No Data
No Data
The focus of AI development is shifting from training to inference, with customized ASICs expected to experience a surge.
① Driven by DeepSeek, CSP providers are expected to more actively develop self-owned ASIC solutions with lower costs, shifting their focus from AI training to AI inference, which is estimated to gradually increase the proportion of AI inference servers to nearly 50%. ② GTJA points out that AI ASIC chips have advantages in power consumption and cost, making them a necessary choice.
Huaxi: Assigning a "Recommended" rating to the SerDes Industry, the significant empowerment of intelligent driving opens up market space for the industry.
Core beneficiary symbols: Longxun Co., Ltd. (688486.SH), Hunan Goke Microelectronics (300672.SZ), Yutai Microelectronics (688515.SH).
Hunan Goke Microelectronics (300672.SZ): Granted 2.6608 million restricted Stocks to the incentive objects.
Gelonghui announced on February 17 that Hunan Goke Microelectronics (300672.SZ) has met the granting conditions outlined in its "2025 Restricted Stock Incentive Plan (Draft)." The supervisory board agreed that the initial grant date for the 2025 restricted stock incentive plan is February 14, 2025, and agreed to grant 2.6608 million shares of restricted stock to 247 incentive recipients who meet the initial grant conditions.
Hunan Goke Microelectronics (300672.SZ): Deeply layout self-developed all-scenario NPU, achieving standard configuration of NPU for the entire series of chips, providing a solid foundation for edge AI deployment.
On February 17, Glonghui reported that Hunan Goke Microelectronics (300672.SZ) stated on its investor interaction platform that the company has deeply invested in developing a self-researched all-scenario NPU, achieving standard NPU in its entire chip series, providing a solid foundation for AI deployment at the edge. The company's next-generation 4K AI visual processing chip GK7606V1 series is equipped with a self-researched AI ISP engine, featuring a dual-core A55 processor, with a maximum computing power of 2.5 TOPS, supporting core technologies such as 4K encoding and decoding, AI ISP, dual 3D noise reduction, image stabilization, multi-camera stitching, and multi-spectral fusion, offering high picture quality, low bitrate, low memory, and low latency.
Hunan Goke Microelectronics (300672.SZ): The AI Edge Computing chip developed by the company is a computing power chip designed based on the underlying architecture of large models.
On February 17, 格隆汇 reported that Hunan Goke Microelectronics (300672.SZ) stated on the investor interaction platform that the AI Edge Computing chip developed by the company is a computing power chip designed based on the underlying architecture of large models. It is equipped with the company's self-developed NPU, which can adapt to mainstream large models in the market, including lightweight LLM language large models, AIGC generative models, CV large models, and multimodal large models. It supports mainstream computing frameworks and development tools and can be applied in Edge Computing, embodied intelligence, Siasun Robot&Automation, industrial vision, and other fields; it can be used in AI PCs, AIoT device terminals, interactive terminals with displays, and AI cameras.
Hunan Goke Microelectronics (300672.SZ): The integrated circuit Fund has cumulatively reduced its Shareholding by 2.6704%.
Gelonghui, February 11 – Hunan Goke Microelectronics (300672.SZ) announced that the company recently received a notice from the Shareholder Integrated Circuit Fund regarding the completion of its Shareholding reduction plan and the change in shareholding that surpassed 1% in integer multiples. It was learned that the Integrated Circuit Fund reduced its holdings of company shares by 2,160,064 shares through centralized bidding trades from November 12, 2024, to December 2, 2024, accounting for 0.9948% of the company's total share capital, and 1.0000% of the total share capital after excluding the number of shares held in the company's repurchase special securities account, which is 1,134,182 shares.