Brokerage morning meeting highlights: DeepSeek reopens its source code, focusing on changes in AI applications.

At today's Brokerage morning meeting, HTSC proposed that DeepSeek be open-sourced again and focus on changes in AI applications; China International Capital Corporation believes that AI+finance is expected to reach a scaling inflection point; China Securities Co.,Ltd. stated that the performance resonance of lithium battery demand is signaling the beginning of a new cycle.

cls.cn19:24 ET

The divergence in the technology stock sector is increasing, pay attention to the rotation opportunities in low-level directions such as Solid State Battery and CSI Consumer 360 index.

Track the entire lifecycle of the main Sector.

cls.cnMar 2 20:19 ET

As high as 545%! DeepSeek has released its cost-profit margin for the first time. What is the impact on computing power?

According to DeepSeek's calculations, the theoretical daily profit of the V3/R1 inference system can reach 3.46 million yuan. The founder of Silicon-based Flow, Yuan Jinhui, commented that the V3/R1 architecture is significantly different from other mainstream models. Brokerage opinions suggest that in the medium to short term, technology giants will continue to make ongoing and large-scale investments in the field of AI computing power, which will remain a highly certain event.

cls.cnMar 2 05:58 ET

The main Communications business is experiencing a downturn! ZTE's revenue and profit both declined last year, with Net income dropping 9.66% year-on-year | Earnings Reports insights.

ZTE faced pressure on revenue last year, with revenue of 121.299 billion yuan, a year-on-year decrease of 2.38%. Profitability also declined, with net income attributable to ordinary shareholders of the listed company at 8.425 billion yuan, a year-on-year decrease of 9.66%. The company stated that it will accelerate the transformation towards "connectivity + computing power" in an attempt to expand new market space.

wallstreetcnFeb 28 07:53 ET

China United Network Communications: Optimized and upgraded the DeepSeek series models, saving an average of 30% in inference computational load.

① China United Network Communications announced an "adaptive slow thinking" optimization upgrade for the DeepSeek series models, saving an average of approximately 30% in inference computation, and it has been open-sourced, making it a first in the industry; ② The specific modifications include difficulty-adaptive fine-tuning, secondary distillation, and difficulty-adaptive reinforcement learning to reduce redundant outputs and resource consumption.

cls.cnFeb 25 19:51 ET

Technology stocks continue to soar, and the dual-main line market of DeepSeek and Siasun Robot&Automation is expected to persist.

Track the entire lifecycle of the main Sector.

cls.cnFeb 23 20:26 ET