Quantum chip gives China's AI data centres '1k-fold' speed boost (www.scmp.com)

A Chinese research team led by Jin Xianmin — founder of Turing Quantum — announced a photonic quantum chip that it says can deliver a "1,000-fold" speed boost for AI data‑centre workloads and won the “Leading Technology Award” at the 2025 World Internet Conference Wuzhen Summit. The group claims several engineering milestones: co‑packaging of photonic components with electronics, chip‑level integration and wafer‑scale mass production of photonic quantum chips — steps the team describes as a world first. They also signalled plans to scale to chips that handle larger numbers of photons. Photonic quantum devices are already being trialled in aerospace, biomedicine and finance, the team added. Technically, the device exploits photons rather than electronic charge for information processing, offering high parallelism, low latency and potentially much better energy efficiency for certain linear-algebra-heavy tasks common in ML (e.g., matrix multiplications, inference pipelines). If the claimed speedups hold under independent benchmarking, the chips could change AI compute architecture by offloading specific kernels to optical hardware and enabling denser, faster inference or accelerator fabrics. Important caveats remain: the 1k‑fold claim is reported by the developers and likely workload‑dependent; practical deployment hinges on error handling, photon‑count scaling, integration with silicon ecosystems and verified benchmarks across training and inference.