In early 2025, the surge of DeepSeek, an open-source AI platform acclaimed for its efficiency and low cost, set the tech world ablazeSince its launch on January 20, the platform quickly amassed a daily active user (DAU) count exceeding 22 million within just 21 days, leading it to be regarded as a phenomenal product in the global AI sector.
This enthusiasm has transcended the realm of technology, triggering subsequent reactions within financial marketsBy February 7, the A-share market experienced a wave of soaring stocks categorized under the DeepSeek phenomenonNotable contenders included Meige Intelligent (002881.SZ), which saw its stock increase by five consecutive limits, alongside Qingyun Technology (688316.SH) and Youkede (688158.SH) that both recorded three consecutive surges (600126.SH). Other stocks like Ankai Micro (688620.SZ), Hangjin Technology (000818.SZ), and Xiling Information (300588.SZ) also noted remarkable increases, all surpassing 5% on opening.
However, the pivotal question arising amidst this capital frenzy is whether these stocks genuinely benefit from DeepSeek's technological breakthroughs or are they merely riding the coattails of a trending topic.
Examining the AI industry chain, it becomes apparent that stocks related to DeepSeek are predominantly concentrated within the upstream foundational layers and mid-level technological frameworksYet, investors still need to scrutinize whether these companies possess true technical competence and operational synergy.
Meige Intelligent, for instance, exhibited particularly brilliant stock performanceAs of the market's close on February 7, its stock surged by 10%, completing a total transaction of approximately 28.12 million yuanNotably, it has seen a dramatic rise of 61.08% over the preceding five trading days and 92.45% over the last thirty.
On January 26, the firm proudly proclaimed its robust AI computing module matrix and experience in deploying large models at the edge via their official WeChat platform
Advertisements
They announced their accelerated effort in developing the DeepSeek-R1 model for practical applications on the edge, alongside an extensive solution that integrates cloud and end-devicesMoreover, the company plans to launch advanced AI hardware in 2025, boasting a single module power of 100 Tops, with a future projection exceeding 200 Tops.
Despite these ambitious claims, skepticism persists regarding the transparency of their technological advancements.
As of the time of reporting, Meige Intelligent has not responded to inquiries about specific collaborations with DeepSeek on the interactive platform provided by the Shenzhen Stock ExchangeOn February 7, a staff member communicated to a reporter that they were “not clear” about specific partnership details.
Currently, the company has deployed several top models, including LLaMA-2 and Qwen, within its high-performance AI modules, affirming its capability across the entire workflow from model compression to framework adaptations.
However, it is essential to note that Meige Intelligent's primary income stream continues to stem from wireless communication modules and Internet of Things (IoT) solutionsTheir half-year report indicated that, between January and June 2024, revenue from wireless communication modules reached 1.236 billion yuan, accounting for 94.64% of total incomeOther business streams contributed a mere 69.95 million yuan, representing 5.36%.
In their official communication, Meige further divulged that their AI product AIMO, founded on Qualcomm Snapdragon’s high-performance computing platform, encapsulates 48 TOPS of AI computing power, catering to the inference demands of 7B parameter-level modelsYet, the specifics of this intelligent product left company representatives hesitant to provide further information.
Chang Qiaoyu, a senior analyst from the Leopard Research Institute, indicated that according to Gartner's market data, the global AI chip market is anticipated to breach $120 billion by 2027, with a compounded annual growth rate nearing 30%. The Chinese market, in particular, is seeing rapid advancements in proprietary computing chips, with firms like Cambricon, Biren Technology, and Denglin Technology actively engaging in the AI inference chip landscape, while major players like Huawei, Alibaba, and Baidu continue to optimize their cost-performance solutions.
However, Huatai Securities analyzes a complex trend in DeepSeek's impact on computing power demands, alleging a short-term suppression followed by long-term growth
Advertisements
Initially, DeepSeek's cost-effective training methods might reduce training appetites, but as model applicability grows, inference demands are poised for significant increases, similar to historical patterns where steam engines reduced coal consumption but ultimately increased overall coal utilization.
In the mid-level technological arena, DeepSeek’s rise has catalyzed a favorable environment for high-quality data processing, storage, and transmission sectors.
On February 6, Qingyun Technology announced through their WeChat platform that AI computing service "CoresHub" has officially integrated DeepSeek Janus-Pro-7B WebUI model, supporting multi-modal understanding and text-to-image generation services, available through container instances within their AI computing cloud service.
Qingyun Technology is a notable provider of enterprise cloud services and digital solutions, strategically positioned in the hybrid cloud market, enabling seamless public and private cloud integration that delivers a cohesive experienceTheir revenue predominantly originates from the sales of cloud products and services versus associated costs.
In light of stock trading anomalies, Qingyun acknowledged in their February 6 evening announcement that while DeepSeek is an open-source large language model available for users to train and customize freely, the contribution of this model to future company performance remains highly uncertain.
Additionally, on February 5 and 6, Youkede Cloud Computing confirmed their alignment with Biren Technology’s domestic chip to adapt DeepSeek's entire model series, while Youkede's private cloud platform UCloudStack has been equipped for local deployment of the DeepSeek series safely and compliantly.
Offering services via public, private, and hybrid clouds, Youkede has also developed three distinct AI-generated content sub-sector products: the intelligent computing platform "Kong Ming", model service platform UModelverse, and UPFS parallel file storage solution.
Similar to Qingyun, Youkede's announcements underline the absence of a deep-seated partnership with DeepSeek, stating that while they have adapted their model series, the actual effectiveness and potential future financial implications remain ambiguous.
Furthermore, Parallel Technology's announcement on February 5 declared they had yet to forge business relations with DeepSeek concerning computing services, although their intelligent computing platform has deployed mainstream models such as ZHIPU and DeepSeek-R1.
On February 6, Hang Steel Co. clarified that any deployments concerning DeepSeek mentioned in media reports involved merely installing DeepSeek software within partner systems to offer computing device leasing services
Advertisements
Advertisements
Advertisements