 6, Canadian photonic-quantum company Xanadu and U.S. chip manufacturer GlobalFoundries announced a multiyear partnership aimed at industrializing parts of Xanadu’s hardware stack on GlobalFoundries’ 300 millimeter manufacturing line in Malta, New York. The first work packages center on ultra-low-loss silicon nitride photonics—the optical pathways that carry quantum light through a chip—and superconducting nanowire single-photon detectors, or SNSPDs, which count the photons used in the machine’s computation.
Those components are not the entire computer. They are, however, unusually consequential components for Xanadu’s architecture. Photonic quantum computing depends on producing, routing, interfering and detecting quantum states of light with extremely low error and extremely low loss. A laboratory can prove that a device works once. A commercially relevant system needs many nearly identical devices, fabricated with consistent performance, integrated into modules, packaged, tested and delivered at volumes that make the overall machine economically plausible.
That is why the location matters. Malta is not a university cleanroom or a one-off pilot line. It is part of GlobalFoundries’ U.S. semiconductor manufacturing network, and the company is explicitly building a quantum foundry business around the idea that multiple quantum architectures will eventually need industrial semiconductor infrastructure. The Xanadu partnership is an early test of whether that proposition can work for some of the most sensitive photonic components in the stack.
The announcement is about industrialization, not a finished quantum computer
Xanadu and GlobalFoundries said the partnership will transfer and scale specific photonic technologies on GlobalFoundries’ 300 mm line. The initial emphasis is on silicon nitride, or SiN, photonics and SNSPDs. Xanadu describes SiN as the low-loss optical “circuit board” that guides photons through the system. The detectors are used to resolve how many photons arrive in a pulse, a capability needed for the company’s measurement-based quantum architecture.
The wording matters. The companies are not saying a utility-scale machine is now in production. They are saying they intend to create a manufacturing foundation for future fault-tolerant demonstrators and, eventually, quantum data-center hardware. That is a forward-looking engineering program with significant execution risk. Process transfer can fail to meet target performance. Yield can be poor. Packaging can erase gains made on the wafer. Detectors that look excellent on a small sample can vary across large lots. None of those problems are unusual in semiconductor manufacturing; quantum hardware simply gives engineers much less room for error.
Still, the shift from bespoke fabrication to an industrial wafer line is strategically important. A quantum computer large enough to do something economically valuable is expected to contain vast numbers of devices and interconnects. If those components require artisanal handling, one-off calibration or unusually low-yield fabrication, the system can become too expensive and too fragile long before it becomes useful. In that sense, “manufacturable” is not a supporting adjective. It is part of the technical definition of a practical machine.
Keep the photons moving
Ultra-low-loss waveguides are intended to carry quantum light across photonic circuits while minimizing absorption and scattering that can destroy the information encoded in those photons.
Count faint quantum signals
Superconducting nanowire detectors are used to detect individual photons with high efficiency. In Xanadu’s architecture, photon-number-resolving detection is central to preparing and processing quantum states.
Replace one-off success with repeatability
Large-wafer semiconductor manufacturing is built around statistical process control, high-throughput metrology, yield learning and repeatable recipes—exactly the disciplines quantum hardware must eventually absorb.
Photonic quantum computing trades one set of hardware problems for another
Quantum computers can be built around different physical carriers of information. Some systems use superconducting electrical circuits, some trap ions or neutral atoms, some encode information in semiconductor spins, and photonic systems use states of light. No architecture has yet established a decisive, general-purpose victory. Each has different strengths, different failure modes and different scaling bottlenecks.
Photonics offers several obvious attractions. Light is naturally suited to networking, fiber-optic infrastructure is mature, and many photonic components can operate near room temperature. That makes modular architectures conceptually appealing: separate racks or modules can be connected by optical links instead of forcing every qubit onto one giant monolithic processor.
But photons are also slippery computational objects. They do not naturally interact strongly with one another, which complicates deterministic logic. Photonic architectures often rely on measurement, interference, feed-forward control and carefully prepared resource states. And the most basic engineering reality is unforgiving: if a photon is absorbed, scattered, misrouted or missed by a detector, the quantum information it carried may be lost. Error correction can tolerate some failure, but only below architecture-specific thresholds. That is why component loss and detector efficiency are not merely optimization metrics. They can determine whether fault tolerance is reachable at all.
Much of the optical routing and control can avoid the extreme refrigeration used by some other qubit platforms. But superconducting single-photon detectors themselves require cryogenic operation. A practical system still has to integrate optical modules, cryogenic detector hardware, electronics, fiber routing and software into one reliable machine.
Aurora showed the architecture could be networked; the next battle is performance at scale
Xanadu’s manufacturing push makes more sense in the context of Aurora, the modular photonic system the company reported in Nature in 2025. The research system tied together 35 photonic chips across four standard server racks and used roughly 13 kilometers of optical fiber. It demonstrated a collection of capabilities needed for a larger measurement-based quantum computer, including resource-state preparation, multiplexing, networking, real-time feed-forward and decoding.
The peer-reviewed paper was careful about what had and had not been accomplished. Aurora was described as a scale model rather than a utility-scale computer, and it remained below the performance required for fault tolerance. The researchers identified optical loss as the dominant and most challenging hurdle to crossing the fault-tolerant threshold. That finding points directly toward fabrication. Better materials, smoother interfaces, tighter control of waveguide geometry, more efficient couplers and better detectors can all reduce the probability that a quantum signal disappears before it contributes to a computation.
That creates a very different development path from simply adding more devices. If the loss per component is too high, making the machine larger can make the overall problem worse. Each additional switch, connector, fiber edge or detector becomes another place where the quantum state can degrade. Scale therefore has to arrive together with quality. The foundry partnership is best understood as an attempt to make those two goals converge.
Prove the device physics
A small number of devices demonstrate the desired optical behavior, detector efficiency or materials performance under carefully controlled conditions.
Translate performance into a fab-compatible flow
Materials, lithography, deposition, etch, cleaning and metrology steps have to fit a production environment without destroying the quantum properties that made the laboratory device attractive.
Measure variation across wafers and lots
A commercial process has to produce many devices within specification, not just a few exceptional parts. Statistical variation becomes a systems-design input.
Package, cool, connect and test the pieces
The final performance depends on interfaces: chip-to-fiber coupling, detector packaging, cryogenic connections, control electronics and network stability all have to work together.
Malta gives quantum hardware something startups rarely own: industrial semiconductor infrastructure
GlobalFoundries’ role is not to decide which quantum architecture will win. Its Quantum Technology Solutions business is explicitly designed to support multiple approaches, including photonic, superconducting, trapped-ion, topological and spin-based systems. The company is positioning itself as the manufacturing layer beneath a fragmented quantum ecosystem.
That is a familiar semiconductor business model with a quantum twist. Most chip designers do not own leading manufacturing plants. Foundries spread the enormous fixed cost of equipment, facilities, contamination control, metrology and process engineering across many customers. Quantum companies can benefit from the same economics if their unusual materials and device requirements can be converted into stable foundry processes.
For U.S. technology policy, the domestic location adds another dimension. A large quantum computer would not be just a scientific instrument. If the technology becomes economically useful, it could touch national-security applications, advanced materials, chemistry, logistics, drug discovery and other strategic fields. Washington has therefore begun treating quantum fabrication capacity as part of the country’s critical technology base rather than as a purely academic research question.
The foundry partnership lands one month after a major U.S. quantum manufacturing award
On September 8, the U.S. Department of Commerce finalized a CHIPS and Science Act research-and-development award of up to $375 million for GlobalFoundries. The stated purpose is to support a secure domestic quantum foundry that can serve multiple quantum architectures and to advance cryogenic CMOS process design kits, advanced packaging and heterogeneous integration. That federal award is broader than the Xanadu partnership and should not be described as a direct subsidy to Xanadu. But the timing makes the policy context impossible to miss.
The Commerce Department’s 2026 quantum portfolio is deliberately multi-modal. Rather than declare a single winner, the government has backed foundry capacity and engineering programs across superconducting, photonic, neutral-atom, trapped-ion and semiconductor-spin approaches. The logic is that the most consequential bottlenecks are not all located inside the qubit itself. They include device reproducibility, optical complexity, cryogenic systems, readout, interconnects, packaging and the ability to manufacture specialized hardware consistently.
That portfolio approach also acknowledges how uncertain the field remains. Quantum companies routinely publish impressive technical milestones, but it is still unclear which architectures can combine error correction, system scale, reliability and cost into something that beats the best classical alternatives on commercially meaningful work. Funding manufacturing infrastructure is a bet that, even before the winning architecture is known, the United States can own more of the process knowledge and production capability required by several plausible winners.
In quantum computing, the “supply chain” is becoming part of the scientific experiment. A design is not truly scalable until the materials, detectors, packages, interconnects and test methods can scale with it.Editorial analysis based on the manufacturing and benchmarking programs now underway in the United States.
A foundry agreement is evidence of a path to scale—not evidence that fault tolerance has been achieved
The quantum industry has spent years wrestling with a credibility problem: laboratory milestones are real, but the distance from a milestone to a useful machine is easy to underestimate. The Xanadu–GlobalFoundries partnership deserves attention precisely because manufacturing is one of those missing bridges. It should not, however, be treated as proof that the bridge has already been crossed.
First, the companies have not disclosed a production volume, a public yield target or a date when the relevant components will meet the performance needed for a full fault-tolerant system. Second, industrial manufacturing can expose defects that small research runs hide: across-wafer variation, thermal stress, contamination sensitivity, packaging loss and long-term drift. Third, even perfect detectors and waveguides would not be sufficient on their own. A fault-tolerant machine also needs sources, switches, fast classical control, decoders, timing, networking and software that meet the architecture’s combined performance budget.
Finally, “utility-scale” is not simply another way to say “large.” DARPA’s Quantum Benchmarking Initiative uses a more demanding concept: a utility-scale system should deliver computational value that exceeds its cost. That definition forces attention onto the full machine—power, refrigeration, calibration, uptime, error correction, networking, maintenance and application performance—not just qubit count.
Three claims to keep separate
Confirmed now: Xanadu and GlobalFoundries have announced a multiyear partnership to industrialize specified photonic components on the Malta 300 mm line.
Targeted next: the output is intended to support Xanadu’s future fault-tolerant demonstrators and high-performance quantum modules.
Not yet established: that the resulting process will meet all loss, yield, integration and cost requirements for a utility-scale quantum computer.
Optical loss is the number the foundry cannot negotiate with
In conventional computing, a weak electrical or optical signal can often be regenerated. Quantum information is different. Copying an unknown quantum state is forbidden by the no-cloning theorem, and aggressive amplification can destroy the state engineers are trying to preserve. That makes cumulative loss especially punishing for photonic architectures.
A single component can look excellent in isolation and still fail at system scale if the architecture requires a photon to survive dozens or hundreds of interactions. Waveguide propagation loss, splitter loss, switch loss, coupling loss and detector inefficiency all multiply. The result is a system-level budget in which many “small” imperfections compound into a large probability that the required photon never reaches the right detector.
This is why low-loss silicon nitride and high-efficiency superconducting detectors sit at the center of the new manufacturing program. They attack the problem from opposite ends of the optical path: keep more quantum light alive while it moves through the chip, then detect more of what arrives. The hard part is doing both at the same time, at scale, without introducing new losses through packaging or fabrication variation.
Even if the wafer works, the system still has to survive packaging, cooling and time
Quantum roadmaps often look deceptively linear on a slide: improve the device, scale the device, assemble the system. Real hardware programs are rarely that orderly. Solving one bottleneck can move stress somewhere else. A detector with exceptional efficiency may impose a tougher cooling load. A lower-loss waveguide may require a new material stack that complicates packaging. A denser photonic circuit may create routing congestion. A faster clock can overwhelm control electronics or data pipelines.
The deeper value of a foundry relationship is that these tradeoffs become measurable in a production context. Process engineers can characterize variation. Packaging teams can design around known tolerances. System architects can use real distributions instead of optimistic single-device numbers. Over time, design rules can encode what is manufacturable, and the quantum team can stop rediscovering the same boundary conditions one prototype at a time.
How many devices meet spec?
Large systems need predictable output across wafers and lots. A process that produces a few outstanding devices and many unusable ones does not scale economically.
Can light get on and off the chip efficiently?
Ultra-low-loss waveguides do little good if fiber coupling, alignment or package interfaces consume the saved optical budget.
Can detectors be cooled as the system grows?
Superconducting detectors require refrigeration. Scaling detector count means scaling thermal design, wiring, mechanical stability and serviceability.
Does performance stay stable over time?
A data-center machine must be calibrated, maintained and operated repeatedly. Drift, contamination and component aging matter even when they rarely appear in benchmark headlines.
The meaningful milestones will be manufacturing metrics, not just bigger qubit headlines
The partnership’s real significance will become clearer only when the companies disclose measurable progress. For readers trying to separate industrial momentum from marketing language, several kinds of evidence matter more than broad promises about future quantum data centers.
One is repeatability: can the partners produce multiple wafers with similar optical loss and detector efficiency? Another is integration: can the detectors, photonic routing and fiber interfaces be packaged together without erasing the gains seen on individual devices? A third is system demonstration: do those production-line components improve a complete fault-tolerant prototype rather than only a component benchmark? Finally, cost and throughput matter. Utility-scale hardware will need a manufacturing flow that can be repeated economically, not a boutique process whose cost rises faster than the computer’s useful output.
Five signals that would make this partnership more consequential
- Published wafer-level yield or uniformity data for the SiN and detector processes.
- Demonstrated detector efficiency and optical loss after full packaging, not only at the die level.
- Evidence that production-line components are incorporated into a larger Xanadu fault-tolerant demonstrator.
- Independent validation or peer-reviewed characterization of the manufacturing process.
- A credible path from engineering lots to sustained volume with known cost, reliability and service requirements.
Washington is increasingly asking a harder question: can any architecture become useful enough to justify its full cost?
DARPA’s Quantum Benchmarking Initiative is one sign that the U.S. government wants a more disciplined answer. The program is designed to evaluate whether quantum architectures have a credible path to utility-scale operation, with an explicit target of determining whether such systems could be built by 2033. DARPA defines the problem broadly: not only whether a device can perform quantum operations, but whether the integrated machine can deliver computational value greater than the cost of building and running it.
That framing is useful for interpreting the Xanadu–GlobalFoundries agreement. Manufacturing scale is necessary because cost is part of utility. If a quantum computer needs millions of specialized components, then wafer throughput, detector yield, packaging labor, calibration time, cooling infrastructure and replacement rates all become part of the computer’s effective price. A theoretically elegant architecture can still lose the commercial race if its hardware cannot be built and maintained economically.
It is also why the foundry’s multi-modal strategy may be more durable than any single quantum roadmap. Even if one architecture falls behind, the process expertise developed for cryogenic electronics, silicon photonics, heterogeneous integration, superconducting materials or advanced packaging can be reused across other quantum systems—and often across conventional high-performance computing as well. The boundary between “quantum manufacturing” and the broader advanced-semiconductor ecosystem may become increasingly porous.
The most important word in the announcement may be “manufacturing.”
Quantum computing is entering a phase in which spectacular experiments are no longer enough. The field needs process windows, supplier relationships, design rules, yield curves, packaging standards, test infrastructure and facilities that can make difficult components the same way twice. That is slower and less theatrical than announcing a new benchmark, but it is how technologies leave the laboratory.
For Xanadu, the GlobalFoundries partnership is a bet that photonic quantum hardware can inherit the discipline of the semiconductor industry without losing the fragile optical performance that makes the architecture work. For GlobalFoundries, it is a bet that quantum systems—whatever modality ultimately wins—will need a domestic industrial base. The next few years will show whether those bets produce something more valuable than another roadmap: repeatable hardware that survives the jump from physics to production.
Quantum manufacturing: quick answers
What exactly are Xanadu and GlobalFoundries planning to manufacture?
The announced first phase focuses on industrializing Xanadu’s ultra-low-loss silicon nitride photonics and superconducting nanowire single-photon detectors on GlobalFoundries’ 300 mm line in Malta, New York. The companies also said they are exploring additional future quantum modules.
Does this mean Xanadu has a fault-tolerant quantum computer?
No. The agreement creates a manufacturing pathway for components intended for future fault-tolerant demonstrators. It does not establish that a complete fault-tolerant or utility-scale system has been achieved.
Why is a 300 mm fab important?
Large-wafer semiconductor manufacturing can support higher throughput, sophisticated process control, metrology and yield learning. The significance is less about wafer diameter by itself than about moving quantum devices into an industrial manufacturing environment built for repeatable production.
Is the federal $375 million award funding Xanadu directly?
The announced CHIPS R&D award is to GlobalFoundries for a secure domestic quantum foundry and related R&D. It is broader than the Xanadu partnership. The two developments are connected by the foundry’s strategy, but they should not be conflated as the same agreement.
Primary sources and further reading
- Xanadu — strategic partnership with GlobalFoundries, October 6, 2026
- U.S. Department of Commerce / NIST — final CHIPS R&D award to GlobalFoundries, September 8, 2026
- GlobalFoundries — launch of Quantum Technology Solutions, May 21, 2026
- Nature — “Scaling and networking a modular photonic quantum computer,” published January 22, 2025
- DARPA — Quantum Benchmarking Initiative
- DARPA — Underexplored Systems for Utility-Scale Quantum Computing
Comments
Post a Comment