Nov 5 AI digest: model releases and chip funding stayed dense while the compute supply side advanced across liquid cooling, supernodes and power-compute coordination — 10 items.

SenseTime launched SenseNova 6.8 Flash: cross-task, cross-session collaboration; outputs now span Web, SVG, Three.js, Blender and PPT, with dynamic slides and Agentic 3D design. (Sina Finance)

Lingchuan Technology closed a 1 billion yuan Series B: the Kuaishou-affiliated AI chip firm won backing from E-Town fund, Qiming Venture Partners and Zhaohui Capital; its fully domestic 3D-stacked chip has returned from fab, with mass shipment due next year and AI video generation as the next target. (Wallstreetcn)

Houmo AI raised billions of yuan in a Series B: its edge-side chip M50 has won over 100 customers as the company scales compute-in-memory into edge AI products. (Shanghai Observer)

Alibaba Cloud launched the Lingjun Zhenwu M890 supernode: first sold in the Ulanqab region; the first supernode in China to run models beyond 2 trillion parameters, with Kimi K3 and Qwen3.8 Max already served on it. (Cailian Press)

The Beijing-area compute corridor is expanding: Langfang's Runze International Information Port sits about 40 km from Beijing's Fifth South Ring Road with 0.33 ms latency, while Zhangbei's green power absorbs compute tasks locally. (Beijing municipal government portal)

Guangzhou's Tianhe district added 7 newly filed large models: a record high for the district as software firms and government scenarios adopt AI. (Guangzhou Daily)

Liquid cooling becomes "mandatory": rising AI rack density is pushing data-center cooling upgrades into a hundred-billion-yuan track. (Cailian Press)

Nvidia reportedly asked partners to limit Chinese customers: its in-China business keeps shrinking under export controls. (media reports)

Wuxi's "AI+manufacturing" matchmaking reached Singapore: supply-demand events landed in the Singapore Science City to link manufacturers with AI vendors. (jiwei.com)

Huzhou advances "power-compute coordination": joint scheduling of green power and compute is becoming standard for intelligent-computing bases. (Xinhua)

Model releases this week cluster on multimodal and agent capabilities, while compute-side attention shifts from single-chip performance to systemic constraints: power, cooling and cross-region scheduling.