TuringQ Gen3: Photonic Quantum Computer in a Standard IDC Rack, Built on Thin-Film Lithium Niobate Chips

Forum topic · 小凯 · 2026-09-12

Summary

On September 12, 2026, at the Pujiang Innovation Forum in Shanghai, Turing Quantum (TuringQ) founder Jin Xianmin unveiled TuringQ Gen3, described as the world's first rack-mounted, chip-level scalable photonic quantum computer. The system uses a standard IDC data-center rack and is built around a high-speed programmable thin-film lithium niobate (TFLN) photonic chip, which the company claims supports scaling to over 10,000 photons and millions of optical modes on a single chip. Gen3 relies on a space-time multiplexing architecture developed over more than a decade, using delay lines and electro-optic modulation to build T×S-scale quantum networks with a fixed set of physical devices. The machine integrates quantum light sources, programmable photonic processors, single-photon detection, and quantum-classical heterogeneous computing, with a full software stack (DeepQuantum, cloud platform, QAgent) and interconnects with domestic Chinese GPUs including Biren, Hygon, and Moore Threads. The company says Gen3 will explore applications such as world models in AI via Gaussian boson sampling, though this remains exploratory. The article also contrasts TuringQ's approach with PsiQuantum and Xanadu, and outlines remaining challenges: detector performance, squeezed-light source quality, and software ecosystem maturity.

Key points

On September 12, 2026, at the Pujiang Innovation Forum results release in Shanghai, Turing Quantum founder Jin Xianmin announced TuringQ Gen3, described as the world's first rack-form-factor, chip-level scalable photonic quantum computer.
The core is a high-speed programmable thin-film lithium niobate (TFLN) photonic chip, supporting single-chip scaling to the 10,000-photon level and millions of modes, per the company.
The architecture keyword is space-time multiplexing: compressed-squeezed-light pulses in S spatial modes circulate through delay-line storage while high-speed electro-optic modulation reconfigures the interferometer network each time step, building a T × S scale quantum network from a fixed set of physical devices.
The key Gaussian boson sampling unit is called "Zhiyuan 3" (致远3号).

Why thin-film lithium niobate

Silicon photonics is mature but limited in modulation speed and loss; TFLN is called the "gold-standard material" for photonic chips — fast modulation, low loss — but the process is hard and was long held abroad. TuringQ built China's first photonic chip pilot line, doing materials, design, fabrication, packaging, and testing in-house, cutting R&D iteration from six months to weeks.

Three generations

| Generation | Positioning | Key change | |---|---|---| | Gen1 | Research-grade dedicated photonic quantum computer | Proof it could be built | | Gen2 | Large-scale hybrid-integrated machine | Modular, rack prototype, compute access for research institutes | | Gen3 | Large-scale chip-level scalable machine | Core optics on a TFLN chip, standard IDC rack, on-demand node expansion |

A note on "chip-level quantum advantage"

The company's claim — chip-level quantum advantage "under strict boundary conditions of system scale and actual loss" — deserves caveats: Gaussian boson sampling is a dedicated sampling task, not universal computation. It shows the output distribution is hard to reproduce classically on specific problems; it does not mean running Shor's algorithm or general optimization. Loss also scales with photon count and mode count, which is why the qualifier exists.

Software stack and domestic compute integration

Hardware: the in-house XLink low-latency interconnect unifies QPU, quantum acceleration units, CPUs, and GPUs in one heterogeneous system.
Software: DeepQuantum framework, a quantum cloud platform, and QAgent agent tooling.
In August 2026, TuringQ reportedly completed interop with domestic GPU vendors including Biren, Hygon, and Moore Threads.
The company says Gen3 will explore world models in AI using Gaussian boson sampling for dynamics prediction of high-dimensional state spaces — explicitly at the exploration stage, with no public benchmark results.

Founder and competitive landscape

Jin Xianmin studied at USTC (with Pan Jianwei), did postdoctoral work at Oxford in one of the earliest photonic-quantum-chip groups (Marie Curie fellow), and returned to Shanghai Jiao Tong University in 2014. After rival PsiQuantum (founded by his Oxford-era competitor) received major Microsoft-backed investment, he founded Turing Quantum in February 2021.

Other photonic players: PsiQuantum (fusion-based, single-photon sources, leading funding) and Xanadu (squeezed states; Borealis reported quantum advantage in 2022). All three bet on photons' low loss and room-temperature operation but differ on encoding and fault tolerance; no route is proven terminal.

What's still missing

1. Single-photon detectors — efficiency, dark counts, crosstalk at million-mode scale. 2. Quantum light sources — squeezing level, brightness, purity. 3. Software ecosystem — need for more reproducible acceleration cases beyond research/pilot users (a "CUDA moment").

Industry side

TuringQ also makes optical interconnect products (CPO co-packaged optics, 1.6T/3.2T AI modulators, 110G satellite modulators) plus post-quantum cryptography and quantum key products. It has filed for IPO tutoring on the A-share market (advisor: Guotai Haitong).

References

1. 南方+《支持量子算力节点按需扩展，图灵量子发布三代机》, 2026-09-12 — https://www.nfnews.com/content/r6d5mmaJoE.html 2. 科创板日报, 2026-09-12 — https://www.chinastarmarket.cn/detail/2481436 3. Tencent News / IPO Early Know, 2026-09-12 — https://news.qq.com/rain/a/20260912A0B8ZQ00 4. TuringQ Unveils Gen3 Photonic Quantum Computer — https://cryptovka.com/news/turingq-unveils-gen3-photonic-quantum-computer-a-leap-in-scalable-computing 5. TuringQ official site and Jin Xianmin team's public space-time multiplexing research