The heart of this AI implementation battle hinges on achieving new breakthroughs in product intelligence and user experience.

Produced by | Business Show

On the evening of February 15, the tech world was shaken by another significant development: WeChat is conducting a grayscale test to integrate the DeepSeek-R1 model. Many users who have secured testing privileges have shared their experiences.

On the 16th, Tencent confirmed to Business Show that a grayscale test is indeed underway, integrating the fully functional DeepSeek-R1 model. Currently, this feature is still in a limited grayscale test phase and will continue to be optimized based on user feedback and experience.

Just days prior, Tencent's AI assistant "Tencent Yuanbao" and AI code assistant also announced the integration of the fully functional DeepSeek-R1 model, supporting both the Hunyuan and DeepSeek models. According to Silicon Valley Beat, more Tencent products will integrate the DeepSeek model in the future.

This undoubtedly poses enormous pressure and challenges to other major companies. With Tencent rapidly integrating AI into its consumer products and ecosystem, will ByteDance, Alibaba, and Baidu remain unperturbed?

At the start of this year, as DeepSeek's popularity soared globally, overseas tech giants like Microsoft Azure and NVIDIA, along with domestic cloud providers such as Alibaba Cloud, Huawei Cloud, Tencent Cloud, Baidu Cloud, and 360 Digital Security, chose to launch the DeepSeek large model on their cloud service platforms.

This effectively makes the DeepSeek model a standard feature of mainstream AI cloud platform services. On the consumer product application front, no company has directly integrated the DeepSeek model into nationally renowned products like Tencent Yuanbao or even WeChat, as Tencent has done.

An industry insider told Business Show that WeChat, with over 1 billion users, exploring the integration of DeepSeek-R1 to enhance AI search directly challenges the traditional position of some major platforms in the search market. Amidst the profound changes triggered by DeepSeek, everyone is accelerating technology iteration, improving search experience, and thereby expanding new business growth avenues. In the backdrop of the AI era, the search market landscape will transform, and commercialization will evolve.

Interestingly, on February 13, Baidu announced that Wenxin Yiyan will be completely free from April 1, 00:00, with all PC and APP users able to experience its latest model, including features like ultra-long document processing, professional search enhancement, advanced AI painting, and multilingual dialogue.

On the same day, OpenAI also announced that the free version of ChatGPT will have unlimited access to GPT-5 for dialogue under standard intelligent settings. Additionally, Google's latest AI model suite was recently announced to be officially open to all users.

According to CCTV News, Gong Zheng, an engineer at the Institute of Technology and Standards of the China Academy of Information and Communications Technology, stated that free basic services will become traffic inlets, achieving a commercial closed loop through value-added services such as technology output and enterprise-level solutions.

So, how will Tencent WeChat's integration of DeepSeek-R1 stir up the Internet business world?

Over the past two years, major companies have actively deployed in the AI field, and the combination of Tencent WeChat and DeepSeek-R1 has compelled tech giants to reassess their AI strategies.

Even ByteDance's Doubao large model, with an average daily call volume exceeding 4 trillion times, faces pressure amidst the rise of DeepSeek-R1 and Tencent's integration.

At a recent internal meeting, Shen Dou, President of Baidu Intelligent Cloud Business Group, commented that the first AI product to be impacted by DeepSeek's aggressive arrival is ByteDance's Doubao, noting that the latter's training and traffic costs are both high.

Although Tan Dai, President of Volcano Engine, responded that the pre-training and inference costs of Doubao 1.5 Pro are both lower than those of DeepSeek V3, with considerable gross profit, it is undeniable that the market popularity of DeepSeek-R1 has impacted Doubao to a certain extent.

Not only does ByteDance need to weigh between independent research and development and cooperative integration, accelerating technological innovation and product optimization to cope with the competitive pressure brought by Tencent and DeepSeek-R1, but Baidu, Alibaba, and other major companies must also consider whether to simultaneously integrate their products with DeepSeek-R1 while developing their own large models.

More importantly, they must strategize on how to enhance product competitiveness and avoid falling behind in market competition. It is evident that the crux of this AI implementation battle ultimately lies in achieving new breakthroughs in product intelligence and user experience.

-Business Show-01 WeChat + DeepSeek: Aiming for AI Search Upgrade?

Business Show attempted to access the search bar at the top of the WeChat chat interface and found an "AI Search" entry. Upon clicking, an input box appears with two options: "Quick Q&A" and "Deep Thinking".

The "Quick Q&A" mode is swift and efficient, catering to immediate needs; "Deep Thinking" involves the DeepSeek-R1 model engaging in prolonged contemplation to provide more comprehensive answers, accompanied by a display of the thinking and reasoning process.

From a content ecology perspective, the WeChat ecosystem has amassed vast content, including public account articles and video content. However, the issue of content "sinking" has been severe, making it challenging for many high-quality contents to be discovered by users.

The advent of AI search rediscovers the value of these long-tail contents through intelligent recommendations, incentivizing creators to produce more in-depth and valuable content.

In terms of user experience, previous WeChat searches relied on keyword matching, leading to issues like inaccurate results and scattered information. After integrating DeepSeek-R1, leveraging natural language processing and deep thinking capabilities, search results are more precise and efficient, enabling users to quickly obtain the required information.

On the commercialization front, the scattered content information within the WeChat ecosystem will be aggregated across scenarios through "conversational" intelligent interaction. When users search for relevant keywords, they can not only obtain information but also directly navigate to related mini-programs, forming a closed loop of "search-service," opening up new commercialization possibilities on WeChat.

In 2025, the explosion of consumer AI applications has become an industry consensus. Andrew Ng, co-founder of Google Brain, Wang Xiaochuan, founder of Baichuan Intelligence, Li Kaifu, founder of 01 AI, and others have predicted that AI applications will experience a boom this year. Amidst this trend, all platforms need to swiftly enhance their products' AI capabilities to attract and retain users.

By deeply embedding AI technology into the WeChat ecosystem, Tencent can expand its business horizons and explore new commercial growth points. For instance, precise AI search can provide more accurate user personas for advertising, thereby boosting advertising effectiveness and revenue; in the e-commerce realm, AI-assisted product recommendations are more precise, potentially propelling the development of mini-program e-commerce and expanding the e-commerce business territory.

Furthermore, this initiative aids Tencent in consolidating its position in the social ecosystem. With the progression of the mobile Internet, users' needs are increasingly diversified, and their expectations for social platform functions and experiences are also rising. The introduction of AI search enables WeChat to offer users more comprehensive and intelligent services, meeting their needs for information acquisition, knowledge Q&A, etc., further enhancing user stickiness and activity, and reinforcing WeChat's leadership in the social domain.

For Tencent Yuanbao, supporting dual-model switching provides users with more options, catering to the needs of different users for AI assistants, which is expected to elevate the user activity and market share of Tencent Yuanbao. After Tencent Cloud's AI code assistant integrates DeepSeek-R1, it can attract more developers, enriching Tencent Cloud's developer ecosystem and bolstering its competitiveness in the cloud computing market.

Moreover, considering costs and efficiency, independently developing large models requires significant investment in funds, manpower, and time, and entails certain risks due to rapid technological iteration. Integrating the mature open-source model DeepSeek-R1 can effectively reduce R&D costs, shorten the product launch cycle, and allocate more resources to product optimization and business expansion.

-Business Show-02 Large Model Battle: From Price Wars to Free Trends

Over the past year, major companies have fiercely competed in the race to develop their own large models.

Baidu began laying out the Wenxin large model as early as 2019. After years of iteration, Wenxin Yiyan boasts mature technology in natural language processing, knowledge graph construction, etc., and is widely utilized in intelligent search, intelligent writing, intelligent customer service, and other fields.

In September 2024, Tencent officially launched the Hunyuan large model and subsequently continuously expanded its applications in gaming, social networking, office, and other domains, providing users with intelligent writing, intelligent customer service, and other services.

Alibaba's Tongyi Qianwen encompasses numerous functions such as generating text, images, videos, etc., playing a pivotal role in e-commerce, logistics, and other business scenarios.

ByteDance's Doubao large model, launched in August 2023, became the first domestic AI large model with over 100 million downloads in September 2024, spanning multiple scenarios like office, learning, and entertainment.

In 2024, large models were still a "hot spot," and no company with strength was willing to miss out. Major companies such as Baidu, ByteDance, Alibaba, and Tencent have released basic large models and also deployed vertical industry large models, which are generally tightly integrated with their own businesses. As large model technology gradually matures and market competition intensifies, price wars have become a critical means for major companies to compete for market share.

The fuse for this price war was lit when DeepSeek took the lead in reducing the price of large models to 1% of GPT-4. This aggressive pricing strategy instantly disrupted the original price equilibrium in the market, triggering a chain reaction in the industry.

Companies like Alibaba, Tencent, and ByteDance swiftly followed suit, lowering the prices of their large models. Alibaba Cloud has repeatedly reduced the prices of its large model products, with the API input price of Qwen-Long, the GPT-4-level main model of Tongyi Qianwen, dropping by 97%; Tencent Cloud has also comprehensively cut the prices of large models, with some functions of the Hunyuan-lite model even being completely free, and the prices of models like Hunyuan-standard have also plummeted significantly.

The escalating price war has intensified competition in the large model market. Against this backdrop, Baidu announced that Wenxin Yiyan will be completely free from April 1 this year, with all PC and APP users able to experience the latest Wenxin series models.

This move is undoubtedly a bombshell, propelling competition in the large model market to a new pinnacle. Behind the free strategy lies Baidu's attempt to attract more users and developers by lowering the threshold for user adoption, swiftly accumulating user scale, and seizing market share.

As leading companies leverage their capital, technology, and resource advantages to take the initiative in price wars and free strategies, further consolidating their market positions, the industry reshuffle is also accelerated, and some weaker small and medium-sized model enterprises are facing immense survival pressure.

It is undeniable that although domestic large models continue to innovate and make small achievements, compared with the global average development speed of large models, it is still premature to halt or reduce training and research. Behind the shift from price wars to free trends lies the persistently high operating costs and subsequent losses of large models.

Take OpenAI as an example; its operating costs exceeded $8.5 billion in 2024, with an estimated loss of approximately $5 billion. It is predicted that the total loss from 2023 to 2028 will reach $44 billion. As for model training costs, OpenAI anticipates they will reach $9.5 billion by 2026.

The aforementioned industry insider told Business Show that entering 2025, after DeepSeek took the lead in reducing the price of large models to 1% of GPT-4, the domestic large model track has entered a brutal elimination phase.

-Business Show-03 Catching the DeepSeek Express on the Eve of AI Business Explosion

The emergence of DeepSeek is undoubtedly a highlight in the AI field this year. Within just 20 days of its launch, its daily active users surpassed 20 million, swiftly topping the US APP Store charts and ranking first in app stores across 140 global markets.

Faced with the robust rise of DeepSeek, the attitude of major domestic and international tech companies is to hop on this "express" and actively integrate DeepSeek.

Even NVIDIA, a company rarely seen making such bold statements, hailed the R1 model as "an exceptional leap forward in AI technology." Intel went further by announcing a deep optimization of the Janus Pro model, while AMD integrated the new DeepSeek-V3 model into the Instinct MI300X GPU. Amazon launched the DeepSeek-R1 model on Amazon Bedrock and SageMaker AI, and Microsoft swiftly deployed DeepSeek-R1 on its Azure cloud service.

In China, the integration of DeepSeek has gained momentum across various industries, from cloud computing giants like Alibaba Cloud, Tencent Cloud, and Baidu Intelligent Cloud to mobile phone manufacturers such as Huawei, OPPO, Honor, Xingji Meizu, and vivo, as well as automakers and internet platforms.

Despite major companies hopping on the DeepSeek bandwagon, they still grapple with numerous bottlenecks in model optimization and computational efficiency amidst the impending AI business boom.

Take GPT-4 as an example; its training process demands vast amounts of data and formidable computational power, resulting in astronomical costs. Despite significant efforts in model architecture design and algorithm optimization, enhancing model training and inference efficiency while maintaining performance remains a colossal challenge.

Regarding computational efficiency, AI computing currently relies heavily on specialized hardware like GPUs, but the utilization rate of GPU power is suboptimal, leading to substantial resource wastage. According to data, during GPT-4's training, OpenAI's computational utilization rate hovered between 32%-36%, with large model training often dipping below 50% effective utilization. This not only escalates AI R&D costs but also stymies the large-scale application of AI technology.

DeepSeek's technical approach could offer fresh perspectives and inspiration to major companies. For instance, DeepSeek's technology empowers models to utilize computational resources more efficiently during inference, minimizing unnecessary computational overhead and thereby boosting computational efficiency.

With the swift advancement of AI technology, domestic large models like Baidu's ERNIE Bot, Tencent's Kunlun, and Alibaba's Tongyi Qianwen are vying for market share. DeepSeek, with its open-source strategy and stellar model performance, swiftly gained traction among developers and users, solidifying its market presence.

For major companies, enhancing product competitiveness is paramount, with user acquisition and retention posing even greater challenges. In the AI market, users enjoy abundant choices and exhibit low loyalty, making it crucial for companies to excel in areas like speech recognition accuracy, response speed, and interaction experience in intelligent voice assistants.

However, the most formidable challenge lies in the unclear profit model for AI products, which hinders commercialization. Despite AI technology's increasing application in various fields, converting these applications into tangible commercial value remains an ongoing exploration.

In certain fields like intelligent security and financial risk management, AI products have commenced commercial applications. Yet, in others like intelligent education and smart homes, AI product commercialization faces numerous hurdles. For instance, in intelligent education, while AI technology can offer personalized learning plans, converting these services into tangible revenue remains challenging.

Currently, some major companies achieve commercialization by providing AI technology solutions tailored to enterprise customers. For example, cloud service providers like Alibaba Cloud and Tencent Cloud offer one-stop services encompassing AI computing power, model training, and application development, aiding enterprises in swiftly implementing AI applications and thereby reaping commercial benefits.

Others have commercialized through AI hardware products, such as smart speakers and watches, merging AI technology with hardware. Baidu's DuerOS smart speaker, for instance, offers voice interaction, music playback, and smart home control through an embedded AI voice assistant.

Additionally, some companies monetize AI products through advertising, paid subscriptions, and other methods. OpenAI's ChatGPT, for example, generates revenue through a paid subscription service offering enhanced features and services.

Over the past few years, technology giants have been actively exploring these commercial avenues, though their effectiveness remains to be seen. The commercialization of AI products is a long and arduous journey with significant responsibilities. "End"