Hong Kong Stock Concept Tracking | OpenAI's first text-to-video model Sora is a hit! Computing power is expected to become a hot investment direction (with concept stocks)

Zhitong App learned that on February 16th, OpenOpenAI released its first text-to-video model "Sora". According to the official website of OpenOpenAI, Sora can generate high-definition videos up to 1 minute long based on text commands. The videos can feature multiple characters, specific types of movements, precise themes, background details, and other complex scenes. Industry analysts believe that Sora is undoubtedly a significant breakthrough in the field of artificial intelligence. This technology not only demonstrates OpenAI's advanced capabilities in understanding and creating complex visual content but also brings unprecedented challenges and opportunities to the content creation, entertainment, and film production industries. Related concept stocks: Baidu Group-SW (09888), ZTE (00763), CHINA MOBILE (00941), China Telecom (00728).

Regarding the launch of Sora, Zhou Hongyi, the founder and chairman of 360 Group, stated that this means the realization of AGI will be shortened from 10 years to 1 year. Video-generating OpenAI with text-to-video capabilities can effectively lower the threshold for creators.

The popular "internet celebrity" Musk also commented on the new model released by OpenOpenAI. In response to a Twitter user's repost of a demo video of Sora with the caption "gg Pixar", Musk commented below the tweet, saying "gg humans".

Another netizen discussed OpenOpenAI's new model and shifted the topic to the film industry, stating, "The film industry will definitely react strongly to this technology, hoping that regulations will not get out of control" and "Unlike most types of OpenAI creations, generative art will not suppress the human spirit." Musk responded to this tweet, saying, "Humans enhanced by artificial intelligence will create the most outstanding works in the coming years."

However, OpenOpenAI also pointed out that the current Sora model has its weaknesses. It may struggle to accurately simulate physical phenomena in complex scenes and may fail to understand specific causal relationships. For example, a person may take a bite of a cookie, but after the bite, there may be no teeth marks on the cookie. The model may also confuse spatial details in prompts, such as left and right, and may have difficulty accurately describing events that occur over time, such as following a specific camera trajectory.

In recent years, OpenOpenAI has been leading the OpenAI track. In early 2021 and late 2022, OpenOpenAI respectively launched the image generation system DALL·E and the chatbot ChatGPT. This has made OpenAI gradually become a tool for various industries and is gradually changing people's views on future work.

OpenOpenAI stated that Sora is built on the research foundation of the DALL-E and GPT models in the past. It adopts the technology of DALL·E 3, which can more faithfully follow the user's text instructions in the generated videos. In addition to generating text-to-video, this model can also generate videos based on existing static images and accurately and meticulously animate the content of the images.The model can also extract existing videos and expand or fill in missing frames.

For the development of Sora, there is a strong demand for computing power. Guotai Junan pointed out that the Sora model is driving a leap forward in the multimodal field of OpenAI. Areas such as OpenAI's creative work will undergo profound changes, expanding the scope of OpenAI's empowerment, and the demand for computing power infrastructure in multimodal training and reasoning applications will further increase.

Similarly, Guosheng Securities also holds the same view, believing that Sora still complies with OpenAI's Scaling Law. As the training computational volume increases, the sample quality significantly improves, further confirming that in the era of multimodality, the demand for computing power will become one of the most critical bottlenecks. OpenAI's computing power is expected to continue to be a hot investment direction in the new year after 2023.

In addition, Huatai Securities also released a research report stating that OpenAI's release of the Sora video model heralds the eve of large-scale applications for OpenAI videos. 1) The Sora model surpasses previous competitors in aspects such as video generation time, semantic understanding, video effects, and stability. With the successive emergence of applications like Sora and Pika, the competition in future OpenAI video applications may become more intense; 2) Although Sora's usage permission has not been publicly disclosed, its potential commercialization is expected to have a profound impact on downstream areas such as short videos, movies, and games; 3) OpenAI's video applications consume far more computing power than text, audio, and images. It is recommended to pay attention to the increase in inference-side computing power demand and whether its degree of commercialization can form a positive feedback loop for income and investment.

Related concept stocks:

Baidu Group-SW (09888): Baidu Cloud Computing (Yangquan) Center is Baidu's first large-scale self-built data center project, which has been running securely and stably for 1533 days. It successfully provides services to important businesses such as Baidu Search, Du Secretary, Intelligent Cloud, Basic Technology Group, Emerging Business Group, Artificial Intelligence, and Intelligent Driving, becoming an important cornerstone for empowering OpenAI and achieving the future.

ZTE Corporation (00763): ZTE Corporation's Intelligent Cloud Card DPU flexibly offloads virtualization, network, storage, security, and other basic service loads, maximizing computing power efficiency. The full range of servers and storage products supports liquid cooling, GPU, and 400G bandwidth. Data center switches are based on ZTE's self-developed high-performance switching, forwarding, and CPU chips, with bandwidth smoothly evolving from 400G to 800G.

CHINA MOBILE (00941): CHINA MOBILE will launch a new "capability + computing power" package that can be flexibly configured and freely combined. It has already created a "3+2+1" computing power terminal product system, where "3" refers to the creation of three thin terminal products: cloud phones, cloud computers, and mobile cloud HD; "2" refers to the creation of powerful terminal products such as flagship computing host and home computing host; "1" refers to the creation of a unified cloud OS platform to achieve unified management and resource scheduling of computing power terminal products.China Telecom (00728): Currently, China Telecom has over 700 data centers, more than 3,000 edge computing centers, 513,000 IDC racks, with a rack utilization rate exceeding 70%. By the end of 2023, China Telecom's total computing power is expected to reach 6.2Eflops.