Sora publicly bids farewell, OpenAI admits defeat?

(Source: TuChong)

On March 25, Sora announced, “We are saying goodbye to the Sora App.”

This video generation model was once very popular and was highly anticipated to become the video version of “ChatGPT” when released in 2024.

In December last year, Disney announced that the company would invest $1 billion in OpenAI, the parent company of ChatGPT, allowing Sora to create AI video content using Disney’s intellectual property. Users would also be able to watch selected Sora-generated videos on Disney+.

Now, before the investment is finalized, Sora has already bid farewell to the market.

However, the AI video generation market still seems quite active.

On X, right after Sora announced its shutdown, Elon Musk said that they will increase investment in the Grok image video generation model.

Pioneers Under Pressure

Sora’s initial release and attention drew early interest in early 2024, which was quite ahead of the curve. In May of the same year, ByteDance’s JiYun officially launched its AI video feature; in June, Kuaishou’s KeLing gradually opened user applications.

As 2025 arrives, competition in the video generation track continues to heat up.

In September 2025, OpenAI officially released Sora 2 as a standalone app for iOS, including a new feature that allows users to create “cameo shots” of themselves, friends, and others.

At that time, the company stated: “With Sora 2, we are directly moving toward what we believe could be the ‘GPT-3.5 moment’ in the video field.”

According to media reports, within five days of its launch, the Sora App surpassed 1 million downloads worldwide, quickly topping the free charts on the App Store in major markets like the US and China.

But over time, this initial excitement seems to have waned.

Media reports indicate that, according to Appfigures, downloads of the Sora App dropped from 2.2 million in December last year to 1.2 million in January this year—a decrease of about 45% month-over-month.

As downloads declined, app revenue also decreased.

Appfigures data shows that user spending on the Sora App fell from $540,000 in December to $367,000 in January.

And the high costs are a more serious problem.

(Source: TuChong)

“OpenAI decided to abandon Sora mainly because the computing power was too expensive to sustain,” said Zhang Xiaorong, Director of DeepTech Research Institute. He pointed out that internal estimates at OpenAI showed that if Sora remained free with a daily active user base of millions, the inference costs alone would exceed $1.5 billion annually—this doesn’t even include model training and iteration costs.

For the AI market, investing more than the output for a period is normal, but more critically, in recent years, OpenAI has shifted from being a non-profit to a company seeking more growth.

A report from Securities Times mentioned that in August last year, OpenAI CFO Sarah Friar publicly stated that the company was considering going public at some point in the future. The same report also noted that OpenAI CEO Sam Altman had expressed confidence that the company would eventually go public. There are also reports suggesting OpenAI plans to list in 2026.

The industry generally believes that whether Sora as a product is successful is still uncertain. The main reason is OpenAI’s broad and ambitious exploration strategy, combined with Sora’s high resource consumption, which has affected the company’s investment in other projects.

Meanwhile, in the U.S., Anthropic has taken a different path. Unlike OpenAI, Anthropic is fully focused on AI programming. Its flagship product, Claude, concentrates on text and code generation, rapidly gaining traction in programming and enterprise markets.

Anthropic disclosed that in less than three years since its founding, it has achieved an annualized revenue of $14 billion. Over the past three years, this figure has grown more than tenfold each year.

With these factors combined, Sora has now reached its farewell moment in the market.

Market Still Has Room

What does Sora’s exit mean for the market?

Internet analyst Zhang Shule believes that Sora’s withdrawal does not signal the decline of the AI video generation market. “For players in the market, this just proves that no one is irreplaceable. Only through technological iteration and leading innovation can one maintain a presence.”

The competition in the AI video generation market is gradually increasing.

In fact, on the very day Sora announced its exit, Elon Musk on X said, “The next version of Grok Imagine will be very impressive. We are doubling down.” Grok Imagine is xAI’s video generation tool.

If the market is healthy, then for companies operating within it, what is the key to development?

Zhang Xiaorong believes that the current core bottleneck for AI video generation projects is the lack of real-world application scenarios, which leads to a lack of genuine demand for scalable paid services.

Zhang Shule thinks that behind the unsuccessful collaboration between Disney and OpenAI lies a huge “IP + AI” track. “In the current era where AI-driven dramas are emerging, the entire industry’s exploration of IP + AI video models will not stop,” he said.

“Domestic companies are exploring revenue models by embedding e-commerce, short dramas, and advertising into B-end applications, creating positive cash flow,” Zhang Xiaorong added.

An industry insider working at an AI company told Times Finance that in the field of animated dramas, introducing AI can reduce single-episode production costs by up to 90%. In e-commerce scenarios, integrating AI into product pages for static and dynamic content can cut costs by about half. “In short, scenarios that prioritize mass production and speed over perfection have space for AI video applications.”

Recently, on the internet, a topic titled “Have you ever rescued a fox on a snowy mountain?” has sparked many discussions. These are AI-generated videos made by netizens, often starting with animals repaying kindness but ending with surprising twists, attracting attention. The tools behind these videos include domestic models like JiYun AI and others.

ByteDance has not gone public yet, but the financial reports of Kuaishou, MiniMax, and others show progress in commercializing domestic large AI video models.

In Kuaishou’s March 25 earnings report, the company reported impressive data: Kexing AI’s revenue in Q4 reached 340 million yuan. In December, Kexing AI’s monthly revenue exceeded $20 million, with an annualized run rate of $240 million.

Kuaishou’s management mentioned that Kexing AI is mainly applied in marketing, e-commerce, film and TV, short dramas, animation, and gaming creative scenes. During the earnings call, executives also noted that Kexing AI is involved in virtual scenes and special effects for the hit drama “Tai Ping Nian.”

Kuaishou CEO Cheng Yixiao stated that as of January this year, Kexing AI’s ARR exceeded $300 million, and the team is confident of achieving over 100% year-over-year growth in revenue this year.

Additionally, MiniMax’s prospectus shows that by the end of Q4 2024, Hailuo AI had 5.735 million users and clients; by Q3 2025, these numbers increased to 42.34 million.

(Source: Screenshot from MiniMax IPO prospectus)

In its 2025 annual report, MiniMax mentioned that AI-native product revenue in 2025 reached $53.075 million, a 143.4% increase year-over-year, mainly driven by increased user engagement, willingness to pay, and ongoing promotion and commercialization of products like Hailuo AI.

(Source: Screenshot from MiniMax 2025 Annual Results Announcement)

In December last year, director Lu Chuan announced that his company, Yuan Dongli, is collaborating with MiniMax and Hailuo AI on AIGC technology, jointly developing AI drama projects.

Of course, the AI video generation market also faces common challenges.

For example, copyright issues. Media reports indicate that ByteDance’s Seedance 2.0 was once sent a letter from Disney, accusing it of using Disney works without permission during training and development. Currently, Seedance 2.0’s global release is paused, and features like real-person material reference, face reference, and IP image generation are also suspended.

Cost is another issue. Kuaishou’s management estimates that by 2026, the company’s overall Capex (capital expenditure) will reach about 26 billion yuan, partly due to investments in computing power for large models like Kexing AI. They also mentioned that the Capex increase is driven by the need for inference computing power due to user growth and revenue scale, as well as preparations for significant model upgrades.

Industry insiders believe that, in order of importance, finding application scenarios and building an ecosystem are the most critical for AI video generation, followed by establishing infrastructure and ensuring copyright compliance.

As long as the business model is viable and the market is large enough, certain investments remain acceptable.

In 2025, Kuaishou’s “Other Services” revenue grew sharply by 27.6% year-over-year to 22.2 billion yuan. The company attributes this mainly to growth in e-commerce and Kexing AI.

A recent report from China Merchants Securities predicts that the AI video generation market will reach $40 billion by 2030. Furthermore, if AI’s penetration rate in production reaches 10%, the B2B market size could hit $36 billion. On the consumer side, with increasing content creators, AI adoption, and willingness to pay, the market could reach about $4 billion by 2030.

“To make breakthroughs in the future, resources must still be invested in refining algorithms—how to achieve the best results with minimal computing power and enable users to get the best products at the lowest cost,” Zhang Shule emphasized.