Sora ultimately couldn't escape the fate of being taken offline.
Even when it first debuted, I had a feeling it wouldn't succeed. After all, what reason does a community with only creators have to survive? Especially since too many similar things have happened in the past two years—a powerful AI product emerges, everyone marvels for a while, and then gradually no one mentions it anymore. Sora is just repeating this "classic path," even though it once topped the AppStore charts.
Many in the industry are discussing why OpenAI abruptly cut multiple product lines, including Sora. Some say it's for an IPO, others blame costs. Of course, OpenAI's official explanation is that it wants to focus on researching world models.
But is that really the case?
We all know that since December 2024, when Sam Altman declared the beginning of the Agent era, OpenAI has shifted its business strategy from model capabilities to a product-first approach. The models it subsequently released lacked the initial wow factor, and the market has shifted from ChatGPT's dominance to a three-way standoff with Gemini and Claude. At the same time, it has also invested heavily in productization attempts. The launch of the Sora app was one of those attempts.
So, rather than saying Sora's shutdown is OpenAI's strategic contraction, it's more accurate to say it's just another failed AI product.
If it were just an ordinary product failure, it wouldn't be worth much discussion, but Sora is a bit different. In the text-to-video赛道, few can match it. Since its release a year ago, it has remained outstanding in terms of effect and stability; there's nothing wrong with its model capabilities. But when it comes to the Sora app, examining it from a product perspective reveals that its biggest problem is that it has done almost nothing over the past year—at least in terms of becoming a "useful tool," it has made no progress.
Anyone who has used text-to-video knows that to generate a decent video, the first step is to write a clever prompt, and the second step is to rely on luck with the model. With extreme luck, you might get the desired effect in one or two tries, but most of the time, it's a random gamble. If you don't end up with a Van Gogh effect, you're already lucky.
This fixed pattern of "high probability of failure" makes it hard to know whether the prompt wasn't good enough, the model was in a bad mood, or you didn't wash your hands after handling a can of surströmming. You have no way of knowing if the next attempt will be closer to your ideal result, let alone gradually steer the outcome in the desired direction like you would with a tool.
This is a major taboo for a product. Once users can't pinpoint the problem, it's hard for them to gain any experience. Using it ten times is no different from using it once. The huge uncertainty creates a cliff-like experience, and having to start from scratch every time you open it is unbearable for anyone.
The ecosystem never developed, the interaction experience saw no improvement, the 30-day retention rate was nearly zero, and on top of that, its operational costs were extremely high. This is the real reason Sora was taken offline.
If growth could have been achieved, or even if OpenAI had used Sora to define the paradigm for AI applications, no matter how much money it burned daily, it would have been a drop in the bucket. But it neither retained users nor formed user habits. Creators couldn't profit, every generation was a one-time consumption, and all the high-quality content was funneled to short-video platforms for monetization. That's very awkward.
In a product with a complete ecosystem, all operational costs should be spread out through economies of scale, but Sora became a tool to help other platforms spread their costs. It was like a one-way funnel, with all costs being net outflows. In this situation, for OpenAI, having more users was painful, and having fewer users was uncomfortable. It couldn't rely on growth to save itself, nor could it cut losses by shrinking.
In the end, the only option was to take it offline.
Looking back at Sora now, aside from those aspects directly covered by model capabilities, its failure shares the typical characteristics of all the "predecessors" that have been washed away: no ecosystem to form a positive cycle, a janky interaction experience, and an imbalanced cost structure.
These structural issues are the key factors determining whether an AI tool can truly become a product. Overcome them, and you reach the sky; fail to, and the only option is to pack up and run.