DeepSeek’s Challenges: fierce competition and unclear monetization of AI models in China

On-the-ground research from Shanghai's Global Developer Conference

Investigative analyst Zhen Li recently attended Shanghai's second Global Developer Conference, an event organized by the city government to promote the advancement of open-source models and vertical industry applications. Despite rumors of DeepSeek's participation, the company was noticeably absent among a mosaic of vertical AI application and AI Agent developers, gadget manufacturers, infrastructure providers, and users of AI-related services.

Nevertheless, DeepSeek was a hot topic among attendees. There are high expectations for the widespread adoption of DeepSeek's model even though our on-the-ground research from the conference seems to indicate possible concerns about its viability and sustainability given the intense competition in China's AI sector and lack of clarity around monetization of AI models. DeepSeek’s model is significantly impacting the pace and approach that governments and enterprises are using to adopt AI. It is also impacting the AI strategies of Chinese big tech companies, such as Alibaba and Tencent.  

Expectations for wide adoption of DeepSeek’s model and vertical applications

DeepSeek has clearly emerged as the top foundational AI model in China, despite being relatively unknown before the end of 2024. With a small R&D team of less than 150 people, the company didn’t have much business and monetization capabilities; however, a developer community and a supply chain ecosystem are quickly growing around it now.

Chip makers, hardware manufacturers, and software developers are collaborating to launch servers or private cloud products deployed with the DeepSeek models and application tools, powered by Huawei Technologies and other Chinese chips. Semi-government industry associations are organizing their member enterprises to help DeepSeek develop vertical applications, using their industry know-how and data for post-training.

DeepSeek’s endorsement by the Chinese government is expected to accelerate the adoption of its model and related vertical applications in governments and State-Owned Enterprises (SOEs), in line with Beijing's directives. This endorsement and advising officials at the local level to use the model should lead to more AI spending and benefit those in the supply chain.

Implications on AI infrastructure investments and the chipmaking industry  

There is general agreement that the implementation of DeepSeek's model will drive growth in AI infrastructure. Alibaba has already committed to significant capital expenditure on AI in the coming years, and governments at various levels are investing billions in building AI data centers.

Out of concern for data security, we believe it is highly probable that Chinese government clients and large SOEs will integrate DeepSeek into their private clouds or those managed by telecommunications companies. Hence, unlike US hyperscalers or cloud service providers, who become clear beneficiaries of rising AI demand, major Chinese cloud providers, Alibaba or Tencent, may not easily harvest deals from those big spenders.

Also, because of DeepSeek’s much lower requirements for hardware and computing power for training and inference, Chinese AI chips, despite being less powerful, are expected to partially substitute Nvidia’s GPU chips, which are indispensable for AI development today. Chinese chipmakers, led by Huawei, said they had successfully adapted their products to support DeepSeek’s models, allowing lower costs for developing and running AI programs. They are promoting their chip products to cloud providers in China, as well as to government and corporate clients who are keen for private AI deployment.

Price wars, fierce competition, and lack of clarity around monetization of AI models in China

For private businesses and smaller companies, the use of API services provided by hyperscalers remains the preferred option to embrace AI. API prices set by China’s cloud providers have been low (many believe they are set below the cost for building the computing power and energy consumption) and the trend toward lower pricing is expected to continue, as cloud platforms aim to retain customers who may be considering shifting to private AI set-ups. Cheaper Chinese chips and lower training and inferencing costs should also contribute.

The battlefield of big AI models in China has been crowded. All big Chinese tech companies and cloud service providers, namely BBAT (ByteDance, Baidu, Alibaba and Tencent) have their own AI models in development, together with several dozens of AI start-ups backed by private and public fundings. Fighting for AI app developers and their potential clients who are gravitating quickly towards DeepSeek will be vital to their future in AI or even their survival.

Currently, all the major cloud service providers have made DeepSeek available on their platforms to capture the exploding demands from small and medium enterprises (SMEs) and individual developers which cannot be handled by DeepSeek due to its server resource constraints. These companies have seen their cloud computing usage soar. On the other hand, they are not giving up on their own models: Alibaba has committed investment in its TongyiQwen, which has been recognized as one of the best in China, rivalling, if not surpassing DeepSeek; Baidu has moved to open-source its model which was previously closed; and Tencent has recently released a new AI model to the public which was said to be used internally to enhance its own gaming and advertising business. These models are claiming their superior performances over rivals and are pricing competitively if not lower than DeepSeek for API uses.

Despite surging demand, monetization of DeepSeek or other AI models remains unclear in China as few customers are willing to pay for AI features, and most vertical AI applications are still in the proof-of-concept stage.

At the conference, vertical application developers expressed difficulties with product-market fit, due to the models’ own suboptimal performance, poor design of interface or display of AI apps, difficulty of use, and inadequate client training (constraints in labor and time to work with AI, difficulties overcoming the learning curve, etc.) A few enterprises said they would prefer the off-the-shelf products instead of making upfront payments for something still in trial, which doesn’t guarantee topline growth months later.

More optimistic attendees noted the accelerated adoption of DeepSeek models, thanks to the government footing the bill. This could help tackle some of the aforementioned issues, and it is expected that we will see some super AI Agent apps and certain viable paths for AI monetization in the second half of this year or early next year.

Implications for Chinese technology companies Alibaba and Tencent

While some believe that other large AI modes in China are losing out to DeepSeek because it enjoys the official endorsement, others have argued DeepSeek’s supremacy has not been secured and it may not be able to maintain its leadership position. In recent weeks, the pace of upgrades in large models in China has accelerated, with several claiming superior performance and efficiency advantages over DeepSeek. It also appears that app developers and their clients can switch between foundational models for their application development and adoption, without incurring significant costs or encountering difficulties.

There are also speculations about whether DeepSeek’s unique culture bolstered under the founder’s vision, commitment and temperament can be preserved, as it has been touted as the key to DeepSeek’s success. The team used to operate away from the limelight at their own discretion.

At the conference, several application developers mentioned using Alibaba’s cloud services and its open-source high-performing AI models for tool use, function calling and corpus access, among other things. Alibaba is said to have created one of the best ecosystems for app development (another being ByteDance). The company’s committed investment in AI infrastructure and model development helps to secure its leadership position, anticipating a potential decline in DeepSeek's popularity.

Tencent, which was scaling back on its cloud business and developing AI models mainly to cater to its core businesses internally, has been swiftly adjusting its AI strategies. The company has seen a surge in traffic to its cloud platform for AI services via its super app WeChat, following the incorporation of the DeepSeek model. Analysts have been excited about the monetization opportunities, as the usage of its cloud computing increases. It remains to be seen, however, whether Tencent will follow Alibaba’s lead and increase investment to meet the potentially exploding demands for cloud services and customers’ willingness to pay improves.

Humanoid robots: the future of AI

The conference also showcased the latest advancements in humanoid robots. Earlier this year, a robotics company Unitree exhibited its dancing robots on China’s LNY celebration gala watched by hundreds of millions of spectators. Unitree’s founder was at Xi Jinping’s high-profile symposium along with other leading private entrepreneurs in the country. Several local governments, including Beijing, Shenzhen and Chongqing, have announced to launch investment funds with billions to support the growth of the AI and robotics industries. While industrial robots are commonplace, the widespread use of humanoid robots for household chores and complex factory jobs may still be years away. However, stocks of companies in China's humanoid robot sector have been rallying.

Investment implications

While there is excitement around pure play AI names and large language model developers, the space is highly competitive, and it is too early to determine who the potential winners will be. Use cases are still developing but monetization is still in very early stages. However, we think that the DeepSeek algorithm can drive material efficiency gains and accelerate the take-up of Cloud services of the larger providers in China, which include our holdings Tencent and Alibaba. We have already seen accelerating take-up of Cloud services for Alibaba in its recent results and increasing demand for its AI models.

However, our investment rationale for Alibaba and Tencent is primarily driven by strong and improving core businesses with increasing Cloud growth as an incremental driver. At Quality Growth, we take a more measured approach to investing in AI. While we are optimistic, it is still in early stages and less predictable. Hence, in the technology space, we are more favorable on businesses that have a strong core foundation and with AI providing incremental upside.