In-Depth: DeepSeek Reignites the Internet Gateway Battle (AI Translation)
Listen to the full version
文｜财新周刊 刘沛林 屈运栩 关聪
By Caixin Weekly Liu Peilin, Qu Yunxu, Guan Cong
中国产业界对DeepSeek的响应从蛇年春节延续至今，引发新一轮人工智能（AI）算力、应用、大模型和云服务的竞争。
The response from China's industry to DeepSeek has continued since the Year of the Snake Spring Festival, sparking a new round of competition in artificial intelligence (AI) computing power, applications, large models, and cloud services.
DeepSeek的2C端（面向个人）应用因为需求过大，自春节开始断续出现“服务器繁忙”问题。互联网公司很快意识到，接入DeepSeek的R1推理模型，为用户提供可支撑应用的算力，才能接住这一轮“泼天的流量”。新一轮AI应用入口之战就此拉开。
The 2C (consumer-facing) application of DeepSeek has intermittently experienced "server busy" issues due to overwhelming demand since the Lunar New Year. Internet companies swiftly realized that connecting to DeepSeek's R1 inference model, which provides the computational power needed to support applications, was essential to handle this massive influx of traffic. Thus, a new round in the battle for AI application entry points has commenced.
短短两周，底层算力服务商密集推出支持DeepSeek模型的产品；中间层模型厂商则跟进开源策略；上层应用更是争先恐后地宣布接入DeepSeek模型，争夺用户。而中美大模型厂商也纷纷推出新模型对标DeepSeek-R1。
In just two weeks, foundational computing power service providers have intensively launched products supporting the DeepSeek model. In the intermediate layer, model manufacturers are adopting open-source strategies. On the upper application layer, companies are racing to announce their integration with the DeepSeek model to capture user interest. Meanwhile, major model manufacturers from both China and the United States have successively introduced new models to compete with DeepSeek-R1.
- DIGEST HUB
Digest Hub
Back
- DeepSeek's significant impact on China's AI industry has prompted intense competition in AI models, computing power, and applications. Major companies like Tencent and Baidu are integrating DeepSeek models to remain competitive.
- The open-source nature of DeepSeek has changed market valuations and led to a surge in interest for AI infrastructure investments, drawing comparisons to a "Sputnik Moment" for China's tech sector.
- DeepSeek's influence is also evident in local governments integrating AI for various applications, highlighting increased demand for computational power and accelerating technological development across industries.
DeepSeek has sparked a fierce competition in China's AI sector, particularly in computing power, applications, large models, and cloud services since the Year of the Snake Spring Festival. Driven by overwhelming demand, DeepSeek's R1 inference model has become crucial for internet companies to manage the surge in AI applications. The model's open-source nature has accelerated its adoption, with major Chinese and US tech companies racing to integrate or develop competing technologies. [para. 1][para. 3]
On February 16, WeChat integrated DeepSeek-R1 into its AI search feature, attracting significant market attention despite initial traffic overloads. WeChat's massive user base exacerbated server demands, prompting Tencent to strategically redirect search traffic to its chatbot Yuanbao. Meanwhile, Tencent continued developing its Hunyuan model to ensure upgrades to its internal systems, underscoring the importance of DeepSeek's technological capabilities in transforming AI applications. [para. 4][para. 6][para. 7]
Globally, DeepSeek's adoption has challenged existing commercial models, pushing companies like Baidu and Alibaba to rethink strategies. Baidu launched "Deep Search" with DeepSeek-R1 support, while Alibaba's DingTalk and Lark office software integrated the model. Alibaba's e-commerce applications, however, stayed aloof from this integration, as they prepare a new inference model rollout in March. [para. 9][para. 10][para. 13]
The emergence of DeepSeek has prompted changes in investor perception towards Chinese tech companies, boosting the Hang Seng Tech Index significantly. It represents a pivotal moment, recognized globally as a significant advancement in China's AI capabilities. The open-source approach has not only democratized access to AI models but has also rekindled competitive pressures among major internet companies and startups. [para. 14][para. 15][para. 19]
Internally, major Chinese tech firms are accelerating their efforts to integrate or compete with DeepSeek. Tencent's swift move to onboard DeepSeek-R1 across various consumer applications highlights the emphasis on real application-level capabilities over rushed or incomplete products. By restructuring its teams, Tencent aims to streamline the deployment of AI across its ecosystem, signaling a strategic shift towards deeper AI integration. [para. 22][para. 23][para. 28]
DeepSeek’s open-source influence has also caused significant recalibration among AI companies globally. The model's capacity to operate with reduced computational resources inspires companies to reevaluate their strategies on computational power investments and open-source dependencies. This has resulted in increased investments by major firms like Alibaba, which plans to invest over 380 billion yuan in AI infrastructure, in response to DeepSeek's success. [para. 30][para. 31][para. 32]
The rapid development and adoption of DeepSeek models have not only challenged tech companies to innovate but also spurred investment interest in infrastructure required for AI development. Chinese firms are now actively pursuing opportunities to enhance capabilities, especially as global models like OpenAI's upcoming GPT-5 promise to push AI limits even further. The dramatic iterations and open-source projects by DeepSeek have reshaped competitive dynamics, positioning Chinese technology closer to par with the US. [para. 39][para. 40][para. 41]
In parallel with model competition, there's a burgeoning demand for AI computing power, driven by applications across industries, from government services to corporate operations. The construction of new data centers and an increase in cloud service offerings are underway to meet this demand. DeepSeek's low-computation needs facilitate its broader adoption in industries and government, suggesting a potential transformation in Chinese industry standards. [para. 42][para. 46][para. 58]
Overall, DeepSeek has catalyzed a narrative change in China's tech sector and capital markets, driving up valuations and invigorating interest in AI-backed technologies. The company's emergence has also impelled firms to explore new business models, highlight the dynamic interplay between foundational model development and application-facing AI advancements. [para. 74][para. 76]
- DeepSeek
DeepSeek - DeepSeek has sparked significant competition in AI infrastructure, applications, and model development in China. Its inference model, DeepSeek-R1, is noted for its advanced reasoning capabilities and has been widely adopted due to its open-source nature under the MIT license. This has led to increased demand for AI-related compute power and shifted focus for many companies toward technology upgrades, impacting China's tech market and drawing international investor interest.
- Tencent
腾讯 - Tencent quickly embraced the DeepSeek model, integrating it into various applications such as WeChat and Yuanbao, while facing challenges with server capacity due to high demand. They also accelerated their AI strategy by announcing the integration of their self-developed Hunyuan T1 model into Yuanbao. In response to increased AI demand, Tencent adjusted its organizational structure to better support AI development and integration across its services.
- Alibaba
阿里巴巴 - Alibaba participated in the AI field by integrating the DeepSeek-R1 model into three B2B applications, including DingTalk and its AI code editing tool Tongyi Lingma. Alibaba has announced plans to invest over 380 billion yuan over the next three years in cloud and AI infrastructure. However, in the AI 2C application domain, Alibaba has been less proactive, missing the recent entry point battle and focusing on using its own models for AI integration.
- ByteDance
字节跳动 - ByteDance is under pressure from the rise of DeepSeek. CEO Liang Rubo emphasized the importance of exploring AI intelligence limits and improving UI interactivity, alongside seeking scale economies. Despite not integrating DeepSeek into its AI assistant "Doubao," ByteDance is testing various deep thinking models. Doubao remains one of China's fastest-growing AI applications, benefiting from ByteDance's in-house promotional capabilities on platforms like TikTok.
- Baidu
百度 - Baidu faced challenges in the AI space, lagging behind the surge driven by DeepSeek. It aims to release the advanced Wenxin Model 4.5 in March 2025 and has embraced open-source strategies to increase market competitiveness. Despite internal restructuring and performance declines, Baidu is increasing AI investments, responding to competition from Tencent and other major players, as highlighted by its plans for cloud and AI infrastructure enhancement.
- Baichuan Intelligence
百川智能 - Baichuan Intelligence released deep reasoning model Baichuan-M1-preview and open-source Baichuan-M1-14B model in the medical field, followed by the Baichuan-Omni-1.5 open-source multimodal model, directly competing with GPT-4o mini. Despite being recognized among leading Chinese AI startups, the presence of DeepSeek is challenging its valuation and positioning in the AI model landscape.
- Moonshot AI
月之暗面 - Moonshot AI, referred to in the article as 月之暗面, is one of the leading AI companies in China, known for its Kimi Intelligent Assistant. The company has been impacted by the rise of DeepSeek, which has changed the competitive landscape. It has shifted focus to technical upgrades and is actively involved in exploring model innovations, such as open-sourcing specific models in collaboration with automotive companies to enhance application capabilities.
- Zhipu AI
智谱AI - Zhipu AI, mentioned in the article, is one of China's leading AI model companies, part of the "six tigers" in the industry, with a valuation exceeding 200 billion yuan. It focuses on building advanced models, and its value proposition is challenged by DeepSeek's success in the AI space, causing a shift in industry dynamics and valuation assessments.
- Step AI
阶跃星辰 - Step AI, under JieYue XingChen, announced the open-sourcing of its Step-Audio and Step-Video-T2V models in collaboration with Geely Auto Group. These models will be available on JieYue XingChen's "YueWen" app. Additionally, Step AI quietly launched an entry point for the DeepSeek-R1 model. This indicates Step AI's ongoing efforts to enhance and diversify its offerings in audio and video AI models.
- MiniMax
MiniMax - MiniMax has incorporated the DeepSeek-R1 model into its "Heluo AI" overseas version. The company is among the major players in the AI generative model field in China, with its app listed among popular AI applications. In response to the advancements catalyzed by DeepSeek, MiniMax and other companies are accelerating their model updates and exploring new applications to maintain competitiveness in the rapidly evolving landscape.
- SiliconFlow
硅基流动 - SiliconFlow is one of the earliest companies in China to offer inference services for the "full-fledged" DeepSeek-R1 model, utilizing Huawei Ascend chips. After the launch, it experienced user growth exceeding its annual target within just four days. On February 14, it partnered with Huawei to provide private deployment solutions and later began offering DeepSeek model inference services through Amazon Web Services' marketplace, following a rental agreement with its domestic data center partner.
- WuWenXinQiong
无问芯穹 - WuWenXinQiong is a service provider that offered model training and inference services on a platform allowing an integrated approach across various domestic chips. Before the Lunar New Year, they had already launched some DeepSeek model services. On February 11, they announced the completion of integrating DeepSeek-R1 on platforms using different domestic chips. This integration allows users to form a unified "compute pool" without compatibility concerns.
- Ice Technology
爱诗科技 - The article does not provide specific information about Ice Technology. It focuses on China's AI industry response to DeepSeek, discussing major companies like Tencent, Alibaba, ByteDance, and Baidu, as well as their strategies and the market impact of DeepSeek's open-source AI model.
- Geely Auto
吉利汽车 - The article mentions that Step-Audio and Step-Video-T2V models, co-developed by Geely Auto Group and Stepwise Star (阶跃星辰), were announced for open source. These models focus on voice interaction and video generation, and they are integrated into Stepwise Star's app "Yuewen," with an additional entrance for the DeepSeek-R1 model.
- 2024-12-31：
- The monthly active user count for WeChat, combined with WeChat, was 1.382 billion
- 2025-01：
- Monthly active users data from Quest Mobile are released
- 2025-02-10：
- DeepSeek's 2C application started experiencing 'server busy' issues due to overwhelming demand
- 2025-02-16：
- WeChat launched an AI search feature integrated with the DeepSeek-R1 model in beta test
- 2025-02-17：
- Tencent announced the integration of its hybrid inference model, T1, into Yuanbao
- 2025-02-18：
- Baidu's 'Deep Search' feature became available on the PC search platform's homepage
- 2025-02-24：
- Alibaba announced an investment of over 380 billion yuan in cloud and AI hardware infrastructure over the next three years
- 2025-02-28：
- Baidu previewed Wenxin large model 4.5, set for release on March 16, 2025
- 2025-02-28：
- Multiple Tencent applications had integrated DeepSeek-R1
- PODCAST
- MOST POPULAR
Sign in with Google
Sign in with Facebook
Sign in with 财新