OpenAI’s Sora Faces Criticism as Kaptur Calls It a Solution Without a Problem

OpenAI’s decision to retire Sora was driven less by a sudden technical failure than by a sustained mismatch between the app’s cost structure and actual user demand. According to the Kaptur report, Sora’s lifetime revenue topped $2.1 million while its estimated daily compute bill hovered around $15 million – a ratio that would require a user base an order of magnitude larger than the roughly one‑million peak downloads it ever achieved to break even. The app’s rapid slide on the App Store, from the top slot to 101st place by January 2026, mirrors a 45 % drop in downloads and a 32 % dip in consumer spending (from $540 k in December to $367 k in January), underscoring a classic novelty curve where curiosity evaporates once the initial “wow” factor fades.

The economics of AI‑generated video are fundamentally different from those of user‑generated content captured on smartphones. Sora’s compute costs stem from running large‑scale diffusion models that synthesize both visual frames and synchronized audio from textual prompts. Each generated minute of video can demand dozens of GPU hours, a factor that inflates operational expenses far beyond the marginal cost of storing or serving a user‑uploaded clip. By contrast, platforms such as TikTok and YouTube rely primarily on ingesting and transcoding already‑recorded footage, a process that is orders of magnitude cheaper per hour of content. Dataportal data cited in the Kaptur brief shows TikTok approaching $33 billion in annual revenue on the back of 1.9 billion monthly active users, while YouTube remains the world’s second‑largest search engine – both thriving on real‑world recordings rather than synthetic generation.

A deeper issue highlighted by Kaptur is the scarcity of “storytelling” intent among the mass market. The 90‑9‑1 rule, first articulated by Jakob Nielsen and validated over two decades of research, predicts that roughly 90 % of participants in any online community are passive consumers, 9 % contribute intermittently, and only 1 % actively create. Sora’s premise—that lowering the technical barrier to video creation would unleash a flood of new content – assumes that the primary obstacle is tooling, not creative impulse. The Kaptur analysis argues that most users lack a narrative to translate into a prompt, and that the skill set required to craft compelling sequences of images and sound remains confined to a small cadre of trained filmmakers and visual artists. This explains why the app’s “TikTok of artificial intelligence” moniker failed to resonate: the platform attempted to replace a human‑centric creative process with a purely algorithmic one, without addressing the underlying demand side.

The collapse of OpenAI’s $1 billion deal with Disney further illustrates the strategic misalignment. The agreement, which would have granted Sora users access to over 200 copyrighted characters, never materialised into a signed contract, according to Kaptur. Without a clear pathway to monetize high‑value IP, the partnership offered little incentive for users to adopt a synthetic video workflow over established, rights‑cleared content creation pipelines. Moreover, the projected revenue from such a deal would have needed to offset the $15 million daily compute outlay, a hurdle that appears insurmountable given the app’s modest $2.1 million lifetime earnings.

In sum, Sora’s shutdown reflects a convergence of technical, economic, and behavioral factors. The compute‑intensive nature of text‑to‑video models imposes a cost ceiling that can only be justified by massive, sustained usage – usage that the 90‑9‑1 participation model predicts will never materialise at scale for a novelty‑driven app. While the appetite for video content continues to grow, as evidenced by TikTok’s and YouTube’s expanding user bases, the appetite for AI‑generated video on demand remains limited to a niche of professional creators. OpenAI’s experience with Sora therefore serves as a cautionary data point: lowering the engineering barrier does not automatically generate a market, especially when the underlying creative demand is fundamentally scarce.