DeepSeek cuts model prices as Chinese AI-compute projects advance
DeepSeek announced lower Flash-series pricing while Chinese companies reported new AI-compute hardware, circuit-board and infrastructure developments.
• DeepSeek will introduce new Flash-series rates at noon Beijing time on Sept. 10, with off-peak cached input at 0.02 yuan, uncached input at 1 yuan and output at 4 yuan per million tokens; its V4.1 Flash interim version is also entering testing with a new architecture and native multimodal support.
• Bichuan Intelligence’s board approved a plan for a new 1.5-million-square-meter annual AI-computing-power high-layer-count and high-end HDI circuit-board project, while Zhongjing Electronics said its 9- to 11-layer HDI products for computing-power accelerator cards are in rollout.
• Beijing-based Deep Space Matrix raised an angel round worth hundreds of millions of yuan to fund space AI-computing technologies, Star Ring Program systems engineering verification and industrial-chain development.
• Arm unveiled the Neoverse CSS N4 compute subsystem for cloud and data centers at its Arm Everywhere China event, with support for LPDDR6 and PCIe Gen7 and CPU frequencies of up to 3.8 GHz.
• Huawei’s Kirin 9050 Pro was described as the company’s first new high-performance Kirin chip in six years and its first to use logic-folding technology.
• Xiaomi’s 18 Fold introduced the company’s in-house Xuanjie O3 AI flagship SoC alongside HyperOS 4 and its MiMo on-device large model.
• Industry reporting linked notices of high-end MLCC price increases exceeding 30% to a widening supply-demand gap as domestic substitution accelerates.