GuShiio.com Wen| Business style
Since the beginning of 2025, enterprises are competing to access and users are madly experiencing DeepSeek. It has become a universal and group-based social phenomenon to be precise.
Statistics released by DeepSeek show that as of February 28, 2025, more than 160 companies and institutions around the world have announced access to its model services, covering AI chips, cloud computing, finance, telecommunications, consumer electronics, content creation, and automobiles. Many industries such as smart driving and intelligent manufacturing have achieved widespread penetration across industries and fields.
Among large Internet companies, Tencent’s products have the widest access range. In addition to Yuanbao APP, which has been ranked on the APP Store through DeepSeek, more than 10 applications such as National Software WeChat, QQ Music, QQ Browser, and Tencent Documents are all connected.
Therefore, users who simply use it as a search tool seem extremely routine. Human imagination collides with AI’s logical thinking, creating one funny and amazing magic way to open.
But in addition to its ease of use, more and more users who have little understanding of technology but have been swept by the wave are asking questions: Does DeepSeek really have that kind of question-and-answer way of using it? Why does it keep displaying “Server busy, please try again later” after I ask a question? Wouldn’t so many enterprises connect to make DeepSeek servers, which are not wealthy, even worse?
The miracle of open source
The reason why DeepSeek is popular among many users is not only based on a more accurate understanding of user questions and potential meanings, but the thinking process it displays is an important factor that makes users smile.
(Source: Internet)
This kind of answer that finds a delicate balance between professionalism and interest has always been pursued by all questions and answers. This led to DeepSeek’s user growth exceeding 100 million in just 7 days after releasing the R1 model on January 20, making it the fastest application in history to exceed 100 million users.
According to DeepSeek statistics, as of February 28, 2025, the total number of daily active users of DeepSeek worldwide reached 119 million, of which 34.94 million domestic APP-side DAUs, 36.85 million overseas APP-side DAUs, and 48 million Web-side DAUs, with overseas users contributing nearly half of the traffic.
The surge in user numbers has naturally attracted the attention of other companies. By connecting their products to DeepSeek, they can attract users for their own businesses through DeepSeek’s traffic, so the wave of DeepSeek’s access quickly surged.
Since not all enterprises will announce the specific details and resource configuration of access after connecting to DeepSeek, and DeepSeek absorbs users too fast, judging from the frequency of DeepSeek outages, some enterprises directly call the cloud services of DeepSeek’s main station.
Therefore, DeepSeek quickly cooperated with domestic cloud vendors to deploy the model to third-party cloud platforms. Users can call DeepSeek’s APIs through the servers of these cloud service providers. Users ‘access consumes cloud vendors’ computing power resources, alleviating the pressure of calling DeepSeek servers.
DeepSeek is highly open source, which means that everyone can download, deploy, and use it. After a brief wave of direct calls, some customized deployments are also coming out. In addition to a number of software companies such as Dingbao, Feishu, and Kingdee, enterprises with higher requirements for data privacy and stability such as government affairs, finance, and medical care have chosen to deploy locally and completely separate themselves from the DeepSeek main station server, reducing their reliance on the DeepSeek main station server.
In other words, if the user is using the customized deployment DeepSeek, whether the server is busy, please try again later depends on the customized deployment enterprise server.
So what should we do for companies that have difficulty deploying locally or are not sufficiently compatible? DeepSeek then provided a customized tool chain and technical support to lower local deployment barriers. For example, DeepSeek has reduced local deployment costs by more than 30% by optimizing the compatibility of models and chips for Huawei’s Cent, Muxi, Cambrian and other chips.
In addition to lowering barriers and customized tools, DeepSeek’s open source has almost no restrictions on enterprises. DeepSeek uses the MIT Open Source Protocol (MIT), which has the feature of allowing anyone to use, copy, modify and distribute software for free, whether for personal or commercial use.
Liang Wenfeng, founder of DeepSeek, also mentioned in public: Our goal is not to make rapid profits, but to promote the development of technological frontiers and fundamentally promote the growth of the entire ecosystem.
External advertising appears?
DeepSeek can quickly top the list of free apps in the Apple App Store and Google Play Store in more than 140 countries and regions. In addition to thinking about performance, it does not have any major company background and has not opened commercial attempts for users.
In other words, DeepSeek, which relies on a free strategy, still needs to balance the surge in user numbers, server expansion and profit model. However, outside DeepSeek, there are already many people or companies that have taken advantage of this trend that are ahead of DeepSeek in terms of profitability.
According to reports, in the early days of DeepSeek’s popularity, it included but was not limited to selling local deployment and account registration services, selling courses on how to use DeepSeek, and even selling courses on how to make money through DeepSeek. Profit-making methods such as induced recharge by copycat domain names have become common. Course prices range from 1 cent to thousands of yuan, and there are many courses with sales exceeding 1,000.
When the use of an application becomes a popular phenomenon in the whole society, such leveraging marketing emerges endlessly. Previously, when ChatGPT swept the world in 2023, many apps, public accounts, and Mini programs used the banner of ChatGPT, smart AI, etc. to induce users to download or register. The usage process was to provide a certain number of free times first. When the number of free conversations was consumed, enter the payment process and charge according to the number of uses or time.
On February 17 this year, the National Computer Virus Emergency Response Center (CVERC) and other agencies issued an announcement announcing the capture of a number of counterfeit DeepSeek official Android apps specifically targeted at domestic users in my country. They contain Android platform mobile phone Trojan virus. After analysis, these apps may be a new variant of the financial theft mobile phone Trojan virus.
From this point of view, preventing speculation or criminals from fishing in troubled waters has become a must-master skill for every DeepSeek user.
At the same time, companies that connect DeepSeek products are also questioned about the inclusion of advertising.
The laboratory of the three fathers of the blogger said that in DeepSeek connected by Tencent Yuanbao, they found that the generated answers contained ad-oriented content such as suggestions to compare prices through regular platforms such as 58 Home, suggestions to make inquiries through fast recycling platforms or local regular recyclers.
Zhang Jun, Tencent’s public relations director, quickly responded via Weibo, saying that theoretically it should not (appear ads) and had already asked technology to check.“” Zhang Jun speculated that the abnormal answer may have originated from the fact that the system quoted the content of the pitch-avoidance guide published by the 58 Home Public Account when it started the online search.“”
(Source: Weibo)
In addition,”Science and Technology Innovation Board Daily” reported that some users reported that after Internet corporate apps connected to DeepSeek, their answers included a lot of content related to the platform. I used Tencent to access R1 ‘s Yuanbao to search for related content on the Douyin platform, but it recommended the content of video numbers.&” rdquo; The answers you get on the DeepSeek App will be different from the ones you get on the DeepSeek full-blood version accessed by Baidu’s search box. Especially for questions involving Baidu itself, the answers will be quite different.
DeepSeek’s open source features and fine-tuning capabilities provide space for content control driven by commercial interests. Other platforms have access to DeepSeek for commercial considerations. When it is unlikely to recommend competitors ‘businesses, it seems reasonable to recommend their own related businesses. It brings traffic gain to your own business, but the drawback is that it overdraws users’ trust in AI tools to a certain extent.
This kind of algorithmic bias may intensify the information cocoon room and violate the original intention of technology for universal benefit, but it is also an inevitable phenomenon in the process of technology commercialization. In the long run, the game between enterprises, users and regulators will continue to exist.
For users, there are currently no auxiliary tools to help users distinguish whether AI-generated content includes marketing content. They can only manually verify or ask several more questions about the same question to cross-verify and get answers that meet their own needs.
This is also in line with the current overall trend in using AI content. Even DeepSeek added at the end of the answer that this answer was generated by AI for reference only and does not constitute a disclaimer for professional advice. All ordinary users can do is to sharpen their eyes and avoid major decisions led by AI.
What is the solution?
Faced with a series of problems such as computing pressure, data security concerns and content manipulation caused by the surge in DeepSeek’s global user base, deploying a private domain model has become a practical solution. Through private domain deployment, enterprises can run the DeepSeek model on their own servers or private cloud environments, thereby reducing dependence on public servers and improving service stability and data security.
Taking the financial industry as an example, a bank wanted to use DeepSeek to process customers ‘loan applications. Then, through private domain deployment, banks can run DeepSeek on their own servers to analyze sensitive data such as customers ‘credit history and income status, thereby improving the accuracy and efficiency of loan approval. At the same time, private domain deployment also ensures data security and avoids risks caused by data leaks.
Judging from the current client needs, the factors considered for the deployment of the private domain large model include data security and privacy protection, customization requirements, high performance and low latency, compliance, cost control, and model stability and controllability.
Therefore, in order to meet all the above needs, the AI large model all-in-one machine has become one of the best ways to deploy large models in the private domain. AI large model all-in-one machine refers to a localized deployment solution that integrates pre-training large models, computing power infrastructure, security modules, industry knowledge bases and application development tools to achieve a full closed-loop data link. It focuses on out-of-the-box use, software and hardware collaboration, taking into account security compliance (national secret algorithm/sensitive word filtering) and efficient reasoning (low latency + high concurrency), reducing the full-cycle cost of enterprises from computing power construction to model tuning.
Take Supercloud’s new AI large model all-in-one series as an example. It covers three products: SuperCube3000, SuperCube5000 and SuperCube7000, and is divided into Xinchuang version and international version. It can provide enterprises with one-stop private domain large model deployment services.
During the private domain deployment process, enterprises usually face problems such as difficulty in selecting computing power, optimizing models, importing data, resource management and control, deploying operation and maintenance, and applying tools. The super-cloud AI large model all-in-one machine can provide countermeasures in terms of computing power, software, ecology and services.
In terms of computing power, the SuperCloud SuperCube series is compatible with mainstream CPUs such as Haiguang, Feiteng, Intel, and AMD. Especially through collaborative optimization of software and hardware of domestic A1 chips, the reasoning performance reaches more than 90% of imported solutions. In terms of software, Super Cloud can realize pre-installation and unified scheduling of computing power for mainstream open source models such as DeepSeek. For example, the DeepSeek-671B full-blood version model, which is currently popular in the market, has been connected to the Super Cloud SuperCube7000 product.
At the same time, Chaoyun also carries out industry-wide ecological cooperation, deeply customizes with industry scenarios, pre-sets industry knowledge bases and fine-tuning tool chains, so that enterprises can train exclusive models at low cost, which reduces the cost of customization compared with closed-source APIs, and solves the last problem of AI applications. One kilometer problem.
In addition, based on the super-cloud AI large model all-in-one solution, the implementation time for integrated delivery has been shortened by 60%, security has been enhanced by 50%, full life cycle operation and maintenance costs have been reduced by 30%, and comprehensive cost performance has been improved by 30%, promoting AI has shifted from universal in the cloud to end-to-side exclusive, accelerating the implementation of the AI large model industry.
Overall, the super-cloud AI large model all-in-one machine has the advantages of controllable data, localized support, out-of-box use, industry customization, cost compression, and high-quality service.
This also means that Supercloud’s AI large model all-in-one machine is not a simple hardware combination, but an overall solution that integrates intensive high-performance computing power clusters, intelligent computing power scheduling engines, and one-stop tracking services, from end-to-end., software and hardware collaboration, computing power resources and other aspects, providing enterprises with a full range of private domain large model deployment services.
The “White Paper on the Deployment of Private Domain Large Models” released by Chaoyun predicts that the future development of the private domain large model will be characterized by a high degree of synergy between supply and demand. Technological breakthroughs on the supply side and deepening of scenarios on the demand side will draw on each other to achieve technological inclusiveness. The two-way rush of technology and the value of scenarios. Taking the AI large-scale model all-in-one machine as the fulcrum, we will build a data-model-business closed-loop, promote the market from technological innovation-driven to scenario value-driven, and reconstruct the underlying logic of enterprise intelligent transformation. Under the resonance of supply and demand, the deployment of large private domain models will move from luxury goods to necessities.
The core of technology inclusiveness lies in reducing the cost curve of technology application. Deploying the private domain model provides a practical path to solve DeepSeek’s current application dilemma, and may also become a key step in promoting the inclusiveness of AI technology and industry intelligence.
When more companies are able to obtain AI capabilities comparable to international top-level levels at controllable costs, this global AI boom triggered by DeepSeek may truly begin to show its power to reshape the industrial landscape.
