Although there is no clarity on how artificial intelligence (AI) in China would be monetised, regulated and publicly consumed, Morningstar Equity Research expects Baidu and Alibaba Group Holding to be the key beneficiaries within its coverage in the near term.
Analysts Kai Wang, Chelsey Tam, Dan Baker and Ivan Su believe the two Internet giants can generate incremental revenue to their cloud or advertising businesses from clientele who are willing to pay for generative AI to enhance their existing products as well as improve operating efficiency.
When the pick-up in AI was seen in financial markets earlier this year, some observers pointed out that it is not exactly a new technology, as AI in various shapes or forms has been around for decades. Similarly, in their July report, the analysts point out that AI is not entirely new to China’s tech industry. However, there is a greater emphasis on the development of foundational AI in large language models (LLMs) recently.
Morningstar believes that earlier movers of foundational AI models, such as Baidu’s Ernie and Alibaba’s Tongyi Qianwen (which is offered under Alibaba Cloud), will benefit the most in the near term due to the faster time-to-market of their AI products.
Additionally, Baidu and Alibaba Cloud already have the hyperscale cloud computing capabilities that are necessary to manage the computing requirements of LLMs. This should give them an advantage over smaller companies. Tencent AI, for one, currently lags behind its peers given the time required to develop these models.
That said, the analysts forecast that Baidu and Alibaba’s first-mover advantage would dissipate in the long run as there will be more large-scale generative AI models launched by other resource-heavy Internet companies like Bytedance and Tencent, who are well placed to capture and dissect large quantum of data and content.
See also: Without regulator buy-in, scaling AI in financial services will be an uphill battle
Ernie versus Tongyi Qianwen
The two key products of Baidu’s flagship AI foundation model Ernie are its existing cloud service and core advertising. Upon speaking to its management, the analysts understand that Baidu’s main strategy is to incorporate generative AI into its cloud business while Ernie will leverage its cloud business to use generative AI in its services to cater to new clients.
Baidu believes that advertising revenue will be enhanced by the addition of Ernie. It expects other industries outside of its top verticals — healthcare, e-commerce, travel, education, and real estate — to become more interested in learning about how Ernie can improve advertising outreach for their campaigns.
“In our conversation with Baidu, management used an example of a media company since media is not normally part of Baidu’s clientele. Media companies can use its generative AI to help their operations, given the co-piloting functions of Ernie where the foundation model is able to turn natural language prompts into suggestions and ideas for its users.
“This enhances the creativity and productivity of its clients, similar to [American AI research firm] OpenAI’s technology. Baidu believes that its co-piloting feature should lead new industries to start using Ernie, which will bring in new clients,” the analysts highlight.
Meanwhile, Tongyi Qianwen is Alibaba’s latest AI LLM. Since its launch on April 11, Alibaba Cloud has received over 200,000 beta testing requests from enterprises across different sectors. Developers can access Tongyi Qianwen to construct their AI applications as the model will be enhanced with multimodal capabilities, such as the ability to understand images and generate images from text.
Tongyi Qianwen will first be deployed in DingTalk, the digital collaboration workplace and application development platform-as-a-service (PaaS) platform. It will also be implemented in Tmall Genie — Alibaba’s consumer electronics brand — and provide a selection of Internet of Things-enabled smart home appliances, including smart speakers, lights and remote controls as per Alibaba’s website.
Alibaba plans to integrate Tongyi Qianwen into all of its businesses to further enhance user experience, similar to Baidu with Ernie. The analysts believe Alibaba could also develop a chatbot product which uses queries to give users personalised shopping recommendations that suit their needs, similar to what Amazon is planning.
“We expect enhanced user experience and incremental revenue in these if there is successful integration of the LLM into Alibaba universally,” they add.
Valuation impact
For more stories about where money flows, click here for Capital Section
Morningstar estimates that Baidu’s foundational AI model can add about 7% to 9% to the analysts’ existing fair value estimate of US$183 ($242). Assuming a 30% incremental growth to the existing revenue CAGR forecast of its advertising segment, the analysts estimate that the four-year CAGR revenue growth from 2024 to 2027 could be around 11% (compared to 8.7% previously).
The Morningstar analysts have a more aggressive projection for Baidu’s cloud business. They assume 50% incremental growth to the current growth rate forecast of 10% CAGR from 2024–2027, which will result in the cloud business CAGR to increase to 15%. This translates to a cloud revenue increase of RMB13.5 billion ($2.5 billion) over four years and valuation increase of 3%.
For Alibaba’s Tongyi Qianwen, the analysts believe the usage among developers may lead to cross-selling opportunities. They also foresee greater adoption of DingTalk and Tmall Genie leading to an increase in cloud revenue in the long term.
Due to the introduction of generative AI to all of Alibaba’s business segments, the analysts expect improved user experience and higher monetisation of all products and services as well as higher cloud revenue. This should lead to a fair value estimate increase of 4% to US$185 per American depository share for Alibaba.
Meanwhile, Tencent has provided few details on its AI progress, Morningstar notes. The analysts believe it will be harder for AI to reshape the human-to-human interactions that are the foundation of Tencent’s core businesses such as social networking and gaming. That said, the company is leveraging AI to improve the quality of its content, the efficiency at which the content is generated, as well as lowering the cost of content creation, the analysts point out.
“Tencent believes that AI is more disruptive in human-to-machine interactions such as search and content recommendations. However, the company stated that it was not going to rush its own chatbot and we believe that AI has been less of a focus given its de-emphasis in its infrastructure-as-a-service (IaaS) products recently as it is incurring losses,” the analysts add.
Separately, Morningstar points out that the Chinese telecom players have also not provided any specific AI solutions to their existing IaaS or PaaS products. However, the telcos remain among the market share leaders in both IaaS and PaaS. Morningstar believes the operators have some advantages in the AI space including strongly cash-generative core businesses enabling investment in new technologies, access to very large computing power infrastructure as well as existing telecom and cloud customer bases.
However, the analysts believe the Internet companies have access to more relevant data with which to train their LLMs. “The telcos will have to evolve as these products become more commoditised, which could be a struggle if they do not have an AI core competency,” they add.
Potential headwinds
The analysts, in an attempt to give a reality check, have flagged two large potential headwinds to China’s AI industry. First is the restriction of semiconductors that are required for training and inference costs. However, the analysts note that most of the training costs have already been incurred, given Alibaba and Baidu both started to build their models in 2019.
Morningstar expects the inference costs to be higher for Chinese companies compared to OpenAI, which has spent hundreds of millions of dollars to enable the processing power that ChatGPT needs. This is as the Chinese players can only access the Nvidia A800 GPU, which runs at 400 gigabytes per second instead of the A100 GPU, which runs at 600 gigabytes per second.
“We do not believe there are currently domestic solutions to this problem as Baidu indicated that it would likely order A800 and H800 chips from Nvidia rather than use its own inhouse Kunlun AI chips,” the analysts add.
Exacerbating the restrictions are the reports that the US is considering banning the sale of A800 chips to China. Morningstar thinks that this could greatly hamper the development of foundational AI, forcing companies to rely on domestic chips, which will take time to be produced on a mass scale.
The second potential headwind for China’s foundational AI models is how they will deal with censorship and sensitive topics. The analysts believe that it will be difficult for foundational AI models to filter out unwanted answers, as one of the risks of foundational AI is “hallucination”, or the generation of plausible answers that are false in nature.
Given the challenge of preventing hallucination in AI models, the AI industry is expected to be heavily regulated in China. As a result, the analysts are unsure whether eventual regulation will affect the performance of AI models, or whether the additional filters mean that there will be greater inference costs to run the models.
“Our contacts indicate that there may be some pressure to maintain a Chinese ideology for the foundational AI products. This should not be surprising given regulatory policies in the country regarding media and other information-type services and products. Currently, all the models are still undergoing mass-testing for filtered content as there is no public widespread use yet,” they conclude.
