How Far Is Photonic Quantum Computing From Industrialization? From Jiuzhang 4.0 to Chip Mass Production

Moore's Law is approaching its physical limits, yet AI, drug discovery, and new materials exploration are hungrier than ever for compute. Quantum computing keeps getting pushed into the spotlight, and after the photonic quantum route repeatedly set records in the lab, the real question has surfaced: how far is it really from industrialization?

Why Photonic Quantum Is Being Bet On: From Quantum Supremacy to Global Optimality

The imaginative space of quantum computing is often summed up as a dimensional strike against classical compute. Qiu Disheng, CTO of Shuyiao Liangsheng, once used a comparison to explain this difference: stochastic gradient descent, commonly used in deep learning, amounts to probing a small area around a local point and feeling out the direction of iteration—a form of local search that easily gets stuck oscillating; quantum computing, by contrast, can scan the entire terrain of the solution space in one pass and directly locate the optimum—what he calls "quantum global optimality."

Jin Xianmin, Chief Scientist at TuringQ, offers a grander perspective. He argues that quantum computing deserves long-term attention not because of any single application, but because it is a foundational direction—following the Newtonian mechanical system—that can systematically empower the entire development of science. In the 1940s and 1950s, the transistor, the laser, and nuclear energy completed the first wave of quantum technology industrialization; today, humanity's leap in the ability to manipulate quantum systems, combined with the explosion of AI compute demand, is what has pushed quantum computing to the threshold of an industrial window.

Quantum Supremacy refers to the demonstration of a quantum computer's ability to outperform the most powerful classical supercomputer on a specific computational task. It typically selects problems that classical algorithms cannot efficiently simulate, such as Gaussian boson sampling or random circuit sampling, and uses experiments to show that a quantum device can complete in an extremely short time a task that would take a classical supercomputer tens of thousands of years or more. It must be emphasized that quantum supremacy verifies "compute advantage on a specific task," not general-purpose computing capability, and it does not directly equate to commercial value. It is an important milestone in quantum computing's journey from proof of principle to engineering, but there is still considerable distance to a fault-tolerant, programmable universal quantum computer.

Multiple Routes on the Same Stage: Where Does Photonic Quantum Stand?

Superconducting, ion trap, photonic quantum, and neutral atom—several technical routes have long competed on the same stage. Among public experiments, those that simultaneously satisfy the three conditions of "manufacturable, scalable, and demonstrated quantum supremacy" are concentrated mainly on the photonic quantum and superconducting routes.

Superconducting started earlier. According to public information, the 105-qubit "Zuchongzhi 3.0" superconducting prototype built by a team at the University of Science and Technology of China was 15 orders of magnitude faster than the fastest supercomputer at the time on a quantum random circuit sampling task, setting a new supremacy record for superconducting systems. The ion trap route has the highest gate fidelity and long coherence times, but system scaling is slow, and it currently remains mostly at the stage of high-precision simulation and proof-of-principle demonstrations; the neutral atom route has great potential for scaling array size, but gate fidelity and speed are still catching up, and engineering maturity is relatively low.

Photonic quantum, meanwhile, shows stronger compute advantage on specific problems. The "Jiuzhang 4.0" programmable photonic quantum prototype manipulated and detected 3,050 photons for the first time, achieving 10⁵⁴ times the speed of the world's fastest supercomputer on a Gaussian boson sampling task, establishing the world's strongest quantum computational supremacy. More critically, this route is moving from "laboratory apparatus" to "manufacturable platform." PsiQuantum's Omega photonic chipset is mass-produced on GlobalFoundries' 45nm line, with single-qubit state preparation and measurement fidelity of 99.98%, inter-chip interconnect fidelity of 99.72%, and plans to build data-center-scale quantum computing centers in Australia and the United States.

Customers Don't Care About Peaks—They Care About Whether It Can Run Stably on Site

Quantum computing does not lack "performance stories"; machines from various vendors have long demonstrated supremacy on specific tasks, and the superconducting route keeps breaking qubit-count records. But what industrial customers really care about is not laboratory peaks, but whether the machine can run stably at the customer site. This is also why the industry is shifting from comparing physical qubits to delivering logical qubits.

In its analysis, CICC noted that quantum computing is still in an early stage of evolution from the noisy intermediate-scale quantum era to fault-tolerant quantum computing, and that physical qubit count, gate fidelity, error-correction thresholds, and system stability remain key uncertainties. No technical route has yet broken through the two major bottlenecks of scale and fault tolerance, and quantum software, algorithms, and engineering capabilities still need to catch up. In other words, for quantum computing to truly enter industry, it must complete the leap from "can demonstrate" to "can run stably."

Photonic quantum has a natural advantage in environmental adaptability. Jin Xianmin said that starting with the second-generation TuringQ Gen2, TuringQ's photonic quantum computers have adopted a standard IDC rack form factor and can operate at room temperature. For industrial customers, there is no need to build a special operating environment—it can go directly into an existing data center like a server.

In September 2026, TuringQ launched the third-generation large-scale chip-level scalable photonic quantum computer, TuringQ Gen3, realizing a fully chip-based architecture based on high-speed programmable thin-film lithium niobate photonic quantum chips, using a space-time multiplexing architecture. Its Gaussian boson sampling compute unit is named "Zhiyuan 3.0," with official specifications of ten-thousand-photon scale per chip and million-mode count, achieving chip-level quantum supremacy within a defined loss boundary.

Logical Qubit is an abstract computational unit encoded from multiple physical qubits via quantum error-correcting codes, with its error rate suppressed low enough to support reliable execution of quantum algorithms. Physical qubits themselves are susceptible to environmental noise and decoherence, and single- and two-qubit gate operations also introduce errors; quantum error correction detects and corrects errors through redundant encoding and syndrome measurements, causing the equivalent error rate of a logical qubit to decrease exponentially as the number of physical qubits increases. It is widely believed in the industry that realizing fault-tolerant quantum computing requires thousands to millions of physical qubits to support hundreds of logical qubits. Therefore, the industry's shift from "comparing physical qubit counts" to "delivering logical qubits" marks that the focus of competition is moving from hardware scale to system reliability and error-correction capability.

Optical Loss: The Real "Process Node" of Photonic Quantum Chips

Unlike electronic chips, where process node matters, in photonic quantum chips, loss is an extremely critical metric. Excessively high loss rates do not merely mean "weaker signal"; they can directly cause photons to be lost and interference to fail, leading to computational errors. The larger the chip scale and the longer the optical path, the more devices a photon passes through, and losses stack up layer by layer—so keeping the loss of the entire link low is the key to whether photonic quantum chips can move toward large-scale production.

The TuringQ team has reduced the photon circuit transmission loss of its thin-film lithium niobate photonic quantum chip to 0.03 dB/cm, with key single-device insertion loss below 0.1 dB, fiber-to-chip single-end coupling loss reaching 0.9 dB/facet, and wafer-level quantum light source spectral purity exceeding 99%. This means that every step of loss—from photon generation, transmission within the chip, passage through devices, to entering and exiting the chip—has been substantially reduced, and the signal is pure enough to lay the foundation for larger-scale, more stable photonic quantum computing.

Behind these numbers is an engineering loop that is truly running. At the SJTU Wuxi Photonic Chip Research Institute, China's first photonic chip pilot line has been completed, forming a wafer-level process loop of "design—manufacture—test—feedback—re-iterate," shortening the chip iteration cycle to 2–3 weeks. It is not just about building a prototype in the lab; it is about trying to use semiconductor engineering methods to continuously reduce loss, improve consistency, and expand scale.

Jin Xianmin likens photonic quantum computing to a "hammer," and the key to industrialization is finding the "nails" where its advantages can truly be brought into play. Currently, TuringQ is exploring diversified business models around complete-machine delivery, compute services, and the construction of quantum intelligent computing centers, and has already carried out application cooperation in biomedicine, finance, and other fields. The criterion for judging whether a photonic quantum route is honest may not be how many papers it has published or how fast its demonstrations run, but whether anyone is continuously paying to use it.

Thin-Film Lithium Niobate (TFLN) is a photonic integration material platform in which sub-micron-scale lithium niobate single-crystal thin films are prepared on a substrate. Lithium niobate has excellent electro-optic coefficients, nonlinear optical effects, and a wide transparency window, and after thin-film formation, the optical field can be highly confined to the subwavelength scale, greatly enhancing light-matter interaction and enabling high-speed electro-optic modulation at low drive voltages. Compared with conventional bulk lithium niobate devices, TFLN chips offer smaller size, lower power consumption, and CMOS-process-compatible wafer-level manufacturing potential, and are regarded as an important integration platform for photonic quantum computing, high-speed optical communications, and microwave photonics.

One Machine Can't Enter Industry Alone, So the Road Must Be Built First

Even if usable quantum hardware is built, that does not equal completed industrialization. For quantum compute to generate commercial value, it must be recognized by scheduling platforms and be openly callable by industry users; a single-point hardware breakthrough cannot close the full loop.

In industrial collaboration on photonic chips, Tiantong Co. supplies materials, Luceda provides design software tools, and the SJTU Wuxi Photonic Chip Research Institute is responsible for manufacturing implementation. Equipment, materials, software, and application parties reached cooperation after long negotiations, with the goal of jointly improving the upstream and downstream ecosystem of photonic chips.

The advancement of the Advanced Intelligent Computing Alliance (FAIC) is another collaborative exploration at the industry level. This alliance was originally initiated in 2022 by four companies—TuringQ, MetaX, UCloud, and NeuroHawk—covering different tracks such as quantum computing, compute chips, compute center construction, and brain-computer interfaces. This year, with TuringQ serving as rotating chairman, the alliance made a directional choice and ultimately chose to expand, bringing in embodied intelligence companies such as Fourier Intelligence; membership grew from the initial four to dozens. Jin Xianmin's judgment is that it is difficult to leverage industry scale by a few institutions huddling together alone; only with more participants from different fields can there be a foundation for the scaled development of new compute chip iteration, intelligent computing center deployment, and compute-enabled industry applications.

The Guangdong-Hong Kong-Macao Quantum-Classical-Intelligent Fusion Computing Center targets the practical difficulties of deploying and calling quantum compute. TuringQ is jointly advancing this project with Asia Computing Power and Zhicheng Suanwang, benchmarking against overseas quantum-classical hybrid computing centers that have already been deployed at scale. Jin Xianmin noted that dozens to hundreds of such compute nodes have been built in Europe and the United States, while China does not yet have a mature benchmark. The construction plan for this project has been finalized, with the core being to achieve joint debugging among QPU, CPU, and GPU and truly incorporate quantum compute into the platform scheduling system. NVIDIA has already launched a quantum-classical fusion solution connecting GPUs to quantum computers, which is also a trend that has already emerged in the industry.

Quantum-Classical Hybrid Computing refers to a computing architecture in which quantum processing units (QPUs) and classical CPUs and GPUs work together through a unified scheduling platform. Its basic idea is to offload subtasks suitable for quantum acceleration (such as combinatorial optimization and quantum chemistry simulation) to the QPU, complete the remaining tasks with classical compute, and achieve dynamic allocation of the two types of resources through high-speed interconnection and task orchestration. NVIDIA has launched solutions connecting GPUs with quantum computers, and dozens of quantum-classical hybrid compute nodes have been built in many countries in Europe and the United States. This model is regarded as an important path for quantum compute to generate practical value ahead of the maturity of fault-tolerant quantum computing, and it also imposes new requirements on scheduling software, interface standards, and data center infrastructure.

What Remains Is Still a Matter of Engineering

The outside world often views photonic quantum computing through the lens of "the boy who cried wolf," which essentially confuses the time gap between scientific breakthroughs and industrial maturity. Today, photonic quantum has moved beyond the pure proof-of-principle stage: there are industry-shaking quantum supremacy results, room-temperature rack-mounted complete machines, a closed pilot-line process loop, paying industrial customers, and even early布局 of quantum-classical-intelligent fusion computing centers, attempting to turn quantum compute into a schedulable standard unit within intelligent computing systems.

But impressive parameters do not equal industrial victory. Stable supply of high-performance photonic materials and key devices, wafer-level manufacturing yield, loss control in complex systems, and large-scale quantum error correction and fault-tolerant computing capabilities all still require sustained breakthroughs. The industry's main battlefield has shifted from publishing papers and breaking photon-count records to production-line yield, long-term system stability, software ecosystems, and customers' willingness to pay.

This does not mean the challenges at the physical level are over. On the contrary, large-scale photonic quantum computing still requires sustained work on light sources, loss, detection, quantum error correction, and system architecture. Scientific breakthroughs and engineering capabilities need to advance in tandem. Quantum computing may not quickly overturn existing compute, but it is by no means a false story—the technological wolf has not immediately pounced on industry, but it has indeed walked out of the lab; it still must climb the high mountains of manufacturing, engineering, and application before it truly reaches the wilderness of commercialization.

One-line Summary

Photonic quantum computing has left the lab, but industrialization still requires climbing the three mountains of manufacturing, engineering, and application.

Paragraph Summary

With quantum supremacy and room-temperature operation advantages, photonic quantum computing is moving from laboratory apparatus to a manufacturable platform. TuringQ has launched the chip-level scalable photonic quantum computer TuringQ Gen3, reduced transmission loss of thin-film lithium niobate chips to 0.03 dB/cm, and built a wafer-level pilot line that shortens the iteration cycle to 2–3 weeks. The focus of industrial competition has shifted from physical qubit count to logical qubit delivery, system stability, and customers' willingness to pay. However, engineering challenges such as high-performance material supply, manufacturing yield, and large-scale quantum error correction remain severe; hardware breakthroughs alone cannot close the commercial loop, and coordinated progress is needed across materials, design software, manufacturing, scheduling platforms, and compute centers.