The Compute Brief · Sep 10, 2026

China compute projects advance as DeepSeek and space-computing updates land

China’s AI compute news centered on new infrastructure agreements, model deployment updates, policy signals and chip-supply-chain developments.

• VNET Group and the Qingyang municipal government in Gansu signed a strategic cooperation agreement for a GW-scale integrated computing and power base, including computing infrastructure investment, green power and computing scheduling, and ecosystem clustering.

• ADA Space said the first eight satellites in the second batch of its independently developed “Star Computing” constellation have departed from its Taizhou computing power satellite innovation base and are expected to launch soon.

• DeepSeek V4.1 Flash went live on the National Supercomputing Internet Center, giving users access through the center’s Model Services API interface.

• DeepSeek said V4.1 Flash is a 552B-parameter MoE model with a Causal-Encoder-Decoder architecture, using 8B active parameters for input and 16B active parameters for output.

• DeepSeek merged its Fast, Expert and Image Understanding modes into a unified Intelligent mode, and delayed the shutdown of V4 Pro service until noon Beijing time on Sept. 14, 2026, after which it will route to V4.1 Flash.

• Zhipu AI, the Hangzhou municipal government and Shangcheng District launched the “Zhipu · Hangzhou Citywide Coding Plan,” described as China’s first district-level support scheme for AI model API usage, running from Sept. 10 to Dec. 9.

• A source denied a report that Moonshot AI is considering listings in both Hong Kong and Shanghai, while separate morning-brief items said sources confirmed DeepSeek is preparing for a STAR Market IPO with CITIC conducting due diligence.

• Policy and supply-chain items said China’s MIIT is boosting computing power infrastructure in the 15th Five-Year Plan, Beijing issued a development plan covering AI, chips and commercial space, Enflame Technology will list on Sept. 11, and a Shenzhen Inovance subsidiary plans to invest 1.85 billion yuan to expand high-end memory chip testing and packaging capacity.