How should we view the "explosive" product Sora, and what are the related investment opportunities?

On February 16, OpenAI released its first text-to-video model called Sora, capable of generating up to a minute of high-definition video from text commands. 

Following text and image generation, OpenAI has extended its advanced AI technologies into the video domain. The official video clips released show that Sora’s strength lies in its ability to generate complex scenes featuring multiple characters, specific types of actions, and accurate details of themes and backgrounds. 

As Sora makes its remarkable entrance, the AIGC (AI Generated Content) concept stocks are once again surging globally, with many concept stocks responding with significant gains. Instantly, AI-generated video has become one of the hottest topics on the internet. 

Is commercial "dawn" already shining? 

Whenever a new product or technology emerges, people tend to focus on its commercial value. Industry insiders believe that despite certain limitations of Sora, such as the inability to precisely simulate common physical movements, inconsistencies in illustrating state changes during video interaction, incoherent development of long-duration samples, or objects appearing suddenly, the showcased effects and supported video parameters establish a foundational basis for commercialization. 

AI video commercialization is no longer a far-off dream and is expected to find applications in various fields like film, advertising, and short videos.

Some institutions assert that on the consumer side, these types of AI video applications could build a proprietary traffic pool through the generated content, leading to the subsequent realization of business-side traffic monetization. Moreover, AI video significantly reduces video production costs, which could benefit advertising, film, gaming, and more, potentially leading to exponential development on the content supply side.

With the ongoing buzz around AI-generated video, numerous companies have quickly stated their involvement and ongoing research and development in the cross-modal domain of text-to-video. There are also public companies rapidly upgrading their large models to support automated commercial video generation.

Zhang Hao, Chairman and CEO of Kunlun Tech Co., commented that Sora's capabilities in three-dimensional coherence, simulating the digital world, long-term continuity and object persistence, and interactive technology mark a particularly significant progression and breakthrough in the AI-generated video field. Sora is approximately six months ahead of its domestic peers, posing a considerable impact on the film, video, and advertising industries. However, he believes that the gap between domestic and foreign manufacturers in AI-generated video is not as large as in other large model domains, and Sora's release will further accelerate domestic investment in this field. 

Overall, Sora's further breakthrough in AI-generated video is expected to catalyze the entire artificial intelligence track. The industry anticipates a blooming era for domestic AI-generated video and faster iterative advancements in multi-modal large-scale models.

What investment opportunities are there? 

Since video training demands significantly more data than text, it is predicted that Sora's large model will increase the gap for computational power.

CITIC Securities points out that Sora combines the diffusion model (DALL-E 3) with the transformer architecture (ChatGPT), allowing the model to process video frames as if they were text, generating each frame of an image in one go. Based on currently available public data, it is speculated that Sora's training set may have been populated with high-quality scenes generated by video game engines, as the quality of the training set remains crucial for the model's output. From the perspective of AI industry investment strategy, three directions are of interest: 

1. An anticipated increase in AI computational power demand; 
2. Video-related applications and generation-oriented AI commercialization are expected to gradually materialize; 
3. Content and IP, as the innovative ways of generating images and videos emerge, the intrinsic value of the content itself will be further highlighted.