The shock wave caused by DeepSeek is still ongoing: not only have multiple platforms announced their integration with DeepSeek, but various industries have also boarded this "fast train" intensively. Although DeepSeek significantly reduces computing power demand through optimization algorithms, its high cost-effectiveness has driven the popularization of AI, resulting in exponential expansion of computing power demand. For the domestic computing industry, DeepSeek's innovation has brought an opportunity to be on par with Nvidia in terms of ecology, ushering in a new wave of development opportunities for domestic computing power. DeepSeek continues to expand its circle, causing a thousand waves with just one stone. In just one month, various industries from technology giants to government departments, from automobile companies to mobile phone manufacturers, are actively embracing DeepSeek. On February 16th, Baidu Search and Wenxin Intelligent Agent Platform announced that they will fully integrate with DeepSeek. Search users can use DeepSeek and Wenxin Large Model's deep search function for free, and developers of Wenxin Intelligent Agent Platform will also call DeepSeek model to create and optimize intelligent agents at any time. Baidu is not the first search platform to integrate with DeepSeek. Recently, some WeChat users have discovered that WeChat Search has launched the "AI Search" function and connected to the "Deep Thinking" service provided by DeepSeek-R1. Tencent has stated that while WeChat's search engine is using hybrid models to enrich AI search, it has also officially launched a gray scale test and integrated with DeepSeek. Users who have been grayed out can search for the entrance at the top of the dialog box and see the words "AI Search". After clicking on it, they can use the DeepSeek-R1 full blooded version model for free and get a more diversified search experience. On February 16th, Shenzhen announced that it will officially provide DeepSeek model application services to all districts and departments in the city based on the government cloud environment, achieving an integrated empowerment upgrade of artificial intelligence government applications based on DeepSeek. Previously, government departments in Guangzhou, Hohhot, Ganzhou, Wuxi and other places have also announced the deployment of the application DeepSeek to promote the intelligence of government services, covering areas such as document writing, public opinion processing, urban governance, etc., improving the quality and efficiency of government processing, and assisting in the construction of smart cities. The access of Internet head applications and government service platforms shows that DeepSeek has reached the widest user group. Data shows that with the rapid increase in usage, DeepSeek has become the fastest application to break through 30 million daily active users. Not only that, multiple cloud platforms including Huawei, Alibaba, Baidu, Tencent, JD.com, etc. have already announced the integration of the DeepSeek big model. About 20 car companies including Geely, Lantu, Zhiji, Great Wall, GAC, Changan, Chery, etc. have announced the launch of DeepSeek. DeepSeek applications have covered multiple fields such as telecommunications, cloud computing, finance, automotive, and mobile phones. Industry insiders believe that DeepSeek's rapid rise to fame lies in breaking the cost curse of "the stronger, the more expensive" in the field of big models, achieving low-cost industrialization of big models with high cost-effectiveness and efficiency, and while reducing costs through underlying architecture innovation, building an ecological moat with an open source strategy, providing a feasible path to break through computing power limitations. It also declares the end of the "violent pile of computing power" era in the global big model industry since ChatGPT sparked a new wave of artificial intelligence boom, promoting the transformation of the entire AI industry chain from basic research to commercial applications. The demand for computing power will continue to grow. DeepSeek has achieved world-class model performance with lower costs and computing power requirements, breaking the path dependence of the global artificial intelligence industry on "heap computing power". At the same time, it is also questioning the direction of industry development: is it still necessary to invest in computing power on a large scale in the future? As economist Jevons has pointed out, although technological progress has improved the efficiency of resource utilization, the increase in demand often leads to an increase in total consumption. Industry experts believe that the "Jevons paradox" proposed during the first industrial revolution will also be realized in the era of big models: Watt's improved steam engine made coal combustion more efficient, but the result was a surge in coal demand. Now, DeepSeek has reduced the inference cost of big models by 97%, while global computing power demand will exponentially expand due to its high cost-effectiveness. The Evaluation Report on the Development of China's Artificial Intelligence Computing Power in 2025 recently released by IDC United Wave Information, an international data company, proposed that DeepSeek, through its advanced algorithm optimization and efficient model performance, has stimulated the demand for new application scenarios and significantly boosted the application development of data center, end side and edge computing. According to a research report by Tianfeng Securities, DeepSeek has reduced hardware dependence, shortened training time, significantly reduced investment, development, and operational costs through technological innovation, providing more downstream application companies with opportunities to enter the big model business. At the same time, the open source concept also breaks down technical barriers, allowing developers to easily access resources and customize development for their own businesses, helping large models to be implemented in various industries. The fact shows that DeepSeek has entered the mobile end of ordinary consumers faster than any previous large model product, and has also landed on the developer or enterprise end faster than other large models. Industry insiders say that on the one hand, DeepSeek's high concentration of technological innovation fully utilizes the effective computing power of hardware, indicating that a stable and reliable computing power base is still the key to model innovation; On the other hand, many small and medium-sized enterprises can afford such large models. The decrease in training costs may temporarily reduce the demand for computing power, but the larger models becoming more economical will attract companies that originally did not plan to use them due to high model costs, thereby increasing the demand for computing power. The capital market, which has always had a keen sense of smell, responds to the wind. As early as the Spring Festival of the Year of the Snake, the upward trend triggered by DeepSeek began to ferment, and the entire computing power industry chain ushered in a wave of upward trend. Several securities research reports reminded investors to pay attention to the opportunities brought by the "computing power shortage" in the industry chain. Experts in the computing industry indicate that in the future, big models will show a parallel development trend of "technological advancement+engineering innovation", and the demand for computing power will continue to grow. Among them, "technological advancement" refers to the firm commitment of AI leading enterprises to innovate pre trained basic models, focusing on efficient, stable, and open computing power bases, ultimate performance, stable and reliable computing power clusters, as well as deeply open computing power platforms and ecosystems; Engineering innovation "refers to new paradigms such as DeepSeek that lower the threshold for post training and model distillation. Enterprises pay more attention to convenient, easy-to-use, and cost-effective computing platforms, distillation and fine-tuning solutions that balance cost and performance, as well as convenient deployment and agile business deployment. Facing the application demand brought by the DeepSeek craze, the domestic computing industry actively invests in the adaptation of DeepSeek series models, further strengthening the domestic software and hardware support capabilities for large models, and accelerating the promotion of software and hardware collaborative innovation. On February 4th, the DeepSeek series of large models was officially launched on the Ascend community. Developers can easily access the DeepSeek series models with just one click and support out of the box use on the Ascend hardware platform. In fact, as a full scenario AI infrastructure for "end, edge, and cloud", Ascend has been deeply involved since the release of the DeepSeek model. On December 26, 2024, DeepSeeker V3 was released to support Ascend, allowing users to achieve efficient inference on Ascend hardware and MindIE inference engine. Since the Spring Festival this year, several companies have successively announced the successful launch of DeepSeek related services based on Ascend. All three basic telecommunications companies have fully integrated into the DeepSeek open source model: Mobile Cloud, based on the Ascend ecosystem, supports the DeepSeek series of products in all versions, sizes, and functions, and has launched 13 China Mobile intelligent computing centers; Telecom Tianyi Cloud's "Xirang" intelligent computing platform supports flexible deployment of DeepSeeker R1 full health version to lightweight distillation models, from Ascend hardware, inference engine to model services; Based on joint innovation and self-developed acceleration engines, China Unicom Cloud has built the Ascend Intelligent Computing Base and deeply integrated it with the DeepSeeker R1 model. In addition, AI infrastructure company Luchen Technology has launched the DeepSeek R1 series inference API and cloud mirroring service based on Ascend computing power, and silicon-based mobile has launched DeepSeek V3 based on Huawei Cloud and Ascend Cloud services DeepSeek-R1； Computing system software service provider Qingcheng Jizhi has also completed the full blood version adaptation of Ascend DeepSeek R1... It is reported that more than 30 software and hardware enterprises have launched DeepSeek all-in-one solutions based on Ascend, and more than 80 enterprises have launched DeepSeek based on Ascend. Ascend has helped more than 20 provinces and cities AICC (Public Computing Service Platform) launch and deploy DeepSeek series models, and services are available in over 70% of regions across the country. It is reported that in terms of training, Ascend is currently the only AI training platform in the domestic industry that is fully adapted from pre training to fine-tuning and reinforcement learning capabilities, DeepSeek; MindSpore supports DeepSeeker V3 pre training, fine-tuning, and inference processes out of the box, achieving day level reproducibility. In terms of reasoning, the DeepSeek small expert innovative architecture brings challenges of communication time and uneven expert load. Ascending dual machine parallel reasoning+expert load balancing helps improve the inference throughput performance of the DeepSeek model by over 30%. In addition, Ascend provides full process AI development services to help enterprises deploy DeepSeek on a weekly basis and achieve rapid deployment. Industry insiders believe that there is now a huge opportunity for the domestic computing power industry to be on par with Nvidia in terms of ecology. Previously, the anchor point of the global computing ecosystem was NVIDIA's CUDA system. Now, DeepSeek has bypassed this anchor point, and all manufacturers are standing on the same starting line. The domestic computing industry has ushered in new development opportunities. Whoever can quickly build new support capabilities based on updated ecological anchors at this time, support good customers, support good partners, and support good developers, may usher in new or even faster development. (New Society)