Zhitong App has learned that CITIC SEC has released a research report stating that the release of the Sora model by OpenAI this time expands the content modes that AI can understand and generate from text and images to videos and 3D, indicating that the upper limit of basic model capabilities is still being continuously improved. This also indirectly proves that the Scaling Law is not over yet, and although it may take some time for popular applications to land, the imagination space has been opened up, and the demand for computing power infrastructure has not stopped. A new round of industrial catalysis has emerged, and the short-term market intensity may be weaker than that of February to May 2023, but the sustainability may be longer.
▍Key points from CITIC SEC:
The boundaries of basic model capabilities are further enhanced, as OpenAI attempts to promote a large-scale chip manufacturing project, further opening up upstream infrastructure and application space.
Event①: On February 15th, OpenAI launched its first text-to-video model, Sora, earlier than expected (previously expected GPT-5 to be released in April-May this year, demonstrating text-to-3D and video-related capabilities). Compared to previous similar models, Sora has significantly improved in two aspects: 1) Longer video generation duration: Sora can generate video clips up to 60 seconds long, while other models generally do not exceed 10 seconds. 2) Better continuity and consistency: In longer content, Sora maintains excellent camera stability and the ability to switch between multiple cameras in a single video. According to OpenAI, Sora can not only understand the meaning of text prompts but also understand how objects exist in the physical world, indicating that the video generation model has the potential to simulate the real world.
Event②: According to sources cited by The Wall Street Journal on February 8th, OpenAI CEO Sam Altman is in talks with potential investors, including the UAE government, to promote a project aimed at enhancing global chip manufacturing capabilities and reshaping the global semiconductor industry. The plan is to raise as much as $50 trillion to $70 trillion.
How to view the space and sustainability of this round of AI market? The short-term intensity may be weaker than that of February to May 2023, but the sustainability may be longer.
The market has basically overcome the liquidity shock, and the new industrial logic will become an important pricing factor and enhance market risk appetite. The core driving force of the current market lies in AI. Pay attention to the possible release time and parameter scale of GPT-5 in April-May, as well as the research and product progress of overseas vertical applications from the second quarter to the middle of the year. In addition, after AI begins to understand and generate the 3D world and video content, the next breakthrough focus will be on the interaction with the physical world, and progress in humanoid robots and autonomous driving fields is worth looking forward to.
<ol>
<li>Compared to ChatGPT and GPT-4V in 2023, the technological breakthroughs in this round have a deeper and wider industrial impact and a longer duration. ChatGPT is just a large language model mainly used for chatbots. Sora adds video and other modes, and since over 80% of human external information comes from vision, the capabilities and generalization of the Sora model have been greatly improved, bringing us closer to visual emergence and AGI.
After integrating multi-modal video capabilities, various industries such as short videos, advertising, gaming, IP conversion, virtual influencers, and entertainment social media could be reshaped. In the future, the understanding and interaction with the physical world may further impact humanoid robots and autonomous driving. From the perspective of industrial influence, the upcoming Sora and subsequent GPT-5 are expected to surpass ChatGPT and GPT-4V.</li>
</ol>
Concerns about the disruption of the industrial chain may weaken the overseas mapping logic in this round. Looking back at the AI market that started in February 2023, the industry went from 0 to 1, with major players entering the field of large models, leading to a frenzy of primary financing and significant valuation increases, with continuous catalytic events. However, the difference in this round lies in: a) the convergence of large model patterns has begun, with winner-takes-all in basic models, while vertical models have limited growth potential; b) due to the limitation of computing power and the gap in domestic basic model capabilities, short-term application landing is not visible, resulting in a lack of performance realization on the application side, likely turning into a thematic game; c) following the pricing model of the US stock market will amplify the impact of fluctuations on stock confidence, causing investors to refrain from significantly increasing their positions.
Under the security theme, focus on the expansion of domestic wafer fabs and the HarmonyOS ecosystem, while under the green theme, pay attention to the policy catalyst for hydrogen energy.
<ol>
<li>
According to DigiTimes, ChangXin Storage may undergo a second phase of capacity expansion. By the end of 2024, ChangXin's monthly capacity will increase by 40,000 wafers. At the same time, ChangXin plans to significantly increase domestic procurement for new expansions, which is expected to bring order increments to local semiconductor equipment manufacturers.
</li>
<li>
Huawei announced on January 18th the opening of the HarmonyOS NEXT developer preview application, accelerating the landing of HarmonyOS native applications.
</li>
<li>
Strong expectations for policy catalysts in the hydrogen energy sector, with high expectations for the bidding for green hydrogen production and the subsidy for fuel cell demonstration.
</li>
</ol>
Strategic Suggestions:
The market has essentially passed the liquidity crisis, and the new industrial logic will become an important pricing factor and increase market risk appetite. The core driving force of the current market still lies in AI. It is recommended to focus on thematic mapping opportunities: gaming, film and television, advertising, short videos, and the metaverse media sector, as well as configuration opportunities: 1) Participation in the global industrial chain division of labor such as optical modules, server OEMs, PCB, etc.; 2) Domestic computing power chain, advanced packaging, HBM, etc.; 3) Other infrastructure: vector databases, synthetic data, etc.; 4) Humanoid robot chain; 5) Autonomous driving chain; 6) XR chain. Relatively speaking, optical modules, PCB, and domestic computing power chains have higher certainty of performance realization. Other industry trends worth noting: 1) Satellite internet; 2) Storage chain; 3) HarmonyOS; 4) Green hydrogen and electrolyzer.
Risk Factors:
Intensification of friction in Sino-US technology, trade, and finance sectors; domestic policy and economic recovery progress falling short of expectations; unexpected tightening of macro liquidity at home and abroad, exceeding foreign capital outflows; escalation of geopolitical risks, and so on.

OpenAI

06030

600030

CITIC Securities released a research report stating that the Sora model released by OpenAI expands AI capabilities to videos and 3D, proving that the boundaries of basic model capabilities are still improving. Although it may take time for popular applications to land, a new round of industrial catalysis has emerged, and short-term market trends may be weaker than from February to May 2023, but with longer sustainability. At the same time, the CEO of OpenAI is promoting a project aimed at enhancing global chip manufacturing capabilities. The potential and sustainability of the AI market are worth paying attention to.

Understanding the Market | How to view the space and sustainability of the current AI market trend?