Image source: Visual China
Text: Lao Fuying; Edited by Ye Jingyan
Produced by DeepWeb and Tencent News Xiaoman Studio
On March 24th, OpenAI announced the closure of the Sora app on overseas social media platform X. OpenAI CEO Sam Altman said that in addition to consumer applications, OpenAI will also discontinue the developer version of Sora and will no longer support ChatGPT’s video features.
Sora was first released in September 2025 and quickly became a sensation, hailed by the industry as a “GPT moment” for video generation. However, just six months later, the product was discontinued. This stark contrast between its meteoric rise and eventual demise raises serious questions for the industry. Why did OpenAI shut down this once highly anticipated product?
Why did we introduce OpenAI? Do you want to shut down Sora?
Technically speaking, Sora has consistently failed to cross the threshold from “great” to “usable.” Song Chunyu, senior partner at Lenovo Capital and Incubator Group, once said: “Personally, I feel that Wensheng Video’s technical framework is not fully developed yet. OpenAI’s Sora is like providing a demo for everyone. The model itself still needs breakthroughs.”
Xiao Shi, an AI video entrepreneur, agreed with this assessment, saying, “The production effect is amazing enough, but it has never reached the standards of stable commercial use and stable distribution. It has never solved the core commercial needs of high controllability, reproducibility, and mass production.”
Market data more directly reflects product distress. When the Sora standalone app was released in September 2025, it received over 1 million downloads in the first 10 days, surpassing even ChatGPT in popularity. But that glory didn’t last long, as downloads fell 32% month-over-month in December and continued to drop 45% in January 2026, and user spending continued to decline.
More importantly, user retention is a key factor. Sora’s 1-day, 7-day, 30-day, and 60-day user retention rates were just 10%, 2%, 1%, and 0%, respectively, according to data disclosed by Olivia Moore, a partner at Silicon Valley venture capital firm a16z. Downloads briefly spiked, but users were quickly losing users.
Cost pressure was another major factor in Sora’s downfall. According to media reports, the company’s monthly computing power costs reached as much as $15 million, and the huge computing resources consumed directly weakened the computing power supply of OpenAI’s other core teams. Years of internal resource allocation conflicts were the main driving force behind its closure.
Sam Altman said in an internal communication that the closure of Sora is an important step in the company’s strategic restructuring, which will see its core computing power, talent and capital fully transitioned to enterprise-level productivity tools in the future.
Additionally, OpenAI plans to launch an IPO in the second half of 2026, allowing it to exit its loss-making Sora business to optimize its financial performance and communicate clear profit projections to capital markets.
The closure also ended Sora’s three-year partnership with Disney, which involved a $1 billion investment and licensing of more than 200 IP.
One investor commented on the event: “Products like AI video generation are like experiments in a transition period of AI development. This round of AI development will continue to have bubbles and culling, and may be more intense than before. This is because expectations were too high at the beginning, and the impact is also large. However, this shutdown process itself is normal.”
For those in the industry, Sola’s death does not seem to be a surprise. “It’s all in the past,” commented one AI video entrepreneur upon hearing the news of Sora’s closure. Since Sora introduced the paid model in December 2025, he and many industry experts have effectively abandoned the product. In his view, Sora only peaked during its early releases. Subsequent models were continually “less intelligent”, resulting in lower quality and falling well below market expectations.
Does Sola’s departure mark a turning point for the video streaming industry?
Just as ChatGPT sparked a wave of large-scale language model startups in early 2023, the release of Sora in February 2024 also accelerated the evolution of text-video models.
Prior to this, AiShi Technology had already released PixVerse V1, Runway Gen1, and Pika 1.0. This was followed by a flurry of startups and tech giants entering the fray. In April 2024, Shengshu Technology released Vidu 1.0, a video platform comparable to Sora, and released several new features during this period. Then Kuaishou released Keling and MiniMax’s Conch Video App was also released worldwide…
Now, due to Sora’s sudden retirement, the competitive environment is facing a new restructuring.
The competitive balance between large corporations and startups is tipping in one direction. Industry insiders point out that large enterprises can leverage their existing business systems to form a closed ecosystem and use AI video as an infrastructure to serve core businesses such as content platforms, achieving large-scale coverage and ecosystem synergies in a short period of time. This inherent advantage is exactly what startups struggle to match, and it directly impacts the speed of the subsequent commercialization process.
The fundamental technical challenges facing all players have not yet been overcome. From a technical point of view, there are no major differences between the mainstream products on the market. A Conch AI employee said, “The duration is typically around 5 seconds, and at most 20 seconds. If the generation time is too long, there is a high chance of a crash later on, which is a common problem of poor stability.” There remains a significant gap between technology maturity and commercial reliability.
Data copyright compliance issues are also an invisible barrier. MiniMax was previously sued by iQiyi. Currently, the main solution for various companies is to cooperate with film and television institutions and video platforms.
Mei Tao, founder of Zhixiang Future, once said that high-quality copyrighted data assets will become one of the core competencies of artificial intelligence companies. “There is reason to believe that by 2028, large-scale models will consume existing off-the-shelf human-generated data. In the face of possible future data shortages, we need to prepare in advance and consider coping strategies.”
While there are clearly opportunities for large companies, startups are not without opportunities as well. “Startups need to build their own interactions and content. The video agent market has not been defined by anyone yet, so it’s still a race for speed and innovation,” one investor noted.
