The Compute Brief · Sep 11, 2026

DeepSeek rollout leads a busy day for China’s AI compute sector

DeepSeek advanced its model lineup as domestic GPU companies gained market attention, government procurement shifted fully to Chinese chip platforms, and new server and space-computing initiatives highlighted continuing infrastructure demand.

• DeepSeek launched V4.1 Flash as the successor to V4 Pro, with reporting emphasizing its lower GPU requirements for serving; the company will also continue V4 Pro API access beyond Sept. 14 at unchanged pricing.

• Moonshot AI attracted a French investment through Cathay Capital at a $30 billion valuation, while reports said Kimi K3 helped push annual recurring revenue above $1 billion in August and put the company on track for $2 billion in annualized revenue by year-end.

• Innocn shares rose more than 220% in their Shanghai STAR Market debut, valuing the Chinese GPU startup at 177.7 billion yuan, while Enflame reached a 170 billion yuan market capitalization on debut and Moore Threads fell limit down after its lockup expired.

• China’s central-government laptop procurement awarded all 10 lots to domestic chip platforms from Zhaoxin, Phytium, Loongson, Hygon and Kunpeng, with Intel and AMD absent from the selected suppliers.

• Loongson chairman Hu Weiwu said the company’s 9A1000 GPU and AI accelerator is expected to go on sale in the first half of next year, with its final price depending mainly on LPDDR4 memory costs.

• The global server market reached a record $166.3 billion in 2026’s second quarter, with revenue up 52% year on year and shipments up 15.4%, implying average server selling prices rose more than 30%; higher prices nevertheless did not curb shipments as enterprise and government buyers joined hyperscaler spending.

• Shanghai unveiled the Pearl Constellation Program, led by Dongfang Tiansuan with Shanghai Jiao Tong University and industry partners, to develop a global service network and integrated space-ground intelligent computing infrastructure by 2030.

• Researchers at the Chinese Academy of Sciences reported angstrom-scale atomic sliding that changes electrical resistance by ten million times, a result they said could support lower-power storage, computing and in-memory architectures once device arrays and supporting circuits are developed.