OpenAI’s recently launched short-form video application, Sora, appears to possess all the elements for widespread viral success. Within mere hours of its Tuesday debut, memes generated using its AI video-creation technology were already circulating across other social networks—including, for example, a video depicting OpenAI CEO Sam Altman rapping from inside a toilet bowl.
The introduction of Sora—complete with a “for you” page reminiscent of TikTok—marks a notable change in stance for Altman. He had previously characterized social media feeds as “an instance of misaligned AI,” whose algorithms are “incredibly effective at keeping you scrolling.”
Altman promptly distanced himself from suggestions that OpenAI had yielded to the temptation of creating what he referred to as an AI-powered “slop feed.” He stated: “The team has put great care and thought into trying to figure out how to make a delightful product that doesn’t fall into that trap, and has come up with a number of promising ideas.”
Nevertheless, OpenAI’s decision to launch a short-form video app is a potential indication that many tech companies now believe that brief, AI-generated video content is a crucial stepping stone towards achieving their goal of artificial general intelligence. Five days prior to Sora’s launch, Meta introduced its own AI-powered short-form video app, called Vibes; meanwhile, Chinese competitors, including TikTok’s parent company ByteDance, are striving to lead in this technological frontier.
AI-generated video is resource-intensive. The AI systems capable of producing video demand considerably more energy than other systems that only generate text and images, such as ChatGPT.
Despite this, OpenAI and Meta now seem to be betting that AI-generated video feeds—although likely operating at a loss for now—will ultimately become a substantial source of revenue. That funding may be critical for financing the training of increasingly large systems that both companies have committed to building.
For OpenAI especially, developing new revenue streams is essential. Meta already possesses a highly profitable business through the targeted advertisements it delivers via Instagram and Facebook. OpenAI, by contrast, is not profitable. Despite its immensely successful product, ChatGPT, which brings in over $1 billion per month, the company’s expenditures on servers and personnel are so substantial that it reported a $7.8 billion loss in the first half of this year, according to the tech news site The Information.
“Much of the current AI landscape revolves around prioritizing growth—essentially acquiring users and subsequently determining how to monetize them,” states Selina Xu, the China and AI policy lead in the office of former Google CEO Eric Schmidt. “Competition is fierce, so the focus is on: ‘Let’s attract individuals who are not yet using Facebook or ChatGPT.’”
There is also a competitive aspect to OpenAI’s decision to develop not just a viral AI model, but also a distribution channel in the form of a new social media application. In March, ChatGPT’s new image generation capabilities gained viral traction on X and Instagram through a wave of impressive posts. However, by creating its own app for Sora, OpenAI now controls the entire user experience. “You prevent your rivals from grasping what influences user behavior, thereby enabling them to build competing products,” says Azeem Azhar, founder of the research group Exponential View.
Furthermore, short-form video apps could also serve as a method for collecting data that may be beneficial for improving future models, Xu suggests. “Launching a product early and making it affordable initiates a virtuous cycle: more users engage with it, more content is uploaded, and your model consequently improves.”
Currently, Sora is free to use, with a superior quality version of the model available to those who pay for ChatGPT Pro. In its introductory blog post, OpenAI implied that they would eventually begin charging for access to the free model. OpenAI chose not to provide a researcher for an interview to discuss the business strategy behind Sora.
The motivation may extend beyond mere revenue. OpenAI also asserts that Sora represents progress towards an AI model capable of precisely simulating the world. The company stated in its blog post announcing the new app that such a “world simulation” will be “essential for training AI models that possess a profound comprehension of the physical world.”
Azhar expresses doubt that Sora 2, the model currently powering the new OpenAI app, will effectively serve this purpose. However, he suggests that if the company can launch an app that draws hundreds of millions of users now, and then deploy an even more advanced Sora model through the same platform in the future, OpenAI could establish a rapid method for gathering vast amounts of valuable data once its models are capable of generating it.
On X, Altman countered a post criticizing him for “releasing AI slop videos marketed as personalized ads” instead of developing AI to cure cancer. He responded, “We primarily require capital to build AI that can conduct scientific research, and we are certainly dedicating almost all our research efforts to AGI.” He added, “It is also pleasing to showcase exciting new technology/products along the way, bring joy to people, and hopefully generate some revenue given our substantial computing requirements.”
