With Sora shutting down, what will become of AI video generation startups?
Image source: Visual China
Text by Rao Fuying; edited by Ye Jinyan
Produced by DeepWeb & Tencent News Xiaoman Studio
On March 24, OpenAI announced on the overseas social media platform X that it was shutting down the Sora app. OpenAI CEO Sam Altman stated that in addition to consumer applications, OpenAI would also discontinue developer versions of Sora and would not support video functionality in ChatGPT.
Sora was first released in September 2025 and quickly became a sensation, hailed by the industry as the "GPT moment " of video generation . However, just six months later, the product was discontinued. This stark contrast between its meteoric rise and its eventual demise raises a crucial question for the industry: Why did OpenAI shut down this once highly anticipated product?
Why did OpenAI shut down Sora?
Technically, Sora has consistently failed to cross the threshold from "amazing" to "usable". Song Chunyu, senior partner of Lenovo Capital and Incubator Group, once said: "I personally feel that the technical framework of Wensheng Video is not yet fully developed. OpenAI's Sora is like giving everyone a demo. The model itself still needs breakthroughs."
Xiao Shi, an AI video entrepreneur, also confirmed this assessment: "Its generation effect is amazing enough, but it has never reached the standard of stable commercial use and stable delivery. It has never been able to solve the core commercial needs such as high controllability, reproducibility, and mass production."
Market data provides a more direct reflection of the product's predicament. After the Sora standalone app launched in September 2025, downloads exceeded one million in the first ten days, and its popularity even surpassed ChatGPT. However, the glory was short-lived—downloads dropped by 32% month-on-month in December, and continued to decline by 45% in January 2026, with user spending also continuing to fall.
More critically, user retention is a key factor: data disclosed by Olivia Moore, a partner at Silicon Valley venture capital firm a16z, shows that Sora's 1-day, 7-day, 30-day, and 60-day user retention rates were only 10%, 2%, 1%, and 0%, respectively. Although downloads surged at one point, users were rapidly churning.
Cost pressures became another major factor that brought Sora down. According to media reports, its monthly computing power cost reached as high as $15 million, and the massive computing resources it consumed directly weakened the computing power supply of other core teams at OpenAI. The long-standing internal resource allocation conflicts became a major driving force behind its shutdown.
Sam Altman made it clear in internal communications that shutting down Sora was a key move in the company's strategic restructuring, and that in the future, core computing power, talent and capital would be fully shifted to enterprise-level productivity tools.
Furthermore, OpenAI plans to launch its IPO in the second half of 2026, and ending the continuously loss-making Sora business will help optimize its financial performance and send a clear profit forecast to the capital market.
Along with the shutdown, Sora's three-year partnership with Disney, involving a $1 billion investment and licensing of over 200 IPs, also came to an end.
One investor commented on the event: "Products like AI video generation are more like attempts during the transition period of AI development. This round of AI development will continue to experience bubbles and shakeouts, and it may be more intense than before. This is because its initial expectations were too high, and its impact is also greater. But this shutdown process itself is normal."
For those in the industry, Sora's demise seems unsurprising. "Just a has-been," commented one AI video entrepreneur upon hearing the news of Sora's shutdown. Since Sora introduced a paid model in December 2025, many industry professionals, including him, have essentially abandoned the product. In his view, Sora's peak was only in its initial release; subsequent models continuously "degraded in intelligence," resulting in declining quality that fell far short of market expectations.
Does Sora's exit signify a turning point in the video streaming industry?
Just as ChatGPT sparked a wave of large language model startups in early 2023, the release of Sora in February 2024 also accelerated the evolution of textual video models.
Prior to this, AiShi Technology had already released PixVerse V1, Runway Gen1, and Pika 1.0. Following this, startups rushed in, and tech giants entered the fray. In April 2024, Shengshu Technology released Vidu 1.0, a video platform comparable to Sora, and launched several new features during this period. Subsequently, Kuaishou released Keling, and MiniMax's Conch Video App was also launched globally…
Now, with Sora's abrupt end, the competitive landscape is facing a new round of reshuffling.
The competitive balance between large companies and startups is tilting in one direction. Industry insiders point out that large companies can leverage their existing business systems to form a closed ecosystem, using AI video as infrastructure to serve their core businesses such as content platforms, achieving large-scale coverage and ecosystem synergy in a short period. This inherent advantage is precisely what startups struggle to match, directly impacting the speed of their subsequent commercialization process.
The fundamental technical challenges facing all players have yet to be overcome. From a technical perspective, the differences between mainstream products on the market are not significant. An employee of Conch AI once stated, "In terms of duration, it's basically around 5 seconds, at most 20 seconds. If the generation time is too long, the possibility of crashing later is very high; this is a common problem of insufficient stability." There is still a significant gap between technological maturity and commercial reliability.
Data copyright compliance issues have become another 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. "By 2028, there is reason to believe that large models will consume the existing, ready-made data generated by humans. In the face of the possible data shortage in the future, we should make arrangements in advance and think about coping strategies."
While opportunities are clear for large companies, startups are not without their chances. "Startups need to build their own unique interactions or content. No one has yet defined the video agent market, so it's still a race for speed and innovation," one investor pointed out.
You May Also Like
Nvidia Faces Class Action Over Alleged Hidden Crypto GPU Revenue
Ethereum koers toont zeldzaam dubbel koopsignaal en richt zich op $4.550
